Corruption in General Equilibrium: Political

Institutions and Bureaucratic Performance in South

America

A dissertation presented by

Daniel W. Gingerich

to

The Department of Government in partial fulfillment of the requirements for

the degree of Doctor of Philosophy in the subject of Government

Harvard University
Cambridge, Massachusetts
November 2006
 Daniel W. Gingerich
All rights reserved.
Dissertation Advisor: Professor Jorge I. Domnguez Daniel W. Gingerich

Corruption in General Equilibrium: Political Institutions and Bureaucratic

Performance in South America

A question of fundamental importance for the wellbeing of democratic governance is how

the format of political institutions may be fashioned in order to prevent electoral victors from

drawing upon the resources of the state to perpetuate themselves in power. This dissertation

addresses this question by examining the consequences of democratic institutional design for levels

of political corruption in three countries in South America: Bolivia, Brazil and Chile.

The specific aspect of institutional design analyzed is the distinction between an open-list

proportional representation (OLPR) electoral system and a closed-list proportional representation

system (CLPR). Theoretical expectations are generated by a game-theoretic model which considers

how the difference between the two systems may facilitate or impede corrupt 'contracts' from

forming between party leaders and politically-oriented militants in the public administration. These

expectations are tested through the statistical analysis of an original public employees survey

conducted in Bolivia (closed lists; 1038 participants in 14 institutions), Brazil (open lists; 1226

participants in 10 institutions) and Chile (open lists; 595 participants in 6 institutions).

Based on the formal model, the dissertation argues that electoral systems which intensify

legislative candidates demand for electoral resources and their willingness to engage in corruption

(such as OLPR), may make extraction of resources from the state more difficult, thereby dampening

the supply of corruption (relative to CLPR). The central idea is that CLPR, which gives parties the

ability to rank candidates on closed ballots, allows party leaders to exert significant control over the

future careers of politically-minded public servants. This, in turn, provides them with the ability to

pressure civil servants to engage in illicit behaviors which redound to the benefit of the party.

iii
 In order to test this proposition, the study develops statistical techniques appropriate for

analyzing the survey data, which utilizes the randomized response survey methodology. Using these

techniques, the study finds that, indeed, politically-oriented bureaucrats in the closed list case

(Bolivia) were more inclined the use the resources of their institutions for partisan benefit than

similar bureaucrats in the open list cases (Brazil and Chile), controlling for other factors related to the

incidence of illicit resource use.

iv
 Table of Contents

Acknowledgements...vii

Institutions and Political Corruption: A General Equilibrium Approach...1

An Approach to Overcoming the Fundamental Problem of Inference in

Corruption Studies..50

Characterizing the Bureaucratic Marketplace in Bolivia, Brazil and Chile: A First

Look at the Survey Data..103

Ballot Structure, Political Corruption and the Performance of Proportional

Representation176

Political Career Paths in the Bureaucracy and the Use of Institutional Resources

In Bolivia, Brazil and Chile.231

Taking Stock of the Findings.276

v
To my wife, Paloma, collaborator in life, love and innumerable other adventures

And

To my parents, Alina and Willard, for introducing me to the joys of social inquiry and academic life

vi
 Acknowledgements

It has been said that a dissertation is like a capital investment, an item whose acquisition

demands considerable personal sacrifice at one point in time but which pays great dividends

thereafter. Yet this simile fails to recognize the fact that the investment represented by a dissertation

is never the authors alone. To the degree that the current work provides dividends to the reader (or

its author), many financiers deserve their due for intellectually bankrolling the project.

Throughout the various stages of the project, I benefited greatly from the advice and

guidance of my dissertation committee. Jorge Domnguez provided steadfast support and invaluable

suggestions throughout. His encyclopedic knowledge of politics in Latin America constantly

challenged me to consider how well my arguments might travel outside the set of cases considered

here. Peter Hall was instrumental in pushing me to see the big picture represented by the project.

His generous feedback on framing my approach helped me better situate the work within the

tapestry of preexisting comparative political research. Steven Levitsky was a source of unfailing

enthusiasm and thoughtful criticism. Steves deep thinking about the informal aspects of party

behavior and organization had a profound impact on my thinking about the consequences of formal

electoral rules and procedures.

Numerous discussants on the various chapters of the dissertation have helped me refine the

arguments herein. I would like to thank Jonathan Hopkin, Susan Rose-Ackerman and David Samuels

for their very helpful input in this regard. My interactions with fellow graduate students at Harvard

have also been a source of great inspiration and learning during the genesis of this project. Gilles

Serra has been an especially perceptive reader, always ready with keen insights on logic and method.

He is also a great friend and kindred soul. Hillel Soifer, Shannon ONeill Trowbridge, Allison Post

and Jose Fernandez-Albertos were also reliable sources of shrewd criticism and commentary.

Various professional and personal contacts in Bolivia, Brazil and Chile had a major impact in

helping me to bring the project to fruition. In Bolivia, I owe a great debt to Andrs Zaratti and

vii
Giovanna Mendoza for taking interest in my corruption survey and ensuring that it received the

support of the Secretara de la Lucha contra la Corrupcin (SLCC). This support was instrumental in

opening doors in Bolivia, and it convinced that similar research could be conducted elsewhere in the

region. To my survey assistants in this country, Natty Nancy Blanco Cusi, Nancy Jurado Laime, Ivan

Jurado Laime, Monica and Vernica Gutierrez Rossy, I offer my sincere thanks for your dedication

and effort. I would also like to thank Walter Guevara Anaya, George Gray Molina, Juan Antonio

Morales, Ren Antonio Mayorga, Bismarck Arevilca and Humberto Zambrana for useful

conversations about politics and corruption in Bolivia, or simply for helping me get my feet on the

ground in Bolivia. I owe a particular debt to the Espinoza family, Ren, Carmencita, Maria Ren,

Anne and, above all, Sra. Carmen Valencia, for bestowing great kindness on a stranger.

In Brazil, David Fleischer was a source of wonderful insights about the workings of political

corruption and party politics in that country. He deserves many thanks for generously allowing me to

use his NGO, Transparncia, Conscincia & Cidadania-Brasil, as a base of operations for the project

in Brazil and for opening up a very useful network of contacts. Paulo Calmon also provided great

support for my research in Brazil. My sincerest thanks go out to him for helping me attain the

support of the Controladoria-Geral da Unio (CGU) for the project and for introducing me to the

excellent group of students at the University of Braslia that became my survey assistants. Within the

CGU, I would especially like to offer my thanks to Yves Zamboni Filho for his collaboration and

interest in the project. My survey research team in Brazil exceeded all expectations in terms of their

professionalism and enthusiasm. Anali Cristino Figueirido and Gustavo Freitas Amora, in particular,

have my deepest gratitude for the crucial management roles they played during various stages of the

survey. I would also like to thank Lucianna Cavalcante Queiroz, Thiago Cortez Costa, Tatiara

Monteiro, Marcelo Negro, Samatha A. Amorim, Isabela Monteiro N. Mota, Janaina Figueira,

Henrique Alves, Keynerson Santiago Silva, Gabriel Lessa Catalo, Adriano Furtado Afonso, Marcilio

Lima Gomide and Rodrigo Algarte for their important participation in the project.

viii
 In Chile, I owe a debt of gratitude to Luciano Tomassini for organizing the support of Chile

Transparente for the project. I also would like to thank Claudio Fuentes, Gonzalo Biggs, Alejandro

Ferreiro and Claudio Orrego for sharing their thoughts about corruption and money-in-politics in

Chile. The Santiago office of Harvards David Rockefeller Center for Latin American Studies

(DRCLAS) was my home base in Chile. I would like to thank Steven Reifenberg and the DRCLAS-

Santiago staff for providing a pleasant atmosphere in which to work and providing me with a locale

for my training sessions with my survey team. Finally, I would like to thank the members of my

excellent survey team for all their efforts on behalf of the project: Briam Chiple, Mitzie Chiple, Juan

Eduardo Dominichetti, Barbara Lopez Figeroa, Alvaro Garcia Marin, Macarena Schweitz, Alejandra

Ferrando, Ghislaine Tisne, Gonzalo de la Barra, and Luna Rosenmann. On a personal level, the great

warmth and generosity shown to me by Maria Angelica Piqueras and Luciano Paulini helped make

my stay an enjoyable one, even during those dark days when it seemed I might not be able to

conduct the survey in this country.

Without question, the individuals to whom I owe the greatest debt are the 2,859 Bolivian,

Brazilian and Chilean civil servants who were willing to set aside time to respond to my

questionnaire. I had the good fortune to personally converse with hundreds of you, and I have

learned a great deal from these interactions. I hope you find the end product well worth your

collective effort.

ix
 Chapter 1
Institutions and Political Corruption: A General
Equilibrium Approach

1
1.1 Introduction

A question of fundamental importance for the health and longevity of democratic

governance is how the format of institutions may be fashioned in order to prevent electoral victors

from drawing upon the resources of the state to perpetuate themselves in power. For the citizens of

the worlds so-called third wave democracies, recently emergent from the dark shadow of military

government and rule by fiat, this question has established itself as perhaps the core democratic

dilemma of the early 21st century.

The considerable moment attributed to this question in new democracies arises from the

fact that these republics are generally characterized by a significant overlap between the electoral and

bureaucratic domains of the polity. This overlap manifests itself in a coalescence of bureaucratic and

political careers, dual loyalties to the conflicting missions of state organs and political patrons, and,

when said loyalties become skewed in favor of the latter, the illicit redirection of public resources into

political activity. The consequence of this perverse conflation of political and bureaucratic power is

that the bureaucratic apparatus in many developing countrieswith much greater potential for good

or ill than its counterpart in the industrialized worldoften abdicates its role as a catalyst for

economic development and social equity.

The potential conflict between the practice of democratic politics and the development of a

modern and effective state bureaucracy has long been of great concern to students of political

development. No less an authority than Max Weber suggested that "democracy creates obvious

ruptures and blockages to bureaucratic organization" (Weber 1946, 231). By this he meant that the

cornerstone of modern bureaucratic organization in the nation-state, the separation between the

political leadership entrusted with shaping policy and the personnel charged with enacting it, may be

undermined both by democratic principles and the exigencies of political competition. In principle,

thought Weber, democracy's egalitarian disposition proscribes the emergence of an insulated 'caste'

of public officials, making it difficult to limit dismissal by election to a narrow band of public

2
servants at the top of the administrative hierarchy. In practice, political parties may be tempted to use

the posts and resources of the state bureaucracy to bolster their electoral support.

Deeply troubled by Webers premonition about the inherent tension between efficient

administration and democratic politics, this monograph examines how changes in formal

institutionsthe written, codified rules of the gamemay impact one of the great scourges of

democratic representation: political corruption. The broad aim of the monograph is to provide a

general theoretical framework by which to clarify the costs and benefits engendered by specific

institutional reform proposals on the level of corruption. The more narrow aim is to examine how

certain key features of democratic institutional design modify the incentive structure of political and

bureaucratic actors in ways that contribute to the level of political corruption found in a given polity.

Which features of democratic institutional design demand attention in a monograph

dedicated to understanding political corruption? There is no lack of options to choose from. The

contemporary democratic experience contains a nearly infinite variety of potential configurations of

institutions which determine the manner in which the ruled delegate power to (or rescind it from)

their rulers. Modern democracies differ along such dimensions as the manner in which the executive

is selected and dismissed, rules which delimit the executives ability to initiate legislation, stipulations

for reelection to legislative and executive offices, the number of legislative seats up for grabs in

electoral districts and the manner in which citizens cast their votesjust to name a few widely

studied components of democratic institutional structure.

In deciding on which aspect of democratic institutional design to focus, I adopted a

pragmatic approach. I choose to cast my sails for waters where the policy debate is both pressing and

unresolved, and where the academic debate is active and highly contested. These criteria led me

naturally to focus on the impact of ballot structure, specifically a move from an open-list

proportional representation electoral system to a closed-list proportional representation system (or

vice versa). In the parlance of contemporary social science, this is the studys primary independent

3
variable. The impact of a reform in ballot structure, in turn, is evaluated in terms of its effect on

political corruption. This is the studys dependent variable.

Ballot structure refers to the way in which citizens cast votes for candidates to legislative

office; depending on the structure of a politys constitution, such offices may be found at the

national, state or municipal level. In the various polities of the world, one can find a wide variety of

mechanisms by which such votes are cast. However, in the Latin American contextthe principal

focus of this studytwo means of electing candidates to the legislature have traditionally dominated:

multi-seat, proportional representation elections combined with closed lists, in which political parties

establish the rank ordering of their candidates before legislative elections are held, and multi-seat,

proportional representation elections combined with open lists, in which voters can indicate their

preference for particular candidates within a party list. This relatively fine-grained difference in the

way in which legislators obtain their seats may not, at first glance, jump out at the casual reader as an

obvious cause of major differences in the quality of democratic representation found across polities.

And yet, according to hundreds of political activists, commentators and politicians in Latin

Americaas well as numerous political scientists around the globethis subtle institutional variation

makes all the difference in the world.

In a public debate organized by the Brazilian newspaper Folha de So Paulo and the Instituto de

Direito Poltico e Eleitoral on September 15th, 2005, Dep. Jos Eduardo Cardozo (PT-SP) encapsulated

the view of many Brazilians when he referred to Brazils open list version of proportional

representation as the entryway of corruption, particularly corruption related to the financing of

political campaigns (Folha de So Paulo, 9/19/05, A-14). This position was seconded by Dep. Ronaldo

Caiado (PFL-GO) (sponsor of a political reform bill that would, among other things, transform

Brazils electoral system from an open list system to a closed list system), who referred to Brazils

electoral system as an open door for caixa dois [illicit campaign financing] and campaign bagmen

(translation mine). Such discontent with open lists in Brazil permeates the popular discourse.

4
Newsweeklies such as Veja, for instance, have decried open lists in Brazil for producing

extraordinarily expensive campaigns that make corruption inevitable (Veja 8/31/2005).

Strong feelings about ballot structure can be found elsewhere in the Americas, however, and

not necessarily in the same way as they are found in Brazil. On May 17th, 2004, more than a dozen

Argentine NGOs and civil society organizationsgrouped under the banner of the movement

Reforma Poltica YA! (Political Reform NOW!)delivered half a million signatures to Argentinas

Senate President in an effort to bring about the end of that countrys system of listas sabanas, or

closed list proportional representation (La Prensa, 5/18/2004, p.3). According to these groups, closed

lists in Argentina are an important contributor to systematic patterns of machine politics and

corruption especially prevalent in the provincial-level politics of that country.

The strongly differing perspectives of activists and policymakers on the merits of open

versus closed listsexemplified by the debates in Argentina and Brazilare reproduced in much

academic work on the subject. One strand of thought in the literature, in line with the arguments of

many reformers in Brazil, suggests that open list systems tend to produce greater levels of corruption

than their closed list counterparts. This is purportedly because open lists spark intense intraparty

competition, which in turn encourages high levels of personalism in legislative campaigns, thus

magnifying their cost and increasing the attractiveness of accepting illicit monies to cover campaign

expenses (cf Mainwaring 1991, Chang 2005, Golden and Chang 2001). Another strand of thought in

the literature, more in line with the arguments of reformers in Argentina, suggests that closed lists

produce more corruption than open lists. The argument here is that closed lists prevent voters from

singling out and punishing individual representatives for bad behavior in office; whereas open lists,

which allow for individual preference votes, do provide voters with the ability to reward or punish

specific legislators (cf Persson, Tabellini and Trebbi 2003, Kunicov and Ackerman 2002).

In light of this state of affairs, a focus on the impact of ballot structure on political

corruption seems warranted, firstly, because millions of citizens in the Americas believe it makes a

difference in terms of the degree to which corruption mars political life, secondly, because much

5
serious academic scholarship also claims that ballot structure has an important impact on corruption

and, thirdly, because political reform proposals advocating the reorganization of ballot structure are

already on the tableyet no consensus exists as to whether it is open or closed lists that produce greater levels of

corruption. It is clear to many that ballot structure matters. Precisely how it matters is much less

certain. Obviously, there is much to be gained by careful reflection and analysis of the precise forms

by which ballot structure exerts an influence on political corruption.

Having established the intrinsic importance of understanding how changes in ballot structure

may induce changes in political corruption, the question remains as to how we might best frame our

analysis for gauging the impact of institutional reform. The very fact that the policy debate on ballot

structure and corruption features such radically opposing viewpoints suggests that something may be

missing from existing frameworks used to think about the consequences of variation in democratic

institutional design. I would submit that this something can be captured in a single word: the

bureaucracy.

To the extent that we seek to account for differences across polities in the degree to which

state resources are illicitly deployed for the purpose of electoral competitionwhat I understand to

be the essence of political corruptionthen it stands to reason that we ought to pay attention to the

state and the people who inhabit it. This means that focusing on the behavior of non-elected public

servants is at least as important as focusing on that of elected ones, and that understanding the

mechanisms of illicit exchange between bureaucrats and politiciansand the role that institutions

play in mediating these mechanismsis crucial.

With these concerns in mind, the present monograph adopts a novel approach for thinking

about the consequences of restructuring democratic structures, one that is attentive to the crucial role

of the bureaucracy in the political life of many developing countries. Borrowing the language of

microeconomics, I refer to the approach as a general equilibrium account of institutional change.

In its traditional usage, general equilibrium refers to the establishment of an equilibrium

schedule of prices and output levels for a set of goods produced in the context of interdependent

6
markets (cf Samuelson 1976). My use of the term in this monograph is conceptually consistent with

this definition but novel in terms of the realm of application. Motivated by the overlap of

bureaucratic and electoral activity in the developing world in general, and Latin America in particular,

I adopt the label to describe equilibrium political and bureaucratic behavior in the context of polities

in which: 1) public bureaucracies are highly politically penetrated; 2) electoral outcomes are highly

contingent on the allocation of state resources. The interdependent markets of concern in this

monograph, then, consist of the electoral arena and the bureaucratic arena.

The monograph conceptualizes the long run, steady state behavior of political and

bureaucratic actors as emerging from highly contingent decision processes, in which the return to

courses of action in one arena of the polity depend on the decisions made in the other. Such a view

has important implications for the analysis of institutional reform in both the electoral arena (e.g.

party and electoral law reform) and the bureaucratic arena (e.g. accountability reforms within the

bureaucracy). Among these is a greater appreciation of the considerable contingency of institutional

reform, i.e. the fact that the impact of institutional change in one arena will be dependent on the

institutions that characterize the other. Equally important is a deepened understanding of the often

subtle mechanisms by which institutional reform exerts its impact. It is in this last respect, the

heightened appreciation for the nuance and subtlety of institutional change, that the present

monograph parts ways with much of the new institutionalist literature that has come to dominate

the study of the quality of government in developing countries.

In terms of understanding the overall impact of a reform in ballot structurea reform

which, in essence, modifies certain rules in the electoral arenathe general equilibrium approach

counsels us to consider not only the ramifications of such a change for the behavior of parties and

politicians on the campaign trail or in the legislature, but also to consider how such change might

modify the interactions between actors in the political and bureaucratic arenas, and thus the

particular channels by which the resources of corruption are actually generated. In this sense, by

explicitly focusing on the consequences of variation in political institutions on bureaucratic behavior, the

7
present monograph represents a fairly radical break from standard analyses of corruption in political

science, which have tended to concentrate most of their analytical energies on the consequences of

institutional variation for the behavior of politicians or voters.

Drawing on the general equilibrium perspective, the core claim of the monograph is that a

reform in ballot structure is likely to produce countervailing causal effects which, in the end, may

cancel each other out. On the one hand, a rich body of theory in comparative politics has established

that electoral systems which encourage high levels of intraparty competition, such as open list

proportion representation, produce costly campaigns and generate a high demand among legislative

candidates for electoral resources. Throughout the monograph, I take the findings of this work

which establish a demand side linkage between ballot structure and political corruptionas given. On

the other hand, this monograph dedicates itself to establishingboth theoretically and empirically

that an alternative channel between ballot structure and political corruption exists. It does so by

providing an in-depth examination of the consequences of electoral rules for the industrial

organization of corruption within the bureaucracy. In this regard, the monograph shows that

electoral systems which invest significant discretionary power over political careers in the hands of

the party leadership, such as closed list proportion representation, tend to be characterized by robust

and well-disciplined political corruption networks within the bureaucracy. In such contexts, the

extraction of resources from the state is facilitated. This is the supply side linkage between ballot

structure and political corruption.

Given the identification of these distinct causal pathways, one can conclude that a reform in

ballot structure will engender two distinct effects which work at odds with one another: a demand

side effect, which refers to the impact of ballot structure on candidates demand for the resources

from corruption, and a supply side effect, which refers to the impact of ballot structure on the ability

of political actors to extract resources from the state. The upshot is that even though ballot reform

has important effects on political corruption, these effects may be more of form than of overall level.

8
This being the case, policymakers and citizens deeply concerned with corruption would do well to

place other reform proposals foremost on the policy agenda.

This is a monograph that advances general claims about the impact of institutional reform

on political corruption. As such, it speaks to a broad body of work in political science on the

relationship between institutions and the quality of government. At the same time, this is a

monograph deeply seeped in the area studies tradition. As such, it seeks to increase knowledge about

bureaucratic quality and the scope of political corruption within the state in a particular region of the

world: Latin America. In this regard, specific propositions which emerge from the monographs

conceptual framework are subjected to extensive statistical tests using data from a large scale public

employees survey I conducted in three countries in the region: Bolivia, Brazil and Chile. In total,

nearly 3,000 public employees in 30 different institutions in three different countries participated in

the project. This was one of the largest corruption surveys ever conducted and the first to use a

specific survey methodology, randomized response, to guarantee the anonymity of respondents

regarding highly sensitive questions. The choice of Latin America as the geographical domain of this

monograph is not coincidental: perhaps more so than other regions of the world, the Latin American

experience cries out for a general equilibrium treatment of political reform.

In sum, the monograph very much shares in the spirit of institution-oriented scholarship

which proclaims, with a wealth of empirical evidence under its belt, that political institutions matter

for the quality of democratic governance in the developing world. Yet I would hasten to add an

addendum to this statement. Institutions and institutional change matters, to be sure, but they often

matter in rather subtle ways and may even produce countervailing effects not readily apparent on the

institutional architects drafting board. An emphasis on general equilibrium, rather than partial

equilibrium, can help bring the subtleties of institutional change out into the light. Armed with a

conceptual framework attuned to the interdependency of electoral politics and the bureaucracy in the

developing world, we can better understand the totality of consequences entailed by political reform.

9
1.2 Defining the Terms of the Debate

No progress can be made in explaining the incidence of corruption, from a general

equilibrium approach or otherwise, until we have established a concrete definition of the

phenomenon we seek to explain. This is a matter of no small consequence: some of the most

methodologically sophisticated studies of corruption have been bedeviled by a certain degree of

ambiguity regarding the precise nature of the dependent variable. Here I hope to clarify what

corruption is (and is not) and to further specify the particular form of corruption with which the

monograph is concerned.

In classical political thought, corruption was principally used to denote the deviation of the

moral and political order from an ideal state of purity (Dobel 1978). It was from this vantage point

that Aristotle refers to tyranny as the corruption of monarchy, oligarchy as the corruption of

aristocracy and so forth (Aristotle, Politics, 3.7, p.79, trans. Ellis). The core division between the

corrupted regime and its ideal, unspoiled counterpart was the degree of public spiritedness with

which the polity was ruled.

Heavily laden with a focus on the moral quality of the intentions of political actors, the

classical conception of corruption brings to mind Saint Bernard of Clairvauxs time honored

admonishment that the road to Hell is, in fact, paved with good intentions. As the world is inhabited

by men and women with a nearly infinite diversity of viewpoints about what constitutes the common

good, the classical conception seems much too generous in terms of behaviors it would exclude from

the rubric of corruption. A president buying the votes of deputies in Congress to shore up support

for his legislative program, to take one example, would not be engaged in corruption according to the

classical definition so long as he felt said program was in the countrys best interest. Neither would a

mayor funneling kickbacks on municipal contracts to his party leadership for use in upcoming

elections so long as he felt a good showing by his party was in the national interest, not just his own

interest.

10
 As it is axiomatic that politicians are highly invested in attaining and maintaining power, and

probably most feel that their having such power is good for their country if not for themselves,

excluding such crimes of power from the definition of corruption is much too limiting. Moreover,

absent a window into the thoughts and desires of public officials, the classical conceptualization of

corruption makes operationalization impossible. We cannot know corruption when we see it because

we can never see itit manifests itself not in particular actions but primarily in the mental state of

the relevant actors.

With these limitations of the classical conception in mind, many scholars have adopted a

definition of corruption that emphasizes process and formal roles and lends itself more readily to

operationalization. This is the so-called public office conceptualization of corruption. The work of

Joseph Nye provides a good example of such a view:

Corruption is behavior which deviates from the formal duties of a public role because of private-
regarding (personal, close family, private clique) pecuniary or status gains; or violates rules against the
exercise of certain types of private-regarding influence (Nye 1967, 419).

From this perspective, the moral value of an agents intentions is absent from the metric of

corruption. The behavior of a public official either corresponds to the relevant set of rules or

guidelines or it does not. Lack of public spiritedness and high regard for personal benefit may

produce behavior which can be categorized as corrupt, but only because the specific actions of public

servants exceed certain thresholds of permissible conduct, not because of the psychological state of

the actors nor due to any lack of loyalty to the polity.

I adopt the above definition of corruption in the present monograph with one very

important distinctionI reject the notion that pecuniary or status gains are necessarily the primary

motivating factors driving corrupt behavior.1 The misguided emphasis on these two stimuli is an

1 For many researchers, especially those coming from anthropology and related disciplines, definitions of
corruption which emphasize formal rules or procedures have been considered highly problematic. Such
scholars have noted that a set of equivalent behaviors might be given a very different status under different
legal frameworks (Scott 1969a). If taken to the extreme, a strictly legalistic conceptualization of corruption
might threaten to produce a stream of works documenting little more than commonalities and differences in
legal statutes with little relevance for understanding the causes of variation in important behavior regularities
across polities. Ultimately, the relevance of this critique depends on the domain of the analysis to be conducted.

11
error in the conceptualization of corruption which has been reproduced many times in empirical

work and one which has had substantial influence on how corruption is measured. The deficiency is

perhaps best illustrated by the unfolding pages of recent history: the major corruption scandals which

have rocked countries as diverse as Brazil, Canada, France, Mexico and South Korea in the few years

preceding the writing of this document have only peripherally, if at all, involved the use of public

resources for the purpose of personal enrichment. Rather, these cases have revolved principally

around the exigencies of financing political activitypolitical campaigns and party operating

expensesand the degree to which the assets or power of the public administration have been

utilized to cover such expenses. Not only does a focus on pecuniary gain paint an incomplete picture

of the type of behavior one might reasonably consider corrupt, it misses the central component of

much of this behavior.

A general definition of corruption should be expansive enough to cover activity oriented

towards personal enrichment as well as activity oriented towards political advancement. Herein

corruption is defined as behavior on the part of public officials, elected or non-elected, which

advances either an individual or groups financial well-being or a political goal through the misuse of

the authority or resources of an official position.

Conceived of in this way, corruption can be disaggregated into two components: personal

corruption and political corruption. Personal corruption simply refers to the first part of this definition:

the abuse of public office for the economic benefit of the officeholder and/or those close to him.

Political corruption refers to the second part of the definition: the abuse of public office to advance a

political goal. More often than not, advancing a political goal entails using public resources directly or

indirectly to support a particular candidacy, faction, party organization or coalition. The lions share

of examples used in this monograph consist of precisely such behavior. Note, however, that the

definition is broader than this: using public monies without the necessary authorization to support

It is especially relevant for scholars who seek to understand the origins of variation in corrupt practice in
different time periods or in radically different cultural contexts. However, for scholars engaged in studies like
this one, where the domain of analysis consists of contemporary democracies with similar legal traditions in a
single region of the world, such concerns do not seem particularly consequential.

12
foreign fighters engaged in a sympathetic cause would qualify, as would the misappropriation of

funds from the defense budget to pay for school lunches.

Readers are forewarned that the definition of political corruption provided here is different

than that they might find in other sources. In a recent publication, Transparency International (TI), a

Berlin-based NGO dedicated to the fight against corruption, defines political corruption as the

abuse of entrusted power by political leaders for private gain, with the objective of increasing power

or wealth (Hodess 2004, 11). In the TI definition, what separates political corruption from other

forms of corruption is the identity of the individuals involved in the transgressions. If political figures

are involved, then the behavior necessarily constitutes political corruption. Conversely, in many

publications one finds reference to bureaucratic corruption as a set of illicit behaviors which arise

primarily as a means to supplement the income of civil servants. Corrupt exchanges that take place in

the public bureaucracy, by implication, necessarily revolve around the personal enrichment of those

directly involved.

While it may be true that politicians are often involved in acts of political corruption, by my

lights it is an error to make their direct participation in a corrupt exchange a necessary and sufficient

condition for that exchange to be labeled an instance of political corruption. To endow the

involvement of politicians with the status of a necessary condition is problematic because the

transgressions of non-elected officials often have explicitly political motivations and important

political effects. To define these actions as being anything other than instances of political corruption

is to place important phenomena, such as machine politics and patronage, outside of any framework

with which we might approach the problem of corruption. To endow the involvement of politicians

with the status of a sufficient condition is equally problematic because certain behaviors, such as the

looting of the public treasury by an elected official in order to bankroll an extravagant lifestyle,

constitute purely personal acts, not related in an obvious way to political competition or the structure

of the political system.

13
 As I understand it, the distinguishing characteristic of political corruption is the end which is

advanced by a particular transgression, not the identity of the transgressor or the institution in which

the transgression occurs. Seen from this light, the public bureaucracies of many countries are indeed

often the venue of political corruption, which manifests itself in exchanges that do not necessarily

(although often do) entail the personal enrichment of civil servants. This is a key component of the

general equilibrium view of corruption.

In order to fix ideas, it may be helpful to outline a number of behaviors which fall under the

rubric of political corruption as I have defined it. Collecting kickbacks on public contracts and

subsequently redirecting them into political channels would qualify, as obviously would the direct

diversion of budgeted funds into political activity. Usually the source of less social opprobrium are

the more quotidian forms of political corruption: the unauthorized use of public property for

electioneering purposes and the forced donations of employee time or wages are representative

examples. In general, political corruption can be conceptualized as consisting of a continuum of

behavior, from high frequency but relatively low impact activities at one end to low frequency but

relatively high impact activities at the other.

Political corruption, defined as above, is the principal explanandum of this work. The

theoretical machinery developed herein is tailored specifically towards explaining how institutions

impact its incidence and volume. Given that much of the existing scholarship on corruption has not

emphasized the distinction between personal and political corruption, the choice of this topic as the

object of sustained scholarly inquiry merits some justification. It is to this endeavor that I now turn.

1.3 Political Corruption: Democratic Growing Pain or Persistent Dilemma?

Why is there a need to study the impact of fine-grained democratic institutions like electoral

rules on political corruption? Are not democratic elections in and of themselves sufficient to reduce

this scourge to nuisance levels with the passage of time?

14
 Most social science research is guided by the hope that the findings it produces, either

through direct application or by influencing future research, will in some way improve the human

condition. This monograph is no exception. However, studies like this one, which analyze political

pathologies, face the unpalatable task of establishing that the topic of study is not akin to Hegels

famed owl of Minerva, comprehensible only as it fades out of existence, even though any right-

minded person would wish that it indeed would do precisely that. Unfortunately, the fact is that

political corruption will not be fading out of existence any time soon, the recent spread of democratic

government notwithstanding.

Political corruption can be persistent in polities where leaders are selected through free and

fair elections, and may even flourish in such contexts. Elections, in and of themselves, are not a

sufficient means of holding elected representatives to account for wayward behavior because the

deployment of resources gotten from wayward behavior can be a successful way of courting voters,

and thus, winning elections. In contrast to illicit personal enrichment among a narrow political elite,

the emergence, deepening and consolidation of democracy may not go hand in hand with a secular

decline in political corruption. The bad news is that political corruption will not fade away by itself.

The good news is that there is much room for factors of intermediate and short-term human design

to improve social welfare.

1.3.1 Democracy and Corruption: Theories and Evidence

Sustained academic concern with the topic of corruption began in the 1960s, as many

scholars sought to get a fix on the root causes of the social and political upheaval afflicting the

developing world at that time. In this regard, political corruption was viewed as a palliative, an

unseemly but necessary ingredient in establishing political order. The often cited work of Huntington

(1968), for instance, put forth this view in unequivocal terms:

15
 Insofar as the governmental bureaucracy is corrupted in the interest of political parties, political
development may be helped rather than hinderedFor an official to award a public office in return
for a contribution of work or money to a party organization is to subordinate one public interest to
another, more needy, public interest (69-70).

The advocates of the corruption-as-political development approach justified their position using a

sequential mode of reasoning: corruption provides a means for parties to establish concrete linkages

with groups of voters; once established, these linkages provide the foundation for the development

of distinctive ideological commitments by parties and citizens, thus institutionalizing the party

system (Huntington 1968, Riggs 1963); finally, once ideological commitments become a central

component of political allegiances, the rationale for corruption disappears, and corruption itself along

with it (Scott 1969b). In this way, went the logic, corruption undermines the conditions of its own

existence (Huntington 1968, 71). Stated differently, this literature viewed corruption essentially as a

democratic growing pain, a temporary feature of new republics inexorably bound to follow the

reform trajectory of the late 19th century United States.

A more contemporary strand of scholarship, focusing on accountability and the electoral

control of politicians, is also compatible with the notion that democratization should bring about a

secular decline in corruption, although this work states its claims in much more general terms.

Beginning with Barro (1973), numerous authors, including Ferejohn (1986), Austen-Smith and Banks

(1989), Persson, Roland and Tabellini (1997) and Fearon (1999) have developed models of political

accountability based on the framework of principal-agent theory. This work envisions political

representation as a mechanism by which citizens, often treated collectively as the principal,

delegate decision making authority to political leaders, their agents. The relationship between the

ruler and ruled is characterized by an asymmetry of information: citizens would like to ensure that

their ruler take some action which is beneficial to the wellbeing of the polity but they are only able to

infer what the leaders action actually is within some margin of error. The inability of the ruled to

perfectly monitor the activity of the ruler, in turn, opens the door for self-serving behavior on the

part of the latter at the cost of the former. The benefits the ruler receives from doing so are generally

16
given the generic label rents, which for all intents and purposes correspond to Nyes notion of

private-regarding pecuniary or status gains.

In the principal agent framework, periodic elections serve as a check on the self-regarding

behavior of political leaders. The clever artifice of an electoral contract between the ruler and ruled

provides the means to this end. In one such formulation, voters collectively establish a retrospective

voting rule by which they commit to reelecting an incumbent if observed performance exceeds a

certain threshold (Ferejohn 1986). If performance fails to meet the threshold, the incumbent is voted

out of office. Because the actions taken by the incumbent are not perfectly observable, the contract

cannot wholly eliminate opportunism. Nevertheless, the electoral contract establishes certain bounds

on the opportunistic behavior of politicians. In this sense, elections are said to impose discipline on

politicians. While not necessarily as sanguine as the earlier generation of writers about the prospects

of democratic government to eliminate corruption, work in this tradition does adopt the position that

where contested elections are complimented by institutions that facilitate the acquisition of

information by voters (such as a free press), the prospects for self-regarding behavior by elected

officials are much reduced.

The aforementioned theories would lead us to believe that there has never been a more

propitious moment for the eclipse of corruption as a permanent feature of public affairs. As a result

of the third wave of democratization that swept Southern Europe and the developing world, a

greater percentage of the worlds population today selects its leaders through contested elections than

at any other time in human history. Contrary to the cynicism of many skeptics, authoritarian reversals

in the new democracies have been rare, and where they have occurred they have been short lived

(Domnguez 1998, Mainwaring and Prez-Lin 2005). More common are governments that

combine elections with marked authoritarian tendencies, the so-called competitive authoritarian

regimes (Levitsky and Way 2002). However, with the passage of time, these governments, too, have

dwindled in number. With this worldwide transformation in governance, it is unsurprising that some

scholars have begun to argue that the era of systematic corruption in government may be reaching its

17
end (Nam 1995). In so doing, they have implicitly grasped upon the well-known Wittman

hypothesis: the idea that the competitive dynamics generated by political markets (read: elections)

produce outcomes which are generally efficient (Wittman 1989, 1995).

And yet, regrettably, recent historical experience paints a more nuanced picture. In line with

expectations, the most grotesque examples of state plunder and personal enrichment are generally

not to be found among polities that have adopted democratic government. Democratic analogs to

the tinpot dictators that proliferated during the Cold War are few and far between. Of course, they

can be found: the governments of Arnoldo Alemn in Nicaragua (1997-2002), Fernando Collor de

Mello in Brazil (1990-1992) and Joseph Estrada in the Philippines (1998-2001) are only a few

examples of how acute personal corruption can emerge within the context of democratic

competition. But these are also good examples of how democratic competition makes it difficult for

such behavior to persist. Although able to finish his term, Alemn was ultimately tried and convicted

of crimes ranging from embezzlement to corruption. He received a twenty year prison sentence as

punishment. Both Collor and Estrada were impeached, the former reduced to a bit player in Brazilian

politics and the latter currently facing charges of plunder and corruption. In all three cases, political

entrepreneurship by former allies or the opposition contributed to the leaders fate.

While democratic leaders such as these are perhaps most notable for their rarity and short

tenures in office, the same cannot be said for non-democratic governments in the same mold. The

notorious dictatorships of men like Ferdinand Marcos (1972-1986), Rafael Trujillo (1930-1961),

Mobutu Sese Seko (1965-1997) and Mohamed Suharto (1967-1998) were not just long-lived

testaments to kleptocracy, they were also fairly representative of a broad swath of the authoritarian

rulers of their day.2 It would be difficult to conceive of how similarly rapacious leaders might

perpetuate themselves in power in competitive political systems. So although it would be an

2 By no means have all authoritarian regimes fallen into this mold. But a sufficient number have to merit study

as a distinct political phenomenon: sultanistic regimes. On the characteristics of such regimes, see Chehabi
and Linz (1998). It bears emphasizing that even some of the most supposedly austere dictatorships, that of
Augusto Pinochet in Chile (1973-1990) being the canonical example, have with time been revealed to also have
been corrupt.

18
exaggeration to call democracy the death knell for leaders that plunder their republics, the notion that

the transition to democratic rule should be associated with an overall decline in the illicit enrichment

of leaders seems to pass the test of face validity.3

The nuance emerges from the fact that the relationship between democratic government and

corruption considered in all its manifestations is much less obvious. The existing cross-national research

has not yet produced conclusive findings on democratic government and corruption one way or the

other. Studies have shown that the relationship between corruption and democracy is highly

contingent on the indicator used to measure the former. To the extent that researchers have been

able to discern any favorable effect of democratic government on corruption, it is only after

democracy has been in place for a substantial period of time or is buttressed by other favorable

conditions such as widespread newspaper readership (Adeser et al 2003, Montinola and Jackman

2002, Treisman 2000).

The likely reason empirical work has been unable to establish a clear cut relationship

between corruption and democracy is that political corruption appears to be flourishing in many new

democracies, even while cases of monumental personal enrichment are becoming less common. At

an anecdotal level, it is easy to make the case that political corruption has not waned much in recent

years and perhaps is even thriving in many new democracies. In this respect, it may be instructive to

consider some vignettes of political corruption taken from the experiences of new democracies

around the world. These accounts offer vivid illustrations of how corruption of this type has

entrenched itself at various times in a number of young democracies in the Americas. As they are

largely based on journalistic accounts, they are biased towards large scale or grand political

corruption as opposed to its smaller scale and more quotidian forms. That being said, they provide a

3It is important to note that the accountability model is not the only theoretical framework which would
predict less personal corruption with the transition to democracy. Selectorate theory, elaborated by Bueno de
Mesquita et al (2003), makes the same prediction. In this case, the theory holds that the capacity of leaders to
extract rents is a decreasing function of the size of the winning coalition necessary to keep a leader in power.
As the transition from autocracy to democracy necessarily entails an expansion of the size of the winning
coalition, democratization reduces the scope for predatory behavior.

19
nice sense of some of the many manifestations of political corruption encountered in the developing

world.

Recent corruption scandals in Brazil are a good place to start. On February 11, 2004, the

Brazilian weekly newsmagazine poca published a front page article called Dinheiro Sujo, or Dirty

Money. The piece exposed a kickback-campaign finance scheme articulated to fund the campaigns

of PT (Partido dos Trabalhadores) gubernatorial and congressional candidates in the 2002 elections. As it

turns out, components of the scheme were captured on videotape. The tape showed then president

of the Rio de Janeiro state lottery Wladimiro Diniz (who later became the principal advisor to Jos

Dirceu, President Luis Ignacio da Silvas Chief of Staff from 2003-2005) explicitly negotiating

payments destined for the electoral campaigns of PT candidates in exchange for on-line and over-

the-phone gambling contracts.

The ensuing scandal was compounded in short order by a floodgate of revelations dealing

with political corruption. Simultaneous with the investigations into the Diniz case, there were

investigations into whether the murder of a popular PT mayor in the city of Santo Andr was related

to his efforts to prevent an associate from skimming money off of the top of a kickbacks-on-

municipal-contracts scheme (which purportedly was pumping money into the PT party coffers).

Shortly thereafter, many feathers were ruffled when it was revealed that the public-private bank Banco

do Brasil had paid tens of thousands of dollars for its employees to attend a private concert at a

Braslia restaurant, the proceeds from which were to go towards the construction of a new PT

headquarters in So Paulo.

In April of the following year, another corruption video was released. This one showed a

rather humble and simple transaction: two individuals were shown paying a bribe of slightly more

than a thousand dollars to the director of contracts in the Brazilian Postal Service. However, the tape,

nearly two hours in length, turned out to be a Rosetta Stone for understanding the inner workings of

corruption; the public employee in question discussed in great detail the corruption network his party

(the PTB) had organized within the Postal Service, his place within that network as well as that of the

20
party leader, Roberto Jefferson, and it provided glimpses of how the party had been using other

public institutions to finance its activities (Veja, May 18, 2005). Hounded by accusations that the PTB

had been charging a monthly tribute of hundreds of thousands of dollars to high placed political

appointees in the public administration, and facing a Congressional inquiry into the matter, Jefferson

tried to settle the score with his erstwhile allies in the PT by releasing a revelation of his own: the PT

party leadership had organized payments to legislators from allied parties of approximately 12,000

dollars a month in order to control their votes in Congress. The fallout from this scandal quickly led

to the resignations of President da Silvas Chief of Staff, the PT party president, the PT general

secretary, the PT treasurer and the Minister of Communications. Even President da Silvas very

popular and powerful Finance Minister was forced to resign due to events related to the scandal.

The case of Mexico provides equally eye-opening accounts of the bureaucracy at the service

of electoral politics. On May 8, 2002, a federal judge in Mexico City ordered the arrest of Rogelio

Montemayor, the former director of Mexico's state-owned oil company Pemex (Petroleos Mexicanos),

along with three other top-level executives of the firm. According to the Comptroller's Office, the

former officials of Pemex had been involved in the diversion of $172 million from the company to

bank accounts belonging to the oil workers' union. Approximately $70 million of this sum

supposedly found its way into the campaign funds of PRI (Partido Institucional Revolucionario)

presidential candidate Francisco Labastida. The officials, headed by Montemayor, allegedly colluded

with Romero Dechamps, leader of the union and a federal PRI deputy, to cover up the money trail.

The voyage of the illicit funds was as humorous as it was circuitous. The money was

transferred from an account at Scotiabank to an account in Houston belonging to the union. The

funds were subsequently transferred to PRI accounts in Mexico, which were reported as proceeds

from raffles in which a suspiciously large number of wives of the PRI members won money. The

cash was then delivered to the party's finance committee and later disbursed to regional committees.

Notably, Montemayor was a purely political hire with no previous experience in the oil industry.

21
 It would be a mistake to think of this type of behavior as a product of the contemporary

political trendspolitical corruption has always presented a threat to the health of democratic

governance. Cubas brief democratic experiment in the 1940s illustrates this state of affairs nicely.

During the government of Ramn Grau San Martn (1944-1948), Jos Manuel Alemn, alternatively

Chief of Budgets and Accounts and Director General of Technical Instruction in the Ministry of

Education, played a role similar to that of Montemayor in Mexico. As Grau's go-to man in the

Ministry, "Alemn used the 'savings' from collapsing relevant programs to bloat the payroll with

patronage jobs and botellas [sinecures] and to become the major source of political campaign funds"

(Ameringer 2000, 35).

These practices had a devastating effect on the quality of the output of the Ministry of

Education. Reviewing the progress of Cuba's education system after the conclusion of the Grau

government, a World Bank mission commented on patronage in the Ministry of Education by saying:

"It would be hard to overstate the adverse effects on educational standards and the morale of the

teachers" (IBRD 1951, 425). The overall assessment of Cuban education was that "the general trend

in the school system as a whole has been one of retrogression" (IBRD 1951, 404). As in the Mexican

case, the amount of public monies diverted into campaign funds was substantial. Stokes (1949)

reports that an estimate of $6 million of public funds spent on the 1948 presidential elections was the

lowest suggested to him by various politicians, which still would have been greater than 10% of the

total budget of the Ministry of Education in the previous year.

Similar activities have been evident in Venezuelas democratic history. On September 29,

1979, the newly inaugurated president of Venezuela, Luis Herrera Campns, gave an address to the

nation on the subject of the public debt. In his speech, he accused the former Minister of the

Secretary of the Presidency, Carmelo Laura, of having authorized an illegal overdraft in the account

of the Ministry of Information and Tourism in the Industrial Bank of Venezuela (BIV) to cover

publicity expenses incurred by the government during the second half of the previous year. Called in

front of the new Minister of the Secretary of the Presidency the next day, Laura and ex-Information

22
Minister Celestino Armas explained that the Council of Ministers had authorized an overdraft of 98

million bolivares [23 million dollars] in the account and ordered a variety of different state agencies

and enterprises to divert their resources in order to fill the breach. After the press began to speculate

that the publicity funds had been destined for campaign coffers of Accion Democratica (AD)

presidential candidate Luis Piera, Armas published a report of the payments made with the funds

from the BIV account. The report only heightened suspicions as many of the payments either went

to particular individuals with no description of why or publicity agencies that had worked for the

Piera campaign (Diccionario de la Corrupcion en Venezuela, vol. 1, 19-23).

Often the mechanisms at work are less dramatic than the above examples, although no less

pernicious for their aggregate effects. In Bolivia, a fairly well-known practice in the past was for

agency heads to demand that the employees which serve under them contribute a fixed percentage of

their salaries to the head's party during an election year. Similarly, many employees were pressured to

devote their time to the campaign activities of their agency head's party. The existence of these

practices, as well the opportunities for more overt forms of corruption that had taken place in

institutions such as the Customs Service and National Road Service (mostly prior to 2000), have been

pointed to as an important explanation for the high degree of politicization in Bolivian public

administration (World Bank 2000a, 25-28).

A colorful description of the same process emerges from the recent Nicaraguan experience:

On the tenth day of every month, the majority of the almost seventy thousand public employees
receive their paychecks, from which is deducted the social security contribution, income tax and a 5%
contribution to the Partido Liberal Constitutionalista (in the government). The deduction is not
voluntary, but obligatoryThe deduction is made through the Human Resources department of each
public institution and is reflected in the payment slip (Vargas 2000, 34, translation mine).4

Only considerations of time and space prevent me from filling dozens of pages with similar

snapshots of political corruption in new democracies in the Americas and around the world. All

4 Forced tithing of public employees is a practice with a long historical pedigree. Probably its earliest

manifestation in democratic politics is to be found in another developing country: nineteenth century United
States. For a considerable amount of time, such tithesreferred to as assessmentsconstituted an
important source of campaign finance for Americas then fledgling political parties. See Johnson and Libecap
(1994, ch. 2) for a description of the practice.

23
things considered, a mountain of anecdotal evidence would suggest that in every region of the world

and in myriad epochs in the history of representative government, competitive elections have co-

existed with various means of illicitly using the state as a source of electoral advantage.

Making the point that political corruption may be especially pernicious in new democracies

with hard data is somewhat more difficult than doing so with anecdotes. Unfortunately, most of the

available cross-national datasets do not allow us to differentiate between personal and political

corruption (see discussion of the measurement of corruption in chapter 2). One source that does

disaggregate corruption into a number of conceptually distinct elements is the World Economic

Forums (WEF) executive opinion survey, conducted as part of this organizations annual Global

Competitiveness Report. The WEF data are generally consistent with the notion that new

democracies are especially afflicted by corruption related to the financing of political activity. Figure

1.1 displays kernel density plots of the responses of business executives regarding the prevalence of

illegal political donations and the frequency of irregular payments in a variety of business activities

for a sample of new democracies (open polities; 67 countries) and polities without democratic

government (closed polities; 18 countries).5

The plots, which present smooth approximations of the distribution of country scores on

the two measures, convey two messages. First, in the eyes of business leaders, new democracies tend

to suffer more than their non-democratic counterparts from forms of corruption related to the

political process. In this regard, illegal donations to parties are considered to be more common in the

former than the latter (p-value of two-sample Kolmogorov-Smirnov test for equality of distributions

= .01). This finding is as one would expect: where political competition is limited to the intrigues and

plotting of relatively few actors, or state power is abused to such an extent that elections become a

mere formality, the exigencies of political finance are mild relative to polities where the process of

electoral competition actually determines who rules. Second, corrupt practices which may entail

5 New democracies were those countries which had experienced a transition from authoritarian rule or

decolonization over the last sixty years. A polity was classified as open if it received a score of 4 or below on
Freedom Houses 2004 Freedom in the World political rights indicator; otherwise, the polity was labeled as closed.

24
either personal or political corruption, such as irregular payments in important public-private

activities, seem to be equally frequent in both types of polity (p-value = .78).6 As these practices may

thrive in very different environments of political competition, it is not surprising that the distribution

of responses for open and closed polities overlap so closely.

Figure 1.1: Political Competition and Type of Corruption (Kernel Density Plots)

Prevalence of Illegal Political Donations Irregular Payments

.5 .4

.4 .3

.3
.2

.2
.1

.1
0
1 2 3 4 5 6 2 3 4 5 6 7

Open Polities Closed polities Open Polities Closed polities

Note: a lower score indicates greater perceived prevalence of corrupt activity

Source: WEF (2004) and Freedom House (www.freedomhouse.org)

Although these data provide somewhat impressionistic indicators of political corruption,

they do raise important concerns. If voters in new democracies come to feel that the illicit

enrichment of caudillos and oligarchs has been replaced by an equal measure of corruption aimed at

campaigns and other political expenses, then this augurs poorly for the commitment of citizens to

the democratic process as the means to make demands on the state.7 As if to make the point, the

6 The variable irregular payments consists of the average of executive responses for each country with respect
to the frequency of irregular payments or bribes by firms for the following: export and import permits, public
utility connections, tax collection, public contracts, loan applications and judicial decisions.
7 In this regard, a number of articles have examined empirically the impact of corruption on satisfaction with

politics and judgments about democracy. In an analysis of a large scale corruption victimization survey
conducted in Bolivia, El Salvador, Nicaragua and Paraguay, Seligson (2002) finds that individuals who have had

25
willingness of disaffected masses to use extralegal, nondemocratic channels of influence to have

demands met seems to be a growing trend, especially in countries where systematic corruption has

left party systems in disrepute. Once elected representatives and political parties are no longer

considered legitimate interlocutors for those they claim to represent, organized groups may begin to

use whatever tools are at their disposal to directly influence policy in their favor. In the words of one

scholar, the praetorian society emerges (Huntington 1968). Unlike in decades past, today the end

result of this process is unlikely to be either revolution or military rule. Rather, the most plausible

scenario is a steady, low level governance trap, marked by the rise of antiparty politicians, rapid

political turnover and widespread voter malaise. This is hardly fertile ground for firm property rights

and the kind of stable and consistent policymaking conducive to economic development.

1.3.2 Why Political Corruption Works

If we take it as given that democratization has not eliminated the scourge of political

corruption (and shows no signs of doing so in the near future), the discussion up to this point still

begs an important question: Why is political corruption thriving when existing theory suggests the

political landscape has never been more favorable to reform? The answer requires a consideration of

how political corruption works in electoral terms; that is to say, we need to understand why rational

politicians, who seek to maintain themselves in office though the imprimatur of the popular vote,

might find it reasonable to divert or misallocate public resources from assigned tasks.

A foray into analytical politics is helpful here. Imagine a world with two parties, an

incumbent party (I) and a challenger (C), each of which competes for the vote of a single voter (V).

The voter evaluates the parties based on two criteria only: the parties' reputations on an ideological

issue and their reputations for corruption. As is standard practice in political science, I construe the

direct exposure to acts of corruption are less likely to have confidence in the basic institutions of government.
Examining the attitudes of Japanese citizens over time, Pharr (2000) finds that corruption scandals tend to
produce high levels of dissatisfaction with politics. Other studies have produced similar findings; see Anderson
and Tverdova (2003), Canache and Allison (2003), Della Porta (2000), Morris (1991) and Shin (1999).

26
term ideology broadlythe ideological issue might reflect matters as varied as a party's association

with a former authoritarian regime, its reputation for neo-liberal economic reform, and so forth.

Now let us assume further that party reputations can be represented by points along the real line.

The voter's ideal point on the ideological issue is v. This means that a party with a reputation for

implementing policy on this issue at a value greater or less than v will provide the voter with less

utility than would a party with a reputation for implementing policy at v. The further a party's

reputation is from v, in either direction, the less utility the voter will get from having that party win

the election.

Figure 1.2: The Tradeoff Between Ideological Proximity and Corruption

The voter's ideal point on corruption is 0; for simplicity, I assume that party reputations for

corruption do not take on negative values (all parties are believed to be corrupt to some degree). The

left panel of figure 1.2 (Before Campaign) displays the situation graphically. The ideological issue is

represented by the x-axis and corruption is represented the y-axis. The semi-circles radiating outward

from the point (v, 0) are indifference curves: any point along the boundary of a given semi-circle is

valued equally by the voter to any other point along the boundary of the same semi-circle. In this

panel, the voter is indifferent between the two competitors. Although the challenger has a relatively

attractive reputation on corruption, it is unattractive on the ideological dimension relative to the

27
incumbent. Conversely, although the incumbent party is ideologically attractive, it has an unfavorable

reputation on the issue of corruption.

Under what conditions will the incumbent party use the public administration to generate

resources from corruption (as in the real world examples given above)? Let us imagine that the

incumbent party in figure 1.2 has decided to draw upon its network of militants in the civil service in

order to fill its campaign coffers. There is a chance it will be caught doing so, in which case its

reputation on corruption will suffer. If this happens, the incumbent party's position will shift from

the point I to the point I' (see right panel of figure 1.2, After Campaign). Having filled its campaign

coffers, however, the incumbent party can now use its ill-gotten resources to launch a media blitz

against its opponent. Political advertisements may be used to `change the subject,' i.e. convince the

voter that the ideological issue on which the challenger fairs poorly is really the matter which should

be weighing most heavily on his mind, not corruption (cf Simon 2002). In more technical language,

the incumbent may use the resources from corruption to increase the salience of the ideological issue

relative to corruption. This process can be seen graphically in right panel of figure 1.2. The

indifference curves, circular in the left-hand-side panel, are elliptical in the right-hand-side one.

Whereas previously the voter weighed distance from his ideal point on the corruption and ideological

dimensions equally, now he weighs distance on the ideological dimension more heavily than distance

on the corruption dimension. In the example shown, using the state to generate resources from

political corruption is a worthwhile investment for the incumbent party because, even though its

reputation on corruption worsens, its relative attractiveness to the voter increases.

Seen from this perspective, political corruption works, then, because resources deployed in

campaigns have real and considerable effects on vote shares, and the state is often an attractive

source for such resources. Importantly, the analytical model suggests that illicitly using the state as a

source of electoral resources may remain attractive to parties even if voter information is sufficiently

high that parties who use resources from corruption suffer reputational costs for doing so. This is

28
because parties may be able use their ill-gotten gains to favorably manipulate their image in the

available media in such a manner as to offset their reputational loses on the corruption issue.

There are good reasons to think such mechanisms are present in real world politics. A

growing literature on the political psychology of voting demonstrates that media exposure during

campaigns may have a substantial impact on vote choice (Krosnick and Kinder 1990, Bartels 1993,

Petrocik 1996). Moreover, consistent with the approach of the above model, the precise mechanism

many authors identify to explain how media exposure drives vote choice is variously referred to as

"agenda-setting" or "priming," by which they mean favorably manipulating the salience of issue areas

(Ansolabehere, Behr and Iyengar 1993, Iyengar and Kinder 1987). In practice, it seems that parties

have been able to use the change-the-subject strategy to considerable effect: in 1940s Cuba, the

Autntico party was able to garner reelection in spite popular disgust with rampant corruption by

focusing its political advertising on the links of its adversaries with the brutal dictator Gerardo

Machado; the Christian Democratic Party in Italy was able to govern Italy for decades in spite of a

well-deserved reputation for corruption by deftly focusing the political discussion on the importance

of combating communism; and more recently, the Fujimori government in Peru was able to win a

landslide reelection victory in 1995 despite concerns about the administrations dirty dealings by

manipulating the media to emphasize the importance of national security and economic concerns

over other aspects of political life.

The empirical evidence suggests that democratization has not brought about a decline in

political corruption. Moreover, logical considerations suggest that the consolidation of democratic

competition in and of itself probably will not do so in the near future. Political corruption is a

persistent democratic dilemma, not a growing pain.

The subsequent chapters of this monograph focus on how institutions mediate the ability of

political actors to use the state as a source of resources for electoral competition. I will not return to

the theme of how political money is translated into votesfrom here on in the idea that money has

some impact on vote shares will simply be taken as given. However, the notion that money and other

29
state resources matter for electoral outcomes is foundational for everything that follows. Imposing

efficiency costs on the public through the operation of corruption networks within the state will only

make sense to office-oriented politicians if the electoral returns from such activity more than

compensates for electoral loses generated by erosion in the quality of public good provision. The

ubiquity and persistence of such networks is indeed a testament to the fact that many political actors

find this a trade off well worth making.

1.4 Institutions and Political Corruption: A General Equilibrium Approach

Given that the diffusion of democratic government around the world cannot be expected to

automatically discipline political corruption out of existence, it is important that political scientists

consider other means of reducing this problem beyond the holding of free and fair elections at

regularly scheduled intervals. A natural focus of research and debate in this regard consists of the

design of the more fine-grained institutional features of the politythe structure of electoral rules

and bureaucratic organization being among these.

Fortunately, the paradigmatic foundations for analysis along these lines already exist. Within

the realm of comparative politics, an important body of work dubbed rational-choice institutionalism

(RCI) has begun to systematically analyze how differences in institutional format across democracies

affect the quality of democratic government. Research in this vein takes as its starting point the idea

that given the same set of preferences among political actors, different methods of preference

aggregation produce different results (Carey 2000). Stated differently, the central supposition of this

work is that formal institutions matter for a host of important political outcomes. They matter, in

turn, because the relevant political actors respond rationally to the matrix of incentives produced by

the formal rules of the game. In order to understand how institutional reform may engender

changes in political outcomes, this work councils analysts to focus on the manner by which changes

30
in the structure of formal rules reconfigure the costs and benefits actors associate with the courses of

action available to them.

Departing from these premises, RCI has already made fundamental contributions to

knowledge about democratic governance in the developing world, and it has been especially

influential in the study of democratic politics in Latin America. In this regard, RCI has made great

strides in tracing out the consequences of presidential versus parliamentary government for the

breakdown of democratic government (Gasiorowski and Power 1998, Linz 1994, Mainwaring and

Shugart 1997, Przeworski et al. 2000, Shugart and Carey 1992, Stepan and Skach 1993, Valenzuela

1994), the impact of electoral rules on the campaign strategies of candidates running for legislative

office and the degree of internal party cohesion in the legislature (Ames 1987, 1995, Carey and

Shugart 1995, Desposato 2006, Figueiredo and Limongi 2000, Jones 2002, Mainwaring 1997,

Morgenstern 2004, Samuels 2002), the consequences of constitutionally-defined presidential powers

for patterns of executive-legislative relations (Alston and Mueller 2006, Carey and Shugart 1998), and

the impact of district magnitude on voting behavior and the number of parties (Cox 1997,

Mainwaring and Shugart 1997).

And yet, as both proponents and critics of RCI are quick to point out, the approach has

neglected certain crucial aspects of political life in the region. One of these neglected areas has been

the bureaucracy (Crisp and Escobar-Lemmon 2001, Huber and Dion 2002, Weyland 2002). In the

words of one critic, RCI pays disproportionate attention to parliamentary politics and stresses its

influence on governmental decision-making, while often neglecting the crucial importance of the

state and its bureaucratic apparatus (Weyland 2002, 64). This is a potentially serious omission,

especially since in Latin America one often encounters a view of the state as a distant fused space

where executive, legislator, politician and public administrator merge into authority figures with rights

of office and discretionary powers and have to be approached in roundabout ways for favors (Spink

1999). Given that a defining characteristic of this distant fused space is its high degree of

politicizationmanifested in bureaucratic politics with a vengeance in which the manipulation

31
of bureaucratic lines of authority is crucial for retaining power (Chalmers 1977), then indeed it

would seem much of the existing RCI literature has excluded from its agenda a central component of

political competition and democratic representation in the region.

A second critique of the RCI approach is that it has not paid sufficient attention to the often

variegated nature of career goals exhibited by political actors nor focused sufficiently on the manner

in which clientelism facilitates or hinders actors ability to attain those goals (Weyland 2002). This,

too, seems like a fair criticism to make. Politicians in various countries in the region are notorious for

jumping to and from elected and bureaucratic posts at various levels of government, and the blessing

of political patrons often plays an important role in determining the success of each leap along the

path.

Here I consider a framework for analyzing the impact of institutional reform on corruption

which directly addresses, and in fact incorporates, the two above criticisms of the RCI approach as

applied to the study of Latin American politics. I begin with the notion that in order to best analyze

the impact of changes in institutional design in the region, it is important to be able to draw upon an

intellectual armature which is attuned to the interplay between electoral and bureaucratic behavior.

To understand why bureaucrats engage in behavior that violates the public trust, it behooves us to

consider the constellation of institutions and outcomes which characterize the political marketplace.

To understand why elected officials are willing to breach norms of acceptable political practice, it is

equally important to consider the features of the bureaucratic market which determine the

profitability of such breaches. Established patterns of contestation in the political sphere are the

provenance of many of the incentives and constraints upon which bureaucratic behavior is

predicated. Conversely, the organization and capacity of the public sector condition the returns

available from various types of political action.

Given the mutually interdependent nature of choices made in the bureaucratic and electoral

domains, this monograph proposes a view of corruption as a topic of inquiry requiring general

equilibrium analysis. To be precise, this means that the level of corruption which obtains in a

32
particular polity should be thought of as emerging from the establishment of simultaneous equilibria

in two separate but related arenas: the political arena, where competitive pressures among parties and

candidates interact with fixed features of the electoral landscape (electoral rules, cleavage structure,

demographic characteristics of the electorate, etc.) to produce the optimal electoral strategies of

political actors and the bureaucratic arena, where the career concerns and economic realities of non-

elected public servants interact with bureaucratic procedures and regulations to produce the optimal

strategies of these actors. Each in their own way, these arenas are characterized by exchange relations

and optimizing behavior, and as such I would suggest that they can be usefully thought of as markets.

Two foundational premises mark my thinking on the operation of these markets. The first is

simply that the markets must clear. This premise reflects the commonplace notion that actors use

the resources at their disposal to do the best for themselves that they possibly can, and that they do

so knowing that everyone else will do the same. The second foundational premise is that the

markets are characterized by mutual interdependency. By my use of this term, I refer to the fact

that the choices made by actors in one market depend on the decisions and institutions which

characterize the other.

Markets must clear. Consistent with the postulates of neoclassical economics, I take

preferences as given and assume that individuals choose the strategies available to them which best

realize their goals. This means different things for the two sets of relevant actors considered in this

monograph: bureaucrats and party leaders.

For bureaucrats, the market clearing assumption implies that civil servants will conduct their

affairs with an eye towards maximizing their career prospects, either inside the public administration

or outside of it. To the degree that the civil service is highly politically penetrated, this may mean that

ascension up the bureaucratic ranks lies at the discretion of party leaders, and that political loyalty

may determine advancement in addition to or as a substitute for good professional performance.

Public servants will choose their courses of action accordingly. For those bureaucrats with an interest

in pursuing a political career, advancement will almost certainly lie to some degree at the discretion of

33
party leaders, although the extent of this dependence will vary as a function of the electoral

institutions in force. For these individuals, the nature of institutions in the electoral arena may play a

major role in shaping incentives. The bureaucratic marketplace clears when each bureaucrat takes

actions which he believes to be in the best interest of his career plans, given the behavior of all other

bureaucrats and the actions taken by party leaders.

For party leaders, the market clearing assumption implies that they select a set of actions

which best advance the electoral prospects of the party (but not necessarily the well-being of the

majority of voters). For these individuals, career advancement is synonymous with the electoral

success of the party. This means they will do their best to parlay electoral advantage out of the power

of persuasion they enjoy vis--vis bureaucrats, although they will be forced to do so subject to

constraints imposed by the bureaucratic and electoral environments. The electoral marketplace clears

when each party leader takes actions which he believes to best further his partys electoral prospects,

given the behavior of other party leaders and given the aggregate behavior of bureaucrats under his

sway.

Markets are characterized by mutual interdependency. My discussion of real world

cases of political corruption in this introduction should have made one point abundantly clearthe

resources of corruption often emerge from processes of political and bureaucratic exchange in which

the career advancement of bureaucratic and political actors is furthered through coordinated action

which defrauds the public fisc. It is certainly not the case that the interdependency between the

electoral and bureaucratic arenas always, or even typically, takes such a form. However, it is especially

prone to in circumstances in which the financing of politics is closely tied to various aspects of

bureaucratic action. To put it bluntly, when the state itself illicitly becomes a purveyor of resources for

electoral competition, a perverse network of political and bureaucratic exchange is practically the

guaranteed result. A steady stream of studies by area specialists in Latin American and European

politics suggest that this state of affairs may in fact be more the rule than it is the exception in many

republics (cf. Malamud and Posada-Carb 2005, Williams 2000). The fact that the state itself, through

34
the operation of such practices, is often a source of electoral resources and, pari passu, a determinant

of outcomes that obtain in the electoral arena, is one reason why the bureaucratic and electoral

arenas ought to be treated as interdependent.

A second basis for interdependency comes from the intermingling of bureaucratic and

political careers, long a topic of scholarly concern in Latin America (Carey 1996, Domnguez 1997,

Grindle 1977, Heredia and Schneider 2003, Schneider 1991, Smith 1979). Much of the region is

characterized by a fusion of the bureaucratic and political elite, to borrow Heredia and Schneiders

phrase, which manifests itself in a revolving door pattern. Former politicians often take up temporary

residence in bureaucracy, only to try their hands at electoral politics again in short order. Similarly,

many first time politicians precede their electoral careers with stints in the public bureaucracy.

Individuals characterized by either description are likely to be political appointees, and many find

themselves in the upper strata of the administrative hierarchy with considerable influence over public

contracts, licenses and the careers of other public employees. The fact that considerable numbers of

public employees have their hearts set on electoral or party politics, and that that the trajectory of

political careers is often at the mercy of party leaders of political patrons, is another important reason

why the bureaucratic and electoral arenas ought to be treated as interdependent.

From the general equilibrium perspective, both bureaucrats and party leaders will condition

their behavior on expectations about outcomes in the market different from the one in which they

find themselves. A moments reflection reveals how this might work. A bureaucrat asked to

misappropriate resources on the behalf of his political party, for instance, would presumably be more

inclined to do so if he anticipates that his party will be strong in electoral terms, and thus have the

wherewithal to repay the bureaucrat for his service in the future. Similarly, the strategies adopted by

party leaders will be conditioned on expectations about the likely patterns of bureaucratic behavior.

Parties will be more inclined to pursue clientelistic strategies, for example, in polities where

monitoring mechanisms within the bureaucracy are fragile or non-existent, and thus the pool of

bureaucrats willing to engage in corrupt exchanges is relatively large.

35
 In constructing forecasts about likely outcomes, agents in both markets will use all the

relevant information at their disposal. As such forecasts are based on systematic features of the two

relevant markets, on average they will be correct. A corruption equilibrium thus emerges when the

strategies and rational expectations of actors in both markets intersect. The resulting outcomes will

ultimately be reducible to the relatively few parameters (read: institutions) characterizing the fixed

features of the electoral and bureaucratic arenas.

Figure 1.3: Direct, Indirect and Feedback Effects

Bureaucratic Arena Electoral Arena

X X
1 (direct effect)
(indirect effect)
Actor 2 Actor
Strategies 3 Strategies
4
5
(feedback effects)

What ramifications does the adoption of the general equilibrium framework entail for the

analysis of the impact of institutional change? Following the general equilibrium approach, I would

expect that institutional change in either the electoral or bureaucratic arenas will lead to a

reconfiguration of all the relevant endogenous quantities of interest in the polity. In other words, a

shift in the key features of one market will not only modify the choices made by the pertinent actors

in that market but will also modify the choices made by actors in the other market. Such

perturbations may result in three different types of effects: a direct effect, which represents the

immediate ramifications of a perturbation in one market for the decision calculus of actors in that

same market, an indirect effect, which represents the how actors in the other market modify their

decisions in response to the anticipated changes in strategies of actors in the first market due to the

36
perturbation, and feedback effects, which represent the way in which actors in one market modify

their decisions in response to anticipated changes in the strategies of actors in the other market,

which are in turn a response to anticipated changes in the strategies of actors in the first market.

Figure 1.3 provides a graphical depiction of these three types of effects.

To illustrate the logic, I consider in the figure an institutional change which has taken place

in the bureaucratic arena. For the sake of argument, let us imagine that the change is a significant

reduction in public employee wages. With lower wages, more public employees may be willing to

engage in corruption schemes benefiting the party in power than before, since the value of a public

sector job is lower than it had been and the cost of being caught engaging in corrupt activities (i.e.

losing ones job) is correspondingly smaller. This is the direct effect of a reduction in wages.

However, note that since more employees are engaging in corruption at the behest of the party, more

money is flowing to the party, which as a result is now correspondingly stronger in electoral terms.

This is the indirect effect of a reduction in wages. With the party now stronger in electoral terms, the

offers it can make to potential co-conspirators in the bureaucracy (such as a future appointive

position in the bureaucracy during the next administration, the party nomination to run for an elected

post, etc.) are now more attractive than they had been previously. For this reason, even more

bureaucrats are willing to engage in corruption for the party. This is a feedback effect. Such feedback

effects will bounce back and forth between the bureaucratic and electoral arena, losing a little force

each time, until a new stable equilibrium materializes. In this particular example, the consequence of

applying the logic of the general equilibrium approach is that a modest reduction in wages may have

a large positive impact on aggregate levels of corruption, much larger than that which would have

been expected based on an analysis of the impact of institutional change in the bureaucratic arena

alone.

In general, it need not be the case that institutional change will result in multiplier effects as

in the wage example. Depending on the details of the institutional environment in the bureaucratic

and electoral arenas, it may very well be that the direct and indirect effects are at odds with one

37
anotheressentially canceling each other out. What the general equilibrium approach does definitely

claim is that in polities characterized by the types of interdependency discussed aboveperhaps the

majority of countries in the developing worldthe electoral consequences of bureaucratic reform are

important; and so are the bureaucratic consequences of electoral reform. Due to the mutual

interconnectedness of the electoral and bureaucratic arenas, institutional reform proposals which

target a single arena and consider the effects of reform in that arena in isolation may produce a

misleading view of the effects of reform.

1.5 Applying the General Equilibrium Logic to Ballot Structure Reform

Electoral rules establish the guidelines by which political careers are made. They delimit the

choices available to voters in the electoral booth, they frame the relationship between politicians and

their constituents, they reinforce or undermine lines of authority that run from the party leadership

to the rank-and-file, and, more mechanically, they provide rules for translating votes into seats in the

legislature. Because of the multitudinous ways in which they determine who will rise to the top of the

political scene and who will lag behind, electoral rules play an important role in defining the types of

investments political aspirants are inclined to make in order to further their careers.

In order to understand why this is the case, it is worthwhile to clarify the mechanics

according to which certain electoral systems function. I will consider these mechanics in the context

of the closed-list PR and open-list PR systems encountered in Latin America. The most

straightforward way to do this is visuallyto present ballot structure as voters themselves see it.

Figure 1.4 presents a demonstration ballot paper from Nicaraguaa country with a closed

list PR system. During legislative elections, voters in closed-list systems can cast a vote for a single

party list among the registered parties in contention for legislative seats. In the Nicaraguan case, if a

voter sought to cast a vote for the Sandanista party, for instance, she would denote her preference

with an X in the space underneath the party symbol (as demonstrated in the figure). As the

38
electoral system is characterized by proportional representation, legislative seats in particular districts

are allocated to parties more or less in proportion to their vote shares, where the degree of

proportionality is determined by the electoral formula and district magnitude. Most importantly from

the vantage point of this monograph, voters in closed list systems cannot denote a preference for a

particular candidate within the list of their choice. Rather, the ranking of candidates within each party

list is established by the party itself, in a process in which the party leadership almost invariably has a

heavy hand. This means that if a party wins less legislative seats than it has candidates on its list

which is always the casethe individuals who will actually attain seats are those candidates who have

impressed their value upon the party leadership in one way or another.

Figure 1.4: Demonstration Ballot Paper from Nicaragua (closed lists)

source:

IDEA (1997)

Closed list systems can be juxtaposed to open list systems, where the relationship between

the candidate and votersand the candidate and party leadersis quite different. Perhaps the

canonical case of open list PR in Latin America is the Brazilian electoral system; figure 1.5 displays a

screenshot from an electronic voting machine used on the Brazilian Supreme Electoral Tribunals

39
website to simulate the voting process. The image shows that voters in Brazil can cast their votes in

one of two different ways during legislative contests. Should they so choose, voters can enter a two

digit number into the electronic voting machinewhich represents the list of a particular party or

coalitionand press the green confirm button. In so doing, they will have cast a so-called voto de

legenda, a list vote, just as is the modality of voting in closed list systems. However, voters also have

the option of entering a three digit number after the two digit party ID to denote their preference for

a particular candidate within the party list. This is where open list PR is fundamentally distinct from

closed list PR. As in closed list PR, seats are allocated to parties in rough proportion to vote shares;

however, within party lists seats are allocated to candidates as a function of their share of preference

votes. Thus, open list systems produce two dimensions of electoral competition: competition among

competing lists and, often more intense, competition among candidates within the same party list.

Figure 1.5: Screenshot of Electronic Voting Machine in Brazil, Voting Simulation (open lists)
party list vote

candidate vote

source: www.tse.gov.br

To simplify the matter greatly, it can be said that political advancement in closed list systems

depends on the regard with which the candidate is held by the party and, especially, the party

leadership while advancement in open list systems depends on the candidates relative attractiveness

to voters. These distinct aspects of the two systems, in turn, fundamentally shape the types of

strategies actors pursue in order to build political careers.

40
 In closed lists systems, the key to political success rests in building a stock of personal capital

with the members of the party leadership most influential in composing party lists, be they at the

national level (as was traditionally the case in Bolivia or Venezuela) or at the local or provincial level

(as is the case in Argentina). There are many ways this might be accomplished. For businessmen

seeking to jump into electoral politics, an advisable strategy might be to make a campaign donation

to the party organization; years of service in leadership roles in union organizations linked to the

party might be another route to prominence; administrative service organizing groups of party

militants during campaign efforts is yet another way to place ones self on the radar screen. The

general point is that a steady accumulation of favors to the set of individuals who control political

careersin addition to faithful service to the party labelwill play a central role in determining who

rises to the top in closed list systems.

In open list systems, aspirants for legislative office need to make a personal mark with the

voters; they need to convince voters that it is they, not the next person on the party list, that should

receive the preference votes. Naturally, there are many ways to go about doing so. Individuals who

have established a certain amount of notoriety in their private lives may be fortunate enough to

capture votes through name recognition alone. Incumbents seeking reelection may court the

electorate by emphasizing the pork they have obtained for the territories to which their constituents

belong. However, the record shows that other mechanisms are often at work as well. Candidates may

feel the need to pay for legions of campaign workers to distribute electoral paraphernalia, they may

choose to host large parties (called showmicios in Brazil)featuring professional DJs or popular

performersto draw attention to their candidacies, they may attempt to purchase the allegiance of

neighborhood associations or local notables to capture the votes influenced by these individuals, or,

in the worst of cases, attempts to cultivate a personal following may degenerate into direct vote-

buying. All of this costs moneyusually considerable amounts of it. Thus an integral part of

advancing ones political career in open list systems consists of coming up with various strategies to

raise political money.

41
 Given the distinct strategies that differences in ballot structure engender among aspirants for

political office, what can we say about the consequences of ballot structure for levels of political

corruption? The general equilibrium perspective prompts us to consider how the difference between

open and closed lists induces change in patterns of corruption both as it relates to the behavior of

politicians in the legislature and campaign trail (politico-electoral arena) as well as how it relates to the

functioning of the public administration (bureaucratic arena).

If we consider the ramifications of ballot structure only as it relates to corruption in the

politico-electoral arena, it would seem that the position of the Brazilian reformers mentioned at the

beginning of this introduction makes the most sense. As described in greater detail in later chapters

of the monograph, a wealth of empirical evidence exists to support the claim that the emphasis on

personalism produced by open list systemsas well as other electoral systems that induce high levels

of intraparty competitiontends to raise the cost of electoral campaigns and creates a willingness on

the part of political actors to accept campaign finance from dubious sources. This state of affairs

produces fertile ground for the flourishing of what might be called a free agent type of political

corruptioncorrupt deals forged between individual legislators, acting more or less on their own,

and entrepreneurs who seek to obtain a state benefit over which said legislators exert some control.

The sanguessugas (leeches) scandal in Brazil is a prototypical example of how this type of

corruption may come to manifest itself in an open list system. According to preliminary

investigations conducted by the Federal Police, scores of sitting federal deputies and various senators

in Brazil had participated in a long-standing (since 1998) public procurement fraud ring orchestrated

by the owner of an ambulance producing firm (Planam). The way that this scheme operated in the

legislative arena is as follows. Planams owner paid bribes to legislators, on a case-by-case basis, in

exchange for inserting amendments into the budget for ambulance purchases in particular

municipalities. The prices of the ambulances in the amendments were substantially overbilled, thus

allowing Planam to turn a handsome profit from the scheme. In the case of the deputies and

senators, it appears that some of the bribe money went directly into legislators pockets and some

42
went into paying for campaign expenses or paying off campaign debts. The legislators implicated in

participating in the ring, over one hundred of them, belong to parties found across the entire

ideological spectrum in Brazil. The scandal is notable for its decentralization, its candidate- (as

opposed to party-) centrism and the apparent irrelevance of party affiliation or ideology. In this

regard, it makes sense to think of the political actors involved in such activities as essentially free

agents. Such free agent types of political corruption had traditionally been the rule in Brazil prior to

ascendancy of the Workers Party in government, as is evident in similar episodes such as the

Budgetgate scandal of the mid-1990s.

Yet if we consider the ramifications of ballot structure only as it relates to corruption in the

public administration, it would seem that the position of the Argentine reformers would win the day.

The reason for this derives from a combination of two factors: the manner in which ballot structure

ties the trajectory of political careers to the preferences of party leaders, discussed above, and the fact

that the public bureaucracies in the Americas are often staffed with significant numbers of patronage

hires. In contexts where party militants with various levels of involvement in party affairs are spread

throughout the public administration, there are invariably considerable numbers of bureaucrats who

are keen on cultivating political careers at one level or another. In closed list systems, these

individuals will be acutely aware of the fact that regional or national level party elites are ultimately

the gatekeepers of their careers. As building a stock of personal capital with party elites is the sine qua

non of political advancement, these individuals may be especially inclined to engage in potentially

risky behavior which imposes costs on public institutions but redounds to the benefit of the party, in

so doing placing themselves in the good graces of the party leadership. As a result, in closed list

systems one is especially likely to encounter what might be called machine politics types of

corruptionrelatively organized groups of party militants embezzling funds, collecting kickbacks or

otherwise using the public resources or manpower under their control to finance the political

activities of their party. In this type of scenario, the public administration is especially likely to suffer

from an extreme level of politicization and considerable inefficiency.

43
 A prototypical example of how this type of corruption may come to manifest itself in a

closed list system is provided by the operation of the Duhalde machine (Liga Federal) in the province

of Buenos Aires in Argentina. Governor of Buenos Aires from 1991 to 1998, Duhalde solidified his

control of the province through the careful distribution of his network of punteros [party workers] at

all levels of provincial and municipal government. Duhaldes punteros were expected to use the public

resources at hand to support the activities of the Liga Federal and to bring out the Peronist vote

during election years. Time and time again, Duhalde used his influence over the composition of the

Peronist partys lists at the national and provincial level to keep the machine running. Punteros

implicated in corruption scandals who began to talk quickly found themselves in political exile;

punteros in the same situation who kept their mouths shut found themselves in high ranking positions

on legislative lists and, ultimately, in legislative office (cf Otero 1997, ch. 6).

Given the two faces of corruption associated with the different configuration of electoral

rules, the general equilibrium perspective would suggest that neither open nor closed lists present an

obvious advantage in terms of the overall level of political corruption induced by the electoral

system.

On the one hand, legislative candidates in open list systems are much more willing to go to

great lengthseven illegal onesto finance their political campaigns. This is because the very nature

of open list systems allows the electoral benefits of individual campaign spending to accrue primarily

to the candidate doing the spending. In closed list systems, the demand that individual candidates

have for campaign finance is much less. This is because, without preference votes, the effect of

campaign spending is diffused across all the members of the party list.

On the other hand, the mechanisms of political control within the bureaucracy are more

solid in closed list systems than in open list systems, making it easier for political actors to extract

resources from the state. This is because the carrots and sticks available to party leaders in closed list

systems are more compelling than they are in open list systems. In the open list case, politically-

oriented bureaucrats know that, when and if they choose to jump into electoral politics, their

44
candidacies will sink or swim on the basis of their ability to cultivate a personal vote. Although there

is certainly space for party leaders to interject themselves into this process, the tools available to party

leaders in open list systems are incomparable with the formal control enjoyed in closed list systems.

By focusing on only one aspect of the relationship between electoral institutions and political

corruption, the arguments enunciated by many reformers in Argentina and Brazil probably overstate

the impact of a reform in ballot structure on levels of political corruption, even though in both cases

reformers have been quite adept at identifying how electoral rules relate to the particular fashion in

which corruption manifests itself in their countries. The error in the prescription of ballot structure

reform as a solution to the problem of political corruption most likely rests with the failure to

consider the relevant counterfactual, namely, how a particular polity might look like after ballot

structure reform, holding all the other aspects of the electoral and bureaucratic environments

constant. In order to get a clear sense of such a counterfactual, one needs to understand how

institutional change impacts upon behavior in both the politico-electoral arena and the bureaucratic

arena. In this respect, this monograph seeks to contribute to a more complete picture of the

relationship between electoral institutions and political corruption by providing the theoretical and

empirical underpinnings for the less explored of the two causal effects of ballot structure just

identifiedthe consequences of ballot structure for corruption networks within the bureaucracy.

1.6 The Empirical Approach of the Monograph

This is a monograph with multiple levels of analysis. In terms of its empirical domain, the

study is simultaneously an N = 3 study, an N = 30 study and an N = 2,859 study. The monograph is

an N = 3 study in the sense that I examine political corruption and bureaucratic performance in three

South American countriesBolivia (closed lists), Brazil and Chile (open lists). However, within these

countries, I evaluate the determinants of corruption and bureaucratic performance using a survey I

conducted in thirty different public institutions (14 in Bolivia, 10 in Brazil and 6 in Chile), thus

45
making the monograph a N = 30 study as well. Moreover, as the survey was completed by 2,859

public servants in the three countries (1,038 in Bolivia, 1,226 in Brazil and 595 in Chile), I am able to

explore how different institutional conditions impact upon individual-level behavior. In this last

respect, the monograph is in fact an N = 2,859 study.

The three levels of observation are designed to complement one other. Variation across the

three country cases provides the study with leverage in evaluating claims about the impact of

national-level electoral institutions on patterns of political corruption. Variation across state agencies

within each country case provides the monograph with the ability to examine how agency-specific

differences in bureaucratic capacity impact upon bureaucratic performance. Moreover, as the

institutional conditions characterizing state agencies vary considerably within each country, the

diversity of said conditions gives birth to very different perspectives among public employees

regarding fundamental matters such as the politicization of bureaucratic tasks, the existence of stable

bureaucratic career paths and the efficacy of corruption control mechanisms. These differences, in

turn, improve the studys ability to make correct inferences about the impact of individual and

national-level characteristics on corrupt behavior. Due to the substantial diversity of employee

perspectives encountered across state agencies, the study is able to match individuals from

different country cases with similar or identical views about various features of their institutions, thus

allowing the study to conduct inference on the impact of individual experiences or national-level

institutions on corrupt behavior, holding key features of the bureaucratic environment constant.

The selection of country cases adheres to a similar logic. Although they differ in a number of

respects, the three country cases included in this monograph share important similarities which allow

the study to hold constant certain potentially confounding background factors which might

otherwise impinge on its ability to trace out the consequences of electoral system design.

Like many countries in Latin America, all three countries share a common legal heritage

based on French civil law. Such a legal tradition has been shown to be related to the presence of high

levels of bureaucratic red tape, extensive tax evasion and weak protections of property rights (La

46
Porta et al. 1999). According to the proponents of the legal origins vein of institutional analysis, the

presence of a legal system based on French civil law is generally reflective of a commitment to weak

constraints on state power by governmental authorities at the founding of new republics. In polities

marked by such a commitment, one might expect the use of state resources for partisan purposes to

flourish.

Perhaps even more important, Bolivia, Brazil and Chile all exhibit the particular

configuration of constitution-type and party fractionalization which has come to be called the

difficult combination by experts in Latin American politicsi.e., presidential government

combined with extreme multipartism. According to various scholars, this combination makes

democratic governability difficult, as the executive in such contexts is forced to negotiate his

legislative program with an amorphous and unruly set of actors in the Congress and often must buy

legislative support through offers of administrative posts and other favors (Linz 1994, Mainwaring

1993). Moreover, the difficult combination also ostensibly makes clean government harder to attain.

Parity in legislative seats and patronage resources becomes less likely as parties increase in number,

thus decreasing the attractiveness of reform for those large parties that enjoy the advantage (Geddes

1994). As it turns out, the party systems of Bolivia, Brazil and Chile are among the most

fractionalized in the Americas, with each country having an effective number of parties over 4

(measured in 2000). So if there are any countries in Latin America that should experience the ill

effects of the difficult combination, one would expect it be these.

The combined weight of legal heritage, constitution-type and party fractionalization would

suggest that Bolivia, Brazil and Chile are all unlikely reformers, and that all are likely to have enduring

problems with corruption. Yet almost all available measures of corruption and government

effectiveness suggest that there is a clear hierarchy of performance which distinguishes the three:

Chile being the regions most stellar performer, Brazil obtaining an intermediate position and Bolivia

generally classified as a laggard. Thus, the differing configuration of electoral laws and party

organization which characterize the cases allow us to examine what effect, if any, these variables

47
might have had on each countrys experience with political corruption, holding the aforementioned

confounding factors constant.

1.7 The Plan of Attack

The remaining chapters in this monograph are organized as follows. Chapter 2 presents the

studys approach to the measurement of political corruption and presents a statistical methodology

for analyzing the impact of factors which contribute to it. Therein, I discuss in detail the organization

of the public employees survey as well as the randomized response methodology used to ask

employees very sensitive questions about public resource use for partisan and personal ends. The

chapter shows how the randomized response methodology can be built upon to conduct individual-

level analysis of the determinants of corrupt behavior.

Chapter 3 draws upon the survey data to provide an in-depth examination of the

bureaucratic marketplace in the three cases. The chapter demonstrates that intrastate, institutional

differences in key dimensions of bureaucratic capacity such as civil service politicization, the

institutionalization of bureaucratic career paths and corruption control are often greater than cross-

national variation in these dimensions. Moreover, the chapter demonstrates that considerable

efficiency gains and a reduction in political corruption can be had in the cases from reinforcing

bureaucratic capacity in public institutions along these three dimensions.

Chapters 4 and 5 constitute the heart of the monograph. Chapter 4 provides a game-

theoretic analysis of the impact of a reform in ballot structure on political corruption, doing so from

the general equilibrium perspective. The chapters formal model, which utilizes a principal agent

framework with hidden knowledge, explicitly considers how the difference between open-list and

closed-list proportional representation may facilitate or impede corrupt 'contracts' from forming

between party leaders and militants in the public administration. The chapter argues that electoral

systems which intensify legislative candidates demand for electoral resources and thus their

48
willingness to engage in corruption (such as open list proportional representation), may make

extraction of resources from the state more difficult, thereby dampening supply (relative to systems

like closed list proportional representation). Consequently, the overall effect of ballot structure on

political corruption is not one of frequency or volume, but rather one of type.

Chapter 5 uses the studys survey data and the statistical methodology developed in Chapter

2 to test the key claim of Chapter 4; namely, that electoral rules have important consequences for the

efficacy of corruption networks within the bureaucracy. Overall, the chapter finds strong evidence in

favor of the purported causal linkage between electoral rules which concentrate power over political

careers in the hands of the party leadership and the willingness of politically-oriented bureaucrats to

engage in acts of malfeasance for the benefit of a party. Institutional change in the electoral arena is

thus shown to have important consequences for behavior in the bureaucratic arena, exactly as the

general equilibrium perspective proposes. Chapter 6 synthesizes the main findings of the monograph

and discusses a number of important unanswered questions that remain.

49
 Chapter 2
An Approach to Overcoming the Fundamental Problem
of Inference in Corruption Studies

50
1 The Fundamental Problem of Inference in Corruption Studies

Before one develops theories about the relationship between institutions and political corruption,

one first ought to establish that the empirical analysis of political corruption is a feasible enterprise.

Doing so is no mean feat. In this chapter of the monograph, I present a new approach to the

empirical analysis of corruption. Although the approach is tailored towards the specific problem

of understanding why public employees might engage in various forms of political corruption, the

methods I introduce are quite general, and they can readily find application in other pursuits in the

burgeoning field of corruption studies.

In defining my approach to the problem, I began with the following question: Were the ideal

dataset available to study the determinants of corrupt behavior, what might the core characteristics

of such a dataset be?

Clearly, one desirable characteristic of such a dataset would be that it consists of observations

at the individual-level. Of all the relevant actors considered in the social sciences, individual human

beings are the sole possessors of choice. Although it is often a convenient (and necessary) shorthand

to speak of aggregations such as unions, parties, nation-states as if they were unitary actorsand

thus possessors of choicein truth the outcomes such entities produce stem from complex processes

of collective choice in which a large number of individual decisionsmediated by formal rules and

cultural valueslead to distinctive patterns of organizational behavior. To put it simply, social choice

percolates up from individual choice. If we want to know why regions, states, and other groupings

dier in interesting ways, we need to know why the individuals that populate these entities make

the decisions they do.

A second desirable feature of such a dataset would be that it exhibits relatively precise measure-

ment of the illicit behavior in question. Although this requirement may seem obvious, it actually

asks quite a lot of the data. As corruption is an illegal and potentially stigmatizing activity, its prac-

titioners have every incentive to keep its incidence from coming to light. The intrinsic furtiveness

of such behavior makes its observation and codification a seemingly Sisyphean undertaking. One

seductive alternative to studying the illicit behavior directly might be to study the determinants of

51
some non- or less sensitive behavior thought to be associated with the true behavior of interest but

more easily observed and recorded by researchers. This is the proxy variable approach. By assump-

tion, explanatory variables found to be responsible for variation in the proxy variable are held to

be responsible for variation in the illicit behavior of interest. The precise measurement requirement

means that we would reject such an approach. In an ideal world, we would have the ability to hone

in on the illicit activity directly.

A third desireable feature of such a dataset would be that it contains information about a wide

range of characteristics of the participants in the study. Of particular importance would be the

inclusion of characteristics relevant for explaining why some individuals engage in the illicit behavior

and others do not. To the extent that all the major determinants of the illicit behavior were recorded

and coded into the dataset, the risk of challenges to correct inference such as omitted variable bias

would be much reduced. Moreover, with a wide range of predictors included in the dataset, we

would be able to establish the relative importance of various factors in determining participation in

the illicit behavior.

Were datasets with these features readily available, it would be possible to make probability

statements of the following form: a randomly selected individual i, drawn from the population ,

would be more (less) likely to engage in illicit activity Z if he had value c1 on individually-

measurable characteristic C than if he had value c0 on individually-measurable characteristic C,

holding all else constant. I understand the elaboration of such probability statements to be the

ultimate goal of (much of) the empirical analysis of corruption. With a critical mass of datasets

satisfying the above criteria, social scientists would have the foundation for making such statements

among a variety of populations, and presumably researchers could come fairly quickly to a consensus

about the features of the institutional environment that are most relevant for explaining variation

in dierent types of corrupt behavior.

And yet such a consensus is elusive, due at least in part to the fact that existing data will not,

in fact, bear such probability statements. The reason existing data cannot bear such statements, in

turn, can be chalked up to what I call the fundamental problem of inference in corruption studies.

52
 The fundamental problem of inference is simply stated. The true influence of an individually-

measurable characteristic C on the likelihood that an individual engages in an illicit activity Z

cannot be calculated unless either all subjects (or a random sample of subjects) are willing to reveal

their involvement in Z. However, some unknown proportion of subjects who have engaged in Z will

never reveal this fact, so long as their behavior can be known to anyone (including the researcher).

If queried, these individuals will choose instead to forego participating in a study on the subject or

they will misrepresent their true experiences if they do participate in such a study.

The fundamental problem of inference thus stems from a combination of two sources of error:

selection on the dependent variable (individuals who have engaged in Z are less likely to participate

in the study) and misrepresentation (some proportion of individuals who have engaged in Z will

participate and lie about their true experiences). In the pages that follow, I advocate for the use of

survey research coupled with the "randomized response" technique as one means of overcoming this

challenge to correct inference. The central idea of this approach is to make as small as possible

by providing a guarantee of privacy which is incorporated into the survey instrument itself. Given

several adjustments to standard techniques of multivariate data analysis, inference can then proceed

as it would for any other subject matter in the social sciences.

A word of caution. I do not wish to imply that the approach that I outline here will produce

ideal datasets for studying corruption. Far from it. However, the approach does provide a means

for making probability statements about the impact of a set of explanatory variables on corrupt

behavior at the individual level. To the best of my knowledge, this is the only approach for which

this is possible, apart from the (somewhat problematic) use of judicial or other ocial records. For

this reason, I believe the approach represents an important step forward in the empirical analysis of

corruption.

The remainder of this chapter proceeds as follows. Section 2 describes the most commonly

encountered approaches toward the empirical analysis of corruption and makes the case for the

diagnostic survey approach adopted in this monograph. Section 3 outlines the organization of the

survey. Section 4 discusses the "randomized response" technique as a means of reducing survey

53
error from lying and non-response in surveys of sensitive behavior, and section 5 makes the case for

the use of this technique to study corruption. Section 6 shows how individual-level inference can be

conducted using randomized response data, thus establishing the feasibility of my solution to the

fundamental problem of inference in corruption studies. Section 7 concludes the chapter.

2 Measuring Corruption: Recent Advances and Stubborn Challenges

For scholars who seek to better understand the causes of variation in corruption, issues of mea-

surement are naturally of great concern. Fortunately, an ever expanding menu of approaches to

measuring corruption provides an unprecedented opportunity for tailoring ones empirical research

design to the key concerns of ones study. As the nature of the propositions to be tested vary greatly

from one study to another, any attempt to identify a single optimal approach to measuring cor-

ruption is likely to be futile. Nevertheless, certain approaches recommend themselves for particular

endeavors and not others. Here I will briefly discuss the options that were available to me at the

onset of the project and present my rationale for adopting the measurement strategy I did: the

diagnostic survey approach.

The majority of available indicators of corruption fall into one of five categories: 1) corrup-

tion perception indicators based on the survey responses of businessmen or country specialists; 2)

aggregate indices based on a combination of the indicators in 1); 3) in-depth country or regional

diagnostic surveys designed to capture the opinions and experiences of citizens or public employees;

4) objective indicators, usually based on the variation in prices paid for similar items in public

procurement; 5) indicators based on judicial or legal records; and 6) measures of corruption gener-

ated by field experiments.1 Table 2.1 lists a number of datasets and/or studies that fall into each

category.

The suitability of a particular approach for a particular research project can be evaluated in

terms of five important criteria: coverage, conceptual precision, level of aggregation, potential for

1 This section provides an overview of the dierent types of indicators available and assesses their strengths and

weaknesses along a number of dimensions. For a more detailed discussion of the indicators themselves, see Di Tella
and Savedo (2002), Kaufmann et al (1999) and Speck (2000). It is also worth noting that researchers have pro-
duced indicators constructed from accounts of corruption reported in the press, although this approach is relatively
uncommon. In this regard, see Pharr (2000) and Fleischer (1999).

54
bias and the potential for harm to subjects. The selection of a measurement strategy invariably

involves a trade o between these criteria.

The aggregate indicators, for instance, produced by collapsing the information available from

dierent polls, have by far and away the greatest cross-national coverage (the latest versions of

the WBI Governance Matters index and the TI corruption perceptions index cover 199 and 133

countries, respectively). As a result, many researchers have found them useful for investigating

variation in aspects of cross-national performance such as economic growth, foreign direct investment

and the like (Lambsdor 2003, Mauro 1995, 1997, Habib and Zurawicki 2002). On the other hand,

the cost incurred by the greater coverage of these indicators is that the concepts they measure

are necessarily more general (or vague, depending upon your perspective) than those measured by

the source variables which contribute to them. In other words, the conceptual precision of these

indicators is low relative to other measures of corruption.

Table 2.1: Indicators of corruption by type

The trade o between coverage and conceptual precision can be illustrated by considering the

variables produced by the Global Competitiveness Survey of the World Economic Forum (WEF),

55
used as source indicators in both of the aggregate indices cited in table 2.1. The WEF survey provides

measures of the perceived frequency of a large array of behaviors that may be classified as corrupt:

money laundering, irregular payments in public contracts, the diversion of public funds, illegal

campaign donations, the exposure of the civil service to political pressure and so on. Upon their

incorporation into the aggregate indices, these measures are dissolved into broad categories such as

corruption perception, government eectiveness, rule of law and the like. In and of itself, such data

reduction can be quite useful, especially for illuminating basic cross-national patterns. However,

researchers interested in conducting more fine-grained analyses of the cause and consequences of

dierent types of corruption will have to turn their attention elsewhere.

Both the perception indicators and aggregate indicators based upon them may be found wanting

with respect to another key dimension: level of aggregation. All of these indicators measure cor-

ruption and/or government eectiveness at the level of the nation-state. That is, they purport to

measure the quality of the government bureaucracy as a whole. The very reasonable objection could

be raised that such a measure actually masks more than it reveals. For example, how might we rate

a public sector for which the vast majority of executive agencies are noted for their eciency and

professionalism but whose Transportation Ministry has been repeatedly marred by scandal? Is this

a corrupt public sector? It is more or less corrupt than a public administration in which corrupt

practices are diused throughout the public sector but in no single institution reach the frequency

or cost of those found in the first countrys Transportation Ministry? These questions go to the

heart of issues social scientists care deeply about. Dierent distributions of corruption in the various

organs of the state can be expected to produce varying outcomes in terms of public good provision in

infrastructure, education, health and so forth, each of which, in turn, may have unique ramifications

for economic growth and overall well-being.2

Equally important, the aggregate perception indicators give us no purchase in understanding the

determinants of corruption at the individual level. As a result, these indicators are of relatively little

utility in helping us to understand the social mechanisms which determine why one set of actors

2 Given the enormous potential importance of cross-institutional, not simply cross-national, variation in bureau-

cratic capacity, I focus extensively on this issue in chapter 3 of the present monograph.

56
engages in illicit behavior in one context and a similarly situated set of actors refrains from doing so

in another. As progress in the social sciences is ultimately best measured by a growing understanding

of the operation of social mechanisms, not just input-output analysis, this is a considerable drawback

(Hedstrm and Swedberg 1998).

For the present study, the crucial importance of conceptual precision and a low level of aggrega-

tion proscribed reliance on the perception-based indicators. As the reader will recall, this monograph

is dedicated to understanding how various types of political institutions aect political corruption, as

distinct from personal corruption. The theoretical expectations I develop in the chapters that follow

do not apply to personal corruption nor to corruption writ large. As the lions share of perception-

based indicators recognize no distinction across categories of corruption, they are ill-suited to the

studys needs.3

The foregoing discussion points to the need for more disaggregated measures of corruption. In

this regard, diagnostic surveys and institution or region-specific objective measurements have an

advantage. The former can be used to ask detailed questions about the sites, forms and frequency

of corruption, thus providing a high level of conceptual precision. The latter can be used to identify

sites of corruption as well and even produce tentative estimates of its cost.

Both also have important limitations. In practice, it can be dicult to distinguish the so-called

the objective measures from simple mismanagement or negligence in the use of public funds. With

these measures, the ultimate use of funds, political or otherwise, is generally unobservable to the

researcher. Many existing diagnostic surveys, on the other hand, probably suer from bias due to

non-response and lying, especially to the extent that participants are asked to provide responses

which may incriminate themselves or their superiors. The coverage of objective indicators and

diagnostic surveys is also limited, due to the unit of measurement in the case of the former (prices

of similar inputs used by multiple institutions in the same sector) and practical limitations of time

and cost in the case of the latter.

3 On this point, I find attempts to gloss over this issue by assuming all bad things are correlated (all forms of

corruption must go together) unconvincing. One need not work too hard to come up with examples of polities where
the overall level of corruption seems to be low but political corruption is rife. Japan and South Korea appear to be
good examples in this regard.

57
 Judicial and legal records present an alternative means of conducting a disaggregated analysis

of corruption. In practice, such indicators are often based on conviction rates for governmental

corruption across subnational units of a polity or judicial procedures initiated against particular

politicians. As is the case with diagnostic surveys, these indicators may allow for individual level

analysis of corruption. Nevertheless, although all corruption indicators suer from bias to varying

degrees, the potential for bias with these indicators may be particularly troublesome. In polities

with highly politicized judiciaries or prosecutors oces, their use may be a nonstarter, as it may

be impossible to dierentiate between legitimate and politically motivated legal action. To the

degree that such measurements are characterized by subnational variation, dierences in enforcement

capabilities at the regional level may add additional complications. Finally, given that such indicators

are based on country-specific legal codes and procedures, cross national comparisons based on such

indicators are generally infeasible.

A final approach to measuring corruption is the field experiment. In conducting field experiments,

scholars have been able to examine in detail the mechanisms by which corruption manifests itself in

particular national contexts and within specific realms of state activity. The pioneering study in this

regard consists of the red tape experiments conducted by Hernando de Soto and colleagues in the

Instituto Libertad y Democracia in Peru and documented in de Soto (1989). In studies of this type,

researchers organize experiments in which project assistants pretend to be engaged in an activity

which requires the concession of some benefit from a public ocial, such as incorporating a business,

establishing legal title to a home or acquiring a drivers license. By monitoring the experiences

of these individuals as they negotiate their way through the bureaucratic maze, these studies are

able to identify important who and hows: to whom do illicit payments need to be made in order

to acquire a benefit? How long do the bureaucratic procedures take?; How much does the benefit

ultimately cost citizens?

Much has already been learned from this very important work. Nevertheless, two important po-

tential drawbacks counseled against the adoption of such an approach in the present study. Firstly,

although field experiments can tell us a lot about how corruption emerges in citizen-state interaction,

58
they are less informative about the going ons within state bureaucracies. The application of the field

experiment approach to address the types of issues which concern me heresuch as the identifica-

tion of factors which determine why bureaucrats engage in party-directed corruptionwould require

something along the lines of an undercover operation on the part of trained actors posing as public

employees. This is a job for the police, not academicians. Secondly, the field experiment approach

to the study of corruption in raises some very thorny ethical issues. In such experiments, project

assistants are often required to engage in some form of deception vis-a-vis public authorities and are

coached to try their best to obtain the benefit in question, even if that requires that they make an

illicit payment to a state ocial or an intermediary. Many researchers might view such practices

as inimical to two core principles of modern research involving human subjects: respect for per-

sons, including informed consent on the part of all research subjects, and beneficence, which entails

the minimization of possible harm for research subjects (National Commission for the Protection

of Human Subjects of Biomedical Research, 1979). The respect for persons is somewhat dubious

in this approach, as the bureaucrats whose behavior is the central focus of the research are not

informed that they are being included in a scientific investigation. Of greater concern, however, is

the potential lack of beneficence, as project assistants and/or subjects may need to engage in illegal

or unethical behavior as part and parcel of the research. In so doing, they may expose themselves

and/or the researcher to legal sanction. Of course, at the end of the day it is the role of university

human subjects committees to decide if the considerable potential benefits associated with this type

of research outweigh the potential risks to participants.

Table 2.2 presents an evaluation of the benefits and drawbacks of the six types of approaches

according to the five criteria under consideration.

59
 Table 2.2: The indicators evaluated

Although far from perfect, the diagnostic survey approach suggested itself as the most appro-

priate empirical strategy for this project. Diagnostic surveys allow for as great a deal of conceptual

precision as the question writer can muster, they permit analysis at the individual level and, with

sucient attention paid to survey design, they can keep bias to minimal levels. Moreover, with

careful attention paid to issues of confidentiality, they can reduce the potential harm to subjects to

practically zero. For these reasons, they can serve as a useful vehicle for improving our understanding

of the mechanisms by which political corruption flourishes or wilts within the public administration.

Equally important, they can be utilized in a cross-national context (barring the inclusion of cases

with insuperable linguistic or cultural dissimilarities), thereby providing researchers with leverage

in gauging the eects of national-level institutions on bureaucratic behavior.

Needless to say, the approach is not without drawbacks. With only three country cases, the

limited cross national variation available to the study necessarily imposes a good deal of humility

on my aspirations to make general claims about the eect of electoral rules or other national-level

60
sources of variation in political corruption. Nevertheless, the fine-grained nature of the dataset

ensures that the key concepts of my theoretical framework match up with the measures and that

I am able to analyze how real institutions perform and how real individuals act. The reader is

ultimately the best judge of the wisdom of the approach.

3 Survey Organization

In the three country cases included in the study, Bolivia, Brazil and Chile, I conducted a public

employees survey with a set of randomly selected public servants in a number of key executive

branch institutions. The institutions selected for inclusion in the study included those responsible for

financing the activities of the government (tax services, customs agencies), guaranteeing the stability

of the national currency (the Central Bank), providing public works and infrastructure (public works

ministries and road services), promoting development projects and agricultural reform (development

ministries, social investment funds, land reform institutions), as well as select regulatory agencies.

Besides being representative of the major activities of the state, the institutions were selected in

order to guarantee variation in a priori expectations about patronage and corruption.

Figure 2.1: The Certificates of Honor

The surveys were self-administered and anonymous. They were delivered to and collected from

individuals selected for participation in the study by trained undergraduate assistants as well as

myself. Each individual who returned a completed survey received a certificate of honor, designed

61
by myself, as recognition of their participation (see Figure 2.1).

Table 2.3 presents the list of institutions selected for inclusion in Bolivia, Brazil and Chile. All

of the institutions selected for inclusion were national-level institutions. Excluded were municipal

or other sub-national level institutions. Only employees that worked in the central oce of a given

institution were included in the survey (regional oces were excluded). In Brazil and Chile, all

individuals classified as auxiliary employees were also excluded (drivers, messengers, food servers,

etc.).4 All told, 2,859 employees in 30 dierent institutions participated in the survey.

Table 2.3: Institutions Selected for Inclusion in Bolivia, Brazil and Chile

4 Fifty percent of the employees who fit the eligibility criteria in each institution were chosen at random to participate

in the survey (up to a maximum of 250)

62
 Access to the institutions selected for participation in the project diered in each country. In

Bolivia, the survey was given the ocial support of the newly created Secretariat of the Fight Against

Corruption (SLCC). Most likely as a result, none of the institutions requested to participate refused

to do so. Similarly, in Brazil the survey was given the support of the Comptroller General of the

Union (CGU). Very few institutions choose not to participate in this country.5 In Chile, similar

attempts to elicit the support of government institutions did not meet with success. Although the

project was given the kind support of Chile Transparente (the Chilean branch of Transparency

International), a number of the institutions asked to participate refused to do so (see Table 2.3). As

a result, the Chilean results should be considered somewhat less representative of the experience of

executive branch institutions in that country than the Bolivian or Brazilian ones.

All things considered, the response rate of subjects selected for participation in the study was

quite high. Response rate by country varied from 69% (Brazil) to 88% (Chile). Response rate by

institution varied from 42% (Instituto Nacional de Reforma Agraria-Bolivia) to 100% (Superinten-

dencia Agraria-Bolivia, Superintendencia General-Bolivia).6

The survey itself was broken up into three sections. The first section contained questions designed

to measure employee perceptions regarding the quality of internal and external control, political

pressure in hiring, promotions and dismissal, the likely behavior of other employees in the institution

in the case of a (known) misuse of funds, their own personal ethics, the degree of job security, and

the reach of their wage. Also included in this section were questions about an employees party

aliation (whether he had one, not to which party he is aliated) and that of his superior (whether

party aliations match, not to which party they both did or did not belong). The second section

contained a set of questions that dealt with personal knowledge, perceived frequency and even

participation in the use of institutional resources for partisan or personal ends. These questions all

5 In both Bolivia and Brazil, ocial support consisted of formal letters sent to the heads of the institutions selected

to participate in the study. The letters simply asked for the cooperation of the institution in question given the
relevance of the project to the mission of the anti-corruption institutions. Selected institutions were under no legal
obligation to comply.
6 In the case of the Instituto Nacional de Reforma Agraria in Bolivia, the low response rate was due to the fact

that the riots and social protests of October 2003 were in full swing at the time the survey was to be collected. For
reasons of safety, the collection of surveys had to be called o prematurely for this institution.

63
used a specific format, "randomized response" (RR), which is discussed in detail below. The third

section contained several questions designed to gauge the participants understanding of how the

RR format guaranteed the confidentiality of their responses to the questions in the second section.

4 A Survey Methodology for Sensitive Questions: Randomized Response

In my evaluation of the existing approaches to the measurement of corruption, I noted a potentially

important drawback of the diagnostic survey approach: respondents, when faced with questions

about illegal or undesirable behavior, may choose to lie about their experiences or not respond,

introducing a potentially serious bias into the survey results. From the standpoint of survey design,

the challenge is twofold. As good academicians, we want to be able to elicit honest responses to

sensitive questions, the answers to which many respondents would not want anyone to know. As

researchers attuned to our ethical responsibilities, we want to be able guarantee that this information,

if obtained, could never be used to harm the individuals who participated in the study.

For many researchers in the social sciences and public health, the solution to this type of dilemma

has been the use of the randomized response (RR) survey. The basic idea behind RR surveys is

to guarantee respondent anonymity to delicate questions by requiring that responses depend on the

realization of a randomizing device introduced by the principal investigator. Since the principal

investigator has designed the randomizing device himself, he can estimate the frequency of delicate

behaviors in the population of interest even though it is impossible to know if any given individual in

the sample has engaged in such behavior. The survey project used one particular variant of the RR

survey (the so-called Warner Model (1965)) in order to make it possible to simultaneously protect

the anonymity of respondents and at the same time estimate the frequency of dierent types of

corruption (and the uncertainty about the estimates of this frequency) for various subclassifications

of the data.7

The survey questions dealing with the illicit use of institutional resources worked in the following

way. For each question in this section, respondents were presented with two dierent statements, a

7 For comprehensive overviews of the randomized response survey methodology, see Fox and Tracy (1986) and

Chaudhuri and Mukerjee (1988).

64
statement marked A and a statement marked B. The two statements were identical, save for the fact

that one was in the positive form and the other was in the negative form. Below the two statements

were two boxes: a box to mark true and a box to mark false. In each survey packet, respondents

were given a spinner containing two dierent regions: a region A and a region B (see Figure 2.2).

They were asked to spin the spinner for each question in the section. If the spinner landed in region

A for a given question, they were asked to respond true or false only with respect to statement

A. Similarly, if the spinner landed in region B for a given question, they were asked to respond

true or false only with respect to statement B. As only they knew the region in which the spinner

had landed for each question, only they could know to which statement their answer corresponded.

Confidentiality was guaranteed.

Figure 2.2: The Randomized Response Spinner

Now note that because I designed the spinner, I know the probability that the spinner lands in

region A and region B. This information can be used to estimate the proportion of employees in

a given institution that have engaged in or have personal knowledge of an illicit activity in their

institution.

65
 Figure 2.3: Example Randomized Response Question

Let us consider an example. Figure 2.3 presents one of the questions in this section. If the spinner

lands in region A, the respondent is directed to answer true or false to a statement indicating that he

had been pressured, at least once, to use the resources of his institution for the benefit of a political

party. This event occurs with probability p. If the spinner lands in region B, which occurs with

probability 1 p, the respondent is directed to answer true or false to the negative form of that

statement. Now let us suppose that the number of armative responses in the institution is n and

the number of total responses is N . The proportion of employees that have been pressured (which

we seek to estimate) is S . Given the survey design, we simply have:

n
= pS + (1 p)(1 S ) (2.1)
N

In words, the proportion of armative responses in a given institution is equal to the proportion

of individuals responding to statement A, on average equal to p, multiplied by the proportion of

individuals that have been pressured, plus the proportion of individuals responding to statement B,

on average equal to 1 p, multiplied by the proportion of individuals that have not been pressured.

Since we know p, we can estimate the proportion of individuals that have been pressured as:

n
+p1
brr
S =
N
, p 6= .5 (2.2)
2p 1

66
 The variance of this estimate is the following (see Warner 1965 for the derivation):

bS (1
bS ) p(1 p)
rr
var(b S )= + (2.3)
N N (2p 1)2

The benefit of using the RR survey design is that it reduces bias in our estimates resulting

from lying and non-response. This benefit does not come at zero cost. The cost of the RR design

is greater variance in our estimator than that resulting from direct survey questions. Is this is a

trade-o worth making? Is it better to produce estimates with a substantial degree of variability

but that are correct on average or to produce estimates that are systematically biased due to lying

and non-response but are calculated with greater precision? Not surprisingly, the answer depends

on how much of a dierence in bias and variance we are talking about. This is where the details of

the research project itself come into play. The following section attempts to provide an outline for

gauging when the use of randomized response might be appropriate for a particular study. There,

I make the case that in many circumstances it is a logical choice for the study of governmental

corruption.

5 The Choice for Randomized Response

To many, it might seem clearly preferable to have greater uncertainty about estimates that, in ex-

pectation, are correct than have greater certainty about estimates that, in expectation, are incorrect.

However, this need not always be the case. Figure 2.4 helps illustrate why not.

In the figure, I consider two hypothetical estimators of the proportion of a population bearing

b1S and
the sensitive trait of interest, b2S . The first estimator is biased; that is, E[b
1S ] 6= S , but has

2S ] = S , but has greater

smaller variance than the second. The second estimator is unbiased; E[b

variance than the first.

Which is to be preferred? In comparing estimators, one is generally concerned with long term

b1S and
performance. Suppose we were to compare b2S by asking the following question: If we were

bjS , j {1, 2}, an infinite number of times to study dierent populations, what
to use the estimator

bjS | < , where is some small positive constant? We might be

proportion of the time would |S

67
inclined to prefer the estimator which produces estimates which are close to the true proportion

with greater frequency.

b1S show the proportion of times

In the graph below, the black segments of the distribution of

b1S ; that is, the proportion of instances in which |S

this condition does not hold for b1S | > .

b2S show the proportion of instances in

Similarly, the grey colored segments of the distribution of

2S | > . The graph shows that although

which |S b b1S is biased, in the long run the frequency with

b1S | < will be greater than the frequency with which |S

which |S b2S | < . In words,
b1S will

b2S . As shown,
be within an arbitrarily small distance from S with greater frequency than will

b1S will be less than away from S is approximately twice the

the proportion of instances in which

proportion of instances in which the latter will be less than away from S . Given these dierences

in long run performance, in this example it might make sense to choose the biased estimator over

the unbiased estimator.

Figure 2.4: Comparing Estimators (Bias vs. Variance)

If an unbiased estimator is not to be uniformly preferred to an biased estimator, how might we

evaluate if the potential bias from survey non-response and lying is sucient to merit the adoption

68
of the RR technique? One common means of comparing dierent estimators consists of evaluating

bjS , the MSE is written:

the mean squared error (MSE) associated with each. For the estimator

jS ) = var(b
M SE(b jS ) + [bias(b
jS )]2 (2.4)

Given this metric of comparison, we can evaluate the circumstances in which it would make sense

brr
to adopt the RR estimator, bdS .
S , instead of the direct question estimator, denoted by

Suppose that a survey is performed with the participation of N respondents. Individuals who

do not have the sensitive trait always respond to a direct survey question about whether or not

they hold the sensitive trait, and they always do so honestly. Individuals who do have the sensitive

trait decide how to respond in the following manner. First, they assess whether or not they are

willing to respond honestly to the question if, in fact, they choose to respond. The probability

that an individual with the sensitive trait is willing to respond honestly is H. Second, given their

assessment about whether or not they can respond honestly, they choose whether or not to respond

at all. The probability that an individual bearing the sensitive trait who is disposed to respond

honestly will actually respond is RH . The probability that an individual bearing the sensitive trait

who is disposed to respond dishonestly will actually respond is RD .

bdS are:
The above assumptions imply that the expected value, bias and variance of

S RH H
dS ] =
E[b (2.5)
(1 S ) + S R

dS ) = E[b
bias(b dS S ]

S RH H
= S (2.6)
(1 S ) + S R
RH H[(1 S ) + S R S RH H]
dS ) = S
var(b (2.7)
N [(1 S ) + S R]2

where R = RH H + RD (1 H) is the probability that a randomly chosen individual who bears

the sensitive trait responds to the question.

Given dierent values of p and assumptions about RH and RD , we are now equipped to ask

69
 bdS is a better estimator of S
the following question: What is the minimum level of H for which

brr
than S ? That is, what proportion of the individuals with the sensitive trait must be willing to

respond truthfully to a direct survey question about having the trait in order to merit using the

direct method instead of the RR method? Figure 2.5 compares the two estimators in terms of the

MSE criterion for various levels of p, RH , RD and H. I assume that because of the confidentiality

guarantee provided by the RR technique, all participants queried in this way will respond and will

respond honestly about whether or not they have the sensitive trait.

Figure 2.5: Direct and RR Estimators Compared (N = 500, S = .3)

Each plot in the figure graphs the relationship between H (x-axis) and the ratio of the MSE of

bdS to the MSE of

brr brr
S (y-axis). Using the MSE criterion, bdS whenever this ratio
S is preferred to

bdS is preferred to
is greater than 1; brr
S whenever this ratio is less than 1. The level of H at which

bdS
the curve of the MSE ratio crosses the horizontal line at 1 is the minimal level of H for which

brr
is preferred to S . In each plot, the proportion of the population bearing the sensitive trait, S , is

70
set to .3 and N = 500.

The parameter values chosen in the figure were conservative, in the sense that they assumed

relatively weak forms of response bias. For this reason, we might expect a priori the direct question

estimator to perform well relative to the RR estimator. Clearly, this was not the case. Even if

we assume the best of all possible worlds for the direct estimatorno non-response among those

individuals bearing the sensitive trait who are inclined to respond honestly (RH = 1) and zero

participation among individuals bearing the sensitive trait who are inclined to respond dishonestly

(RD = 0)the level of truthfulness meriting the use of the direct estimator over the RR estimator

still varies between .55 (p = .6) and .86 (p = .8). If we assume relatively mild non-response on the

part of those with the sensitive trait (RH = .8), then even perfect honesty among respondents is

insucient to justify using the direct estimator over the RR estimator in some instances (p = .8).

Thus, in contexts in which researchers have legitimate concerns about both lying and non-response

on the part of subjects bearing a sensitive trait, the magnitude of priors about the size of these

biases need not be particularly large to merit the use of randomized response.

What expectations should we have about the potential for biases of this sort in direct surveys

about governmental corruption? Luckily, a series of diagnostic surveys conducted by the World Bank

since the late 1990s provide an anchor for expectations in this regard. One survey in particular,

Cambodias Governance and Corruption Diagnostic (World Bank 2000a), provides a particularly

illuminating view into the potential diculties associated with asking public employees directly

about corruption. In this survey, respondents were asked to estimate how the receipts from bribes

were shared (with their superiors, colleagues and politicians). Non-response for these questions

varied from 68% to 72%. Another question in the survey prompted respondents to estimate the

prevalence of the sale of public positions. Non-response for this question was 95% (the estimated

proportion of jobs bought being only 7%).

Beyond non-response, there was also reason to doubt the truthfulness of some of the responses

provided. Citizens and enterprises surveyed as part of the diagnostic reported having to pay bribes

with much greater frequency than the public employees own estimate of the frequency with which

71
bribes were collected. The dierences between the estimates were separated by orders of magnitude.

For instance, foreign enterprises reported having to pay bribes in 68% of their contacts with public

ocials, as opposed to an estimate of 7% produced by public employees. Similarly, urban households

reported having to pay bribes in 53% percent of their contacts with public ocials, whereas public

employees estimated that citizens had to pay bribes in only 9% of interactions.

Without some credible mechanism to protect public employees from the potential consequences

of responding honestly to questions of this stripe, it would seem that direct surveys of corruption face

very serious challenges. There is reason to believe that RR may help in overcoming some of those

challenges. In the survey conducted for the present monograph, non-response for the questions asked

in RR format was quite low, varying from 4% to 5%. Certain questions which utilized this format

were at least as sensitive as those mentioned above for the Cambodia diagnostic, as they queried

respondents about their own participation in instances of resource use for ones personal benefit or

the benefit of political parties. Moreover, it does not appear that the dierences in response rates

are solely due to cultural dissimilarities between the Latin American cases included in the present

study and Cambodian case. The most sensitive question included in the present study that did not

use the RR technique asked participants to estimate (hypothetically) by how much an individual

of their rank within their institution could increase annual income if were they inclined to accept

bribes on a regular basis (see chapter 5). Non-response for this question was 29%.

The preceding discussion notwithstanding, the question remains as to whether or not randomized

response in practice elicits greater truthfulness to sensitive questions than do other survey methods.

A recent meta-analysis of studies using randomized response and other survey techniques (face-to-

face, self-administered surveys, computer-aided surveys and telephone) suggests that this is indeed

the case (Lensvelt-mulders et. al. 2005). This work analyzes the results of both validation studies

and comparison studies, and finds that RR generally provides the best estimates of the prevalence

of sensitive behaviors. Importantly, the authors find that the performance of RR improves as the

sensitivity of the topic under study becomes greater. As political corruption is about as sensitive as

one is likely to find in political science, the choice for randomized response in this project was clear.

72
 One potential drawback of the randomized response approach is that it places a greater cognitive

burden on survey respondents than the traditional, direct question survey format. For this reason,

the instructions for the RR segment of the survey were provided verbally by project assistants as well

as included in written format within the survey instrument itself. Nevertheless, in order to verify that

participants understood the logic of the RR approach, the final two questions of the survey queried

them directly about their understanding of the RR technique as applied in the survey. The first of

these questions simply asked respondents whether or not they understood the manner in which the

RR procedure guaranteed the anonymity of their responses to the sensitive questions. The second

question asked those respondents who responded in the armative to indicate how the anonymity

guarantee was provided by the RR procedure. They were given three potential responses: 1) "No

one else besides me could ever know to which statements my responses in this section correspond

since only I know whether the spinner landed in section A or section B for each question" (correct

response); 2) "a machine reads the survey responses in such a way that it is impossible to know

what my responses were" (incorrect response); 3) "If the spinner lands in section A, I respond True.

If the spinner lands in section B, I respond False. The manner in which I respond depends only on

the spinner" (incorrect response).

Approximately three fourths of the participants indicated that they understood how the random-

ized response procedure guaranteed the confidentiality of their responses to the delicate questions

(see figure 2.6). Moreover, the vast majority of respondents were also able to identify the exact

mechanism by which the RR procedure guaranteed their confidentiality. Although this latter pro-

portion was far from 1 (as it ideally would be), it is nevertheless clear that the lions share of survey

participants were cognizant of the basic logic of randomized response. Thus, while it is true that

RR imposes a slightly higher cognitive burden than other survey methodologies, it is nevertheless

clear that this burden is also a superable one.

73
 Figure 2.6: Participant understanding of confidentiality guarantee produced by RR procedure

6 An Approach to Overcoming the Fundamental Problem of Inference

The RR methodology as presented thusfar constitutes a means of inferring the proportion of a

given population that bears a sensitive trait. The presentation up to this point has shown how one

can calculate the proportion of individuals bearing the sensitive trait in any aggregation: a public

institution, a country, etc. However, we would like to do more. We would like to conduct individual-

level analysis. In this section, I will elaborate the means by which inference can be conducted about

the eect of a set of explanatory variables on the likelihood that a bureaucrat engages in a particular

act of corruption. In order to do so, I will proceed in three steps.

First, I present a means of estimating the proportion of a population exhibiting a sensitive trait

when observations of the dependent variable consist of a mixture of RR data and direct response

data. As a small proportion of respondents in the study chose to respond directly to the questions

presented in RR format, this was an important step in modeling the probability that a randomly

74
selected bureaucrat had engaged in a particular illicit behavior.

Next, I consider how the method of maximum likelihood can be utilized to estimate the eect

of individually-measured explanatory variables on the likelihood that a bureaucrat had engaged in

an illicit behavior. This approach consists of modifying standard binary response models in order

to conform to the probabilistic structure of the studys data generating process.

Finally, I consider how matching techniques can be utilized to do the same thing (albeit with

weaker assumptions): namely, estimate the eect of a binary predictor (treatment) variable on the

likelihood that a bureaucrat had engaged in a particular illicit behavior. Since the proportion of

respondents bearing the sensitive trait can be estimated for any subclassification of the data, such

techniques can be utilized in their standard forms with few modifications required to adjust for the

RR data generating process.

6.1 Estimation with Direct and Randomized Responses

A small proportion of the studys respondents chose to circumvent the randomized response proce-

dure and respond to the sensitive questions directly.8 Those who did so either made a note on the

questionnaire denoting where the spinner had landed for a particular question, or they underlined,

circled or wrote out their response. Naturally, the direct survey responses were coded distinctly from

the randomized responses.

Although direct responders constituted a fairly small group, issues of self-selection eliminated

the option of simply removing the observations in which respondents circumvented the randomizing

device. The concern is simply stated. In making his choice about whether or not to use the protection

aorded by the randomized response procedure, a given respondent will likely condition his choice on

whether or not he is the bearer of the sensitive trait in question. Respondents who do not bear the

sensitive trait will be less inclined to use the protection of randomized response (they have nothing

to hide) while those who do bear the sensitive trait will be more inclined to use the protection

aorded by randomized response (they do have something to hide).

The consequence of self-selection into randomized versus direct responders is that neither the

8 The proportion of direct responders varied from 14% to 16% of respondents, depending on the question.

75
 brr
randomized response estimator S , calculated for the respondents that used the randomized re-

sponse procedure, nor the proportion of direct responders who admit to having the sensitive trait,

bdS , is an unbiased estimate of s . The randomized response estimate will likely be biased upward, as

respondents with the sensitive trait are more likely to use the protection of the randomizing device.

Similarly, the direct response estimate will likely be biased downward, as those who do not have the

sensitive trait will be more apt to forego the protection of randomized response.9 Fortunately, ob-

taining an unbiased estimate of s with both randomized and direct responses can be accomplished

through a rather a straightforward extension of the methods presented in the previous section. In

what follows, I will derive the method of moments estimator of s with mixed data and show that

this is indeed also the maximum likelihood estimator.

6.1.1 Method of Moments Estimation

Method of moments estimation entails equating various sample moments to the corresponding popu-

lation moments then using the resulting system of simultaneous equations to solve for the parameters

of interest. Although not advertised as such, the derivation of the randomized response estimator in

section 4 was based on method of moments estimation. We can use the same strategy to estimate

s with a combination of direct and randomized responses.

Note that each of the N respondents provides one of j dierent responses, where j {R1, R0, D1, D0}

and element R1 denotes the event that the randomizing procedure is used and a response of true

is given; R0 denotes the event that the randomizing procedure is used and a response of false is

given; D1 denotes the event that the response is direct and a value of true is given; and D0 denotes

the event that the response is direct and a value of false is given. The number of responses in each
X
category is nj , where nj = N .
j
Let us define the following set of dummy variables:

9 In b dS was uniformly smaller than

line with this discussion, b rr
S for all the RR questions asked in the survey.

76
 yi,j = 1 if the ith individual exhibits response j

yi,j = 0 otherwise

Assuming that responses are i.i.d., we can write:

P (yi,R1 = 1) = p R|S S + (1 p) R|S (1 S ) (2.8)

P (yi,R0 = 1) = p R|S (1 S ) + (1 p) R|S S

P (yi,D1 = 1) = (1 R|S )S

P (yi,D0 = 1) = (1 R|S )(1 S )

where R|S is the probability that a randomly chosen individual chooses to use the randomizing

device given that he has the sensitive trait and R|S is the probability that said individual uses

the randomizing device given that he does not bear the sensitive trait. Let us denote R as the

probability that a randomly chosen individual uses the randomizing device. Using the fact that
R R|S S
R|S = 1S , we can rewrite the above as:

P (yi,R1 = 1) = R (1 p) + R|S S (2p 1) (2.9)

P (yi,R0 = 1) = R p R|S S (2p 1)

P (yi,D1 = 1) = (1 R|S ) S

P (yi,D0 = 1) = 1 S R + R|S S

Given the structure of the choice process, it is appropriate to assume that the N values of yi,j are

selected from a population with a multinomial probability distribution. Using information about the

first sample moment and its population analog, we produce the following system of three equations

in three unknowns:

77
 N
X
nR1 = E[yi,R1 ] = N [R (1 p) + R|S S (2p 1)] (2.10)
i=1
N
X
nD1 = E[yi,D1 ] = N [(1 R|S ) S ]
i=1
N
X N
X
nR1 + nR0 = E[yi,R1 ] + E[yi,R0 ] = N R
i=1 i=1

Solving explicitly for the parameter vector V = (b bR|S ,

S , bR ), we calculate the method-of-

moments estimator VbM M as:

p nR1 1 p nR0 nD1

bS =
+ (2.11)
2p 1 N 2p 1 N N
p nR1 (1 p) nR0
bR|S =

p nR1 (1 p) nR0 + nD1 (2p 1)
nR1 + nR0
bR =

N

bS is simply the weighted average of the randomized response estimator

The reader can verify that

(weighted by the proportion of respondents that used the randomizing device) and the proportion of

direct responders that state that they have the sensitive trait (weighted by the proportion of direct

bS is:
responders). The expected value of the estimator

N N N
1 p X 1p X 1 X
S ] =
E[b { E[yi,R1 ] E[yi,R0 ]} + E[yi,D1 ] (2.12)
2p 1 N i=1 N i=1 N i=1
1
= {p[R (1 p) + R|S S (2p 1)] (1 p)[R p R|S S (2p 1)]}
2p 1

+(1 R|S ) S

= S

thereby establishing that the moment estimator of S is unbiased.

As the moment estimator consists of non-linear functions of random variables, the multivariate

78
delta method provides a helpful means of calculating the asymptotic variance of the parameter

vector. Section A of the appendix to this chapter uses this method to establish that the following

are the asymptotic variance estimators of each element in the parameter vector:

bS )
bS (1 p(1 p)
var(b
S ) = bR
+ (2.13)
N N (2p 1)2
N (2p 1)2
R|S ) =
var(b {b bS (1
[1 bR|S )](1 b2S
bR|S ) 2 + 2nD1 bR|S (1
bR|S )(2p 1)
[ + (2p 1)nD1 ]4 S
+[(2p 1)2 + (1 p)p
bR ] n2D1 }

bR (1
bR )
var(b
R ) =
N

bS
where p nR1 (1 p) nR0 and bR|S (1
bS
bR|S ).

As the reader can see, the mixed randomized response/direct estimator of S is more ecient
(1p)p 10
bR )
than the pure randomized response estimator. The eciency gain is equal to (1 N (2p1)2 .

Although the presence of direct responders in the present study was unplanned, the fact that

unbiased and more ecient estimation is possible with the mixed randomized response/direct es-

timator points to its potential usefulness in future research. In order to harness these eciency

gains, future studies may consider the possibility of formally providing respondents with the option

of using the RR procedure or responding directly.11

6.1.2 Maximum Likelihood Estimation

Maximum likelihood estimators have a number of properties which have made them a popular

choice among political methodologists (cf King 1998), including consistency, asymptotic eciency

and asymptotic normality.12 Should the moment and maximum likelihood estimators of V be equal,

this last property implies that standard confidence intervals and tests of significance (e.g. t-tests)

1 0 Chauduri and Mukerjee (1985) were the first to note that RR survey schemes incorporating an option for direct

responses could result in eciency gains.

1 1 Providing an option for direct responses is only practicable, however, in contexts in which researchers are sure

that participants survey responses will not be viewed by anyone else in the institution to which they belong. If the
opposite is true, one could imagine scenarios in which participants are pressured to respond directly and in a way
that casts the institution in the best light. Pure RR obviates such scenarios.
1 2 The asymptotic normality property simply refers to the fact that under certain regularity conditions, maximum

likelihood estimates are asymptotically normally distributed. See Greene (2000) for details.

79
are applicable to the estimated parameter vector Vb . For this reason, it is important to examine if the

moment estimator is the maximum likelihood estimator as well. Moreover, as maximum likelihood

constitutes one of the two inferential approaches used to conduct individual-level analysis, it will

serve us well to establish now the foundation for what follows later.

The method of maximum likelihood selects as estimates the values of the parameters that max-

imize the likelihood of the observed sample, given the specification of an underlying probability

distribution. As our data are generated by a multinomial distribution, we can write the likelihood

of the sample as:

N
Y
L= P (yi,R1 = 1)yi,R1 P (yi,R0 = 1)yi,R0 P (yi,D1 = 1)yi,D1 P (yi,D0 = 1)yi,D0 (2.14)
i=1

= P (yi,R1 = 1)nR1 P (yi,R0 = 1)nR0 P (yi,D1 = 1)nD1 P (yi,D0 = 1)nD0

The log of the likelihood is:

log L = nR1 log[P (yi,R1 = 1)] + nR0 log[P (yi,R0 = 1)] + nD1 log[P (yi,D1 = 1)] + nD0 log[P (yi,D0 = 1)]

(2.15)

The maximum likelihood estimator (MLE) of the parameter vector V is the solution to:

max {log L}
S ,R|S ,R

The gradient vector constitutes the score equations for the MLE:

log L R|S (2p 1) R|S (2p 1) nD1 (1 R|S )

= nR1 [ ] nR0 [ ]+ nD0 =0 (2.16)
S P (yi,R1 = 1) P (yi,R0 = 1) S P (yi,D0 = 1)
log L (2p 1) (2p 1) nD1 nD0 S
= nR1 [ S ] nR0 [ S ] + =0
R|S P (yi,R1 = 1) P (yi,R0 = 1) 1 R|S P (yi,D0 = 1)
log L nR1 (1 p) nR0 p nD0
= + =0
R P (yi,R1 = 1) P (yi,R0 = 1) P (yi,D0 = 1)

80
 which, in principle, can be solved explicitly for the maximum likelihood estimator VbML . As

it turns out, the complicated structure of the above system of equations does not allow one to

derive a closed-form solution for the maximum likelihood estimator. However, we can evaluate the

hypothesis that the maximum likelihood estimator is the same as the moment estimator by plugging

in the parameter values of the moment estimator into the equations above. If the estimators are the

same, the score equations will equal zero when the values of the moment estimator are introduced.

Plugging the values of VbMM into the above system, we have:

log L(VbM ) p nR1 (1 p) nR0 + nD1 (2p 1)

= N [b bR|S (2p 1)] + N [
R|S (2p 1) ]N
S p nR1 (1 p) nR0 + nD1 (2p 1)

=0 (2.17)

log L(VbM ) 1
= N [b bS (2p 1)] +
S (2p 1) {[p nR1 (1 p) nR0 + nD1 (2p 1)]
R|S 2p 1
[p nR1 (1 p) nR0 + nD1 (2p 1)]}

=0

log L(VbM )
= N (1 p) + N p N
R
=0

thus establishing the equivalence of the moment and maximum likelihood estimators.

6.2 Individual-level Inference

The vast majority of studies using the randomized response technique have limited their attention to

estimating the proportion of individuals that bear a certain sensitive trait. While understanding the

prevalence of sensitive behaviors in a population may be of considerable interest, ultimately the goal

of social science research is to make inferences about the determinants of human behavior. In the

present study, I will be interested in estimating the eect of a set of individual characteristics on the

likelihood that bureaucrats illicitly use the resources of their institutions for the benefit of a political

party. How can I go about doing this with the survey data? This section presents two approaches

81
to the estimation of individual-level eects: parametric estimation via maximum likelihood and

non-parametric estimation through matching techniques.

6.2.1 The Parametric Approach

Estimation A number of social scientists have developed maximum likelihood methods to estimate

the eect of directly observed covariates on the likelihood that an individual exhibits a sensitive trait

in the context of randomized response data. Maddala (1983) was the first to do so, addressing the

issue in the context of the so-called forced-response model. Scheers and Dayton (1988) present a

framework for estimating covariate eects in the context of the Warner Model (my model of choice)

as well as the so-called unrelated question design (cf Greenberg et al 1969). van der Heijden and

van Gils (1996) develop a regression model for Kuks (1990) procedure, which they later applied in a

study of unemployment benefit fraud (2000). All of the aforementioned models involve a modification

of the basic logistic regression model to conform to the particular nature of the randomized response

data in question. My framework follows suit, although I consider various link functions and provide

a means to incorporate both the randomized response data as well as the direct response data.

We would like to model the probability of having the sensitive trait, S,i , as a function of an N k

covariate matrix xi . In the vernacular of generalized linear models, the covariate matrix is related

to S,i through an invertible link function (cf McCullagh and Nelder 1983). The link function maps

the expectations about S,i , generated by xi , from Rk (the space spanned by xi ) to the range of

S,i , in this case [0, 1]. With binary data, three suitable link functions are the logit, the probit and

the complementary log-log. The most common of the three is the logit, which allows us to write:

1
S,i|xi =
1 + exp(xi )

where is a parameter vector with k elements. Alternatively, we may wish to use the probit

link function, in which case we have:

S,i|xi = (xi )

82
 where (.) is the standard-normal cumulative distribution function. Finally, the complementary

log-log functional form can be used, which specifies that

S,i|xi = 1 exp[ exp(xi )]

Having chosen the link function, we can substitute S,i|xi for S in equations (2.8) and (2.9). In

so doing, we continue to assume that the yi,j are independently distributed (i.e. one respondents

response does not influence that of another). Moreover, we assume that conditional on having the

sensitive trait, an individuals choice of using the randomizing device is independent of xi .

The log-likelihood to be maximized is written:

N
X N
X
log L = yi,R1 log[P (yi,R1 = 1|xi )] + yi,R0 log[P (yi,R0 = 1|xi )] (2.18)
i=1 i=1
N
X N
X
+ yi,D1 log[P (yi,D1 = 1|xi )] + yi,D0 log[P (yi,D0 = 1|xi )]
i=1 i=1

b = (b
Our vector of parameter estimates, b 0 , has k+2 elements and comprises the
bR , )
R|S ,

solution to:

max{log L}

Maximization of the log-likelihood can be accomplished using an iterative optimization algo-

rithm.13

One of the advantages of the maximum likelihood approach is that it oers a straightforward

b by
approach to asymptotic variance estimation. We can estimate the asymptotic variance of

evaluating the second derivatives matrix of log L at the maximum likelihood estimates. Specifically,

we can estimate the asymptotic variance as the inverse of the observed Fisher information matrix:

b
2 log L()
b = [I()]
asy.var() b 1 = [ ]1
b b

1 3 This study used the algorithm developed by Broyden, Fletcher, Goldfarb and Shanno, "BFGS," available through

the R statistical computing language.

83
 bR|S ,
the diagonal elements of which constitute the asymptotic variance estimates of b
bR , and ,

respectively.

Diagnostics: Hypothesis Testing and Goodness of Fit Two important components of

parametric modelling consist of testing hypotheses about the value of specific model parameters and

evaluating how well the assumed model accounts for overall variation in observed outcomes. Both

of these tasks can be accomplished through the use of the likelihood ratio statistic.

Suppose we are interested in testing the null hypothesis that the true parameter vector belongs

to 0 , which is a subset of the permissible parameter space . We can test the null using the

likelihood ratio statistic. It is written as:

(xi ) = max L(|xi )/ max L(|xi )

s.t. 0 s.t

Note that the denominator of (xi ) is always greater than or equal to its numerator, thus guaran-

teeing that (xi )[0, 1]. Given certain regularity conditions imposed on the probability distribution,

the statistic 2 log (xi ) converges to a chi-squared distribution as sample size approaches infinity

(Casella and Berger 1990). Thus, under the likelihood ratio test (LRT), we reject the null if and

only if:

2 log (xi ) 2,

where is the degrees of freedom of the test (the total number of free parameters minus the

number of free parameters under the null) and is the size of the test, commonly referred to as the

confidence level. The p-value associated with a particular LRT is written

Pr(2, > 2 log (xi ))

which has the interpretation of being the smallest at which we would reject the null hypothesis.

The LRT can be used to considerable eect in testing hypotheses about the contributions of

covariates to the observed patterns in the data. Suppose, for instance, we wish to test the hypothesis

that S,i is independent of xi , i.e. our covariates are unrelated to the sensitive trait. In this case, we

84
could write the null hypothesis as H0 : = 0. We would simply perform the test above with (xi )

equal to:

=0 = max L(|xi )/ max L(|xi )

s.t. =0

and = k 1.14 In this instance, the magnitude of the likelihood ratio statistic gives us a

measure of how well our chosen xi predict the yi,j .

Alternatively, we may be interested in testing the hypothesis that the influence of two covariates

(x1 and x2 ) on the the probability of exhibiting the sensitive trait is the same, so that 1 = 2 . In

this case, we perform the LRT with (xi ) equal to:

1 = 2 = max L(|xi )/ max L(|xi )

s.t. 1 = 2

and = 1. More complicated hypothesis tests could proceed according to the same logic.

6.2.2 The Non-parametric Approach

The maximum likelihood approach to individual level inference is applicable in contexts in which

explanatory variables are continuous or binary. An alternative approach to estimating the eect of

a single predictor variable, appropriate when the variable is binary in form, consists of the use of

matching estimators.

Matching estimators have gained substantial popularity of late among economists and political

methodologists due to the relatively weak assumptions they require in order to correctly identify

marginal eects. The underlying idea of these techniques is to reproduce as closely as possible

for observational data the conditions of a controlled experiment with randomized assignment of

treatment and control groups. According to this approach, observations are grouped into matched

sets (the basis of artificial treatment and control groups) which are as similar as possible in terms of

potential confounding variables but dier in terms of whether or not they receive the treatment.

In order to better understand the matching approach, the introduction of some terminology is

1 4 For this test, we set all of the s to 0 except the intercept term.

85
in order. In using matching, one is often concerned with estimating the average treatment eect

(ATE), the expected eect of the binary predictor variable (the treatment) on a randomly drawn

individual from the population. Let us denote the binary treatment variable by w, where wi = 1

indicates that individual i is assigned to the treated group and wi = 0 indicates that individual i

is assigned to the control group. Similarly, let us introduce two random variables (r1i , r0i ), which

we refer to as the potential outcomes. These variables have the following interpretation: r1i is the

value of the response variable for individual i were he assigned to the treated group and r0i is the

value of the response variable for individual i were he assigned to the control group. The ATE is

defined as the expected value of r1i r0i in the population, or

AT E E[r1i ] E[r0i ] (2.19)

This expression can be expanded to:

AT E {E[r1i |wi = 1] E[r0i |wi = 1]} + (1 ){E[r1i |wi = 0] E[r0i |wi = 0]} (2.20)
| {z } | {z }

average treatment eect for treated (AT T ) average treatment eect for controls (AT C)

where Pr(wi = 1) is the unconditional probability that a randomly selected individual

belongs to the treated group. Thus, the ATE consists of a weighted average of two components:

the average treatment eect for the treated (AT T ), the average impact of the treatment for those

individuals who belong to the treated group, and the average treatment eect for the controls (AT C),

the average impact of the treatment for those individuals who belong to the control group.

The challenge in estimating the ATE is that in any given dataset, one observes r1i or r0i , but not

both. Specifically, we observe (r1i |wi = 1) and (r0i |wi = 0), the value of the response variable for a

particular member of the treated group given that he is exposed to the treatment and the value of

the response variable for a particular member of the control group given that he is not exposed to

the treatment, respectively. However, we do not observe (r1i |wi = 0) or (r0i |wi = 1), the value of the

86
response variable for a particular member of the treated group in the counterfactual scenario that

he is not exposed to the treatment and the value of the response variable for a particular member

of the control group in the counterfactual scenario that he is exposed to the treatment. Seen from

this perspective, estimating the ATE is essentially a missing data problem.

With randomized assignment of the treatment, overcoming this problem is a straightforward

matter. In this case, with a standard response variable (not RR data), the ATE could be esti-

mated simply by comparing the dierence in means between the treated and control groups. With

a response variable consisting of RR data, the ATE could be estimated simply by comparing the

dierence in estimated proportions between the treated and control groups, using an appropriate

formula such as (2.2). Depending on the context, both approaches are possible because random

assignment of the treatment implies that, in expectation, the distribution of observable and unob-

servable characteristics of the members of the treated and control groups will be the same.

In an observational study, where the treatment is not randomly assigned by the researcher, a

simple comparison of means (or estimated proportions) is an invalid measure of the ATE. The reason

is that there is likely to be self-selection into the treated group: individuals with a certain set of

characteristics may be more likely to accept the treatment, and these characteristics, in turn, might

be correlated with the value of the response variable. In this case, the dierence-in-means estimator

(or dierence-in-proportions estimator, as the case may be) provides an incorrect measure of the

causal eect of the treatment because the treatment assignment itself is correlated with the outcome.

Rosenbaum and Rubin (1983) have shown that causal inference can proceed with the introduc-

tion of a key conditional independence assumption, which they called the ignorability of treatment

assignment (ITA). According to ITA, conditional on a set of observed covariates zi , the treatment

assignment and the response variable are independent. In other words, once we condition on zi , the

assignment of the treatment is essentially random, as it would be in a randomized trial. Importantly,

the plausibility of the ITA assumption requires that zi include all covariates that determine treat-

ment assignment and are possibly related to the response variable. To the degree that covariates

related to both the treatment assignment and response variable are excluded from zi , then estimates

87
of the ATE will suer from hidden bias (cf Rosenbaum 2002).

With standard (non-RR) data, the ATE can be estimated as:

N N
[ 1 X 1 X
AT b[
E = b )[
(r1i r0i )|wi = 1, s(zi )] + (1 (r1i r0i )|wi = 0, s(zi )] (2.21)
N i1 N i1
N N N N
1 X 1 X 1 X 1 X
b [(
= r1i b )[(
r0i )|wi = 1, s(zi )] + (1 r1i r0i )|wi = 0, s(zi )]
N i1 N i1 N i1 N i1

b is the proportion of treated observations in the sample and s(zi ) is a function of the
where

confounding variables.

If the response variable consists of RR data, we can rewrite the above as:

[
AT 1S
b [(b
E= b0S )|w = 1, s(zi )] + (1 1S
b )[(b b0S )|w = 0, s(zi )] (2.22)

1S
where [(b b0S )|w, s(zi )] is the estimated dierence between the proportion of individuals in

the treated group that exhibit the sensitive trait and the proportion of individuals in the control

group that exhibit the sensitive trait, conditional on the treatment condition and s(zi ). The only

change that has taken place from equation (2.21) to equation (2.22) is that the dierence between

the average value of the response variable for the treated and control observations has been replaced

by the dierence in estimated proportions between the treated and control observations, where the

proportions are calculated using a suitable formula for RR data.

Section B of the appendix to this chapter presents two methods of conditioning on zi : exact

matching and nearest neighbor matching based on the propensity score. With suitably chosen zi ,

both strategies provide a means to estimate the marginal eect of a binary predictor variable with-

out imposing the functional form assumptions that characterize the maximum likelihood approach.

Therein, I develop the rationale for each technique and show how these methods can be used in the

context of the studys randomized response data.

88
6.3 Taking the Techniques for a Test Run

Having presented my statistical framework for conducting individual-level inference into the deter-

minants of corrupt behavior, here I take the techniques for a test drive. In this section, I conduct

some basic statistical analyses using both the maximum likelihood model developed in (6.2.1) as well

as the matching estimators presented in the appendix. In so doing, my aim is to give the reader a

taste of how these techniques can be utilized to make probability statements about the impact of the

determinants of dierent types of corruption, and to demonstrate the plausibility of the relationships

uncovered by the survey data.

Let us begin with the maximum likelihood approach. I examined the impact of a small set of

covariates on the likelihood of three outcomes: 1) that a bureaucrat had used the resources of his

institution for the benefit of a political party; 2) that a bureaucrat had been forced to use his own

resources (wage, time, etc.) for the benefit of a political party; and, 3) that a bureaucrat had used

the resources of his institution for his own benefit or that of his family. The aforementioned out-

comes were examined through the use of three distinct randomized response questions contained in

the survey.15 The covariates I considered included an indicator variable equal to 1 if the respondent

indicated that he belonged to a political party (0 otherwise), an indicator variable equal to 1 if the

respondent had a political career (0 otherwise), a four-level ordinal variable capturing the respon-

dents perception of the strength of formal control mechanisms within his institution, a three-value

categorical variable capturing the respondents perception of the ability of the legal process to pun-

ish acts of corruption, an indicator variable equal to 1 if the respondent indicated that his wage

was sucient to cover the costs of basic necessities (0 otherwise) and a three-level ordinal variable

denoting the respondents level of formal education.16 The exact wording of the questions on formal

1 5 The format of these questions was as in figure 3 with the two statements for each question being as follows:
(institutional resources used for a political party) A: I have never used, not even once, the resources of my institution
for the benefit of a political party, B: I have used, at least once, the resources of my institution for the benefit of a
political party; (personal resources used for a political party) A: I have never been forced during my job, not even
once, to use my own resources (wage, time, etc.) for the benefit of a political party, B: I have been forced during
my job, at least once, to use my own resources (wage, time, etc.) for the benefit of a political party; (institutional
resources used for own benefit or that of family) A: I have never used, not even once, the resources of my institution
for my own benefit or that of my family, B: I have used, at least once, the resources of my institution for my own
benefit or that of my family.
1 6 For the Bolivian and Chilean data, an individual was classified as having a political career if he responded that

he was currently a member of a party and was either actively involved in the aairs of the party or had political
aspirations for the future. In the Brazilian case, an additional set of questions allowed me to adopt a slightly more

89
control, the ecacy of the legal process and the suciency of wages is presented in table 5.3. The

link function used for the estimation was the logit (as is true for all such estimations conducted in

this monograph).

I approached the estimation with a set of well defined expectations about the impact of the

covariates. Membership in a political party or, even more strongly, having a political career, should

be positively associated with the use of institutional resources for a party and with having been

forced to use personal resources for a party. This is because individuals in these categories are likely

to have political aspirations which may be advanced in exchange for providing good service to the

party.17 Individuals who indicate that the formal control mechanisms in their institution do a good

job of investigating instances of illicit resource use should be less likely to have used resources for a

party or for their own benefit, as the subjective risk of detection associated with doing so would be

greater than that for individuals who indicate that such control mechanisms do a poor job. Similarly,

individuals who indicate that the legal process is likely to produce a sanction for those who engage

in corrupt behavior should be less likely to have used resources for partisan or personal benefit, since

the potential cost of doing so would be greater than that perceived by those who indicate otherwise.

The ability of ones wage to cover basic costs should also have a dampening eect on illicit resource

use. The higher a civil servants public sector wage relative to the private sector wage he could get

if he left the public sector, the less likely he is to engage in corrupt acts, since the long term income

loss associated with being booted from the public service may be considerable. My expectations for

the education variable were weaker than those for the other variables included in the estimation,

but I expected that individuals with higher levels of education would be less likely to have used

institutional resources for partisan advantage, and also less likely to have been forced to use their

expansive definition. For this country, an individual was classified as having a political career if the above condition
held or, additionally, if he was currently a party member and had run for elected oce in the past or was a party
member and expressed interest in doing so in the future.
The education variable was coded into three categories: low, medium and high. In all countries, an individual
was categorized as having low formal education if he indicated that his maximum educational attainment was a
secondary school diploma or below. In Bolivia and Chile, an individual was categorized as having medium formal
education if he had a bachiller or licenciatura. In Brazil, medium formal education corresponded to a course of
studies completed at the nvel superior. In Bolivia and Chile, an individual was categorized as having high formal
education if he had obtained a masters, doctorate or a professional degree (e.g., medicine, engineering). In Brazil,
high formal education corresponded to a masters, doctorate or other post-graduate degrees.
1 7 The dynamics and institutional roots of such exchanges is the primary focus of chapter 4 and 5 of the present

monograph.

90
own resources for political purposes. This is because highly educated individuals are more likely to

have been selected for their posts on meritocratic grounds, and thus less likely to feel they need to

repay a debt to a party or political patron.

What did I find? All of the covariates were signed according to expectations and many were

statistically significant at conventional levels.18 Table 2.4 presents coecient estimates for six dif-

ferent models using the three response variables described above. Belonging to a political party, and

even more strongly, having a political career were clearly very strong predictors of partisan resource

use. Not surprisingly, party membership was also an important determinant of whether or not a

bureaucrat had been forced to use personal resources for a political party. Faith in the strength of

formal control mechanisms was negatively related to the partisan and personal use of institutional

resources, although more strongly so for the former than the latter. Faith in the ecacy of the legal

process was also negatively related to the partisan use of institutional resources, as was having a

high level of formal education. Moreover, individuals who felt that their wage covered basic costs

were less likely to have used institutional resources for partisan or personal benefit.

In and of themselves, tables of coecient estimates and their associated standard errors only tell

us the direction (sign) of the relationship between a predictor variable and the response variable and

whether or not said relationship is statistically significant at conventional levels. In this regard, it

bears repeating the well worn phrase that statistical significance is not scientific significancei.e.

simply because a variable is statistically significant does not mean that it is substantively important.

In addition to uncovering the sign of coecient estimates and measures of uncertainty, we would

also like to measure the magnitude of the impact of variables found to be statistically significant. In

the maximum likelihood approach, we can do this through simulation.

1 8 In all tables in this monograph, a single star next to a coecient estimate denotes that the null hypothesis that

the estimate is zero can be rejected at the 90% confidence level, two stars means the null can be rejected at the 95%
level and three stars means the null can be rejected at the 99%.

91
 Table 2.4: Impact of covariates on three types of corruption (maximum likelihood estimation)

Figure 2.6 presents a simulation analysis showing the impact of having a political career on the

likelihood that a bureaucrat had used the resources of his institution for the benefit of a political

party (based on model 2.2). In the analysis, all covariates other than the political career indicator

variable are held at their mean values in the sample. Using the vector of coecient estimates and

its covariance matrix, the distribution of the predicted probability of having used resources for a

party is then calculated for the mean bureaucrat in two scenarios: in the instance that he does

not have a political career and the instance that he does have a political career. Figure 2.6 graphs

the two distributions. The expected value of the distribution in the "no political career" scenario

is .16, whereas the expected value of the distribution in the "political career" scenario is .38. This

92
means that the expected impact of a political career on the probability of having engaged in partisan

resource use (for the mean bureaucrat in the sample) is .38-.16 = .22, a very considerable amount.

Figure 2.6: Impact of political career on likelihood of partisan resource use (maximum likelihood estimation)

The matching techniques outlined in the appendix present a dierent means of doing essentially

the same thing: calculating the eect of a binary predictor variable on the response variable for

a randomly selected member of the samplethe average treatment eect (ATE). Using the same

covariates as in the maximum likelihood estimation, I calculated the impact of having a political

career on the likelihood of partisan resource use using the exact matching and nearest neighbor

propensity score matching algorithms. The results are presented in table 2.5. The increase in

the probability of using institutional resources for partisan ends due to having a political career is

estimated by exact and nearest neighbor matching to be .14 and .17, respectively. These figures are

smaller than the estimate calculated using maximum likelihood, but they are nonetheless still quite

93
considerable.

Table 2.5: Impact of political career on partisan resource use (matching estimators)

7 Conclusion

This chapter has outlined the monographs approach to the empirical analysis of corruption. My

principle goal here has been to develop a novel methodology for studying the determinants of corrup-

tion and other types of illicit or socially stigmatizing behaviors. In developing the approach, I have

woven together intellectual advances made by actors at international lending institutions (such as

the use of diagnostic surveys of public employees, pioneered by the World Bank) and important con-

tributions in statistics (such as RR survey design for studying sensitive traits, pioneered by Stanley

Warner, and matching techniques for causal inference in observational studies, pioneered by Paul

Rosenbaum and Donald Rubin).

The result of the endeavor is not principally to produce a particular measure or set of measures

of corruption, nor is it to produce a set of statistical estimators for studying corruption. Rather, the

contribution of the chapter is to present a new framework for organizing survey research and con-

ducting statistical analysis so that researchers interested in the determinants of sensitive behaviors

can take a first step towards overcoming the "fundamental problem of inference" outlined at the

94
beginning of the chapter. Although my aim in developing the approach was rather narrow, namely,

to study the determinants of political corruption within the state in Bolivia, Brazil and Chile, the

approach is flexible enough that I anticipate it may travel well to dierent substantive problems and

various regions of the globe. Only time will tell.

8 Appendix

8.1 Section A: Covariance Estimation of Vb

Here I derive the asymptotic variance estimators presented in equation (2.13). Let us begin the

derivation by noting that each element of the parameter vector Vb is a function of one or more

element in the vector n = (nR1 , nR0 , nD1 )0 . According to the multivariate delta theorem, N (Vb

V ) Normal(0, n 0n ), where =asym.var(n) and

b
S b
S b
S
nR1 nR0 nD1

b
n = R|S
nR1
b
R|S R|S
b
(A1)
nR0 nD1


b
R b
R b
R
nR1 nR0 nD1

Now note that as n is distributed according to a multinomial distribution, we have:

P (yi,R1 = 1)[1 P (yi,R1 = 1)] P (yi,R1 = 1)P (yi,R0 = 1) P (yi,R1 = 1)P (yi,D1 = 1)


= N
P (yi,R1 = 1)P (yi,R0 = 1) P (yi,R0 = 1)[1 P (yi,R0 = 1)] P (yi,R0 = 1)P (yi,D1 = 1)



P (yi,R1 = 1)P (yi,D1 = 1) P (yi,R0 = 1)P (yi,D1 = 1) P (yi,D1 = 1)[1 P (yi,D1 = 1)]
(A2)

Plugging in values into the above matrices using (9) and (11) and extracting the diagonal elements

of n 0n , we produce the asymptotic variance estimators in (2.13).

95
8.2 Section B: Two Matching Algorithms for RR Data

8.2.1 Exact Matching

The idea behind exact pair matching is to provide an estimate of the influence of a binary predictor by

breaking the data up into exactly matched pairs that are identical in terms of the values of observed

covariates but dier in the value of the binary predictor. Such a strategy is generally only feasible

when the zi are categorical and consist of a small number of covariates. Thus, exact matching is

particularly well suited to the analysis of survey data in contexts where theory emphasizes the need

to control for a small set of potentially confounding variables. With regard to the function s(zi ),

exact matching is equivalent to conditioning on zi directly, so we can write s(zi ) = zi .

I consider here an exact pair matching estimator with three key characteristics: 1) information

from all observations in the treated and control groups with one or more exact match in the opposite

group is utilized in estimating the ATE; 2) the estimator has a sensible method of dealing with ties,

i.e. the existence of multiple exact matches for any given member of the treated or control groups;

3) the same individual can be used multiple times as a match for members of the opposite treat-

ment group. The approach I present here is essentially the same as the simple matching estimator

presented in Abadie et al (2004), with the two distinctions being that I limit my attention to exact

matches only and I tailor the approach to the studys RR survey data.

Let zi refer to the set of observed covariates for individual i. We define ||u z|| as the distance

between the vectors z and u, where u represents the covariate values for a potential match for obser-

vation i. With exact matching, we are interested in evaluating how identically matched individuals

dier as a function of the application of the treatment. Thus, from the treated and control groups,

we retain only the M t and M c sets, defined as:

X
M t = {i : wi = 1, 1[||zl zi || = 0] > 0} (A3)
l:wl =0
X
M c = {i : wi = 0, 1[||zl zi || = 0] > 0}
l:wl =1

96
 where 1[.] is the indicator function, equal to 1 if the expression in brackets is true, 0 otherwise.

In words, the M t set contains all individuals in the treated group with one or more exact matches

in the control group. Similarly, the M c set contains all individuals in the control group with one or

more exact matches in the treated group.

In the exact matching approach, only observations with one or more exact match in the opposite

treatment group are treated as comparable. Thus, we utilize the observed value of the response

obs , as our estimate of (r |w = 1, z ).

variable for a given member of the M t set, which we denote by r1i 1i i i

Similarly, we utilize the observed value of the response variable for a given member of the M c set,

obs , as our estimate of (r |w = 0, z ). However, in order to estimate the ATE,

which we denote by r0i 0i i i

we must also provide estimates of the counterfactual outcomes that correspond to each of these

individuals, i.e. the outcomes that would have obtained if these individuals had been placed in the

opposite treatment group. How can we estimate these counterfactual outcomes: (r0i |wi = 1, zi ) and

(r1i |wi = 0, zi )?

For each i with one or more exact match, our approach is to estimate the value of the unobserved

outcome by taking the average of the response variable for all of is exact matches in the opposite

treatment group. Let us consider how this is accomplished with the RR survey data.

Given the nature of our RR survey data, note that ri can be written as a 1 4 column vector

with a single element in the vector equal to 1 and all other elements equal to 0. The entry in each

column of the vector denotes individual is response to the sensitive question under consideration,

where an entry of 1 in the first column denotes a response of R1, an entry of 1 in the second column

denotes a response of R2, an entry of 1 in the third column denotes a response of D1 and an entry

of 1 in the last column denotes a response of D0. Let IMc (i) denote the set of indices for individuals

in the M c set that are exact matches for individual i in the M t set. Similarly, let IMt (i) denote the

set of indices for individuals in the M t set that are exact matches for individual i in the M c set.

For each individual i in the M t set, the response values of his |IMc (i)| exact matches in the M c

set can be written as a |IM c (i)| 4 response matrix r[IMc (i)]. Each row of this matrix constitutes

the response value of a distinct individual, with the entries in each column following the protocol

97
just described. Similarly, for each individual i in the M c set, let the response values of his |IMt (i)|

exact matches in the M t set be written as a |IMt (i)| 4 response matrix r[IMt (i)]. Finally, let us

introduce the BMc (i) and BMt (i) matrices, defined as follows:

1
BMc (i) r[IM c (i)] (A4)
|IMc (i)|
1
BMt (i) r[IMt (i)]
|IMt (i)|

For each individual i in the M t set, the unobserved potential outcome that would have obtained

had i not received the treatment is estimated as:

!
|IMc
P(i)| |IMc
P(i)| |IMc
P(i)| |IMc
P(i)|
cf
r0i = BMc (i) [l, 1] BMc (i) [l, 2] BMc (i) [l, 3] BMc (i) [l, 4] (A5)
l=1 l=1 l=1 l=1

where, for any matrix A, A[l, s] denotes the element located at the lth row and sth column of

matrix A.

For each individual i in the M c set, the unobserved potential outcome that would have obtained

had i received the treatment is estimated as:

!
|IMt
P(i)| |IMt
P(i)| |IMt
P(i)| |IMt
P(i)|
cf
r1i = BMt (i) [l, 1] BMt (i) [l, 2] BMt (i) [l, 3] BMt (i) [l, 4] (A6)
l=1 l=1 l=1 l=1

Given this averaging procedure, we can characterize all four potential outcomes. They are esti-

mated as:

obs for i M t
(b
r1i |wi = 1, zi ) = r1i
rb1i = (A6)


(b cf
r1i |wi = 0, zi ) = r1i for i M c



(b cf
r0i |wi = 1, zi ) = r0i for i M t
rb0i =

obs for i M c
(b
r0i |wi = 0, zi ) = r0i

98
 With expressions for the potential outcomes in hand, we can now easily estimate the four compo-

nents of (2.22). Let robs

1 denote the |M t| 4 response matrix for all the members of the M t set and

rcf
0 denote the |M t| 4 matrix of estimated counterfactual outcomes for each of these individuals.

Similarly, let robs

0 denote the |M c| 4 response matrix for all the members of the M c set and rcf
1

denote the |M c|4 matrix of estimated counterfactual outcomes for each of these individuals. Using

(2.11), our exact matching estimator estimates the components of the ATE as:

|Mt|
P |M
Pt| |M
Pt|
robs
1 [l, 1] robs
1 [l, 2] robs
1 [l, 3]
p l=1 1p l=1 l=1
1S |wi
[b = 1, zi ] = + (A7)
2p 1 |M t| 2p 1 |M t| |M t|
|Mt|
P |Mt|
P |Mt|
P
rcf
0 [l, 1] rcf
0 [l, 2] rcf
0 [l, 3]
p l=1 1p l=1 l=1
0S |wi = 1, zi ] =
[b +
2p 1 |M t| 2p 1 |M t| |M t|
|Mc|
P |Mc|
P |Mc|
P
rcf
1 [l, 1] rcf
1 [l, 2] rcf
1 [l, 3]
p l=1 1p l=1 l=1
1S |wi = 0, zi ] =
[b +
2p 1 |M c| 2p 1 |M c| |M c|
|Mc|
P |Mc|
P |Mc
P
robs
0 [l, 1] robs
0 [l, 2] robs
0 [l, 3]
p l=1 1p l=1 l=1
0S |wi = 0, zi ] =
[b +
2p 1 |M c| 2p 1 |M c| |M c|

Plugging the above into the expression in equation (2.22), we calculate the ATE.

8.2.2 Nearest Neighbor Matching

If the covariates that researchers would like to control for are large in number or continuous in nature,

exact matching techniques like the one presented above become infeasible. For this reason, many

researchers have turned to matching techniques that produce matched pairs which are as close as

possible in terms of the observed covariates but not perfectly matched. One method of doing so

consists of propensity score matching.

Propensity score matching entails the construction of matched pairs based on a one-dimensional

specification of s(zi ) called the propensity score. It has been shown that creating matched pairs

with the same probability of receiving the treatment (i.e. having the binary predictor variable equal

1), is equivalent to comparing cases with similar values on all of the covariates (Rosenbaum and

Rubin 1983). So long as the probability of receiving the treatment is bounded away from 0 and 1

99
for all individuals in the sample (the so-called common support condition), then, conditional on the

probability of treatment, the treatment assignment and the response variable are independent. As

opposed to conditioning directly on zi , which may have high dimensionality, this result implies that

one need only condition on the probability of treatment, given zi , in order to correctly estimate the

ATE.

Building on this finding, the propensity score matching approach consists of two steps. The first

step consists of the estimation of sb(zi ) = Pr(wi = 1|zi ) for each individual in the sample. This

is the propensity score: the propensity of individual i towards treatment given the covariates zi .

Traditionally, sb(zi ) is estimated using the predicted values from a binary response regression model

such as logit or probit where the treatment is regressed onto the confounding variables. The second

step consists of the construction of matched pairs based on sb(zi ) and the consequent estimation of

the ATE. Let us consider how estimation can proceed in the context of our RR survey data.

I adopt a particular matching algorithm, nearest neighbor matching, to produce the matched

pairs. The nearest neighbor matching estimator considered here has three characteristics: 1) infor-

mation from all of the observations in the sample is used in estimating the ATE; 2) as above, the

estimator has a sensible method of dealing with ties, i.e. the existence of multiple matches for any

given member of the treated or control groups; 3) the same individual can be used multiple times

as a match for members of the opposite treatment group.

In the nearest neighbor matching approach, all observations satisfying the common support

condition are treated as potentially comparable. Thus, we can utilize the observed value of the

obs , as our estimate of

response variable for a given member of the treated group, denoted by r1i

[r1i |wi = 1, s(zi )]. Similarly, we can utilize the observed value of the response variable for a given

obs , as our estimate of [r |w = 0, s(z )]. The counter-

member of the control group, denoted by r0i 0i i i

factual outcomes are estimated in a manner similar in spirit to the method presented for the exact

matching estimator above.

Let IM (i) denote the set of indices for the |IM(i) | individuals in the treatment group opposite to

that to which i belongs, who: 1) exhibit the propensity score closest to sb(zi ) among all the members

100
of the treatment group to which they belong; 2) share the same propensity score. In colloquial

terms, the members of IM (i) are is nearest neighbors. If the propensity score of is closest match

is not tied with that of any other member of the same treatment group, then |IM(i) | = 1 (i has a

single nearest neighbor in the opposite treatment group). With continuous confounding variables,

having a single nearest neighbor will generally be the rule.

Denote the response values of is |IM(i) | nearest neighbors as the |IM(i) | 4 matrix r[IM(i) ]. The

elements of the matrix are defined as in the discussion of the exact matching estimator above. The

BM(i) matrix, in turn, is defined as:

1
BM(i) = r[IM(i) ] (A8)
|IM(i) |

The nearest neighbor matching estimator estimates the four potential outcomes as:

obs
r1i |wi = 1, sb(zi )] =
[b r1i
!
rb1i = |IM (i)| |IM (i)| |IM (i)| |IM (i)|

cf = P P P P
r1i |wi = 0, sb(zi )] = r1i
[b BM (i) [l, 1] BM (i) [l, 2] BM (i) [l, 3] BM (i) [l, 4]
l=1 l=1 l=1 l=1
!

|IM
P(i)| |IM
P(i)| |IM
P(i)| |IM
P(i)|
cf
r0i |wi = 1, sb(zi )] = r0i =
[b BM (i) [l, 1] BM (i) [l, 2] BM (i) [l, 3] BM (i) [l, 4]
rb0i = l=1 l=1 l=1 l=1



[b
r0i |wi = 0, sb(zi )] = obs
r0i

Using these potential outcomes, redefine robs

1 as the NT 4 response matrix for all the

members of the treated group and rcf

0 as the NT 4 matrix of estimated counterfactual outcomes

for each of these individuals, where NT is the number of individuals in the treated group. Similarly,

redefine robs
0 as the NC 4 response matrix for all the members of the control group and rcf
1 as

the NC 4 matrix of estimated counterfactual outcomes for each of these individuals, where NC is

the number of individuals in the control group. Given this reformulation of notation, the nearest

neighbor matching estimator estimates the components of the ATE as in (A7), with zi replaced by

sb(zi ), |M t| replaced by NT and |M c| replaced by NC .

101
8.2.3 Variance Estimation

The two matching estimators presented above do not lend themselves to a simple expression for the

estimator of the variance of the ATE. For this reason, the monograph follows common practice and

estimates the variance of the matching estimators through bootstrapping. The interested reader is

referred to Efron and Tibshirani (1993) for the details on bootstrap variance estimation.

102
 Chapter 3
Characterizing the Bureaucratic Marketplace in
Bolivia, Brazil and Chile: A First Look at the Survey
Data

103
3.1: Introduction

In the chapters which follow, I will use formal political analysis to generate expectations

about the relationship between aspects of a bureaucrats institutional environment and his propensity

to engage in certain acts of political corruption. These expectations will be evaluated using the

statistical techniques presented in the preceding chapter. However, before diving into the process of

formal hypothesis construction and evaluation, it is worthwhile to take a step back and ask what the

survey data itself can tell us about the basic features of the institutional environment confronted by

public employees in the country cases. In this chapter, I provide a first look at the survey data,

employing the tools of exploratory data analysis to examine key points of variation across the cases,

as well as key points of variation across the public institutions included in each country case.

My agenda in the chapter is simple. I seek to convey an elementary understanding of the

core features of the bureaucratic marketplace in the cases. The focus here is primarily descriptive,

and the emphasis lies on accurately depicting the principal dimensions of cross-national and cross-

institutional variation. An examination of the causes of variation in the individual choice to engage in

illicit resource use awaits the reader in Chapters 4 and 5.

The fact that my approach to the data in this chapter is primarily exploratory does not mean

that issues of broad interest to the discipline of political science are not at stake. There are two major

disciplinary contributions made herein. Firstly, the chapter addresses debates about the utility of

using cross-national measures of corruption and bureaucratic effectiveness as yardsticks by which to

evaluate the performance of entire public sectors. In this regard, the analysis is guided by the

concerns of authors such as Aberbach and Rockman (1987), who were acutely aware of the potential

pitfalls of characterizing public sectors by their central tendencies, especially contexts in which

institutional diversity is considerable. These concerns have been reflected in much empirical work by

scholars concerned with the causes of variation in state capacity and the genesis of the

developmental state, many of whom noted patterns of intense politicization and clientelism in

104
certain public institutions in their regions of expertise but distinctive patterns of Weberian practice in

others (cf Evans 1995, Geddes 1994, Kang 2002, Kohli 2004, Schneider 1999).1

The mantra of the chapter in this respect is simple: we ought to take institutional diversity

seriously. Even in instances in which cross-national differences in state capacity are considerable,

these differences may be swamped by internal variation in capacity. For many readers, this point may

not seem like a contentious one. Be that as it may, an appreciation of this point nevertheless entails

important consequences for how we go about organizing research on the performance of the state.

In particular, significant intrastate variation in capacity might recommend a departure from the

nation-state-as-unit-of-analysis paradigm which has come to dominate the quantitative study of

bureaucratic performance in comparative politics.

Instead of posing the guiding research question as Why does the public administration of

country X underperform relative to that of country Y?, researchers in the future may find it equally

illuminating to pose the question as Why does institution A in country X underperform relative to

institution B in country X? or Why does institution C in country X underperform relative to its

functional equivalent C in country Y? The general line of attack would be to take Snyders (2001)

exhortation to scale down our units of analysis seriously and organize research accordingly. Here I

propose an even greater scaling down than the one advocated by that author, as my emphasis in this

chapter is on cross-institutional (functional) variation, not subnational (regional) variation.

With the academic enterprise focused on institutional variation, the search for appropriate

models of bureaucratic organization is one that is as likely to bear fruit if oriented internally as it is if

oriented externally. That is to say, academics and policymakers may find it easier to identify and

reproduce successful patterns of bureaucratic organization that they encounter in their home country

1 Perhaps the starkest portrayal of intrastate variation in bureaucratic quality emerges from the contending

scholarly accounts of the public bureaucracy in post-WWII Japan. By far the most influential of these accounts
is Chalmers Johnsons 1982 MITI and the Japanese Miracle, which ties Japans postwar economic miracle to
interventionist measures implemented by a core of highly trained, politically autonomous bureaucrats in the
Ministry of International Trade and Industry. Later work on the Japanese bureaucracy during this period, most
notably Brian Woodalls 1996 Japan under Construction (but see also Ramsayer and Rosenbluth 1993), paints a
picture of the Japanese bureaucracy completely at odds with the one depicted by Johnson. Above and beyond
any differences of interpretation, the drastically different accounts can be chalked up to the particular
institutions chosen for analysis: MITI (Johnson) versus the Ministry of Public Works (Woodall).

105
than those they encounter abroad and which are emergent from very different legacies of state-

society interaction. A focus on cross-institutional variation makes internal searches for bureaucratic

models potentially fruitful because we can in effect hold constant country-specific fixed effects

culture, legacies of colonialism and distinctive patterns of social organizationwhich often bedevil

even the most careful cross-national analyses.

The second disciplinary contribution of the chapter is that it addresses debates about the

merit of deepening Weberian reforms within the public bureaucracies of developing countries. With

the widespread dissemination of the core postulates of the New Public Management (NPM)

philosophy, many developing country governments have seen fit to restructure the administration of

public institutions in ways that depart sharply from the Weberian model of legal-rational bureaucracy.

In practice this has meant undergoing a process of debureaucratization, including but not limited to

the introduction of measures to provide greater flexibility in personnel management and public

procurement.

Proponents of such measures have argued that the time when those great enemies of legal-

rational administrationclientelism and political patronageposed the primary threats to the

smooth functioning of state institutions is now past, and that there are great efficiency gains to be

had from instilling a more entrepreneurial spirit within the public sector (Bresser-Pereira 1998,

Osborne and Gaebler 1992). Opponents of such measures have argued that by chipping away at

some of the cornerstones of Weberian bureaucracythe prudential separation of the political and

bureaucratic spheres of the polity, relatively strong guarantees of employment tenure and ascension

through expertise and seniorityNPM reforms may have the unsalutary consequence of eroding

technical competence and reopening key areas of the state to the spoils system (Suleiman 2003). This

would be a particularly unsavory outcome in the many (all?) developing countries in which the spoils

system had never really been dismantled in the first place (Schick 1998). In such contexts, argue such

authors, it may be greater doses of Weberian bureaucracy, not less, that are associated with high

institutional efficiency.

106
 The chapter engages with this debate by examining the relationship between the placement

of public institutions in the cases on three crucial axes of Weberian bureaucracythe politicization

of the civil service (or lack thereof), the institutionalization of bureaucratic career paths, and the

strength of mechanisms to prevent corruptionand two measures of performance: employee

perceptions of institutional efficiency and levels of political corruption. The broad conclusion of the

empirical analysis is that there are still great gains to be made from reinforcing the basic principles of

Weberian administration within the public bureaucracies of many developing countries.

3.2: Taking Institutional Diversity Seriously

The current state of the bureaucratic marketplace in Latin America is one of searing

disparities in capacity, resources and opportunities for professional advancement. To place such

disparities into relief, it might be helpful to imagine the lot of two fictional but not unrepresentative

characters: Reynaldo and Oscar.

Every day of the week, Reynaldo arrives to work in an aging government building. He enters

through a side entrance, as the main portal and faade remain charred black and impassible from a

firebombing received during street protests a year ago. He shares a computer with four other

colleagues, and he shares 10% of wage with the party to which he owes his current employment.

Apprehensive about the future, rumors of a cabinet reorganization and the impending departure of

his department head have led Reynaldo to pursue the possibility of employment in a number of other

public institutions. Such a move would be a shame, he thinks, as he only finished the requisite

training for his post a few months ago.

Oscars circumstances are very different from Reynaldos. Every day of the week, Oscar

arrives to work in a towering, state of the art facility. The buildings infrastructure for information

technology is top of the line, and his personal workspace is wired with a flat screen monitor, high

speed internet access and a fax machine. On Thursday afternoons, Oscar and a number of his

107
colleagues receive several hours of English language instruction, on-site and paid for by his

institution. Just last month, he concluded a two month training session designed to familiarize him

with recent developments in his field. Given his hard work and dedication to honing expertise, Oscar

recently received word to he has been selected to receive a promotion. Next week, he will be

celebrating his tenth year with his institution.

The lexicon of comparative politics is rich with labels that might be used to describe the two

public administrations that Reynaldo and Oscar find themselves working in: the neo-patrimonial

state versus the Weberian state, the spoils system versus the mandarin bureaucracy, the

desarrollista state versus the developmental state, and so on and so forth. The utility of these

taxonomies aside, the reality of the Latin American experience is this: the Reynaldos and Oscars of

the world are just as likely to be found working in different institutions in the same public

administration, in the same capital city and separated by a few small city blocks, as they are to be

found working in distinct public administrations in different capitals, separated by mountain ranges,

deserts, jungles or oceans.

That this is so should have profound implications for scholarly inquiry into the ability of

states to provide high quality public goods and generate greater prosperity for their citizens. With the

collection and diffusion of large cross-national datasets such as the Political Risk Services Group

International Country Risk Guide, the World Bank Institutes Governance Matters dataset and its Database of

Political Institutions, the World Economic Forums Global Competitiveness Survey and Transparency

Internationals Corruption Perceptions Index, scholars interested in the relationship between political

institutions, governance and development have had an unprecedented opportunity to submit much

theorizing to rigorous statistical tests. Many have put the aforementioned datasets to good use, some

analyzing the effect of legal traditions, electoral rules, federalism, constitution type and other

108
institutions on governance, while other scholars, in turn, have analyzed the effect of governance on

economic growth, patterns of government expenditure and outcomes in education and health.2

As a general development, the creation and dissemination of such data sets has been a great

boon for the discipline of comparative politics, as these datasets have opened up to systematic

empirical inquiry a range of issues long thought impossible to examine scientifically. Nevertheless,

one could argue that the greater availability of such cross-national data has had an unfortunate side

effect, that being the emergence of a perhaps exaggerated scholarly emphasis on cross-national

variation in governance. Regional or cross-institutional variationwhich may be equally important

has received less attention.3 Although an earlier generation of work on state capacity recognized the

considerable heterogeneity within the state (cf Putnam 1993, Tendler 1997), the current generation of

scholarship on the subject rarely does.

So what? Why should institutional diversity matter for how we conduct research on the

performance of the state? One reason it matters is because institutional diversity impels us to

consider potentially important aspects of variation among states that may go unnoticed in studies that

rely exclusively on cross-national measures of state capacity. It may be that variation in the quality of

public goods and services in certain categories of state activity is more important for explaining the

outcomes of interest than variation in others. All else equal, if we were interested in explaining the

incidence of banking crises, for instance, we would do better to collect indicators of the quality of

banking regulation than indicators of the quality of policing or public health services. In this example,

it would seem self-evident that neither of the two latter indicators would serve as an adequate proxy

for the former.

2 A non-exhaustive list of references in this regard includes: Gupta et al (2001), Knack and Keefer (1995),
Kunicova and Ackerman (2002), La Porta et al (1999), Montinola and Jackman (2002), Mauro (1995, 1998),
Persson and Tabellini (2003) and Treisman (2000).
3 Studies of regional variation in governance in developing countries do appear to be picking up steam,

however. See Murillo and Calvo (2004), Sinha (2003) and Wantchekon (2003). Moreover, it should be said that
regional variation has always been key to the study of corruption in the United States (cf Alt and Lassen 2003,
Goel and Nelson 1998, Nice 1983).

109
 But just as having good information about policing and public health does not translate into

good information about banking regulation, neither should, ipso facto, information about the overall

(national) level of corruption or government effectiveness. To bring this point home with a case

included the present study, it was during the Pinochet dictatorship in Chile, within the context of a

public administration considered by many to be among the most austere and professional in Latin

America, that one of the most costly banking crises of recent decades came to pass. Overall levels of

corruption or the lack thereof certainly do not explain this outcome; the inability of specific

regulatory institutions to prevent connected lending at least partially do. The moral of the story is

that once we allow for the possibility that state capacity is distributed in different ways within polities,

cross-national summary measures may be found wanting. Clearly, there can be no guarantee that

levels of state capacity measured at the national level will reflect levels of capacity in the particular

institutions most important for the outcomes we wish to explain.

Another reason institutional diversity matters is because in its extreme form it renders

untenable classification schemes applied to entire public sectors. Steeped in the Weberian tradition,

many students of state capacity have found it helpful to develop ideal types of public bureaucracies.

Proximity to such ideal types, in turn, is often used to explain certain types of political or economic

outcomes.4 The utility of such classifications schemes, however, rests on their ability to capture in a

more or less accurate way the general properties of the bureaucratic systems to which they are being

applied. High levels of institutional diversity make it likely that such general taxonomic schemes will

misclassify significant segments of the public sector.

In order to visualize why this is the case, it is helpful to piggyback on the strategy of

Aberbach and Rockman (1987) and think of these matters in terms of probability distributions. For

the sake of argument, let us classify public bureaucracies on the basis of a one-dimensional scale. We

can think of this scale as representing degrees of bureaucratic professionalism, with the Weberian

ideal type bureaucracy located at one end of the scale and the patrimonial ideal type bureaucracy

4For very careful work along these lines, see Evans (1995), Evans and Rauch (1999), Kohli (2004), Rauch and
Evans (2000) and Schneider (1999).

110
located at the other. Now assume that all countries contain large numbers of public institutions, each

of which has a different value on the professionalism scale. Thus, a countrys overall performance on

this scale can be represented by a continuous probability distribution function. A vertical line located

at the midpoint of the scale separates Weberian institutions from patrimonial ones. Institutions

which have values of bureaucratic professionalism to the left of the vertical line are considered

Weberian, while institutions to the right of the vertical line are considered patrimonial.

The classification scheme I consider here is simple but not wholly unrepresentative of much

thinking on state capacity. If the mean value of a countrys distribution of institutions is located to

the left of the vertical line, then that countrys entire public bureaucracy is classified as Weberian.

Similarly, if the mean value of a countrys distribution of institutions is located to the right of the

vertical line, then that countrys entire public bureaucracy is classified as patrimonial. In this way,

countries are categorized according to their administrative systems central tendencies.5

Whats the problem with classifying bureaucracies in this way? In certain circumstances,

there may not be much of a problem at all. For countries in which institutional diversity is relatively

mild and the central tendency of the public sector is clearly to the left or to the right of the vertical

line (that is to say, clearly Weberian or patrimonial), the classification scheme will provide a fairly

accurate portrayal of the experiences of the vast majority of public institutions. This is the case for

country A in the figure below, whose entire public bureaucracy would be classified as Weberian and

nearly all of whose institutions also fall under that heading.

5 In a text which is generally quite conscious of the heterogeneity within the Brazilian state, Torres (2004)

adopts such an approach: In this way, when we utilize these concepts [the Weberian, patrimonial and
managerial models], we will be referring to the mean tendency of the public sector, with all the variation and
specificities that it entails (141).

111
Figure 3.1: Misclassification error in a simple taxonomy of public bureaucracies, bell shaped (normal)
distributions

The problem emerges when institutional diversity is considerable (and/or the central

tendency of the public sector lies close to the dividing line). Country B in the figure above is an

example of a case with high institutional diversity. Given the classification scheme just described, the

public bureaucracy of this country would also be classified as Weberian. Nevertheless, a considerable

segment of institutions in this country fall under the patrimonial heading (25% as depicted in the

figure). Thus, the Weberian label as applied to the bureaucracy of this country is highly misleading.

Country B contains a considerable number of institutions that could be characterized as patrimonial

as well as a considerable number of institutions that approach the Weberian ideal. In countries such

as this, where institutional diversity is considerable, measures of central tendency are not particularly

informative guides to behavior and perceptions within any given constitutive unit of the

administrative system.

The problem is compounded when the distribution of bureaucratic professionalism across

institutions takes irregular shapes, such as is the case for multimodal distributions. Figure 3.2 depicts

the classification scheme as applied to two countries in which bureaucratic professionalism is

distributed across institutions in this way. Country C shows the case of a country with a bimodal

112
distribution of bureaucratic professionalism. In this case, one finds a large contingent of institutions

which fall well within the rubric of Weberian bureaucracy, a smaller but still sizable contingent of

institutions which fall well within the rubric of patrimonial bureaucracy and relatively few institutions

located between the two extremes. As was true for countries A and B, the classification scheme labels

country C a Weberian bureaucracy, even though a large proportion of institutions is misclassified and

there appear to be two quite distinct sets of institutions in this country. Country D depicts a slightly

more complicated case: a country with three separate modes, each corresponding to fairly distinct

institution types. Again, using the value of the mean institution in order to categorize the entire

public sector, country D would be classified simply as Weberian.

Figure 3.2: Misclassification error in a simple taxonomy of public bureaucracies, multimodal

distributions (mixture of normals)

113
 Lest some readers think such irregular-shaped distributions of bureaucratic professionalism

are an unrealistic portrayal of the distributions across institutions found in the real world, it is worth

remembering that the enclave or islands of administrative competence approach to administrative

organization, much touted in academic and policy circles as a crucial means to effect economic policy

reform, corresponds directly to a bimodal pattern like that of country C (Burki and Perry 1998,

Waterbury 1992, World Bank 1997). Along these lines, recent work on the success of state

developmental policy in South Korea and its failure in the Philippines points precisely to such a

bimodal patterndubbed the bifurcation of the stateas a crucial ingredient in the former

countrys success (Kang 2002).

The perception of a certain bimodality of the state, moreover, manifests itself in the

informal language of citizens throughout the developing world. In various countries in Latin

America, for instance, certain sets of public institutions are commonly referred to as botines polticos

literally, political spoils. The perception is that in these institutions, posts and resources are there to

exploited as a lever for partisan political advantage. The botines are juxtaposed to the so-called

autonomous or autarkical institutionspublic entities which enjoy high levels of independence in

personnel management and the budget (usually Central Banks and regulatory agencies; sometimes

revenue-generating institutions such as tax or customs services). In these institutions, meritocratic

patterns of recruitment and career advancement are the rule. Other institutions, such as Ministries of

Finance, national statistics institutes and the like, which may lack formal autonomy but whose

technical demands proscribe a patronage-based approach to personnel management, often fall under

the same rubric. The distinction between the spoils institutions and the technical institutions

sometimes has a legal basis, sometimes not, but it is well appreciated and often reflects the existence

of a tacit elite agreement about which areas of the state are up for political grabs and which are off-

limits.6

6
A skeptical reader might respond to the preceding discussion by arguing that a better wayand, in fact, a
more common wayof measuring overall state capacity is to use a continuous measure of capacity, not a
dichotomous one such as the Weberian/patrimonial categorization presented above. However, replacing the

114
 A third reason that we should take institutional diversity seriously is because it affords us

opportunities for inference unavailable at the cross-national level. In any research design based on

exploiting cross-national variation in order to identify the causal effect of a particular variable on

bureaucratic performance, one must always contend with existence of country-specific fixed

effects. Used in this context, this term refers to unmeasured or unobservable characteristics of

nations that exert a causal influence on bureaucratic performance and that may also be correlated

with one or more predictor variables. Unless a cross-national dataset is characterized by observations

over an additional dimension, such as time (as in panel or time-series cross section datasets), regions

or institutions, the presence of such fixed effects will lead to omitted variable bias. Because such

country-specific effects can be assumed to affect the performance of all institutions in the same

country in the same way, a cross-institutional analysis allows for the possibility of controlling for such

effects. To put the matter in other terms, potentially relevant elements of national history such as

legacies of authoritarian government, the imprint of colonization and the like can be held constant in

a cross-institutional analysis, whereas they cannot in a strictly cross-national analysis.7

The final reason why it is worth taking institutional diversity seriously goes beyond the realm

of academic debate and takes us into the realm of policymaking. Taking into consideration the

existence of institutional diversity is important in this respect because recent trends suggest that such

diversity may have important ramifications for the allocation of development assistance.

As international lending institutions such as the International Monetary Fund and the World

Bank have devoted greater energy to matters relating to the quality of governance, these concerns

dichotomous classification with a continuous measure does not eliminate the problem of unrepresentativeness
I have identified. In the best of cases, a continuous, one-dimensional summary measure of bureaucratic
performance would accurately capture the average tendency of each countrys administrative system. But even
if we had a perfectly accurate measure of the average tendency, this would not solve the problem. As drawn,
the distributions for country A, B, C and D not only lead them all to be categorized as Weberian but also
produce exactly the same score on a continuous measure of the average level of bureaucratic professionalism. This
is because all the countries share the same mean, despite having very distinctive concentrations of
professionalism throughout the public sector. The problem, therefore, is not one of an oversimplified
taxonomic scheme per se, but rather the notion that average performance can provide us with all the information
we might need to evaluate an entire public sector.
7 Datasets which contain variation across institutions within a given region may be particularly useful in this

regard, as cross-regional variation in norms and patterns of governance found in large federal states like Brazil,
India or the United States may be as marked as those found across some countries.

115
have come to manifest themselves in the targeting of development funds to countries in need. The

clearest demonstration of this trend is the recent establishment of the Millennium Challenge

Corporation (MCC) by the US Congress in January 2004. The MCCs duty is to oversee the

disbursement of funds for development assistance to countries that are suffering from economic

want but whose governments are committed to improving the quality of public goods and services

for their people. The core idea behind the founding of the MCC is that development assistance ought

to be tied to observable improvements in welfare and governance. To this end, sixteen quantitative

indicators measuring three components of governanceruling justly, investing in people and

promoting economic freedomare used to select recipients from a pool determined by income level

(and statutory restrictions limiting the potential recipients of development assistance).

Only countries that meet the following requirements are eligible for funds: 1) they perform

above the median of their income group on at least half of the indicators for each component of

governance; 2) they do not perform substantially below the median on any indicator; 3) they perform

above the median of their income group on a corruption control indicator produced by the World

Bank; and 4) any supplemental information favors their inclusion.

With $1.5 billion in aid disbursed in 2005 (and $5 billion being the goal for the future), it is

clear that much rides on the quality of the indicators, especially the corruption control indicator. To

produce a just allocation of funds, it is important not only that this indicator be a relatively accurate

yardstick of the average level of corruption in the public administration, but that the average itself be

reasonably informative as to the distribution of corruption in a variety of public institutions. If notif

a high level of institutional diversity is presentthen difficult questions about eligibility raise their

heads. Are there certain areas of the state in which a little bit of corruption is more tolerable than

others? Is a candidate country more attractive if it has a low to moderate level of corruption spread

throughout the public administration than it would be if, instead, corruption was highly concentrated

in certain small pockets of the state but absent from the rest? As unpalatable as answering such

116
questions might seem, ignoring them simply institutionalizes a degree of arbitrariness in the process

of targeting funds.

3.3: Perspectives on the Bureaucratic Marketplace in Bolivia, Brazil and Chile

In the pages that follow, this chapter compares cross-national and cross-institutional

variation in bureaucratic capacity in Bolivia, Brazil and Chile. Why are these appropriate cases to

employ in order to underscore the importance of cross-institutional variation in bureaucratic

capacity? The stylized facts which emerge upon a reading of standard references on bureaucratic

performance in these countries point to one important reason: the cases are depicted as ideal types of

certain types of bureaucracies. The Bolivian public administration purportedly approaches the neo-

patrimonial ideal type, the Chilean public administration purportedly approaches the Weberian ideal

type (at least relative to other bureaucracies in the region) and the Brazilian public administration

approaches an ideal type bifurcated or dualist state. To the extent that we find that intra-state,

institutional variation overwhelms interstate variation in this set of cases, then we will have a strong

basis for making the case that analysis of variation in capacity at the institutional level deserves as

much a place at the academic table as analysis of variation at the national level. Let us consider the

standard descriptions of the bureaucracy in the cases.

In order to describe the basic contours of the Bolivian state, observers from the World Bank

have made extensive use of the term informality (World Bank 2000b). Much like economic

exchange in the open air markets of La Paz, the Bolivian public sector is purportedly characterized by

the primacy of informal norms and practices over formal rules and standard operating procedures.

Entry into the public administration and advancement is supposedly based largely on partisan

affiliation, not merit, and the assiduous cultivation of political contacts is characterized as the

principal means of rising up the ranks. In this context, the public sector has come to be characterized

by high rates of turnover, low morale and low institutional memory (Leyton 1992).

117
 Patronage indisputably plays a central role in the Bolivian political system, with the allocation

of bureaucratic posts having been a critical means of cementing alliances in pre-electoral, post-

electoral and mid-term coalitions (Gamarra and Malloy 1995). Certain authors have attributed the

importance of patronage to a peculiarity of the Bolivian constitution, which throws the selection of

the president to the Bolivian Congress in the event that the first round electoral contest does not

produce a majority winner (Gamarra 1997). This important institutional peculiarity notwithstanding,

historical treatments of the Bolivian state have noted that the dispensation of patronage played no

less a role during the myriad military governments from the 1960s to the early 1980s than it did in

democratic epochs before and after military rule (Domingo 2003, Gamarra 1987, 2003). In fact, the

ostensible patrimonial nature of the Bolivian public administration is generally written about as if the

core sets of practices which defined the Bolivian state half a century ago remain in place today, little

changed by the passage of time. In this vein, Domingo (2003) argues that the patrimonial character

of the Bolivian state since 1952 is one of the most pervasive and long-lasting legacies of the

revolution (373).

Just as time has purportedly not dislodged the key features of Bolivias patrimonial state,

most accounts of the Bolivian public administration do not emphasize distinctions across

institutions.8 As one high level Bolivian functionary put it: take 100, subtract the percentage of

employees who say their institution belongs to the cuoteo poltico [the spoils system orchestrated by

Bolivias major parties] and there you have the percentage that are lying.9 To summarize, the stylized

depiction of the Bolivian bureaucracy which emerges from the literature is one of a homogenous,

8 There are a few important exceptions to this characterization. Juan Antonio Morales, former president of the
Bolivian Central Bank, has written: the overall quality of the bureaucracy has improved over time, although
possibly not with results commensurate to the resources invested. The more technical ministries and agencies
of the Bolivian government and the upper-level civil servants are as good as their counterparts in any of the
neighboring countries (Morales 2003, 225). The important study by the World Bank (2000b) emphasizes the
problems encountered in the Bolivian customs, road and tax services prior to the onset of major reforms in
these institutions at the beginning of the new millennium (see also Lanyi et al. 2000 on the customs service).
The aforementioned study also underlines the exceptional performance of Bolivias Superintendencies
(regulatory agencies), a theme to which I will return in the empirical analysis which follows.
9 Confidential interview with an official belonging to the Unidad de Anlisis de Polticas Sociales y Econmicas

(UDAPE), La Paz, Bolivia, July 2003. After making this statement, the official then explained to me that the
institution to which he belonged was not part of the cuoteo politico, and that he only accepted his post after
having received assurances that it would not be.

118
inert and inefficient apparatus, long treated as a teat from which patronage-minded parties have

suckled.

Academic accounts of the nature and evolution of the Brazilian public administration share

certain commonalities with the Bolivian case as well as important differences. The logic of

patrimonialism is acknowledged to have installed itself in wide swaths of the Brazilian state and to

have had a historically important role in establishing bureaucratic careers (Graham 1990, cf Schneider

1991). Moreover, as in the Bolivian case, the exigencies of constructing and maintaining governability

coalitions has been pointed to by observers as one of the root causes of persistent patronage politics

within the executive branch (Abranches 1988, Torres 2004, Weyland 2000). However, unlike the

conventional picture of the bureaucracy painted in Bolivian case, the Brazilian bureaucracy is nearly

always depicted as exhibiting great diversityboth over time and across institutions.

In terms of variation over time, numerous observers have remarked upon a distinct cyclical

pattern. In this view, bureaucratic professionalism constituted an overriding concern for authoritarian

governments committed to state-led industrializationboth under the Vargas dictatorship from

1930 to 1945 and the military government from 1964 to 1985but the democratic transitions which

succeeded them had the unpalatable consequence of reopening previously insulated areas of the

state to clientelist networks. In the earlier period, the archetypical example of this process was

Vargas establishment of the Departamento Administrativo do Servio Publico (a civil service organ

responsible for propagating Weberian best practice) and its subsequent dismantlement by the

democratic governments of the Republican period (Geddes 1994, Graham 1965, Nunes 1997, Siegel

1966). The cycle repeated itself in the latter period, as clientelistic tendencies evident in the waning

years of the authoritarian regime intensified themselves after the transition (Hagopian 1990, Weyland

2000).

As much as the priorities of different regimes impacted upon bureaucratic professionalism, it

is widely accepted that throughout its history the Brazilian public administration has always been

characterized by a marked dualism. This phrase is used to underline the fact that, at different times

119
and to different degrees, institutions considered to be critical to the core functioning of the state,

such as the economic bureaucracy, the foreign service, the national development bank and certain

sets of state-owned enterprises, have been carefully protected from patronage politics and

corruption, while others, such as the Ministries, social agencies and institutions governing

transportation infrastructure, have been consistently subjected to them (Gouva 1994, Hochman

1992, Prado 1997, Torres 2004, Willis 1995).

Another aspect in which the image of the Brazilian bureaucracy departs from that of the

Bolivian lies with the purported impact of formal rules and procedures. Whereas the Bolivian

administration is characterized as suffering from excessive informalism, a common criticism of the

Brazilian bureaucracy is that it suffers from the opposite type of ailment: excessive and unwieldy legal

formalism in personnel management (Bresser-Pereira 1998, Daland 1981). Although an important

reform effort in the late 1990s made inroads in redressing this problem (cf Bresser-Pereira 2003),

legal strictures remain onerous by international standards. The peculiarity of the Brazilian public

administration in this regard lies with the apparent conviviality of rigid legalism and clientelism within

the same state apparatus (Daland 1981, Nunes 1997, Torres 2004).

To summarize, the image of the Brazilian bureaucracy which emerges from the literature is

that of a heterogeneous, bifurcated stateone characterized by patrimonialism in some spheres and

Weberian administration in others. Although the bureaucracy has been much affected by the reform

initiatives of different governments and regimes, an emphasis on formal procedures has been a

constant, as has been the resulting clash between formalism and reality, to borrow Grahams

(1965) provocative phrase

Of the three cases in the present study, the Chilean public sector is perceived as most closely

approximating the conditions of a highly professional, Weberian bureaucracy. According to

observers at the World Bank, the Chilean state is effective in delivering public services, its public

servants are professionalizedcompetent and honestand its operations are predictable (Burki

and Perry 1998, 132). As in Bolivia and Brazil, the necessity of maintaining electoral and governability

120
coalitions has required a careful distribution of bureaucratic appointments among coalition partners,

but the standard view is that this process has not had a negative effect on the functioning of the

bureaucracy. A string of recent corruption scandals has led many to qualify this position somewhat

but the general perception of the Chilean state as an exemplar of probity and efficiency, at least

within Latin America, remains unchanged.10

As in Brazil, the passage of time and the elaboration of different policies by elected and

authoritarian governments have left a major imprint on the Chilean bureaucracy. Prior to the onset

of the military government in 1973, the Chilean bureaucracy required postulants to satisfy a rather

stringent set of educational standards for ingression to middle-level and technical posts, yet at the

same time it was widely understood that advancement up the ranks depended to a large degree on

partisan affiliation and contacts (Petras 1969, Valenzuela 1984).11 By the time the Pinochet regime

took power in 1973, standard accounts of the period suggest that the politicization of the state was

becoming more and more widespread, as the Christian Democratic and Popular Unity governments

expanded the scope of patronage beyond previously established limits (Rehren 2002, Valenzuela

1984).

In line with its ideological bent, the military government reacted to this state of affairs by

drastically slashing public employment, cutting wages and rescinding employment tenure for all

administrative posts (Garretn and Cceres 2003). Once it had become clear that Pinochet had lost

the plebiscite on the military governments continuance in power, the military government then

changed course, promulgating the so-called ley de amarre decree which made it nearly impossible to

dislodge the appointees Pinochet had placed in the public bureaucracy. In so doing, the military

government radically reduced the appointment power of the incoming democratic administrations

(Garretn and Cceres 2003, Marcel 1999, Orellana 2004). Given this highly circumscribed legal

10 A forceful dissenting view is provided in Orellana (2004), who argues that the Chilean public administration

has become feudalized by political parties in recent years. Although this position has gained currency of late,
it still remains far from the majority viewpoint.
11 Unlike Brazil, a classical civil service (with a general system of exams and training) was never introduced in

Chile. To this day, each institution remains responsible for recruiting its own personnel (Marcel 1999).

121
environment, the potential scope for patronage in democratic Chile became much narrower than it

was for most other countries in Latin America.

Although journalistic accounts of corruption in Chile occasionally touch upon differences

across public institutions (cf Brescia 2001), most discussions of the Chilean bureaucracy treat it as a

relatively homogenous entity, well-described by the constituent features of the ideal-typical Weberian

bureaucracy. One of the origins of the relatively high level of probity encountered in the Chilean

public sector supposedly rests with the considerable power and independence of its supreme auditing

institution (SAI), the Comptroller General of the Republic, or CGR (Marcel 1999, Siavelis 2000,

Valenzuela 1978). The head of Chiles CGR enjoys lifetime tenure (unlike any other SAI head in

Latin America) and the institution itself has formidable statutory powers (Groisman and Lerner

2000).

Along with the beneficial formalism represented by the strong external control provided by

the CGR, the contemporary Chilean bureaucracy is also characterized by a more negative type of

formalismhighly rigid rules governing career advancement. Authors have noted that these rules

place a heavy emphasis on seniority, thus downplaying the importance of individual effort and

reducing incentives for high performance (Marcel 1999).

To summarize, the stylized depiction of the Chilean bureaucracy which emerges from the

literature is that of a fairly homogenous state, which approximates in many respects the features of a

Weberian bureaucracy. The degree of formalism which characterizes the bureaucracy is high, both in

beneficial forms such as the strong external control provided by the CGR but also in negative forms

such as overly rigid rules for personnel management.

The above accounts of bureaucratic quality in the three cases find resonance in various

cross-national measures of bureaucratic effectiveness and corruption control. Table 3.1 below

presents rankings from indicators produced by three different organizations: the World Economic

Forum, the World Bank Institute and Transparency International. On all of the indicators, Chile

scores highest among the three cases, followed by Brazil, which is in turn followed by Bolivia. On the

122
World Economic Forums indicator of civil service independence (i.e., the independence of the civil

service from political pressure), the difference between the Chilean and Brazilian cases is slight.

However, for all other indicatorsone designed to capture the frequency with which public funds

are diverted from their official purposes (WEF), one designed to measure overall government

effectiveness (WBI) and two designed to capture overall levels of corruption (WBI and TI)the

Chilean bureaucracy is ranked well ahead of the Brazilian bureaucracy, which in turn, is ranked well

ahead of the Bolivian bureaucracy.

Table 3.1: Ranking of cases on cross-national measures of bureaucratic effectiveness and corruption

World Economic Forum World Bank Institute Transparency

International

Civil service Diversion of Government Corruption Corruption

independence public funds effectiveness control perceptions
(2000) (2003) (2004) (2004) index (2003)

Bolivia 57th of 59 73rd of 80 146th of 209 152nd of 204 106th of 133

Brazil 39th of 59 52nd of 80 88th of 209 96th of 204 54th of 133
Chile 32nd of 59 24th of 80 28th of 209 24th of 204 20th of 133

Given the very different characterizations of the overall level of bureaucratic quality in the

three cases, they provide us with a unique opportunity to evaluate the claim that differences

encountered across institutions in the same polity may be equally or more important those

encountered across entire public sectors. In essence, the use of the three cases here follows in broad

strokes the crucial case strategy elaborated by Eckstein (1975). If we find evidence that the level of

institutional diversity within these cases exceeds the diversity encountered between the public sectors

considered in their entiretiesa surprising finding given the well-established national level

reputations of the casesthen it is likely that this state of affairs obtains more generally.

The following section of the chapter uses the studys survey data to investigate this claim.

Therein, I consider three dimensions of bureaucratic capacity: civil service politicization, the

123
institutionalization of bureaucratic career paths and the strength of institutional mechanisms

designed to control corruption.

3.4: What Do the Data Tell Us? Examining Three Dimensions of Bureaucratic Capacity

Civil Service Politicization

I begin my exploration of the data with a focus on employee perceptions regarding the

politicization of the civil service. Politicization is understood to be low when the following

conditions obtain. Firstly, hiring decisions depend primarily on the education, training and work

experience of candidatesand only secondarily or not at all on partisan affiliation or political

contacts. Secondly, the criteria for promotions are clearly delineated to public employees, and consist

primarily of indicators of job performance and/or experience, not on political or personal affinities

with superiors. Thirdly, dismissals occur as a result of poor performance or unethical behavior, and

perhaps occasionally as a result of budgetary shortfalls or the shifting priorities of the government.

As a rule, dismissals of significant numbers of employees are generally not tied to cabinet

reorganizations or the electoral cycle. Finally, the role played by elected officials in relation to the

bureaucracy is primarily one of establishing the direction of broad policy initiatives. Political intrusion

into the day-to-day implementation of policy is kept to a minimum.

Examining the Survey Responses

How do the cases fare in terms of these aspects of civil service politicization? We begin our

evaluation with the visual inspection of responses to two questions concerning hiring and

promotions. One of these questions asked respondents to what extent they agreed or disagreed with

the following statement: political connections are important for determining who obtains a position

124
in my institution. The other asked respondents to what extent they agreed or disagreed with the

statement that promotions in my institution depend more on performance than political affiliation

or political influence. Potential responses for both questions were: strongly agree, agree, disagree

and strongly disagree.

Figure 3.3 displays the distribution of answers to these questions for each country in the

form of mosaic plots. Such plots visually represent the counts in the cells of a contingency table by

rectangles whose size is proportional to the cell frequency (Hartigan and Kleiner 1981, 1984, Friendly

1994). As a guide to interpretation, note that the larger are the light gray areas of the above plots, the

greater the proportion of respondents who felt that political factors were important in the two ways

just mentioned.

Figure 3.3: Perceptions of the importance of political connections in obtaining a post and rising up
the ranks (mosaic plots)

The plots demonstrate that the breakdown of responses by country are fairly similar. In both

Bolivia and Chile, the plurality of respondents indicated that they disagreed with the statement that

political connections were important for obtaining a post in their institution and that they agreed with

the statement that promotions in their institution depend more on performance than political

connections or influence (22.3% in Bolivia and 20.5% in Chile). In Brazil, the plurality was pretty

125
much evenly split between two categories of respondents: those who jointly indicated that they agreed

that political connections were important for obtaining a post in their institution and that they

disagreed that promotions in their institution depend more on performance than political connections

or influence (17.0%) and those who jointly indicated that they agreed that political connections were

important for obtaining a post in their institution and that they also agreed that promotions in their

institution depend more on performance than political connections or influence (16.9%). In all three

countries, a relatively large contingent of respondents fell into the former categoryi.e. respondents

who agreed that political connections were important for obtaining a post and disagreed that

promotions depended more on performance than political factors (17.7% in Bolivia and 15.0% in

Chile).

Naturally, the question arises as to what degree these aggregate results mask variation in

perceptions at the institutional level. Do respondents in the same country offer different answers to

the above questions because different individuals situated in similar institutions perceive their

institutions differently or because different institutions in the same country really behave quite

differently? The data suggest quite clearly that the latter proposition is the case.

Figure 3.4 conveys a sense of the internal variation in the cases. The first two columns of the

figure show mosaic plots for two selected institutions in each country. The last column displays a bar

chart depicting the proportion of respondents who indicated that they agreed or strongly agreed that

political connections were important for obtaining a post and that they also disagreed or strongly

disagreed that promotions depended more on performance than political factors for each institution in

the cases. As both the mosiaic plots and bar charts make clear, the range of internal institutional

variation is considerable.

126
Figure 3.4: Perceptions of the importance of political connections in obtaining a post and rising up
the ranks, results by institution

For each country, the institution depicted in the first cell of a given row is the institution for

which respondents indicated political considerations were most relevant. The institution depicted in

the second cell of each row is the institution for which respondents indicated political considerations

were least relevant. Employee evaluations of the importance of political considerations in the two

institutions are radically different for all the countries included in the survey. In Bolivias Ministry of

Sustainable Development, for instance, 58% of respondents jointly indicated agreement or strong

agreement that political connections were important in obtaining a post in that institution and

disagreement or strong disagreement that promotions depended more on performance than politics.

127
This compares to Bolivias Superintendency of Banks (SBEF), where only 7% of respondents

responded in the same manner.

Cross institutional differences were equally marked in the Brazilian and Chilean cases.

Interestingly, in all instances it was a countrys Central Bank or Superintendency of Banks which was

found to be most insulated from political pressures, a finding which, as we shall see, holds up for

similar questions worded in a variety of different ways.

Tapping on the same dimension, several questions contained in the survey asked

respondents to rank the factors of greatest importance in promotions and dismissals in their

institutions. With respect to dismissals, respondents were asked to rank the importance of the

following four factors: poor performance, unethical behavior, insubordination and political factors.

With respect to promotions, respondents were asked to rank the importance of good performance,

seniority, friendship/personal connections and political connections. Figure 3.5 displays in mosaic

form the proportion of respondents who indicated that a given category was the most important.

The responses provided by the Bolivian and Brazilian respondents appear to exhibit a

marked bimodality. In the Bolivian case, for instance, there was a large contingent of bureaucrats that

indicated that performance was the most important determinant of both promotions and dismissals

in their institution (28% of respondents). Taken by itself, this finding would suggest that the Bolivian

public administration contains a much greater dose of Weberian practice than it has been credited

with, and that the traditional view of the Bolivian state as a receptacle for patronage appointees may

be somewhat exaggerated. Nevertheless, the Bolivian case also featured a large contingent of

bureaucrats who responded exactly as the traditional view might lead one to expect: 20% of

respondents indicated that in their institutions political factors were the most important determinant of

dismissals and promotions.

128
Figure 3.5: Most important factors in dismissals and promotions (mosaic plots)

The Brazilian results conform to this same basic pattern. In this country, a plurality of

respondents jointly indicated that unethical behavior was the primary cause of dismissals in their

institution and that good performance in ones job was the primary determinant of promotions

(18%). Yet very close behind was the proportion of bureaucrats who indicated that political factors

were the primary determinants of both dismissals and promotions (17%).

In this last respect, the Chilean results were quite distinct from the Bolivian and Brazilian

ones. Whereas relatively sizeable contingents of bureaucrats in the Bolivian and Brazilian cases

indicated that political factors constituted the foremost criteria for dismissals and promotions in their

institutions, only 6.4% of the Chilean respondents responded in this way. As we shall see below, this

difference emerges not so much from the fact that the median public institution in Chile was

perceived as being much less politicized than the median in Bolivia or Brazil, but rather from the fact

that none of the Chilean institutions were perceived as being anywhere near as politicized as the most

politicized institutions in the other two countries.

These important cross-national differences notwithstanding, as before, the aggregate results

conceal significant levels of cross-institutional variation. Figure 3.6 decomposes the results by

129
institution. Shown in the first two cells of each row are the mosaic plots for the institution in each

country in which political factors were felt to be most important in dismissals and promotions as well

as the institution in which such factors were considered to be least important. Shown in the final cell

is a barplot depicting the proportion of respondents in each institution that indicated that political

considerations were the single most important factor in both dismissals and promotions in their

institution.

Figure 3.6: Ranking of the importance of various factors in dismissals and promotions, results by
institution

130
 Upon examination of the barplots, five institutions immediately stand out from the rest

Bolivias Ministry of Sustainable Development, Ministry of Peasant Affairs (read Agriculture),

Ministry of Economic Development and Productive Social Investment Fund (FPS) as well as Brazils

National Department of Transportation Infrastructure (DNIT; read Road Service). In no other

institution included in the survey was the importance of political connections reported by employees

anywhere near that for even the least politicized of these institutions.

In Bolivias Ministry of Sustainable Development, for instance, 64% of respondents jointly

indicated that political factors were the most important criteria in dismissals and promotions. This

percentage was nearly six times greater than that for the median institution in Bolivia, for which the

corresponding percentage was only 11%. A similar distinction between extreme cases and the median

emerged for Brazil. In Brazils DNIT, 50% of respondents jointly indicated that political factors were

the most important criteria in dismissals and promotions. This compares to the Brazilian median of

only 14%.

Two distinguishing features of the Chilean case seem to be the relatively low value of the

maximum and the fact that the distinction between maximum and median institutions was much less

marked than the other cases. The largest percentage of employees in a Chilean institution indicating

that political criteria were key was found in the Instituto de Desarollo Agropecuario (INDAP), where

only 13% of respondents considered political factors to be of foremost importance in dismissals and

promotions. This was slightly more than double the percentage for the median Chilean institution

5%yet still only represented a difference of 7% between maximum and median. In the Bolivian

and Brazilian cases, on the other hand, the difference between the maximum and median institution

was 53% and 36%, respectively.

To what degree do public employees in the cases perceive the everyday actions of civil

servants to be subject to political manipulation by the leadership of the governing parties? A question

touching on this issue asked respondents to what extent they agreed or disagreed with the following

statement: Party leaders can easily punish (perjudicar/prejudicar) functionaries that do not obey the

131
orders of the governing parties. Figure 3.7 displays the distribution of responses to this question for

the three cases.

Figure 3.7: Perceptions regarding the political control of the bureaucracy in the cases

Although a majority of employees in all of the cases felt that party leaders could easily punish

wayward bureaucrats should they chose to do so, the Bolivian case clearly stands out in terms of the

degree of consensus on this point. Overall, 88% of the Bolivian respondents felt that party leaders

could easily punish bureaucrats that disobeyed their directives. Forty three percent strongly agreed

that this was the case. Substantial numbers of bureaucrats in the Brazilian and Chilean public

administrations, on the other hand, felt that there was a greater degree of insulation from political

control. In these countries, 38% and 34% of respondents, respectively, indicated disagreement or

strong disagreement with the notion that party leaders could easily punish bureaucrats.

Generating Politicization Scores

All of the survey questions that have been discussed thus far in this section share a very

important feature in common. They can all be thought of as observable manifestations of a single

132
latent (unobservable) variable. The latent variable common to all is the survey respondents

underlying perception of the politicization of the bureaucratic environment in which he works. Given

the conceptualization of the survey responses as emanating from this variable, the tools of

confirmatory factor analysis (CFA) can be utilized to estimate a politicization score for each

individual included in the survey; the score being a one-dimensional representation of the latent

politicization variable. The distribution of politicization scores can then be compared across

countries and across institutions, with the purpose of evaluating how general perceptions of

politicization vary across these units.

The first step in the estimation of the politicization scores consists of the construction of a

path diagram. The purpose of such a diagram is to illustrate the fundamental assumptions underlying

the estimation of the CFA model. In our case, they are: 1) the survey responses for each question are

determined by the influence of only a single factor (the latent politicization variable); and 2) the

errors in measurement are uncorrelated. Figure 3.8 presents the path diagram alongside a description

of the observable variables used in the analysis.12

Figure 3.8: Confirmatory Factor Analysis Model of Politicization

12 The parameters of the above CFA model were estimated using John Foxs sem package for the R statistical

computing language. As the observable indicators consist of ordinal variables, the correlation matrix for the
observable indicators was calculated using the aforementioned authors polycor package for calculating
polychoric correlations.

133
 The results of the CFA analysis are displayed in table 3.2. Each of the factor loadings (the

estimates of the relationship between the survey responses and the latent variable to which they

correspond) was of the correct sign and statistically significant by conventional standards. Various

measures of the overall fit of the model suggested that the single factor CFA model corresponded

exceptionally well to the data.

Table 3.2: Estimates of factor loadings for the CFA model

Given the estimates of the factor loadings above, politicization scores were calculated with

Barletts scoring method and normalized to lie between 0 and 1.

Having a one-dimensional, quasi-continuous measure of perceptions of politicization allows

us to tackle a number of questions with which we begun the chapter.13 With our summary measure

of politicization in hand, we are now equipped to answer the following: How does the distribution of

13Strictly speaking, the estimated politicization score is not continuous, since the observable variables upon
which it is based are ordinal. Nevertheless, due to the large number of potential configurations of these
variables, there are 662 distinct values of the politicization score observed in the sample (which is comprised of
2408 individuals without missing data on the observable indicators). Because of the extraordinarily large
number of distinct values taken on by the politicization score, I treat it as a continuous in what follows.

134
perceptions about institutional politicization differ across the country cases? How does it differ

across institutions within each country? Is cross-institutional variation greater than cross-national

variation? Do the country cases differ significantly if we limit our comparisons to institutions with

similar functions?

The top row of figure 3.9 illustrates how the scores are distributed in the three country cases.

For each country, the figure depicts the density of the politicization scores using histograms overlaid

with kernel density plots. As the reader can see, in all countries the distribution of politicization

scores is skewed to the left, indicating that a preponderance of public employees find themselves in

institutions which they consider to suffer from relatively low levels of politicization. However, the

Bolivian and Brazilian distributions also appear to have multiple peaks, indicating the presence of

substantial masses of employees who consider politicization levels to be quite high relative to the

perceptions held by their colleagues.

Figure 3.9: Distribution of the perception of politicization by country (density estimates and

QQ-plots)

135
 The bottom row of figure 3.9 addresses the issue of how different the distributions are. It

does so through the use of quantile-quantile (QQ) plots. The QQ-plots show the quantiles of the

distributions for the two indicated countries plotted against one another. To the degree that the

observations in the plot are arrayed on or close to the 45 line, then this constitutes evidence that the

distributions for the two countries are similar to one another. To the degree that the observations

arrayed in the plot lie above the 45 line for certain ranges of the politicization score, this indicates

that there is a greater concentration of observations in those ranges for the country indicated on the

y-axis than the country on the x-axis. Similarly, to the degree that the observations arrayed in the plot

lie below the 45 line for certain ranges of the politicization score, this indicates that there is a greater

concentration of observations in those ranges for the country indicated on the x-axis than the

country on the y-axis.

The Q-Q plot for the comparison of the Bolivian and Brazilian cases reveals that the

distributions of perceptions about civil service politicization for these cases are nearly

indistinguishable (by visual inspection, at least). The same cannot be said for the QQ-plots which

compare the Chilean cases to the other cases included in the study. In juxtaposing the Bolivian case

to the Chilean one, the QQ-plot shows that there is a relatively greater concentration of employees in

Bolivia who consider their institutions to suffer from a moderate to high level of politicization (i.e.

have a politicization score higher than the full sample median, .37) than in the latter. An examination

of the QQ-plot for the Brazil-Chile comparison reveals the same pattern, with the Brazilian case

exhibiting a higher concentration of moderate to high politicization scores.

To formally test the hypothesis that the distributions of politicization scores for two

countries are equal, I used the two sample Kolmorgov-Smirnov (KS) test. Table 3.3 shows the results

of this test for each pairing of countries in the study.

136
Table 3.3: Kolmogorov-Smirnov Two Sample Tests

Brazil vs. Bolivia Chile vs. Bolivia Brazil vs. Chile

H0: FS(Brazil) = FS(Bolvia) H0: FS(Chile) = FS(Bolivia) H0: FS(Brazil) = FS(Chile)

HA: FS(Brazil) FS(Bolivia) HA: FS(Chile) FS(Bolivia) HA: FS(Brazil) FS(Chile)

Test stat. = .058 Test stat. = .119 Test stat. = .139

p-value = .085 p-value = .0003 p-value = 3.2 10-6

According to the KS test, there is some weak evidence in favor of rejecting the null

hypothesis of no difference in the cumulative distribution functions of politicization scores for

Bolivia and Brazil. However, for this pairing the median score for the Brazilian case is almost

identical to that of the Bolivian case (.38 versus .37). The differences between the Chilean case and its

two counterparts, on the other hand, were found to be quite strong. The tests corresponding to these

comparisons revealed that the differences between the distributions of politicization scores were

significant at any reasonable threshold of statistical significance. Thus, the data suggest that overall

perceptions of politicization are distributed similarly in our sample of employees in the federal public

administration in Bolivia and Brazil but that these distributions differ quite drastically from

perceptions in our sample of employees in the federal public administration in Chile.

How does the distribution of perceptions about politicization vary within the state? Does the

variation across institutions tend to swamp the overall variation across entire public sectors? An

examination of politicization scores by institution suggests that the answer to the latter question is

yes. Figure 3.10 below shows box-and-whisker plots depicting the distribution of politicization

scores for each institution included in the survey. A vertical black line in each box represents the

median score for a particular institution while the width of the box contains 50% of the observations

for that institution.

As our earlier examination of the mosaic plots would lead one to suspect, the variation

encountered across institutions is considerable. In the Bolivian case, the median politicization score

varies from .175 (Superintendency of Banks) to .699 (Ministry of Sustainable Development). In

Brazil, internal variation was also large, the lowest median score for an institution being .204 (Central

137
Bank) and the highest .609 (DNIT). While still noteworthy, the level of cross-institutional variation in

the Chilean case was more muted than the other cases, median scores varying from .209

(Superintendency of Banks and Superintendency of Pension Fund Managers, both tied) to .474

(Institute of Agrarian Development).

Figure 3.10: Perceptions of politicization by institution (box-and-whisker plots)

Note: vertical gray line represents median score for all institutions in indicated country

How might we quantify the difference between the variation encountered between

institutions within a given public sector and that encountered between public sectors considered in

their entireties? One way to do so is compare the absolute value of the difference in median

politicization scores across countries with a similar measure of variation across institutions. For a

138
country j, the latter quantity of interest is calculated as the mean absolute deviation (MAD) from the

median:

1 m
MAD j = | xij median(x ij ) |
m i =1
(3.1)

where xij is the median politicization score in institution i in country j, xij is the vector of median

politicization scores for all the institutions in j and m is the number of institutions surveyed in j. Table

3.4 presents a comparison of the differences across countries with the MADs for each country.

Table 3.4: Variation in perceptions of politicization within and between public sectors

Differences across countries MAD Ratios

M1 |median(BOL)- MADBOL = MADBOL /M1 = 8.69 MADBOL /M2 = 2.97

median(BRA)| = .013 .113

M2 |median(BOL)- MADBRA = MADCHL /M2 = 2.29 MADCHL /M3 = 1.71

median(CHL)| = .038 .090

M3 |median(BRA)- MADCHL = MADBRA /M1 = 6.92 MADBRA /M3 = 1.76

median(CHL)| = .051 .087

As the table illustrates, the variation in politicization scores across institutions in a given

country is uniformly greater than the variation encountered in any pairwise comparison of the overall

country medians. Measured in this way, the cross-institutional variation in Bolivia is nearly nine times

the absolute value of the difference between the Bolivian and Brazilian medians. Similarly, the cross-

institutional variation in Brazil is nearly seven times the absolute value of the difference between the

Bolivian and Brazilian medians. Even in the case in which cross-institutional variation is least

pronounced, Chile, the level of cross-institutional variation is still over twice as large as the absolute

value of the difference between the Bolivian and Chilean medians.

Comparing Sister Institutions

139
 The analysis thus far has shown that cross-institutional variation in perceptions of

politicization is greater than cross-national variation. However, it is important to recognize that

comparing the distributions of perceptions of politicization across country cases may be a somewhat

misleading guide to overall differences in employee perceptions. This is because the sets of

institutions which participated in the study do not match exactly in terms of their functional

specialization. In Bolivia and Chile, for instance, 5 out of 14 and 3 out of 6 institutions surveyed,

respectively, were regulatory agenciesabout one third of the institutions in each case. In Brazil,

only 2 out of 10 institutions surveyed were regulatory agencies. Moreover, in both Bolivia and Brazil

national road services participated in the survey, whereas the corresponding institution in the Chilean

casethe Ministry of Public Workschose not to participate in the survey.

Although the initial list of institutions selected to participate was drafted with an eye towards

guaranteeing the inclusion of institutions with comparable functions across the cases (and in fact,

such inclusion was attained in many instances), certain refusals to participate and intrinsic differences

in the organization of the state conspired against the possibility of perfect pairings across countries.

For this reason, it is important to examine differences in the country cases by matching pairs of

sister institutions, i.e. institutions in the different countries that execute similar tasks. In so doing,

we can see if overall cross-country differences in perceptions correspond well with the differences

encountered in specific areas of state activity.

Table 3.5 specifies the institutional pairings for each comparison across countries. The

scores for more than a single institution are pooled in cases in which multiple institutions perform

the task(s) denoted under the category heading. In Brazil, for instance, both monetary policy and

banking regulation fall under the purview of the Central Bank, whereas in Bolivia these two activities

are the separate domains of the Central Bank and Superintendency of Banks, respectively. Thus, in

comparing employee perceptions of politicization in institutions dealing with monetary policy and

banking regulation, the scores for the two Bolivian institutions are pooled. Instances in which such

pooling was necessary should be clear from the table.

140
 The take home message from the case by case comparison of sister institutions is this:

knowing how the cases fare at the aggregate level tells us remarkably little about cross-national

differences in the views of employees dedicated to specific areas of state activity. Simply because

greater or lesser concentrations of employees in one country believe that political connections are an

integral part of getting ahead in the bureaucracy than in other countries, this does not mean that the

same cross-national relationships will hold for all dimensions of state activity. States which perform

poorly relative to their counterparts in the aggregate may still outperform their counterparts in

specific tasks. Not unlike academic departments, states may occasionally be molded in the boutique

model, with improvements in certain aspects of public good provision receiving considerable time

and effort, and others left by the wayside.

Table 3.5: Sister institutions (as determined by administrative function)

Bolivia/Brazil Bolivia/Chile Brazil/Chile

sister institutions sister institutions sister institutions

category Bolivia Brazil category Bolivia Chile category Brazil Chile

Monetary Central Bank, Central Bank Banking Superintendency Superintendency Banking Central Bank Superintendency
Policy and Superintendency Regulation of Banks of Banks Regulation of Banks
Banking of Banks
Regulation
Customs Federal Revenue Federal Revenue Customs Service
Revenue Service, Tax Service (RF) Customs Customs Service Customs Service Customs Service (RF)
Service
Insurance,
Insurance Superintendency Superintendency Securities Superintendency Superintendency Insurance Superintendency Superintendency
Regulation of Pensions, of Private and Pension of Pensions, of Pension Fund Regulation of Private of Securities and
Securities and Insurance Fund Securities and Managers, Insurance Insurance
Insurance Regulation Insurance Superintendency
of Securities and
Insurance
National National National National National National
Agrarian Institute of Institute of Agrarian Institute of Institute of Agrarian Institute of Institute of
Reform Agrarian Colonization Reform Agrarian Agrarian Reform Colonization Agrarian
Reform and Agrarian Reform Development and Agrarian Development
Reform Reform
National Road National
Roads Service Department of
Transportation
Infrastructure
Ministry of Ministry of
Ministries Sustainable Social
Development, Development,
Ministry of Ministry of
Economic Agrarian
Development, Development
Ministry of
Peasant Affairs

141
 This pattern can be observed in all the cross-national comparisons. Take the comparison of

sister institutions for Bolivia and Brazil, for instance, conducted via the use of Q-Q plots and

presented in figure 3.11 below. In three types of institutions, those that govern monetary policy and

banking regulation, those charged with generating state revenue and the government ministries,

Bolivian civil servants perceive their working environments to suffer from significantly greater levels

of politicization than their Brazilian colleagues. On the other hand, in national road services and

institutions that regulate the market in private insurance, the opposite is the case.

Figure 3.11: Comparison of perceptions of politicization in sister institutions, Bolivia and Brazil
(QQ-Plots)

Note: p-val(KS) denotes p-value from two-sample Kolmogorov-Smirnov test

142
 A similar scenario obtains when one compares sister institutions in the Bolivia and Chilean

cases. Figure 3.12 show the relevant plots. Here again one finds a mixed bag. Perceptions of

politicization are markedly lower in Bolivias newly reformed customs service than they are among

employees in Chiles customs service but they are markedly higher in the institutions regulating

pensions, security and insurance in Bolivia than they are in similar institutions in Chile. Moreover,

there is clearly no difference in the distributions of perceptions about politicization in the institutions

that provide banking regulation in the two countries.

Figure 3.12: Comparison of perceptions of politicization in sister institutions, Bolivia and Chile

(QQ-Plots)

Note: p-val(KS) denotes p-value from two-sample Kolmogorov-Smirnov test

When one analyzes perceptions of politicization across sister institutions in Brazil and Chile,

the situation is no different. Figure 3.13 shows the relevant comparisons. In two types of institutions,

those dedicated to agrarian reform and those responsible for banking regulation, there were no

discernible differences between the country cases. For institutions that provide government revenue

through customs levies, perceptions of politicization were more marked in the Chilean case than the

Brazilian one. On the other hand, for employees in institutions that regulate the market in private

insurance, perceptions of politicization were more marked in the Brazilian case than the Chilean one.

143
Figure 3.13: Comparison of perceptions of politicization in sister institutions, Brazil and Chile

(QQ-Plots)

Note: p-val(KS) denotes p-value from two-sample Kolmogorov-Smirnov test

Politicization Scores and Party Militancy

Up to this point, we have examined the pervasiveness of civil service politicization by

drawing upon the evaluations and perceptions of public employees regarding the institutions in

which they work. In order to reduce cultural biases from creeping in, a number of these questions

asked participants to make explicit comparisons about the importance of political factors versus

other determinants of obtaining a public job, moving up the ranks and being separated from ones

institution. Nevertheless, it is important to recognize that in addition to reflecting concrete

differences in the functioning of public institutions, there exists the possibility that participants

responses to some of the survey questions presented above may also reflect country-specific norms

about what constitutes acceptable levels of political interference. For this reason, it is worthwhile to

check how the perceptions of politicization relate to more objective measures of politicization.

Perhaps the most direct such measure is the proportion of respondents in each institution

that belong to a political party. In institutions where partisanship is judged to be a necessary criterion

for obtaining a post and moving up the bureaucratic ladder, it would seem axiomatic that we should

144
also find relatively high concentrations of party militants. In fact, the data do suggest that this is the

case. The correlation between an institutions median politicization score and the proportion of its

employees that belong to a party is quite high (.69).

The bar charts in figure 3.14 present the proportion of respondents belonging to a political

party for each institution included in the survey. Overall, 15% of the Bolivian respondents indicated

that they belonged to a political party, followed by 9% of the Chilean respondents and 7% of the

Brazilian respondents. Eying the barcharts, it is clear that not only do perceptions of politicization

vary considerably across institutions but that actual concentrations of militants do as well. The three

Bolivian ministries included in the surveythe Ministry of Sustainable Development, the Ministry of

Economic Development and the Ministry of Peasant Affairsstand out for special mention in this

regard. These institutions had concentrations of party militants which were unparalleled by any of the

other institutions participating in the study, varying between 24% and 62% of all respondents.

Remove these institutions from the Bolivian case, and the distribution of partisanship across

institutions looks more or less like it does in Brazil or Chile. Keep them in, and the Bolivian public

administration clearly evinces a disproportionate penchant for partisanship.

It is worth noting that perceptions of politicization were not keyed to the concentration of

party militants in the same way for all the country cases. The left-hand side cells in figure 3.14 show

the bivariate scatterplots for each country. In Bolivia and Chile, there was clearly a very strong linear

relationship between the proportion of party militants found in an institution and the institutions

median politicization score. In Brazil, this relationship was much weaker, as institutions such as

DNIT and INSSperceived to be the most politicized of the Brazilian institutions surveyed

contained proportions of militants not very different from the Brazilian average. This finding points

to the possibility that perceptions of political intromission in some of the Brazilian institutions may

be keyed more to patterns of elite level rent-seeking and favors-for-contracts than the more

traditional manifestations of patronage politics, a point to which I will return in later chapters.

145
Figure 3.14: The concentration of party militants across institutions and its relationship to median
politicization scores

Note: dark solid lines denote fitted values from OLS regression; dotted lines denote fitted values from
locally weighted regression.

Summary of Civil Service Politicization Findings

Having examined a variety of indicators of civil service politicization as well as a one-

dimensional, summary indicator of the phenomenon, it appears safe to conclude that variation across

institutions generally overwhelms the variation found across public administrations considered in

their entireties. Thus, in this respect, the central tendencies of the country cases do seem to be

overwhelmed by their own institutional diversity.

However, this conclusion does not imply that cross-national patterns were absent in the

data. In terms of the ranking of political considerations in promotions and dismissalsas well as

overall perceptions of politicizationthe Chilean respondents painted a picture of considerably less

political intromission in the bureaucracy than did their Bolivian and Brazilian counterparts. Self-

reported partisanship in Bolivia was significantly higher than in the other two cases, and it was in this

146
country that the most politicized institutions (measured in a variety of ways) were consistently found.

However, these cross-national differences notwithstanding, on nearly all of the survey items reviewed

above the cross-national differences proved to be substantially less marked than the cross-

institutional ones.

Institutionalization of Bureaucratic Career Paths

Whereas the take home message of the previous section was that the perceptions of

politicization may vary more across institutions than across public sectors, an examination of the

institutionalization of bureaucratic career paths paints a very different picture of the cases. On this

dimension of state performance, the distinction between country cases is of paramount importance.

Here the cases appear to fall into two distinct groups. On the one hand, the Bolivian public

administration evinces a clear pattern of highly fluid, unstable bureaucratic career paths. On the

other, the Brazilian and Chilean administrations demonstrate markedly more stable and long-lived

bureaucratic career paths.

Perhaps the most direct way of measuring the institutionalization of career paths is by simply

examining the longevity of bureaucratic careers across the cases. Relevant in this regard is the amount

of time served by employees in the public sector overall, in the institution where they currently work

and the position which they currently hold. Figure 3.15 shows the proportion of employees that have

served for indicated time intervals in each of the above.

Immediately apparent upon inspection of the figure above is the experience gap of the

Bolivian public sector relative to its counterparts. Whereas 64% of the Brazilian respondents

indicated that they had served for 10 years or more in the public sector and 56% of the Chilean

respondents did the same, only 34% of Bolivian respondents indicated that they had served for this

period of time. These differences carry through when one examines tenure in ones institution and

post as well. A majority of the Bolivian respondents (51%) indicated that they had worked three years

147
or less in their institution, raising potentially serious concerns about the extent of institutional

memory in the public administration of this country. To juxtapose this with the other cases, a near

majority (46%) of respondents in Brazil indicated that they had worked 10 years or more in their

institution and an outright majority (51%) did so among the Chilean respondents.

Figure 3.15: Longevity of bureaucratic careers in the cases

Importantly, the cross-national differences on this survey item do not disappear when we

examine variation in responses at the institutional level. Figure 3.16 displays the institution-level

breakdown of responses regarding time served in ones institution. Although the distribution of

responses varies quite a bit among the Bolivian institutions, variation among institutions in the

Brazilian and Chilean cases is minor.14 Especially noteworthy in the Bolivian case is the fact that

several institutions which have atypically high proportions of employees in their first year of service

14 Moreover, in the Brazilian institutions in which employee tenure in the institution seems to be rather short,

such as the National Agency of Land-based Transport (ANTT), this was due to the recent creation of the
institution itself.

148
to the institutionthe Ministry of Sustainable Development and the Ministry of Agricultureare

also institutions which experienced a change of leadership as a result of a cabinet reorganization

which took place several months before the survey was administered. As the transfer of leadership in

both cases involved an incoming minister with a different partisan affiliation than his predecessor,

the large proportion of neophytes in these institutions is consistent with a spoils-type approach to

personnel management.

Figure 3.16: Proportion of employees denoting that they had served the indicated number of years in
their institution

Another way of examining the institutionalization of career paths is to gauge perceptions

about job instability. It is reasonable to expect that institutions characterized by high levels of

turnover would also be characterized by an employee base highly apprehensive about losing their

posts. The data indicate that this is indeed the case. Figure 3.17 displays the proportions of

149
respondents in each country who indicated that it was highly probable, probable, improbable or

highly improbable that they would lose their post in the near future.

Figure 3.17: Perceptions of job security across the cases

As before, there is a clear distinction between the responses provided by the Bolivian public

employees and their counterparts in Brazil and Chile. Moreover, this survey item also produced fairly

distinctive responses from the Brazilian and Chilean respondents. In the Bolivian case, the vast

majority of respondents68%indicated that it was highly probable or probable that they would

lose their posts in the near future. This was almost four times the corresponding figure for Brazil and

nearly twice that for Chile. The Brazilian and Chilean respondents, on the other hand, diverged in

terms of their perceptions about just how unlikely it was that they would lose their posts in the near

future. A majority of the Brazilian respondents (51%) felt job security was essentially guaranteed,

these individuals responding that it was highly unlikely they would lose their posts in the near future.

The modal response for Chilean respondents was a bit more guarded: 45% of the Chilean

respondents indicated that it was improbable that they would lose their posts in the near future,

whereas only 15% indicated that such an outcome was highly improbable.

150
 As was true for the longevity of bureaucratic careers, cross-national variation in responses

about job stability was not overwhelmed by variation across institutions. Figure 3.18 displays

participant responses by institution. Out of the fourteen Bolivian institutions surveyed, in thirteen a

majority of respondents felt it was probable or highly probable that they would lose their job in the

near future. In Brazil and Chile, a majority of respondents responded likewise in only one of the ten

and two of the six institutions surveyed, respectively.

One of the important distinguishing features about the tenuousness of public employment in

Bolivia is its tight relationship with perceptions of politicization. As a general rule, institutions which

Bolivian bureaucrats characterized as suffering from high levels of political intromission in personnel

management and other daily tasks were also institutions in which large numbers of employees were

highly apprehensive about losing their posts in the short term. Importantly, this was not the case for

the Brazilian and Chilean institutions surveyed.

Figure 3.18: Perceptions of job security by institution

151
 Figure 3.19 shows the scatter plot of the median politicization score for each institution (x-

axis) by the proportion of employees indicating that they felt it was highly probable they would lose

their job in the near future (y-axis). In the Bolivian case, there is a clear positive relationship between

the two variables. The more institutions become politicized, the less stable bureaucratic career paths

become. In this respect, the data point to enduring patterns of machine politics in certain segments

of the Bolivian state, in which the litmus test of partisanship retains considerable importance in

staffing decisions deep within the administrative apparatus. In Brazil and Chile, on the other hand,

the stability of career paths does not seem to be keyed to levels of politicization. Thus, to the degree

that political pressure and influence in these cases impact upon personnel management, they would

seem to do so for relatively small segments of the employee base.

Figure 3.19: The relationship between perceptions of politicization and job instability

Note: dark solid lines denote fitted values from OLS regression; dotted lines denote fitted values from
locally weighted regression.

Another important component of the institutionalization of bureaucratic careers consists of

the frequency of internal versus external promotion (cf Evans and Rauch 1999). Public employees

who have the opportunity to work their way up the ranks in their institution should be more inclined

to invest in institution-specific skills and develop an espirit de corps than employees who find

themselves constantly transferred from institution to institution or who jump back and forth between

the public and private sectors.

152
 In this aspect of bureaucratic capacity, the clearest cross-national difference was that

encountered between the Chilean case, on the one hand, and the Bolivian and Brazilian cases, on the

other. Chilean bureaucrats were substantially more likely to have held previous posts in the same

institution before beginning their current job than were the Bolivian and Brazilian respondents. In

Chile, bureaucrats drawn from within the ranks of their current institution were encountered twice as

often as those drawn from other institutions in the public sector. Those employees not previously

holding a post within their institution were by in large recruited from the private sector. Horizontal

recruitment, by which I refer to the filling of posts with employees drawn from other institutions in

the public sector, was relatively rare.15 By contrast, the Brazilian bureaucrats surveyed were twice as

likely to have been recruited horizontally as they were to have been selected internally. In Bolivia,

horizontal recruitment was also more common than internal promotion, although the difference was

not as marked as in Brazil. Figure 3.20 displays the relevant proportions for the cases.

Figure 3.20: Internal vs. external recruitment in the cases

A final aspect of the institutionalization of bureaucratic career paths worth considering is the

adequacy of employee remuneration. This is conventionally thought to be an important predictor of

This finding fits with expectations. According to Marcel (1999), no provisions exist for horizontal
15

movements in the public service. While all institutions are governed by the same employment statute, each
must conduct its own personnel management (304).

153
the incidence of corruption. Public servants receiving a wage which covers the costs of basic

necessities are ostensibly less tempted to supplement their incomes through illicit means or otherwise

engage in risky activities which might threaten continuance in their posts (Becker and Stigler 1974,

Rijckeghem and Weder 2001).

Publicly available data might lead us to expect that civil servants in Bolivia, Brazil and Chile

would be rather satisfied with their wages. According to official figures, the ratio of the average

public wage to GDP per capita is 3.4 in Bolivia, 3.9 in Brazil and 4.2 in Chile.16 These figures are well

above the Latin American average of 2.5 calculated in the international survey by Schiavo-Campo et

al. (1997) and they underscore the value of a public sector post in the cases relative to other forms of

employment. Nevertheless, when queried in the survey about the reach of their wage, in all three

countries a majority of respondents declared that their wage was insufficient to cover the cost of

necessities64% of bureaucrats responding in this way in Bolivia, 69% in Brazil and 57% in Chile.

As real wage rates vary considerably across institutions, so do employee responses about the

ability of ones wage to cover basic costs. Figure 3.21 presents the breakdown of responses to this

question at the institutional level. In both Bolivia and Brazil, there was a remarkable level of variation

in the perceived adequacy of wages across institutions. In Bolivias Superintendency of Banks, for

instance, 69% of respondents declared that their wage was sufficient to cover the cost of necessities.

This compares to the Ministry of Sustainable Development, in which only 17.5% of participants

responded the same way. In Brazil, the highest reported satisfaction with wages was found in the

National Agency of Land-based Transport (ANTT), in which 55% of respondents declared that their

wage covered necessities. The low was found in the National Institute of Social Security (INSS), in

which only 14% of respondents said that their wage covered such costs. In Chile, the variation across

institutions was much more muted in this regard, although the National Institute of Agrarian

16 The figures for Bolivia and Chile come from Schiavo-Campo et al. (1997). The figure for the average public

sector wage in Brazil (in reais) comes from Ministerio de Planejamento, Oramento e Gesto (July 2004). To
produce the ratio of average public sector wage to GDP per capita, this figure was converted to dollars using
the (average) exchange rate for July 2004 and divided by GDP per capita for Brazil in 2004 according to data
provided in the World Banks World Development Indicators CD-ROM.

154
Development (INDAP) was a notable outlier in terms of the low proportion of respondents

indicating that their wages covered basic costs (26%).

In examining the various indicators of the institutionalization of bureaucratic career paths, it

is clear that there are important cross-national distinctions to be reckoned with. Overall, the Bolivian

public administration is characterized by significantly greater tenuousness in bureaucratic careers than

the other cases. This tenuousness, in turn, appears to be related to enduring patterns of patronage

politics. The Chilean case, by contrast, is clearly characterized by greater rates of internal promotion

than the other two cases. However, these important cross-national patterns notwithstanding, the

cross-institutional variation encountered for these indicators was still considerable, especially within

the Bolivian public administration. Moreover, for certain indicators, like the perceived adequacy of

employee wages, the cross-institutional variation clearly overwhelmed the cross-national variation.

Figure 3.21: Perceived adequacy of employee wages by institution

Corruption Control

155
 A final dimension of bureaucratic capacity on which we might evaluate the cases consists of

the efficacy of mechanisms providing for accountability and financial control. Here I consider two

aspects of such control: formal control, by which I refer to the ability of internal or external auditing

bodies to detect instances of wrongdoing, along with the capacity of prosecutors and the judicial

branch to bring culpable individuals to justice in a timely fashion, and informal control, the

willingness of public employees to keep their eyes and ears open for acts of corruption and report

wrongdoing themselves.17

The perceived strength of formal control was measured by two questions: one querying

respondents about the likelihood of an investigation taking place after a misappropriation of

resources and another querying respondents about the likely consequences of such an investigation

were it to take place. Figure 3.22 displays the two questions and the breakdown of responses by

country.

Bolivian public employees were the least convinced that formal institutions would react in

such a way as to detect and punish corruption. Nevertheless, it bears emphasizing that employee

responses were in general quite different than this countrys reputation for rampant informalism

might lead one to believe. Presented with the hypothetical scenario that an employee had

misappropriated resources from their institution, 38% of Bolivian respondents declared that an

internal audit, external audit or another type of investigation was highly probable as a result.

Another 43% declared that such an investigation was probable. All in all, only 19% of respondents

declared that an investigation was an improbable or highly improbable outcome.

17 Readers may note that my use of the term formal control is in many ways similar to the widely used concept

of horizontal accountability, defined as the existence of state agencies that are legally enabled and
empowered, and factually willing and able, to take actions that span from routine oversight to criminal
sanctions or impeachment in relation to actions or omissions by other agents or agencies of the state that may
be qualified as unlawful (O Donnell 1999, 38).

156
Figure 3.22: Employee perceptions of formal control across the cases

When asked about the likely consequences of such an investigation were it to take place,

55% declared that the culpable individual would probably be judged in a court of law (enjuiciada) and

punished (sancionada) for the misdeed, 32% declared that the culpable individual would probably be

judged in a court of law but that the individual would not be punished for his action and 14%

declared that the most probable result of the investigation would be that the culpable individual

would neither be judged nor punished for the misdeed. Overall, about half (51%) of the Bolivian

employees surveyed believed that a misappropriation of resources would probably lead to both an

investigation and a legal proceeding culminating in some form of punishment.

Brazilian public employees were slightly more convinced that formal institutional channels

would be used to actively investigate corruption. Moreover, they were substantially more convinced

157
that if an investigation were launched, the legal process would produce some form of punishment for

the transgressor. An internal audit, external audit or another type of investigation was considered a

highly probable outcome in the event of a misappropriation of resources by 46% of Brazilian

respondents and a probable one by 41%. When asked about the likely consequences of such an

investigation, 70% declared that the culpable individual would be judged in a court of law (julgada)

and punished (punida) for the misdeed, 24% declared that the culpable individual would be judged in

a court of law but that the individual would not be punished for his action and 6% declared that the

most probable result of the investigation would be that the culpable individual would neither be

judged nor punished for the misdeed. Overall, about two thirds (68%) of the Brazilian employees

surveyed believed that a misappropriation of resources would probably lead to both an investigation

and a legal proceeding culminating in some form of punishment.

The Chilean public employees were the most convinced that formal institutional channels

would be used to actively investigate corruption, and that, if an investigation were to be launched, the

legal process would produce some form of punishment for the transgressor. An internal audit,

external audit or another type of investigation was considered a highly probable outcome in the

instance of a misappropriation of resources by a hefty majority of Chilean respondents (62%) and a

probable one by 28% of respondents. When asked about the likely consequences of such an

investigation, 81% declared that the culpable individual would be judged in a court of law and

punished for the misdeed, 14% declared that the culpable individual would be judged in a court of

law but that the individual would not be punished for his action and 5% declared that the most

probable result of the investigation would be that the culpable individual would neither be judged nor

punished for the misdeed. Overall, 79% of the Chilean employees surveyed believed that a

misappropriation of resources would probably lead to both an investigation and a legal proceeding

culminating in some form of punishment.

When survey responses about formal control are decomposed at the institutional level, one

finds a great degree of variability within the Bolivian public administration but considerably less

158
variability within the Brazilian and Chilean public administrations. Figure 3.23 depicts the proportion

of individuals in each institution who replied that an investigation would be probable or highly

probable in the event of a misappropriation of resources and that, if an investigation occurred, the

process would likely culminate in a trial and legal punishment for the guilty party. In essence, these

figures represent the proportion of employees that believe formal institutions would react as they

ought to in the event that a serious act of corruption (a misappropriation) had taken place.

The figure below places in relief the wide range of viewpoints that exist within the Bolivian

administration regarding the efficacy of formal control mechanisms. The usual suspects inhabit the

high and low extremes. At one extreme, we again find the Superintendency of Banks outperforming

other institutions in Bolivia, 82% of respondents in this institution indicating that formal institutions

would react as they ought to in the event of a misappropriation. At the opposite extreme, we again

find the Ministry of Sustainable Development underperforming relative to all other institutions in

Bolivia, only 28% of respondents in this institution indicating that formal institutions would react as

they ought to. In between these two extremes, we also find a significant degree of variability across

institutions.

Variation within the Brazilian public administration is more muted than in the Bolivian case,

although here, too, one finds a significant divide between the top and bottom ranking institutions.

Employees in the newly created ANTT felt most convinced formal institutions would perform well,

85% of respondents in this institution indicating that the action of formal institutions would

probably produce an investigation and ultimately the punishment of the guilty party. At the opposite

end of the spectrum one finds the Superintendency of Private Insurance (SUSEP) and DNIT, 48%

and 49% of whose employees, respectively, indicated that the aforementioned outcome was the most

probable one. Half of the Brazilian institutions surveyed showed very little variation in this respect.

For the ANTT, the Ministry of Agrarian Development (MDA), the National Institute of

Colonization and Agrarian Reform (INCRA), the Central Bank (BC) and the Federal Revenue

159
Service (RF), the proportion of respondents replying in the indicated manner varied only between

77% and 85%.

Figure 3.23: Employee perceptions of formal control across institutions

The Chilean public administration exhibited even less cross-institutional variation in

perceptions about the efficacy of formal control. Employees were least confident that formal

institutions would work as they ought to in the Customs Service, where the vast majority (69%)

nonetheless believed that a misappropriation of resources would trigger an investigation and the

subsequent legal sanctioning of the perpetrator. Thus, in terms of faith in formal control

mechanisms, all of the Chilean institutions performed quite well.

160
 Formal institutional channels of monitoring and sanctioning are a crucial means of deterring

corrupt activity within the public sector. However, they are by no means the unique deterrents of

such activity. Also important in this regard is the corporate culture of the institutions themselves.

Public institutions whose employees are attentive to potential breaches of ethical and legal standards

and who are willing to come forward if aware of the existence of such breaches can be expected to

suffer lesser degrees of corruption than institutions not imbued with a similar corporate culture.

One of the questions included in the survey attempted to tap into employee perceptions

regarding the existence of this type of informal control within their institutions. Participants were

asked whether or not they believed that the majority of employees in their institution would file a

complaint (denuncia) if they were aware of an inappropriate use of public funds. Figure 3.24 displays

the results for this survey item by country and institution.

Two observations can be made from the figure. Firstly, the differences across countries are

relatively minor, with the proportion of respondents declaring that the majority of their colleagues

would file a report varying from 41% (Bolivia) to 51% (Brazil). Secondly, the differences across

institutions are relatively large, especially within the Bolivian and Brazilian public administrations.

In these countries, a number of institutions stand out as outliers in terms of the responses

provided by their employees. Not surprisingly, one again finds the SBEF well ahead of it peers in

Bolivia. In this institution, 72% of respondents felt that the majority of their colleagues would file a

complaint if privy to the occurrence of a misuse of funds. Next in order in was the Productive Social

Investment Fund (FPS), in which only 54% of respondents replied similarly.

In Brazil, the Central Bank and DNIT clearly stood apart from the rest. In the former, 76%

of respondents declared that the majority of their colleagues would file a complaint. The percentage

for the institution with the second highest placement was substantially lower63% (SUSEP). DNIT,

on the other hand, was noteworthy for the low percentage of respondents who believed a majority of

their colleagues would file a complaint. Only 26% of participants in this institution responded this

161
way. The institution with the penultimate placement on this survey item in Brazil was significantly

higher45% (MDS).

Figure 3.24: Employee perceptions of informal control across and within the cases

The examination of the various indicators of corruption control leads one to conclude that

cross-institutional variation is at least as important as cross-national variation. Although the Chilean

case exhibited rather moderate internal variation with regard to perceptions of the efficacy of formal

and informal control, the Bolivian and Brazilian cases exhibited substantial variation on both counts.

162
Thus, at least for Bolivia and Brazil, the central tendencies of the country cases on corruption control

seem to be overwhelmed by their own institutional diversity.

Taking Stock of the Survey Responses

I began the chapter by elaborating some of the consequences of institutional diversity for

studies of bureaucratic quality. Given the well established reputational rankings of the Bolivian,

Brazilian and Chilean states in this regard, I argued that should we find that within-state, institutional

diversity in these cases overwhelms cross-national variation, then we would have strong prima facie

evidence in favor of the view that institutional diversity probably overwhelms national diversity more

generally. Given such a finding, the potential pitfalls associated with a strictly national-level approach

to the study of bureaucratic capacity would be well worth considering.

What did I find? On two of the three dimensions of bureaucratic capacitycivil service

politicization and corruption controlthe diversity encountered within the three public

administrations was generally greater than that encountered between them. This was especially so for

perceptions of the politicization of the civil service. On one dimension of bureaucratic capacitythe

institutionalization of bureaucratic career pathscross-national differences appeared to outweigh

institutional differences. Even in this instance, however, certain survey items exhibited greater cross-

institutional variation than cross-national variation. Overall, the survey responses thus suggest that

concerns about the nation-state-as-unit-of-analysis approach to the study of state capacity are well

founded.

Nevertheless, the empirical examination I have conducted thus far begs an important

question: namely, how do the three dimensions of capacity we have identified contribute to the

performance of public institutions? That is to say, taking as given the fact that institutional diversity is

considerable, how do the differences encountered among the state institutions affect their ability to

provide high quality goods and services to citizens? I take up this question in the following section.

163
Therein, I examine the impact of civil service politicization, the institutionalization of bureaucratic

career paths and mechanisms of corruption control on two measures of performance: perceptions of

institutional efficiency and the prevalence of political corruption.

3.5: The Impact of Bureaucratic Capacity on the Performance of Public Institutions

Bureaucratic Capacity and Institutional Efficiency

Imagine for the moment that we had the power to transform the basic structure of public

institutions in our cases in whatever way we saw fit. With such power, we would be able to answer

very important questions about the impact of different institutional arrangements on bureaucratic

performance. One question we might seek to address would concern the magnitude of the efficiency

gains associated with a transformation of a given institution from an ideal-typical botin poltico to an

ideal-typical Weberian bureaucracy.

Note, however, that even if we had the power to effect such a transformation, the exercise

might tell us very little about the efficiency gains to be had in the real world from efforts to introduce

or reinforce features of Weberian public administration. The reason for this is that certain constraints

built into a countrys political system and/or social structure may make either the ideal-typical botin

politico or the ideal-typical Weberian bureaucracyor bothunattainable in that context. The most

useful way to proceed would be to make local comparisons instead. In this regard, we would like to

estimate the efficiency gains associated with a transformation from a hypothetical institution A,

corresponding to the least Weberian institution possible in a given national context, to a

hypothetical institution B, corresponding to the most Weberian possible in that context. In this

section, I draw upon the survey data to accomplish this task for our three cases.

One of the survey items confronted respondents with the following question: How would

you characterize the efficiency with which your institution performs its duties? Potential responses

164
were highly efficient, efficient, inefficient and highly inefficient. In order to examine the

determinants of (perceptions of) institutional efficiency, I performed an ordered logistic regression in

which participants responses to this question constituted the dependent variable.

As independent variables, I utilized a number of the survey items contained under the

rubrics of politicization of the civil service, institutionalization of career paths and corruption control.

Under the rubric of the politicization of the civil service, the predictor variables were the following:

the proportion of employees in the respondents institution that belong to a political party, the

relative ranking given by the respondent to political factors as determinants of dismissals in the

institution, the relative ranking given by the respondent to political connections as a determinant of

promotions in the institution and the respondents perception of the importance of political

connections in hiring in the institution.

Under the rubric of the institutionalization of bureaucratic career paths, the predictor

variables were: the proportion of employees in the respondents institution having served less than 1

year (a measure of turnover), the respondents assessment of the probability of losing his/her job in

near future, the proportion of employees in the respondents institution having been promoted

internally (i.e. whose previous job was in the same institution) and the respondents assessment of the

adequacy of his/her wage to meet basic costs.

Finally, under the rubric of corruption control, I included a variable measuring the

respondents perception of the likelihood of investigation resulting from misappropriation of

resources in the institution, an indicator variable denoting whether or not the respondent believed

that the majority of employees in his/her institution would file a complaint if aware of

misappropriation and an indicator variable denoting whether or not the respondent professed faith in

the ability of internal control mechanisms in the institution to hold both high level and low level

employees to account.

165
 In addition to the aforementioned covariates, I also included country dummy variables in

order to control for the fact that employee perceptions of what constitutes an efficient institution

might differ substantially across nations.

Table 3.6 presents the results of the ordered logistic regression analysis. Employee responses

on the dependent variable were coded in ascending order (highly efficient=1, efficient=2, etc.)

so variables with positive coefficients are associated with greater perceived inefficiency, whereas variables

with negative coefficients are associated with greater perceived efficiency.

Nearly all the variables in the politicization category were of the expected sign and

statistically significant (greater politicization being associated with greater institutional inefficiency) as

were all of the variables in the corruption control category (greater corruption control being

associated with greater efficiency). The variables under the institutionalization of career paths

heading were mixed in terms of their relationship to perceptions of efficiency. As expected, high

turnover was associated with greater bureaucratic inefficiency. Nevertheless, respondents who

declared it was improbable that they would lose their post in the short term were significantly more

likely to give their institution high ranks for efficiency than individuals who declared such an

outcome was highly improbable. This latter finding points to the possibility that extreme levels of

job tenure are likely to undercut institutional performance, as the incentive to perform at a high level

is weakened by unconditional tenure.

How much of an efficiency gain is feasible in public institutions in the cases through the

strengthening of Weberian features within said institutions? Using the statistical model estimated

above, I attempted to address this question through simulation analysis. The analysis proceeded in

two steps.

166
Table 3.6: Ordered logistic regression of perceived institutional efficiency on covariates

167
 In the first step, each institution-varying covariate was held at the value observed for the

worst scoring (least Weberian) institution in a given country (on that covariate) while each

individual-varying covariate was held at the median value encountered for the worst scoring (least

Weberian) institution in said country (again, on that covariate).18 These covariate values characterize

the working conditions and perceptions of a public employee in the least Weberian institutional

environment conceivable within the public administration of a given country. Given these covariate

values for each country, I calculated the probability that an individual so categorized would classify

his institution as highly efficient, efficient, inefficient and highly inefficient.

In the second step, each institution-varying covariate was held at the value observed for the

best scoring (most Weberian) institution in a given country (on that covariate) while each individual-

varying covariate was held at the median value encountered for the best scoring (most Weberian)

institution in said country (on that covariate). These covariate values characterize the working

conditions and perceptions of a public employee in the most Weberian institutional environment

conceivable within the public administration of a given country. Given these covariate values for

each country, I again calculated the four relevant probabilities. Having calculated these probabilities

in the two scenarios for each country, I was then able to estimate the maximal impact that Weberian

reform was likely to have on perceptions of institutional efficiency in each country.

18By worst scoring or least Weberian, I refer to the institution that is most politicized, has the weakest
institutionalization of bureaucratic career paths or has the weakest corruption control mechanisms, depending
on the particular question.

168
Figure 3.25: Weberian reform and institutional efficiency, simulation results

most Weberian
- - least Weberian

Figure 3.25 presents the results of the simulation analysis. The potential impact of

institution-strengthening, Weberian reform in Bolivia and Brazil was estimated as being quite large. A

Bolivian public employee in the least Weberian institution possible in that country, for instance, was

169
estimated as having a .02 probability of classifying his institution as highly efficient, a .48

probability of classifying his institution as efficient, a .47 probability of classifying his institution as

inefficient and a .03 probability of classifying his institution as highly inefficient. The median

employee in the most Weberian institution possible, on the other hand, was estimated as having a .34

probability of classifying his institution as highly efficient, a .63 probability of classifying his

institution as efficient, a .04 probability of classifying his institution as inefficient and a .001

probability of classifying his institution as highly inefficient. To put these results another way, the

odds are 1 in 2 that the median employee in the least Weberian institution possible would classify the

institution as inefficient or highly inefficient while the odds are 1 in 24 that the median employee in

the most Weberian institution possible would do so. The impact of a Weberian transformation in the

Brazilian case was of a similar magnitude, as can be seen in the figure. In the Chilean case, the

potential impact of Weberian reform was estimated as being much smaller than in the other two

cases, but nonetheless important in shifting upwards the proportion of employees who classify their

institutions as highly efficient. Clearly, the institutional diversity found within the cases on the

three dimensions of bureaucratic capacity has a major substantive impact on perceptions of

bureaucratic efficiency.

Bureaucratic Capacity and Political Corruption

The question remains as to whether or not the institutional outcomes we have identified--

politicization, bureaucratic career paths and corruption control--matter in terms of facilitating or

inhibiting political corruption. This is an issue of major concern in the cases. Bolivia and Chile, for

instance, have undergone similar efforts of institutional engineering in recent years to address the

unsavory nexus between political patronage and corruption. These efforts are exemplified by the

Institutional Reform Project (PRI) and the Political Parties Law (both under the rubric of the

National Integrity Plan) in Bolivia and the New Labor Accord and the Law on Transparency, Limits

170
and Control of Campaign Spending in Chile. In both cases, architects of the reforms identified the

nature of political financing as the root cause of certain shortcomings in public sector performance.

Along these lines, it was argued that civil service reform and political finance reform are necessarily

related, as one of the underlying motives of the politicization of the state has been the desire of

political leaders to exploit their powers of appointment in order to funnel state resources to political

campaigns. Reformers have voiced similar concerns in Brazil as of late, due in no small part to a

major Congressional vote-buying and illicit campaign financing scandal that has enveloped the ruling

Workers Party government.

The next two chapters of the dissertation devote themselves exclusively to the theme of

illicit party financing through the public bureaucracy and its relationship to national-level, electoral

institutions. In this section of the chapter, by contrast, I briefly examine the roots of cross-

institutional variation in corruption. Specifically, I examine here how intrastate differences in the

three dimensions of bureaucratic capacity I have identified affect levels of political corruption found

in public institutions.

I addressed this question in the following way. One of the randomized response questions

queried respondents about the frequency with which the resources of their institution were used for

the benefit of political parties.19 I used the randomized response estimate for this question for each

institution as a dependent variable in a series of linear regression models. Given the uncertainty in the

estimation of the dependent variable, I conducted both simple OLS regressions as well as variance-

weighted least squares regressions.20

The independent variables consisted of factor scores (normalized to lie between 0 and 1)

generated from a three-factor confirmatory factor analysis. The three factors consisted of the three

dimensions of bureaucratic capacity that I have considered in this chapter. Figure 3.26 displays the

path diagram for the CFA model along with variable descriptions for the observable indicators. A

19 The question format was as follows: A: The resources of my institution are commonly used for the benefit of

political parties, B: The resources of my institution are not commonly used for the benefit of political parties.
20 The variance-weighted regressions weighted observations in accordance with the inverse of the standard

error of the RR estimate for each institution.

171
table with estimates of the factor loadings and descriptions of model fit is included in the appendix

of this chapter.

Figure 3.26: Three factor confirmatory factor analysis model of politicization, career paths and
corruption control

In the OLS estimation, the politicization variable and a constant alone explain 47% of the

variation in the aggregate responses about partisan resource use (see figure 3.27). The coefficient on

this variable was positive and statistically significant for each estimation performed. Introducing the

other two indicators and country dummies to the regression equation increased explained variation

by only 5%. The variance-weighted regressions produced similar findings. The coefficient on

172
politicization was positive and statistically significant in both regressions, with the marginal impact of

politicization dwarfing that of the other indicators.

Thus, it seems fairly safe to conclude that the politicization of state institutions was indeed a

prelude to the partisan use of institutional resources in our cases. The preoccupations of the Bolivian

and Chilean reformers appear to be well-founded in this regard. Somewhat surprisingly, I was unable

to detect any statistically significant influence of the institutionalization of bureaucratic career paths

or levels of corruption control in the cross-institutional analysis. Nevertheless, the patient reader will

discover in Chapter 5in which I examine the determinants of illicit partisan use at the individual-

levelthat certain variables which fall under these headings do indeed have a significant impact on

the individual choice to engage in partisan resource use.

Figure 3.27: Politicization of institutions and partisan resource use

3.6: Conclusion

This chapter departed from a simple premise: in order to understand the basic features of

the bureaucratic marketplace in our cases, it is important to examine cross-institutional differences as

well as cross-national ones. In exploring such cross-institutional variation, the chapter has made two

173
main points. Firstly, cross-institutional variation in bureaucratic capacity is often greater than cross-

national variationeven when one limits oneself to the analysis of national bureaucracies with very

different reputations for bureaucratic capacity. Secondly, institutional reforms of the old-fashioned,

Weberian kind still hold great promise for unlocking significant efficiency gains within the public

sector in Latin America.

As discussed in the text, one of the factors responsible for the diffusion of NPM reforms

was the belief that the epoch in which clientelism posed the primary threat to honest and efficient

government had passed, and thus the dangers associated with dismantling certain core features of the

Weberian state was minimal. The data lead me to view such claims with skepticism. In section 3.5, I

showed that the manifestations of Weberian bureaucracy were generally associated with greater

institutional efficiency, not less, and that institutions which suffer from significant political

intromission in basic tasks are also likely to be cash cows, consistently milked to provide resources

for political parties.

The subsequent chapters of the monograph shift the focus from the features of the

bureaucratic marketplace to those of the political marketplace. The approach therein is both more

macro and more micro than that adopted in the current chapter. It is more macro because my

primary concern rests with addressing a major debate about how certain aspects of national-level

electoral institutions impact the prevalence of political corruption in the countries of Latin America.

Thus, the explanatory emphasis in this section of the monograph shifts gears to the nation-state level.

However, these chapters are also more micro because I model theoretically and examine empirically

the behavior of individual bureaucrats, not aggregations such as nation-states or public institutions.

In so doing, I develop a general equilibrium approach to the problem of political corruption, one that

endogenizes difficult choices about illicit resource use to the relevant features of both the political

and bureaucratic environments. Let us now turn to this endeavor.

174
Appendix

Table 3.7: Estimates of factor loadings for the three factor CFA model

175
 Chapter 4
Ballot Structure, Political Corruption and the
Performance of Proportional Representation

176
1 Introduction

Throughout the Americas, issues of democratic institutional engineering are at the forefront

of heated political debate. In numerous capitals in the hemisphere, one can find lively

arguments about how specific changes in the configuration of formal institutions may lead

to lasting improvements in human welfare. Perhaps none of these debates has attained the

level of prominence, nor demonstrated as great a degree of schizophrenia, as the arguments

about the relationship between electoral institutions and clean government.

The cases of Brazil and Argentina are illustrative examples of this state of aairs. Among

political commentators, social scientists and legislators in Brazil today, few topics are the

source of as much sustained attention as the relative merits of political reform. A cen-

tral component of the reform proposals under consideration consists of the replacement of

Brazils current electoral system, characterized by open list proportional representation,

with a closed list version of the same. Many exponents for reform argue that political fi-

nancing in Brazil has become irredeemably corrupt, and that this corruption, in turn, owes

it origins to the exigencies of political marketing inherent to an open list system. Recent

revelations of widespread illicit campaign financing involving Brazils ruling Workers Party

(PT) and its coalition partners, apparently coordinated by individuals in the upper echelons

of the party organizations, have added fuel to the fire for supporters of electoral reform.

And yet the perspective on the other side of the Paran River is very dierent. In

Argentina, the issue of political reform is no less salient than it is in Brazil. Nevertheless,

the stance of clean government activists in this country towards electoral reform is exactly

opposite that of their Brazilian counterparts. According to many Argentine reformers, the

countrys system of listas sabanas, or closed list proportional representation, is the principle

culprit to blame for entrenched political corruption and machine politics to be found there.

177
Revulsion with corruption schemes in which the public administration has been used as an

illicit source of party financing, principally orchestrated by factions of the ruling Peronist

party (PJ), has made this a hot button issue in Argentina.

In no way is the schizophrenia characterizing the debate about electoral reform limited to

Brazil and Argentina. The depiction of open list proportional representation and other so-

called candidate-centric electoral systems as catalysts for corruption related to the financing

of politics can be found in past and present debates in countries as diverse as Italy, Japan,

Colombia and Taiwan. The portrayal of closed list proportion representation (and party-

centric electoral systems more generally) as a hindrance to accountability and a facilitator of

illicit financing, on the other hand, can be found in past and present debates in countries such

as Bolivia, Honduras and Venezuela. The one constant in these debates is that reformers

generally agree that their own electoral system is the least attractive among the menu of

potential options. Clearly, they cannot all be right.

What is going on here? What is the real relationship between electoral reform and

political corruption and why is the policy debate so seemingly schizophrenic? This chapter

seeks to cut a way through the fog. Specifically, the chapter presents a novel account of the

relationship between ballot structure (i.e., the manner in which citizens cast their votes) and

corruption related to the financing of politics. It seeks to explain why the received academic

wisdom tying candidate-centric electoral systems to political corruption is as compelling as it

is, but also shows how an incomplete picture of the causal pathways linking ballot structure

to corruption has led the predictions generated by this paradigm to miss their mark.

The central argument is that the eects of a reform of list structure can be decomposed

into two components: a demand side component and a supply side component. In essence,

the two components represent countervailing eects of electoral reform.

178
 On the one hand, a rich body of scholarship in comparative politics demonstrates that

electoral systems which encourage intraparty competition, of which open list PR is one

variant, tend to produce costly campaigns and generate high demand among politicians for

electoral resources. I will discuss this work in more detail in what follows. However, the

important point to remember about this literature is that it establishes a link between ballot

structure, campaign expenses, and the willingness of politicians to accept illicit monies to

meet those expenses. This is the demand side linkage between ballot structure and political

corruption.

On the other hand, this chapter will demonstrate that there is another channel through

which ballot structure influences levels of political corruption. The channel consists of

the linkage between ballot structure and political career control, and the implications that

such control holds for the behavior of Latin Americas politicized bureaucracies. Electoral

systems which magnify leadership control, such as closed list PR, give party leaders a

means of leverage vis--vis politically oriented bureaucrats in the public administration.

With such leverage, it is easier to convince bureaucrats to engage in risky behavior for the

benefit of their party, thus facilitating political corruption. This is the supply side linkage

between ballot structure and political corruption. It considers the relationship between

ballot structure and the ability of parties to extract resources from the state. The linkage

is concerned with the manner in which the resources of political corruption are actually

supplied, not simply the level at which they are demanded.

From this perspective, a move from open to closed lists will dampen the demand side

but intensify the supply side. A move from closed to open lists will have the opposite

eect. Due to the existence of countervailing pathways through which list structure aects

political corruption, the schizophrenia characterizing the policy debate is perhaps not all

179
that surprising after all.

The chapter is organized as follows. Section 2 presents the dominant paradigm for

thinking about the relationship between ballot structure and political corruption. Section

3 discusses a number of country cases in Latin America that expose serious shortcomings

in the paradigm. Section 4 argues for the importance of the supply side perspective and

presents a formal model of supply side political corruption. The model, which utilizes a prin-

cipal agent framework with hidden knowledge, explicitly considers how dierences in ballot

structure may facilitate or impede corrupt contracts from forming between party leaders

and militants in the public administration. Section 5 details the observable implications of

the model. The final section concludes.

2 Ballot Structure and Corruption: The Traditional Paradigm

With injustice to the wide variety of ballots found around the world, one can categorize elec-

toral systems as falling into two types: those which allow voters to choose among candidates

and those which limit voters to a choice among closed party lists. Electoral systems which

allow voters to cast their votes for one or more candidates, such as open-list proportional

representation (OLPR), the single non-transferable vote (SNTV), the single transferable

vote (STV) and single member simple plurality (SMSP) have been associated with can-

didate strategies based on cultivating a personal vote and generating narrowly focused,

parochial policy outputs (Cain et al. 1987, Carey and Shugart 1995, Ames 1995, Mainwar-

ing 1999). Moreover, scholars have singled out two variants of the above systems, OLPR

and SNTV, as being highly propitious to political corruption as I have defined it in chapter

1.1 The purported link between ballot structure and political corruption is based on the

1
Political corruption is defined as being the abuse of public oce to advance a political goal, usually
entailing illicit financial support of a favored candidate, faction, party organization or coalition.

180
high demand for campaign resources that these systems generate.

In OLPR and SNTV systems (as well as STV systems), candidates for legislative posts

may compete simultaneously against candidates from other parties as well as members

of their own party list. Forced to dierentiate themselves from opponents who bear the

same party label, candidates must impress upon voters the unique virtues which only they

bring to the table. In Cox and Thies words, in such systems candidates face a product

dierentiation problem (Cox and Thies 1998, 271). The mechanisms by which candidates

dierentiate themselves from fellow listmates often involve the clientelistic distribution of

state resources, the financing of costly local-level get out the vote eorts by political

operatives, or both. Importantly, candidates generally cannot dierentiate themselves based

on programmatic dierences because the stiest competition they face is from members of

their own party. Similarly, relying on common pool resources of the party (including the

party label) is a losing strategy because within-list competitors have access to the same

resources. As a result, intraparty competition pushes candidates to cultivate a personal

following among voters, which in general is quite costly and often involves recourse to

corrupt exchanges. The argument can be illustrated using examples from Brazil, Italy and

Japan, the three cases which have spawned the literature on ballot structure and corruption.

2.1 Brazil: OLPR and the Politics of Clientelism

In Brazil, elections for the Chamber of Deputies, state level assemblies and municipal coun-

cils provide voters with the option of casting a vote for a particular party or coalition as

well as an individual within the list of their choice. Seats are apportioned to lists using the

dHondt method of proportional representation (in accordance with the sum of votes cast

for the list and individual candidates within the list). Within lists, seats are distributed

181
to candidates in accordance with their relative vote shares. The vast majority of Brazilian

voters exercise the option to cast a vote for a particular candidate (Samuels 1999). Not sur-

prisingly, studies of voting behavior in Brazil have noted the predominant role of candidate

characteristics in vote choice (Meneguello 1995, Silveira 1998). Due largely to the influence

of OLPR, most Brazilian parties are relatively weak political actors and the Brazilian party

system has been characterized as feckless (Mainwaring 1995).2

Brazilian deputies are elected in large, multimember districts whose boundaries are

isomorphic with those of the state in which they serve. District magnitude varies from 8 to

70. As a response to the institution of OLPR and high magnitude districts, deputies fashion

the boundaries of informal bailiwicks and transform them into their personal electoral turf.

Cultivating a loyal following in ones bailiwick is the be all and end all of political survival for

Brazilian deputies (Ames 1995). Because bailiwicks are under constant threat of invasion

by competing candidates, especially candidates from a deputys own party, providing a

steady stream of resources to ones home base is a crucial component of maintaining political

support (Ames 2001, Mainwaring 1991, Geddes and Neto 1999).

The resources used by deputies take various forms, although two of the most common

are narrowly targeted public works programs and public sector jobs for supporters. The

hijacking of social programs at the local level and the politicized distribution of need-based

benefits are also common. Corruption, too, is often an intrinsic part of the process of bring-

ing resources to ones home base. The budgetary powers of deputies give them substantial

leeway to tack particularistic amendments onto the national budget. One consequence of

this has been the development of a formalized system of influence peddling between Brazils

large civil construction firms, members of Congress and bureaucrats in the federal pub-

2
For an important critique of this position, especially with respect to the power of party leaders in the
legislative arena, see Figueiredo and Limongi (1999).

182
lic administration (Bezerra 1999, Carvalhosa 1995). Two common manifestations of this

relationship are quid-pro-quos between firms and deputies based on the drafting of favor-

able amendments in exchange for campaign donations and lobbying eorts on the part of

deputies designed to keep budgeted resources flowing to projects of interest to construction

firms. Along these lines, recent research shows that having public works funds allocated

to a deputys bailiwick can be especially valuable since contracts are often eectively lever-

aged for campaign donations (Samuels 2002). Moreover, investigations have shown that in

a number of instances deputies have set up philanthropic and other organizations as fronts

in order to directly capture budgeted resources themselves (Krieger et al. 1994, Fleisher

1997).

Keeping a bailiwick loyal is an expensive proposition. In this regard, campaign spending

in Brazil is often deployed to great eect through networks of local political operatives

(cabos eleitorais) who canvass the streets on the candidates behalf (Diniz 1982, Gay 1994).

These individuals engage in myriad activities designed to create a loyal support base for the

candidate, from distributing campaign pamphlets to lobbying neighborhood associations for

endorsements. Occasionally spending is also dedicated to buying o local political bosses,

thus transferring to the candidate votes under the bosses control (Ames 2001, cf. Nunes

Leal 1975). A consensus exists that much of the money funneled into such activities is

unreported, perhaps as much as two thirds of the total (Aguiar 1994, Kinzo 1998). It should

thus be no surprise that in Brazils OLPR system, political financing is often considered to

be coterminous with corruption (Dias 2004).

183
2.2 Italy: Preference Voting and the Persistence of Patronage

Research on the impact of intraparty preference voting in Italy paints a similar picture.

From the end of WWII until 1994, elections for representatives of the Italian Chamber

of Deputies utilized a combination of proportional representation in multimember districts

with an intraparty vote. Under this system, Italian voters were required to indicate their

preference for a party competing in their constituency. Additionally, they had the option of

indicating a preference for up to three or four candidates within the list they had chosen.

The existence of intraparty preference voting in Italy has been associated with low

party cohesion and highly personalistic linkages with voters (Katz 1985). Authors have

argued that various manifestations of bad government found there were endogenous to

politicians need to cultivate personal followings (Golden 2003). From this vantage point,

a number of features of the Italian bureaucratic environment which made bad economic

sense endured precisely because of their political value. Complex bureaucratic procedures

and excessively detailed legislation were invaluable to legislators because they created a

high demand for constituency service. Breaking through self-created bureaucratic obstacles

became an important means of political advancement.

In addition to making bureaucrats act in favor of constituents, an important component

of political success consisted in transforming constituents into bureaucrats, i.e. procuring

public sector jobs (Cassese 1993, Chubb 1982). Personal networks established by politicians

in the public bureaucracy became an important mechanism for raising illicit moniesgotten

primarily from kickbacks and embezzlementin order to finance political campaigns (della

Porta and Vannuci 1999). Such resources were a critical input in expensive campaigns

designed to dierentiate candidates from their listmates. In fact, the lions share of recent

academic work on corruption in Italy has focused precisely on the connection between the

184
demand for campaign funds created by intraparty preference voting and political corruption

(Chang 2005, Golden and Chang 2001, Rhodes 1997). So dependent had the Italian political

process become on illicit sources of finance that soon after the Tangentoli investigations of

the early 1990s dried up these funds, a number of parties went bankrupt (Newell 2000). As

the pressures to cultivate a personal vote were particularly intense for members of the long

dominant Christian Democratic Party, this organization was especially riven by accusations

of corruption related to campaign funding and, in the end, destroyed by them (Golden and

Chang 2001).

2.3 Japan: The SNTV and Money Politics

The literature on money politics in Japan shows remarkable congruence with the accounts

of political corruption in Brazil and Italy. From 1955 to 1993, Japans electoral system

for the lower house combined medium-sized, multimember districts with a SNTV ballot.

Japans larger parties, especially the hegemonic Liberal Democratic Party (LDP), routinely

fielded multiple candidates in each district. Since voters could cast only a single vote

and were routinely presented with more than a single candidate from the larger parties,

the personal characteristics of politicians came to play an important role in vote choice

(Miyake 1999).3 Intraparty competition encouraged by the SNTV system with multimem-

ber districts produced a remarkably personalistic and expensive style of conducting political

campaigns (Curtis 1988, Hrebenar 2000, ch.3, Fukui and Fukai 1999). In response to the

need to dierentiate themselves from their listmates (and stringent regulations which pro-

hibited routine forms of campaigning), politicians developed extensive networks of personal

supporters called koenkai (Bouissou 1999).

3
Unlike Brazil and Italys OLPR systems, in Japans SNTV system voters were forced to cast their votes
only for candidates. Party label voting was not an option.

185
 The koenkai are made up of hundreds of small groups devoted to wide ranging activities

and interests (climbing, karaoke, etc.) which share the common feature of being financed

by a single elected representative. They are ubiquitous in Japanese political life.4 Koenkai

members often enjoy substantial benefits of membership, including bus tours and cruises

provided at a charge well below the market rate (the politician pays the remaining balance).

Although not the norm, occasionally koenkai are used as a means of distributing cash to

voters (Fukui and Fukai 1999).

The system of generating personal support through koenkai is obviously one that

entails great cost. Numerous authors have noted the dubious origins of funds that make

their way into the koenkai (Bouissou 1997, Hrebenar 2000, ch.3, Reed 1996, Woodall 1996).

By one estimate, as much as two thirds of the money going into the Japanese political

system comes from illegal contributions (Bouissou 1997). Much of this money originates in

Japans massive construction industry. Not surprisingly, favorable intervention in the pro-

cess of public procurement was a commonplace means of raising campaign funds, especially

for members of the LDP (Woodall 1996). So systematized was this process that retiring

bureaucrats from the Ministry of Construction often drew upon their connections with con-

struction firms to finance their own political careers and improve the electoral prospects of

their faction (op cit., ch.2).

2.4 Synthesis

Using the stylized facts which emerge from our discussion of intraparty competition and

political corruption in Brazil, Italy and Japan, we can flesh out the main components of the

traditional paradigm. There are four steps to the argument. First, ballot structure gives

4
According to one authors assessment, almost all Japanese politicians have their own koenkai (Bouissou
1999). This author presents data showing that membership in koenkai hovers slightly below 20% of the
Japanese electorate.

186
rise to a particular mode of evaluating candidates. Ballots which present voters with a

choice of multiple candidates from the same party for the same oce increase the weight of

individual characteristics in vote choice. Second, given the mode of evaluation adopted by

voters, candidates choose a campaign style which is optimal given the anticipated campaign

styles of their opponents, especially opponents belonging to their same party. If voters

care about individuals, not party programs, and they evaluate favorably those candidates

who demonstrate considerable largesse with the electorate, then campaigns based around

the particularistic distribution of resources are the predictable result. Third, the campaign

style adopted by candidates determines the cost of conducting campaigns. Campaigns

based on cultivating a personal vote, according to this logic, are resource intensive and

significantly more costly than those based on programmatic grounds. Finally, the cost of

campaigns determines the propensity of politicians to engage in acts of corruption. The

more demanding the requirements of campaign finance, the more likely are politicians to

engage in corrupt behavior. Figure 4.1 displays the various linkages in the traditional

paradigm.

Figure 4.1: Ballot Structure and Political Corruption: The Traditional Paradigm

The traditional paradigm obviously has much to recommend it. It identifies a specific

trait of certain electoral systemsintraparty competitionand shows that in a number of

important countries which share this trait, remarkable commonalities in candidate behavior

187
and the incidence of political corruption are to be found as well.5 These commonalities are

all the more remarkable when one considers the vast geographic, cultural and socio-economic

dierences among the cases which inspired the traditional paradigm. However, obtaining

valid inference about the independent eect of the trait requires more than showing that in

otherwise distinct contexts, the appearance of the trait is associated with similar outcomes.6

If intraparty competition is indeed the principal catalyst for political corruption, then we

should expect to find that in countries whose electoral rules proscribe intraparty competi-

tion and place list order in the hands of party leaders, all else equal, political corruption is

a less prominent part of the political process. Most importantly from the vantage point of

the traditional paradigm, we would expect to find that electoral systems of the latter type,

to the extent that they do exhibit patterns of political corruption, should exhibit political

corruption in spite of their electoral rules, not because of them. That is to say, the tra-

ditional paradigm would lead us to believe that in closed-list, party centric systems where

political corruption is evident, its incidence must be related to background factors orthog-

onal to the electoral rules themselves. Observations of electoral systems which depart from

these expectations are anomalies which present a fundamental challenge to the traditional

paradigm. It is to these anomalies that I now turn.

5
Arguments about the link between intraparty competition and money politics are not made only by
students of the Brazilian, Italian or Japanese political systems. Similar arguments have been made about
the eect of the SNTV in Korea and Taiwan (Grofman et al. 1999, Cheng and Haggard 2001). Pervasive
political clientelism in Colombia, too, has been attributed to the eects of intraparty competition resulting
from this countrys factional list system and largest remainders version of PR (Archer and Shugart 1997).
6
Considered as a collective enterprise, the contributions to the traditional paradigm have selected their
cases based on the value of the dependent variable (Brazil, Italy and Japan are ostensibly all cases of high
political corruption). Selection bias of this sort can make it very dicult, and in some cases impossible (no
variance designs), to make inferences about causal eects (Geddes 1990, King et al. 1994). Nevertheless, a
number of studies within this vein have ingeniously exploited variation across politicians in corruption and/or
campaign spending in order to test postulates of the traditional paradigm (Cox and Thies 1998, Golden and
Chang 2001, Samuels 2001). Unfortunately, as these analyses are large-n studies limited to a single country,
the principal culprit for financial corruption, the electoral system, remains constant throughout.

188
3 Cracks in the Faade of the Traditional Paradigm: Party-

centric Electoral Systems in Latin America

Irrespective of their region of expertise, contributors to the traditional paradigm have done

their readers the service of being quite explicit about their causal claims regarding intraparty

competition and political corruption. Thus, Mainwaring says of the Brazilian experience:

Because there is such a premium on individual campaigning and because significant benefits

accrue to winning, PR with an open list has encouraged massive individual spending and

financial corruption" (1991, 27). In his study of corruption in Italy, Chang states political

corruption becomes a necessary evil because intraparty competition triggers candidates

needs for illegal resources to finance their campaigns (1). Reed echoes the point in the

Japanese context: The need to develop and sustain ones personal vote is the basic cause

of Japans problems with the high cost of campaigning, which, in turn, is the primary cause

of political corruption (25).

The claims of the authors cited above invite us to consider more carefully issues of

causality and inference. Following King et al. (1994), let us think of these claims in the

following way. Imagine that we could go back in time and rewrite Italian history an ar-

bitrary amount of times in any way we wanted. We might be tempted to conduct the

following experiment. We would go back to the end of WWII and, leaving everything else

unchanged, implement closed list PR (CLPR) instead of OLPR. We would perform this

intervention numerous times, each time carefully measuring the level of political corruption

which emerged with the passage of time. Similarly, we would also go back in time and

implement OLPR, as conforms to historical reality. Again, we would perform the inter-

vention multiple times, each time measuring political corruption. Due to random elements

189
of the social environment, we might expect observed levels of political corruption to dier

to some degree across each intervention we perform. Nevertheless, we could examine the

dierence between the average levels of political corruption estimated for the two types of

interventions. In so doing, we would calculate the systematic eect of OLPR vs. CLPR

on political corruption in Italy. To the extent that the former is found to be significantly

larger than the latter, we could conclude that an important cause of political corruption in

Italy was ballot structure.

All claims of the sort the cause of X in country Y is A rest on counterfactual thought

experiments like the one above.7 Needless to say, such experiments belong exclusively to

the realm of abstraction. Controlled experiments with randomly assigned treatment and

control groups can approximate these conditions to some degree, but they are infeasible for

many applications in social science, including ours.8 The best we can do, and our minimal

responsibility, is to exploit the leverage of comparison.

However, all comparisons do not give us equal leverage in evaluating the adequacy of

a theory. Cases which are evenly matched in terms of all potentially relevant confounding

variables provide greater mileage in testing a theory than those which are poor matches

(Rosenbaum 2002). Finding good matches can be a formidable problem for much case

study research, as the universe of cases is often small relative to the number of contending

theories which need to be addressed. However, even somewhat imperfect matches can

nevertheless be extremely useful in identifying deficiencies in a theory. To the extent that

we identify cases which exhibit neither the postulated outcomes nor the causal mechanisms

put forth by the theory, then we have just cause for serious doubt. The reason is that

7
This statement assumes that use of a covering law model is a requirement for valid explanation in the
social sciences (cf Hempel 1965). Most mainstream scholarship in political science adheres to this approach,
as do the various works which have contributed to the traditional paradigm.
8
The papers by Gerber and Green (2000) and Wantchekon (2003) are examples of exemplary work using
field experiments in political science.

190
although background factors may be responsible for observed dierences in the outcome of

interest, if the theory in question is valid we would expect that the same mechanisms would

be operative in dierent cases. If the mechanisms linking the independent variable to the

outcome are inoperative or take a dierent form in cases not originally under the purview

of the theory, then adjustment or abandonment of the theory is warranted.

It is with this discussion in mind that we turn our attention to the party-centric, closed

list and mixed-closed list electoral systems of Latin America. Of Latin Americas eighteen

presidential democracies, in all but five some portion of candidates for the lower chamber

run on closed party lists. According to the tenets of the traditional paradigm, these systems

should be favorably disposed against the incidence of political corruption. A consideration

of the historical record reveals otherwise.

At one end of the spectrum, one finds systems such as those of Bolivia and Honduras,

which for many years combined closed list PR and a fused ticket for lower house and

presidential elections. The party structures which evolved in the two countries were highly

centralized and revolved around the personal authority of a small number of elites. These

countries also exhibited persistent political corruption, the heart of which consisted of the

eorts of party activists in the public administration to use public resources for campaigns

and party finance.9 The existence of closed party lists actually reinforced corruption of

this stripe because it provided a mechanism of control for party leaders vis--vis political

hopefuls in the bureaucracy.10

Similar dynamics have been at play at various times in the Mexican and Venezuelan

9
On Bolivia, see Gamarra (1997), Lanyi et al (2000), Mayorga (1998) and World Bank (2000). On
Honduras, see Taylor (1996), Meza et al. (2002) and Salomn (2004).
10
With respect to the Honduran case, Salomn voices concern about this linkage explicitly: The way in
which [high level Honduran bureaucrats] assure their inclusion in the lists is with the financial, political and
institutional support of the electoral campaign of the presidential candidate of their party, with which is
created a nexus of complicity with very negative eects for the healthy exercise of public administration
(Salomn 2002, 157).

191
party systems. The most important parties in these countries, the PRI in Mexico and AD

and COPEI in Venezuela, shared the high degree of centralization of their Bolivian and

Honduran counterparts, albeit with greater routinization within party bureaucracies.

The primacy of political careerism (relative to bureaucratic careerism) within the Mex-

ican public administration under the PRI is well documented (Arellano and Guerrero 2003,

Grindle 1977). As one might expect, the asymmetric allegiance owed by bureaucrats to

their political benefactors, in detriment to that owed to the institution in which they work,

has been associated with pervasive clientelism and political corruption (Morris 1991). Mex-

icos electoral system, which combines plurality elections in single member districts and

closed list PR in multimember districts for elections to the lower chamber, has probably

contributed to this pattern.11

The Venezuelan case illustrates the same pattern more vividly. For thirty years, from

1958 to 1988, the Venezuelan electoral system was characterized by a single, closed list

ballot for elections to the Senate, Chamber of Deputies, state legislative assemblies and

municipal councils (Crisp and Rey 2001). The career trajectories of aspirants to political

oce remained firmly under the control of national party leaders throughout this period.

Leadership control was so absolute that legislative votes in the Congress practically never

resulted in roll calls: once the party leadership indicated their stance on the issue in ques-

tion, it was assumed all legislators in the party would follow suit (Coppedge 1994). Such

control naturally extended into the public administration as well, where partisan aliation

was the sine qua non of obtaining ones post. For political hopefuls and activists in the

bureaucracy, of which there were many, allegiances clearly rested with the party leadership

11
It is important to underline the point that in the Mexican case, the eect of ballot structure is limited
to reinforcing such behavior since CLPR in multimember districts was introduced relatively recently. On
electoral reform in Mexico, see Horcasitas and Weldon (2001).

192
that would determine the subsequent stages of ones career. In this context, the misappro-

priation of public resources, proselytizing in public institutions and the payment of party

commissions by entrepreneurs in exchange for public contracts flourished (Njaim 1981,

Rangel 1973). Political corruption as a means to finance the electoral machinery of AD and

COPEI became a preoccupation of many concerned with the Venezuelan political system

(Alvarez 1997).

At the opposite end of the spectrum, we find countries which combine closed list PR

with relatively fluid and decentralized party structures. Argentina is a good example of this

combination. In Argentina, provincial elites, not national party bosses, are the gatekeepers

of political careers (De Luca et al. 2002, Jones et al. 2002, Levitsky 2003). Consequently, in

this country the patterns of political corruption sketched above tend to manifest themselves

most visibly in provincial level institutions. Argentinas most important grass roots political

organizations, the agrupaciones, are financed principally from illicit resources drawn from

the state (Levitsky 2003, ch. 7). The heads of these organizations more often than not have

jobs in the provincial public administration (op cit., p. 195). As a general rule, provincial

elites exert considerable influence over the both the drafting of candidate lists for elected

oce and job placement in the public administration. As a result, with control over who

gets in the public administration and their prospects of having a political career after they

get out, a number of provincial elites have furthered their political ambitions by creating

elaborate political machines financed by the state (Lpez Echague 1996, Otero 1997).

A more general sense of the challenge that the Latin American experience presents to the

traditional paradigm can be gotten by considering a map of corruption in the entire region.

Figure 4.2 presents such a map. Countries which have open list proportion representation

systems are colored white and those with closed list proportion representation systems are

193
colored light gray.12 The vertical bars associated with each country on the map show a

countrys corruption score, measured in two dierent ways. The bar on the left-hand side

depicts perceptions regarding the prevalence of illegal political donations.13 The bar on the

right-hand side is a composite measure of problems with corruption control.14

Figure 4.2: List PR and Corruption in Latin America

The traditional paradigm would lead us to expect that these measures should be pos-

12
Colombia and Uruguays list PR systems, treated as sui generis for our purposes, are given a dark gray
coloring.
13
This measure comes from the Senior Executive Survey conducted by the World Economic Forum for its
2002-2003 Global Competitiveness Report. It was normalized to lie between 0 and 1.
14
This measure is the 2004 World Bank Institutes aggregate governance indicator for corruption control,
inverted and normalized to lie between 0 and 1.

194
itively related to having an open list system, especially the first. In fact, one finds that

both measures are on average higher in the closed list systems than the open list systems

(although the KS-test for dierence in distributions fails to reach statistical significance). It

thus would seem that the Latin American experience with CLPR casts on doubt the central

propositions of the traditional paradigm. For many countries in the region, party centric

electoral rules have gone hand in hand with persistent and pervasive political corruption.

How might we reconcile the dominant social science theory with the anomalous histor-

ical experience of Latin Americas list PR systems? The existing literature suggests two

potential responses to this state of aairs.

One response would be to defend the traditional paradigm using the rationale of potential

omitted variable bias. In this regard, defenders of the traditional paradigm could reasonably

argue that some of the cases used to illustrate the symbiosis of party centric electoral rules

and political corruption are poor matches for the cases that provided the inspiration for the

traditional paradigm. Had the former cases been characterized by party centric electoral

rules and political and economic environments otherwise roughly equivalent to those of

the latter, then one might argue that a low political corruption outcome would likely have

obtained.

This is not an unreasonable position to take, especially in light of the importance of

natural resource rents generated by petroleum extraction in two of the anomalous cases

(Mexico and Venezuela). Recent research has shown that such rents tend to encourage the

incidence of corruption, due in part to the fact that government resources so gotten do not

require the taxation of domestic firms and workers and thus are less exposed to scrutiny

from civil society (Ades and di Tella 1999, Leite and Weidmann 1999).

Another response to the challenge posed by Latin Americas experience with party-

195
centric electoral systems might be to abandon the traditional paradigm entirely. In this vein,

two important recent papers argue the relationship between electoral rules and corruption

is unambiguously the opposite of that postulated by the traditional paradigm: open lists

being associated with less, not more, corruption than closed lists (Kunicova and Rose-

Ackerman 2002, Persson et al. 2003). The logic behind this claim is principally derived

from two stylized facts. Firstly, voters can punish individual legislators for bad behavior

in OLPR systems, whereas they can only punish entire party lists in CLPR systems, thus

creating a "team production" problem in the latter (Persson et al. 2003). Secondly, due

to the substantial control of party leaders over the rank and file, intra-party monitoring

is a dead letter in CLPR systems, while it might not be in OLPR systems (Kunicova and

Rose-Ackerman 2002). From the perspective of these papers, the challenge posed by the

Latin American experience to the traditional paradigm is easily explained: the paradigm is

simply wrong, both as applied to Latin America and elsewhere.

In this paper, I pursue a third response to the Latin American challenge, namely, the

incorporation of the traditional paradigm within a more general theory of the relationship

between ballot structure and political corruption. In this regard, I would argue that the

principal contribution of the Latin American experience with party-centric electoral rules is

not so much to show that such rules can coexist with high levels of political corruption but

rather to suggest a totally dierent causal pathway whereby ballot structure contributes to

the same: the supply side pathway. With two causal pathways at work, the demand side

pathway elucidated by the traditional paradigm and the supply side pathway elucidated

in the pages that follow, we can make much better sense of Latin Americas anomalous

experience with political corruption and ballot structure.

196
4 A Supply Side Model of Political Corruption

The principal weakness of the traditional paradigm can be stated concisely: access to re-

sources generated by corruption is treated as a deus ex machina. That is to say, parties

and politicians who seek to finance campaigns or political activity through illegal means

are assumed to be able to do so with equal facility and frequency. No dierences across

electoral systems are theorized to exist in terms of the supply of illicit resources which make

their way into politicians hands. Rather, the only dierences of import are dierences in

the demand of politicians to obtain access to these resources.15

Both logically and empirically, ballot structure can be shown to have important sup-

ply side eects. I will concern myself here with developing a logically consistent, supply side

account of the relationship between ballot structure and political corruption. Constructing

this account requires first establishing a core set of postulates upon which the account will

build. For my purposes, there are three: 1) the locus of political corruption is the public

administration; 2) a non-negligible percentage of employees in most public administrations

consist of individuals with well established partisan aliations and political aspirations for

the future; 3) individuals so characterized face a dual/conflicting principals problem. Let

us consider each in turn.

The locus of political corruption is the public administration. The resources of

political corruption must come from somewhere. In the first chapter of the monograph, I

discussed in detail the origins of the resources of political corruption. As the discussion

15
This is not to say that important works which have contributed to the traditional paradigm have not
provided empirical discussions of the supply side, but rather that the relationship between ballot structure
and the supply of financial corruption has remained untheorized. Woodall (1996), in particular, provides
an excellent description of the origins of illicit campaign finance in the dealings between bureaucrats in
Japans Ministry of Construction (MOC) and large construction firms. However, he fails to consider the
counterfactual question of how this process may have diered if LDP party leaders had had (closed) list
positions to hold over the heads of MOC bureaucrats looking to establish political careers.

197
of cases above and innumerable corruption scandals demonstrate, the lions share of such

resources originate in the state. Although elected politicians may have some sway over

the budget and may be able to lobby contacts in the public administration, ultimately the

resources most valuable to parties and politicians are at the discretion of civil servants. The

relationship between the civil service and political parties is thus key to understanding how

political corruption becomes an important part of the political process.

Politicized bureaucracies. For better or worse, the Weberian ideal of meritocratic

and apolitical bureaucracy remains far from reality in many parts of the world. Many bu-

reaucracies, especially in countries with presidential forms of government, are characterized

by a large number of administrative posts whose occupants serve at the exclusive confidence

of the executive (Schneider 1999). Not uncommonly, these posts become a repository for

also-rans, politicians who lost a bid for elected oce immediately prior to assuming their

position (Carey and Siavelis 2005). To these cases, one can add public administrations

where the division between the appointive and career civil service is not respected in prac-

tice and in which political criteria color stang decisions at all levels of the bureaucracy.

In either of these two scenarios, political loyalists and hopefuls for elected oce permeate

the ranks of the bureaucracy.

Conflicting principals. Individuals who obtain posts in the public service by virtue

of political connections are often exposed to opposing pressures. On the one hand, they

may experience pressure from their political benefactors to use the resources at their dis-

posal in a way that redounds to the benefactors electoral advantage. In this way, political

corruption threatens to rear its head. On the other hand, they may be periodically held

to account for procedural improprieties or misallocations of institutional resources. Con-

trol organs help produce this countervailing pressure towards universalism. With varying

198
degrees of ecacy, internal control units, supreme auditing institutions and anti-corruption

institutions have helped ensure that political corruption is an activity entailing some de-

gree of risk for participants. In essence, many bureaucrats must adjudicate between the

dictates of principals who make sharply conflicting demands on their behavior: political

operators who control future career trajectories and, in certain circumstances, permanence

in administrative posts and representatives of oversight organs and the judicial branch who

hold sway over functionaries reputations and liberty.

These postulates lay the foundation for a rigorous examination of the relationship be-

tween ballot structure and and the supply of illicit resources originating in the state. I

proceed to conduct such an examination here through a formal model of political corrup-

tion in the public bureaucracy. Corrupt transactions are modeled according to a principal

agent framework with hidden knowledge, in which a corrupt party leader is the principal

and party militants in the bureaucracy are the agents. As in the seminal papers by Cadot

(1987) and Andvig and Moene (1990), multiple corruption equilibria emerge as a result of

complementarities between the actions of dierent bureaucrats. However, unlike these pa-

pers, complementarity is not imposed by assumption (Andvig and Moene) nor is it a result

of bribe splitting by bureaucrats of dierent hierarchical levels (Cadot). Rather, comple-

mentarity emerges from the link between illicit finance and party strength, and the positive

relationship between party strength and the value of nominations used to induce militants

to engage in corruption. Moreover, multiple equilibria are shown to exist only in certain

contexts, while in others a single high or low corruption equilibrium prevails. The crucial

role of ballot structure in encouraging equilibria of various forms lies at the heart of the

analysis which follows.

199
4.1 Corruption in the Public Administration

4.1.1 Players, Timing and Actions

Imagine a polity in which a party leader (L) interacts with a group of N party militants

within the public administration. L oers the militants the opportunity to run for legislative

oce in exchange for extracting some amount of resources, [0, +), from the institution

in which they work. In this context, we can think of extraction as constituting any type

of corrupt action which imposes pecuniary damages on the public sector but financially

benefits the militants party in the amount . Having been presented with the oer of a

nomination in return for , each militant i must decide whether or not to accept Ls oer.

If a militant decides to extract , there is a probability q() that he will be caught doing

so by an administrative control organ outside Ls direct control. If caught, the militant will

receive punishment (, 0] and will be unable to run for oce. The probability of

being caught is a convex increasing function of the amount extracted with q(0) = 0.

If a militant extracts and is not caught, which occurs with probability 1 q(),

one of two outcomes will obtain. With probability pe , i will be elected and receive utility

disbursement ri . The term pe represents the expected probability of obtaining oce if given

a nomination (shared across all i) and ri represents the personal ego rent that militant i

receives from being in oce. With probability 1 pe , i will lose his bid for elected oce.

In this case, i simply retains his administrative post, which is normalized to have utility 0.

If a militant chooses not to extract , he maintains his administrative post with certainty.

Figure 4.3 presents the timing and actions of the game.

200
 Figure 4.3. Timing and Actions

4.1.2 The Party Militants Choice Problem

Given the potential consequences of his choice, a militant i will extract if and only if:

[1 q()][pe ri + (1 pe )(0)] + q() > 0 (4.1)

The above equation states that militant i will extract if the expected utility of accept-

ing Ls oer exceeds that of refusing it. The expected utility of accepting the oer consists

of the probability of not being caught extracting resources, 1 q(), multiplied by the ex-

pected utility of being nominated to run for oce, pe ri + (1 pe )(0), plus the probability

of being caught extracting resources, q(), multiplied by the value of the punishment, .

The utility of refusing the oer is simply the value of retaining the administrative post with

certainty.

Ego rents are private information. L does not know the ego rent that any given militant

associates with holding oce. However, L is aware that the value of holding oce is

distributed among militants according to a continuous probability density function f with

support [0, ]. I assume f has a nonnegative first derivative defined along every point in its

support.16 Let us denote as the proportion of militants willing to accept Ls oer. This

16
Stated in laymans terms, this assumption means that high values of ri are at least as common as low

201
proportion can be written:

q()
= 1 F ( ) (4.2)
[1 q()]pe

Equation (4.2) illustrates three important aspects about the scope of participation in

corrupt exchanges. Firstly, participation in the corruption scheme is negatively related

to the magnitude of the sanction given to militants caught engaging in illicit behavior.

Secondly, the scope of participation is negatively related to the amount of tribute demanded

in exchange for a nomination. To be more precise about this relationship, the proportion of

militants participating in the scheme is concave decreasing in the level of tribute. This means

that the larger L sets the level of tribute demanded in exchange for a nomination, the more

rapidly will further increases in the level of tribute reduce the proportion of militants willing

to participate in the scheme. Finally, equation (4.2) shows that the proportion of militants

willing to participate in the scheme is concave increasing in the expected probability of

victory if given a nomination. Thus, as the expected probability of victory increases, the

proportion of militants willing to extract does so as well, although at a decreasing rate.

In certain circumstances, no militants will be willing to participate in the corruption

scheme. Equation (4.2) establishes a lower boundary on the expected probability of victory

necessary to ensure that the proportion of militants willing to participate in the scheme is
q()
greater than zero. This requirement is: pe > p [1q()] . Until the expected probability

of victory reaches the point p, changes in pe will have no eect on the proportion of militants

willing to participate.

As the level of tribute plays an important role in determining the proportion of militants

willing to engage in Ls corruption scheme, it makes sense to think of this variable as being

values.

202
optimally selected by L in order to maximize the returns generated by the scheme. It is to

this component of the model that I now turn.

4.1.3 The Party Leaders Choice Problem

The party leader is devoted to advancing the electoral prospects of his party. Whether

he does so because of an ideological commitment to the party program or a desire for

self aggrandizement is immaterial. In either case, he will require considerable resources to

realize his goal. Thus it is reasonable to assume that the party leader will attempt to select

the level of which maximizes the total revenue extracted from the state.

There are several ways in which L might attempt to realize the goal of extracting

maximal revenue from the state. In the best of all worlds, he would be able to tailor the

level of tribute to the characteristics of each militant, choosing an optimal i instead of a

single . Were this possible, L would select the tribute level such that each militant would

be indierent between extracting the tribute and running for oce, on the one hand, and

not extracting the tribute and remaining in the public administration, on the other. In so

doing, he would be able to extract the maximum amount feasible from the state. This is

the first best level of .

Informational constraints make such a strategy impossible. Because ri is unknown

to him, L must design a strategy which maximizes total revenue given the information

available. Such a strategy consists of selecting a single value of which maximizes total

revenue given Ls knowledge of the distribution of ego rents among the rank and file. This

is Ls second best choice of .

Let us denote total revenue by . The second best level of is the solution to the

following:

203
 max{ = N } (4.3)

The first order condition which defines this value of is (subscripts denote partial

derivatives):

+ = 0 (4.4)

Using the fact that is concave decreasing in , the reader can easily verify that the

second order condition for a maximum is indeed met by .

The trade-o facing L is as follows. He would like to set as high as possible, maxi-

mizing the amount of tribute gotten from militants willing to extract resources from their

institutions. However, because is concave decreasing in , a considerable increase in

will cause a precipitous decline in the proportion of militants willing to extract resources

on the partys behalf. In order to maximize revenue for his party, the party leader must

choose a level of tribute which balances out the joint considerations of ensuring significant

participation and garnering considerable tribute from each participant.

Having previously established the relationship between and pe , we can also characterize

the relationship between the maximum level of revenue extracted from the state, , and

the expected probability of obtaining elected oce for militants given the opportunity to

run. The appendix proves formally that is a concave increasing function of pe . Its graph

is shown in Figure 4.4. As pe increases, total revenue will remain at 0 until pe crosses the

threshold p. After that point, further increases in pe will drive up the value of until pe

reaches its maximum value of 1.

204
 Figure 4.4: Revenue from Corruption and the Expected Probability of Gaining Oce

4.2 Ballot Structure and Electoral Competition

4.2.1 Overview

Up to this point, the probability that a militant oered a chance to run for elected oce

actually obtains the post he seeks has been treated as an exogenous feature of the political

environment. In this section, I bring the model into general equilibrium by considering

the facets of the electoral arena that determine this probability. I do so by developing two

stylized models of list proportional representation: one that corresponds in its key features

to a CLPR system and another that corresponds to an OLPR system. In the latter case,

I further divide the analysis into two subsections: one which examines the outcomes which

obtain under the simplifying assumption of candidate homogeneity within party lists and

205
one which examines the outcomes which obtain when candidate heterogeneity is imposed

by targeted electoral support provided by L for the militants participating in the corruption

scheme. Given the electoral rules in force and a number of fixed features of the electoral

environment, I deduce the objective probability of victory in all three scenarios. Because

the expectations of militants are rational, I equate the expected probability of victory with

the actual probability in each case. In doing so, I characterize the equilibrium values of p ,

, and that emerge under each electoral system.

Endogenizing the value of a nomination in terms of the key features of the political

environment reveals the presence of strategic complementarities between the choices made

by militants within the public administration. Complementarity emerges as the result of the

impact of bureaucratic choice, in the aggregate, on the electoral viability of the governing

partys list. As was shown above, a militants choice of whether or not to extract resources

on behalf of his party in exchange for a nomination will depend on the expected probability

of obtaining a seat. This probability, in turn, will depend on the amount of resources the

party is able to extract from the state, which ultimately rests on the proportion of militants

who extract and the level at which they do so. As a result, the return for extraction for

any bureaucrat is an increasing function of the remaining bureaucrats that have decided to

extract. In this sense, militants can be said to face a large scale coordination dilemma, one

whose resolution may in certain circumstances rest at the disposal of party leaders.

4.2.2 Voter Preferences and Vote Shares

Imagine that in our polity there are two parties, denoted by h {G, O}: the government

party (G) and the opposition party (O). The parties compete in simultaneous elections

in a large number of identical districts. In each district, there is a continuum of voters,

206
subscripted by j. In any given election, an individual voter j will prefer will prefer the G

list to the O list if and only if:

vj > 0 (4.5)

where vj vjG vjO . The term vjh represents the utility voter j associates with party hs

platform. The term vj , in turn, measures voter js preference for the party platform of G over

O. Party leaders do not know vj for any particular voter but they do know its distribution in

the electorate. It is assumed to be uniformly distributed with support [ 1+2()

2 , 1+2()
2 ].

The term is a mean zero random variable which represents unforeseeable events which

have a common eect on voter preferences: an economic downturn, the outbreak of war

1 1
and so forth. It is uniformly distributed with support [ 2 , 2 ].

Given this setup, the fraction of votes which will go to the G list is:

1
s= + () (4.6)
2

The function () represents the electorates preference for the platform of party G

over that of party O. The greater the amount of revenue L is able to extract from the state,

the more money can be spent on manipulating the salience of issues in a way that redounds

to the G partys advantage, the more money can be funneled towards the establishment

of clientelistic links with voters, and so forth. Thus, it is logical to suppose that () is

increasing in . I assume that it takes the following linear functional form: () = + .

Given this characterization, if = 0, () = . Therefore, the variable [1, 1] indicates

the electorates "natural" preference for G over O. That is, it represents the electorates

preference for G when L is unable to extract any resources from the state. The variable

207
 R+ , in turn, indicates the marginal eect of such resources on party vote shares.

The variable represents the (inverse of) the diversity of preferences in the electorate.

The greater it is, the less votes depend on party activities and the more they depend on

exogenous factors. Given that is a uniform random variable, is a uniform random

variable as well, the latter having support [ 2 , 2 ]. The ratio is the length of this

support. The greater this ratio, the more exogenous factors will have the potential to

create large shifts in voter preferences for or against the G party. This ratio can be thought

of as a measure of the degree of electoral volatility in the polity.

4.2.3 Proportional Representation and the Corruption Equilibrium

In order to decide whether or not to partake in Ls corruption scheme, a militant must

generate an expectation about his probability of attaining elected oce given all the infor-

mation at his disposal. The fundamental inference regards the probability the he obtains

a seat given his nomination to the list of the G party. Here I will characterize the proba-

bility of obtaining a seat generically under proportional representation without specifying

the unique features of CLPR or OLPR. After having conducted equilibrium analysis in the

general case, I will characterize the equilibria which emerge in the three scenarios mentioned

earlier as well as the dierences between them.

In the polity under consideration, all districts share a common district magnitude, de-

noted by k. Moreover, following approximately Coxs (1997) k + 1 rule, each party list runs

exactly k candidates in each district. Irrespective of whether a candidate runs in a CLPR

system or in an OLPR system, he will have a list position, denoted by l {1, 2, ..., k}, which

establishes his rank vis-a-vis listmates. In a CLPR system, list position is established before

the election by L. In an OLPR system, the position emerges after the election as a function

208
of the within list vote share captured by the candidate.

In general, an lth ranked candidate will win a seat if and only if:

l
s (4.7)
k+1

The term on the RHS of inequality (4.7) is the party vote threshold (PVT). Only if the

vote share of the G party exceeds this threshold will an lth ranked candidate attain a seat

in the legislature.

The probability that a candidate obtains a seat will depend on both the vote share of

his party and his list position. It can be written as:

k
X m
Pr(seat) = Pr(s ) Pr(l = m) (4.8)
m=1
k+1

The probability that a candidate receives a seat is equal to the product of probability

that the G list vote share exceeds the party vote threshold for a given list position and the

probability that the candidate attains that position, summed over all k possible positions.

Using (4.6), we can rewrite the expression above as:

0 if D + (pe ) < 0


Pr(seat) =
D + (pe ) if 0 < D + (pe ) < 1 (4.9)





1 if D + (pe ) > 1

where D 1
2 + ( 12 + E[ k+1
l
]). The last term in the expression for D is the expected

party vote threshold (EPVT). It is the weighted average of the G list vote shares necessary

for the militant to obtain a seat for each for each of the k positions on the list, weighted by

the probability that the militant attains each position. It is written:

209
 k
X
l m
E[ ]= Pr(l = m) (4.10)
k+1 k+1
m=1

As will be made clear later, it is essentially restrictions on the value of Pr(l = m) which

dierentiate the two forms of PR considered in this chapter.

As party militants have rational expectations, the expected probability of victory will

equal the true probability, which in turn is a function of the expected probability. As all

militants have access to the same information and are aware that all other militants have

rational expectations, the following equation will define the equilibrium values of p, and by

extension, the equilibrium values of , and :

p = D + (p ) (4.11)

Graphically, equilibrium values of p consist of all points of intersection between the

objective probability of victory, Pr(seat), and the expected probability, pe . Let us note that

D, the intercept of the Pr(seat) curve, is a decreasing linear function of the EPVT. As the

l
intercept of Pr(seat) depends on the level of E[ k+1 ], this variable will determine the number

and location of the equilibria which emerge. There may be one, two or three equilibrium

values of p. Figure 4.5 displays the equilibria which emerge at dierent levels of the EPVT.

210
 Figure 4.5. Proportional Representation and Equilibrium Outcomes

l {Pr(seat)}
Let us define the cutpoint E = {E[ k+1 ]: p = 1}. This is the value of the EPVT

for which the slope of Pr(seat) equals 1 at an equilibrium point. The lower gray dotted

l l
curve in figure 4.5 depicts the graph of Pr(seat) for E[ k+1 ] = E. If E[ k+1 ] > E, there will

be a single equilibrium value of p. This equilibrium corresponds to a low value of p and a

correspondingly low value of and . It represents a unique, low corruption equilibrium.

I will denote this value of p which emerges in this scenario by p0 .

Let us define a second cutpoint E = ( 12 + ) (p 12 ). This is the value of the EPVT

for which D = p. The upper gray dotted curve in figure 4.5 depicts the graph of Pr(seat)

l l
for E[ k+1 ] = E. If E[ k+1 ] < E, there will also be a single equilibrium value of p. In this

211
case, the equilibrium value of p will be "large" and will induce a high corruption outcome.

I will denote this value of of p by p4 .

l
If E < E[ k+1 ] < E, there are three potential equilibria: one corresponding to a low

corruption outcome, one corresponding to an intermediate corruption outcome and one

corresponding to a high corruption outcome. In this instance, I will denote the lowest

equilibrium level of p by p1 , the intermediate level by p2 and the highest level by p3 .

l l
In the rare event that E[ k+1 ] = E or E[ k+1 ] = E, two equilibrium values of p emerge.

These equilibria are knife-edged equilibria, so called because they exist only for exact values

l
of E[ k+1 ], not for any range of values of this variable. Because their existence is essentially

a zero probability event, these equilibria play a minor role in what follows.

It is important to note that the equilibrium values of p are ranked in the sense that

p0 < p1 < p2 < p3 < p4 . The lowest of the three potential equilibria which emerge when

l l
E < E[ k+1 ] < E is higher than the unique equilibrium which emerges when E[ k+1 ] > E.

l
Similarly, the highest of the three potential equilibria which emerge when E < E[ k+1 ]<

l
E is lower than the unique equilibrium which emerges when E[ k+1 ] < E. This natural

l
ranking of equilibria based on the value of E[ k+1 ] leads us to the following assumption:

(A.1): Any increase (decrease) in D leads to and increase (decrease) in p

Assumption (A.1) follows directly from (4.11), with the exception that it rules out two

pathological cases emergent from the knife-edged equilibria discussed above.17

17 l
The two pathological cases are as follows. Firstly, if E[ k+1 ] = E, a slight increase in the value of D will
change the number of equilibria from two to three. If for some reason militants had coordinated around the
l
higher of the two equilibria when E[ k+1 ] = E and they coordinate around one of the two lower equilibria
when the Pr(seat) curve shifts upwards, then the shift will not have resulted in a higher equilibrium level of
l
p but a lower one. Secondly, if E[ k+1 ] = E, a slight decrease in the value of D will also change the number

212
 Having characterized the equilibria which emerge in the general case, we can discuss

comparative statics which hold under all of the three specifications that follow. Firstly,

the greater the electorates natural preference for party G, the more valuable Ls oer of

a nomination in exchange for tribute and the greater the proportion of militants that are

willing to extract resources from their institutions. In this sense, the model leads us to

expect that parties which are strong in electoral terms without recourse to illicit resources

will also have greater access to these resources because they can establish more favorable

bargains with their militants in the public administration. Machine parties, which thrive

o of the partisan use of public resources during electoral contests, may at some point have

to establish themselves as viable electoral vehicles before they are able to capture the public

administration.

Secondly, electoral volatility will have a contingent eect. If the G partys expected

l
vote share is above the EPVT (E[s] > E[ k+1 ]), then an increase in volatility will reduce

the probability of victory, thereby reducing the proportion of militants willing to extract.

Conversely, if the G partys expected vote share is below the EPVT, then an increase

in volatility will increase the probability of victory, thereby increasing the proportion of

militants willing to extract. The logic underlying this result is straightforward. As volatility

rises, the systematic components of the probability of victory diminish in importance. This

means that the electorates natural preference for G, the total revenue extracted from

the state and the expected party vote threshold will cease to have much of an impact on

Pr(seat). As volatility gets very large ( +), the probability of victory reduces to

a simple coin toss (Pr(seat) = 12 ). This disadvantages candidates from parties with strong

of equilibria from two to three. If for some reason militants had coordinated around the lower of the two
l
equilibria when E[ k+1 ] = E and they coordinate around one of the two higher equilibria when the Pr(seat)
curve shifts downwards, then the shift will not have resulted in a lower equilibrium level of p but a higher
one. As these cases are the result of knife-edge equilibria, their substantive importance is limited.

213
electoral support and/or candidates who foresee having good list positions and it advantages

candidates from parties with weak electoral support and/or those who foresee having poor

list positions. In essence, high volatility levels the playing field between party G and party

O and decreases the electoral value of the corruption scheme.

4.2.4 Closed Lists

In a CLPR system, L is able to rank the candidates that run on the G list in each district

in advance of the election. Therefore, if militant is nominated and given the rank b
l, Pr(l =

b b
l) = 1 and Pr(l = m) = 0 for all m 6= b l
l. As a result, E[ k+1 ]= l
k+1 . Stated verbally, because

a candidates rank is chosen ahead of time by L, each militant knows his future list position

with certainty. Because the list position is known with certainty, the PVT necessary to

obtain a seat is also known with certainty. The probability that a given candidate obtains

a seat is simply the probability that the party meets or exceeds the threshold.

The crucial point to notice about CLPR is the leverage it aords L to encourage a unique,

high corruption equilibrium. Under CLPR, L will be able to guarantee the emergence of

such an outcome as long as the electorates natural preference for the G party is suciently

strong and district magnitude is suciently large. To see this, note that the condition for

l 1
the emergence of a single high corruption equilibrium is: k+1 < 2 + (p 12 ). Now

b 12 + ( 12 p). So
note that the RHS of this inequality will be greater than 0 i

l
b and district magnitude is large, there will be some
long as > k+1 which satisfies the

inequality.18

This discussion points to the importance of district magnitude as a determinant of the

scope of corrupt exchanges. For any given level of the electorates natural preference for G,

18
The reader can verify this by setting l = bl and = b + where is some small constant. The inequality
b
will be satisfied if and only if k b
k 1 +b+l ( 1 p) 1.
2 2

214
oering militants a better position on the list makes the opportunity to run more attractive

and encourages militants to coordinate their expectations around an outcome where both

the probability of victory and the amount of resources extracted from the state are high.

However, the electoral value of a particular ranking is only relative to the number of fellow

listmates running, itself a function of district magnitude. In our two party model of PR,

oering a militant the opportunity to run on the party list as the first ranked candidate in a

polity where each district has a magnitude of 3 means that the party must claim 25% of the

district vote in order for the militant to obtain a seat. On the other hand, oering a first

ranked candidacy in a polity where district magnitude is 99 would mean the party would

only have to claim 1% of the vote in order to guarantee that the militant would obtain a

seat. Clearly, magnitude has a major eect on the attractiveness of the oers party leaders

can make under CLPR. As we shall see later, under certain conditions it has a major impact

on the bargains that can be struck under OLPR as well.

With no supply constraints on the number of positions he can oer (many districts), in

a CLPR system L would clearly do best to oer militants the number one rank on the G

list in exchange for extracting resources. Under no condition can the oer of a nomination

in an OLPR system beat this oer. However, there are informal mechanisms that L can

use to make the oer of a nomination under OLPR more attractive than it otherwise would

be. In certain circumstances, these mechanisms may allow L to make an oer under OLPR

that comes arbitrarily close to the value of a first ranked position under CLPR. It is to the

consequences of OLPR for the corruption equilibrium that I now turn.

215
4.2.5 Open Lists

Candidate Homogeneity within Lists In an OLPR system, candidates for oce re-

ceive no ocial ranking from party leaders. Seats won by party lists are allocated in strict

accord with the vote shares won by each candidate. Therefore, according to the formal

structure of electoral law, all candidacies begin on equal footing. The consequences of the

formal equality imposed onto candidacies by OLPR will be the focus of the analysis which

follows.

To begin, I consider a reduced form characterization of OLPR where each of the k

candidates competing on list G are identical. Because of the homogeneity of candidates,

within-list vote shares are also identical and the allocation of G list seats is accomplished

by lot. This state of aairs might emerge in a context where candidate vote shares depend

on personal resources and the endowments of candidates are the same. Alternatively, a

similar outcome would obtain in a setup in which candidates propose platforms along a

single ideological dimension and voters with single-peaked preferences cast within-list votes

sincerely based on these platforms.19

Because seats are allocated by lot, we have Pr(l = 1) = Pr(l = 2) = ... Pr(l = k) = k1 .

Therefore, we can rewrite the EPVT as:

k k
l 1 X m 1 X 1
E[ ]= = m= (4.12)
k+1 k m=1 k + 1 k(k + 1) m=1 2

Not knowing for certain what his future list position will be, under OLPR with identical

candidates a militant knows that on average the G list must attain a vote share greater than

1
or equal to 2 in order to guarantee that he gets a seat. Therefore, a single high corruption

19
For a discussion of the potential of equilibria to emerge in the latter case, all of which result in equal
vote shares, see Cox (1990).

216
equilibrium will emerge if and only if > (p 12 ). Note that in this context there is nothing

L can do to bring this equilibrium about: either the electorates natural preference for the

G party is strong enough to ensure that it exists or it is not. Moreover, the equilibrium

values of p which emerge will always be below those which emerge under CLPR as long

as in the latter scenario L sets the list position oered to militants to a value less than or

k+1
equal to 2 . Since it is unlikely that a party leader would seek to entice militants to join

a corruption scheme by oering them list positions below the median, the superiority of

CLPR as a conduit for corrupt exchanges seems clear.

Targeted Support and Candidate Heterogeneity within Lists In this section of

the paper, I will consider an extension of the above model of OLPR in which L is able to

target electoral resources to militants campaigns, thus motivating more militants to extract

resources from their institutions and increasing the total amount of revenue which can be

claimed by the G party.

Let us suppose that upon presenting the level of tribute required to obtain a nomination,

L also informs potential co-conspirators that militants who participate in the corruption

scheme and are eligible to run for oce will not only receive the G party nomination

but also some amount of electoral support from the party, R+ , to use in order to

dierentiate themselves from fellow listmates in the upcoming elections. This electoral

support can be conceptualized in a number of dierent ways: as direct financial transfers

to candidates campaign funds, preferential allocation of party television or radio time to

favored candidates, stumping by party leaders on behalf of particular candidates and so

1
on and so forth. As a result of such eorts, Pr(l = m) will not, in general, be equal k as

assumed above.

Rather, Pr(l = m) can be represented using the Kumaraswamy probability distribution.

217
In this setup, a militants potential list position, kl , is represented by a Kumaraswamy dis-

tributed random variable y, distributed according to a Kumaraswamy pdf with parameters

a = 1 and b = a + . This distribution function has the property that when = 0

1
(no targeted support), it reduces to the uniform distribution and Pr(l = m) = k for all

m {1, 2, ...k}, as in the model considered in the previous section. For > 0, the mass

of the distribution shifts to the left; the interpretation being that, holding party vote share

constant, higher results in a greater probability of having a good list position (low kl ).20

Figure 4.6 presents the graph of the distribution at varying levels of .

Figure 4.6. The Kumaraswamy Probability Distribution over List Positions ( = 10)

m m1
Note that Pr(l = m) = Pr(y < k) Pr(y < k ).Given this more general frame-

work, we can rewrite the EPVT as:

20
The Kumaraswamy distribution is a continuous and double-bounded distribution like the Beta, but it
has a pdf and cdf which are much easier to work with and thus it recommends itself for modelling exercises
like the current one. See Kumaraswamy (1980) for details.

218
 k
X
l m m m1
E[ ] = [Pr(y < ) Pr(y < )] (4.13)
k+1 m=1
k+1 k k
Xk
m m ab m1 a b
= {1 [1 ( ) ] } {1 [1 ( ) ]}
m=1
k + 1 k k
k
X m m1 b m
= {[1 ( )] [1 ( )]b }
m=1
k+1 k k
k
X
1 m1 b
= [1 ( )]
k + 1 m=1 k

The parameter R+ represents the magnitude with which party administered electoral

support aects the popularity of a candidate vis-a-vis his listmates. If is large, then

militants who participate in the scheme and are eligible to run for oce can expect to be

among the first candidates in line to receive seats won by the G list. If is low, then the

partys targeted support will have little eect and participating militants can expect to do

no better or no worse than the majority of their listmates.

To appreciate how the targeted allocation of electoral resources aects the corruption

equilibria, we must revisit the expression for the objective probability of attaining a seat

given that resources can now be funneled into individual candidacies. It is written:

l
Pr(seat) = D(E[ ]()) + (pe ) (4.14)
k+1

l
where D(.) is as defined in section 4.2.3 and E[ k+1 ](.) takes the form given in (4.13)

above. As before, we require that expectations be rational, thus defining the equilibrium p

by the expression:

l
p = D(E[ ]()) + (p ) (4.15)
k+1

219
 The corruption equilibria defined by (4.15), under the assumption of OLPR with tar-

geted electoral support to conspiring militants, is uniformly greater than that obtained

under the assumption of OLPR with within-list candidate homogeneity. To see this, recall

that according to (A.1), any shift upward in the intercept of Pr(seat) curve results in the

establishment of higher p , , and . Now note that the intercept is increasing in , a

fact which can be easily established by dierentiation:

k
X
m1 b m1
D = [1 ( )] ln[1 ( )] > 0 (1)
(k + 1) m=1 k k

Thus, it is clear that the targeting of electoral support to compliant militants is one

potentially fruitful way of consolidating party corruption networks in the often candidate-

centric environment of OLPR. In essence, this is because targeted supported support allows

L to improve the quality of the oer he makes to militants in the public administration.

Note, however, that irrespective of the amount of support L is able to provide to compliant

militants, he can only approximate to a greater or lesser degree the oer of a first ranked list

position. No finite amount of targeted electoral support will allow him to oer a genuine

equivalent to a nomination to a first ranked list position, as he could do under CLPR.

l 1
Formally, as +, E[ k+1 ]() k+1 ; the party leader would have to commit

himself to an infinite amount of support to establish an EPVT equivalent to that he could

establish under CLPR by promising first place list positions to compliant militants. For this

reason, it is safe to conclude that, all else equal, CLPR more readily lends itself to machine

politics than OLPR, even if party leaders are able to make use of informal mechanisms to

channel party funds or other resources to favored candidates.

Two parameters determine the attractiveness of targeting electoral resources to compli-

ant militants: the impact of said resources on within-list vote shares, , and the district

220
magnitude, k. For = 0, an examination of the cross-partial derivative of the Pr(seat)

curve establishes that the marginal impact of raising from 0 to any positive level (i.e.

initiating targeted support) is increasing in the level of ,

k
X
m1 m1
D ( = 0) = [1 ( )] ln[1 ( )] > 0
(k + 1) m=1 k k

thus demonstrating that the more targeted support by party leaders impacts upon

within-list vote shares, the more party leaders can perturb upward the corruption equi-

libria by commiting themselves to favor compliant militants.

As in the case of CLPR, under OLPR with targeted support district magnitude also

plays an important role in determining the attractiveness of the oer L is able to make to

militants. Figure 4.7 shows how the expected party vote threshold varies with for two

dierent levels of district magnitude, k1 and k2 , where k2 > k1 . Where district magnitude

is large, an increase in targeted support leads to a more rapid reduction in the EPVT than

it does where it is small. Moreover, figure 4.7 shows that it is possible for a militant in

an OLPR system and large district magnititude to have a lower EPVT than he would in a

system with CLPR and small district magnitude. The horizontal lines show the party vote

threshold for a first ranked candidate in polities where district magnitude equals k1 and k2 .

Note that once exceeds the point 0 , the expected party vote threshold for a polity with

OLPR and k = k2 is lower than the party vote threshold for a polity with CLPR and k =

k1 . Thus, it is entirely consistent with the model to expect more pervasive machine politics

in a polity with OLPR and large district magnitude than one with CLPR and small district

221
magnitude.

Figure 4.7. EPVTs and District Magnitude

5 Implications of the Model

The model generates two sets of empirically observable implications: one set oriented to-

wards the eects of the electoral system and another set oriented towards the characteristics

of political parties. In terms of the electoral system, the model makes three predictions.

Firstly, the model predicts that closed list systems lend themselves to a greater supply of

corruption than open list systems, all else equal. The prediction stems from the fact that

in closed list systems, party leaders can present would be coconspirators with a nomination

to run for elected oce as well as a particular position on a closed party list. In an open

list system, party leaders can oer a nomination, and perhaps financial resources generated

by targeted support, but they cannot guarantee that a militant participating in the cor-

ruption scheme will get a particular position on the party list. List placement is ultimately

222
determined by candidate vote shares in popular elections, only one of whose determinants

is the level of political resources the candidate can bring to bear. In essence, closed list

proportional reputation provides a superior technology for corrupt deal making than open

list proportional representation because of the greater discretionary power it invests in the

upper ranks of the party leadership. This means that the party leadership can oer a better

bargain to each militant, and also that, since the bargain is good, a considerable number

of any given militants peers will join the scheme, generating more resources for the party

and further increasing the value of the nomination.

The second electoral system level prediction deals with district magnitude. The greater

the average district magnitude in a polity with closed lists, the greater the supply of cor-

ruption, all else equal. Again, the prediction stems from the impact of district magnitude

on the quality of the oer the party leadership is able to make to militants in the public

administration. In a polity with high magnitude districts, a high ranking list position is

worth more (in terms of the total list vote share necessary to attain a seat) than it would

be in a polity with low magnitude districts. As long as party leaders are able to engineer

a means to funnel party resources to militants that participate in the corruption scheme,

then the relationship between district magnitude and the supply of corruption holds for

open list systems as well. In the open list case, greater district magnitude means that the

expected party vote threshold necessary to obtain a seat is lower than it would be if district

magnitude was low, and that increases in targeted support are more eective in reducing a

militants expected party vote threshold (see figure 4.7).

The final system level prediction applies only to open list systems. In such systems, the

greater the impact of party resources on within-list competition, the greater the impact of

targeted support, and by extension, the greater the supply of political corruption. If said

223
resources have a large impact on a candidates standing vis-a-vis fellow party members, then

this means that targeted support will produce a high expected list position for militants

participating in the scheme, thus increasing the attractiveness of the nomination and the

proportion of militants willing to participate and, pari passu, increasing total party resources

and the value of the nomination even further. The result may be a significantly greater

supply of corruption than would be present were the impact of party resources on intraparty

competition low.

In terms of party level characteristics, the model generates two sets of expectations.

Firstly, it predicts that parties which have had strong electoral support without illicitly

using state resources will be more eective in generating resources through corruption once

they have acquired access to the state. The reasoning underlying the prediction follows the

same pattern of logic as above. Parties which have a significant natural vote share are able

to present a given militant with a good chance of attaining elected oce, even if no other

militants participate in the scheme. However, given that all militants know this, many will

choose to participate, bolstering the partys resources and electoral strength and increasing

the value of the nomination even further.

It is worth noting that this prediction paints a picture of party evolution which is strictly

at odds with an influential hypothesis developed by Martin Shefter (1977). Shefters ar-

gument is that a partys inclination to use patronage is determined by the partys "foun-

dational experience." To be more precise, he argues that parties which initially cultivate

their electoral following without having had access to state resources are unlikely to rely on

patronage as an electoral strategy if and when they do get access to the state. According

to the author, this is because a mobilizational experience which takes place without the

benefit of patronage is one which selects for leaders and constituents who hold ideological

224
commitments dear relative to material ones. This chapters perspective is really quite dif-

ferent. The model leads us to expect that it is precisely those parties which have developed

substantial electoral bases prior to attaining power which will be most eective in orga-

nizing corruption once they actually do come into power. As the model assumes strictly

oce seeking politicians, I cannot make claims about how the ideological commitments of

leaders do or do not aect this process. Nevertheless, I believe it is a mistake to assume

that ideologically oriented leaders would keep themselves from using state resources for

electoral advantage, especially when the characteristics of their party would allow them to

do so with particular eciency. The more committed the leadership to particular ideals and

programs, the greater weight they will place on the importance of the party winning future

elections, thus opening the door to justifications of political corruption. Similarly, the more

committed the party base to the ideals held by the leadership, the more willing they may

be to turn a blind eye to unsavory practices deemed necessary to attractive floating voters.

The final expectation generated by the model deals with party organization. Although

the chapters principal concern up to this point has consisted of the formal structure of elec-

toral law, party statutes and informal practices might magnify or retard some of the eects

of ballot structure. Examples of relevant aspects of party organization include the rules

which govern candidate selection and the control exercised by the party leadership over the

allocation of financial resources or advertising time. The model provides a straightforward

way of thinking about the eects of these features of party organization. The more either is

centralized in the hands of the party leadership, the greater is the potential for corrupt deal

making as we have conceptualized it. Similarly, the more either is decentralized, the weaker

the commitments leaders can make reward militants with plum nominations in exchange

for services rendered in the public administration.

225
 Table 4.1 summarizes the implications of the model. Closed list systems with high

district magnitude have the greatest potential to generate substantial political corruption,

while open list systems with low district magnitude and scant within-list returns to campaign

finance have the least. Two configurations have the potential to produce an intermediate

to high supply of political corruption: closed list PR with low district magnitude and

open list PR with high district magnitude and significant within-list returns to campaign

finance. Without ancillary information about the organization of parties in these systems,

a precise ranking cannot be given to one configuration relative to the other. A similar

situation obtains for the two configurations categorized as having the potential to produce

an intermediate to low supply of political corruption: open list systems with low district

magnitude and high within-list returns to campaign finance and open list systems with high

district magnitude and low with-list returns to campaign finance. The ranking of these two

configurations relative to the other configurations presented in Table 4.1 is clear; however,

the model in and of itself does not make a strong prediction about where one stands relative

to the other.

Table 4.1. Structural Features of Electoral Systems and the Supply of Political Corruption

226
6 Conclusion

How does our discussion of ballot structure and the supply side dynamics of political cor-

ruption inform our understanding of the overall level of political corruption likely to emerge

in a given polity? Ballot structure partially determines the degree to which control over

political careers is centralized in the hands of party leaders. Electoral systems which cen-

tralize control of political careers in the hands of relatively few elites create patterns of

career advancement based on cultivating a high level of personal capital with the gate-

keepers of elected oce. Within the legislative arena, this often has the salutary eect of

solidifying party discipline (Coppedge 1994, Crisp 1997, Jones 1997). Within the public

administration, this dynamic can take on more sinister dimensions. For politically oriented

bureaucrats making a pit stop in the public administration before heading on to other pur-

suits, the leverage exerted by party leaders can be considerable. The nature of party-centric

electoral systems provides party elites with political bargaining chips which are dicult to

match in systems that encourage intraparty competition. The result is that in party-centric

systems, colonizing public institutions with loyalists who seek to build a political career is

a relatively ecient means of redirecting state resources to explicitly partisan activity. All

else equal, convincing bureaucrats to run the risk of legal sanction by engaging in acts of

political corruption is a more challenging task in systems where career control is weak or

decentralized.

Because patronage is relatively ecient in party centric systems, it follows logically that,

holding demand for resources constant, its incidence should be greater. Let us imagine that

governing party elites face a trade o in assigning posts in the public administration (cf

Geddes 1991). On the one hand, they can assign available posts primarily along meritocratic

grounds and relegate partisan criteria to the backburner. In so doing, the party may

227
experience a boost of electoral support due to the aggregate welfare benefits associated with

ecient administration. On the other hand, they can assign posts to loyalists and exert

pressure to have institutional resources devoted to party activities. In so doing, the party

may experience an electoral boost from having militants reroute institutional resources. In

the case of party-centric electoral systems, party elites can eectively condition advancement

within the party ranks, and thus political careers, to the fulfillment of obligations imposed

upon the rank and file. As a result, for any given number of militants assigned to the public

sector, the potential for extraction should be greater in these systems than candidate-

centric ones. Since the reputational cost of assigning posts to militants is equivalent in

either scenario, patronage will be a more attractive strategy in party-centric systems than

candidate-centric ones. In colloquial terms, patronage provides greater bang for the buck

(or buck for the post, if you will) in party-centric systems. Figure 4.8 illustrates the linkages

in both the demand and supply side accounts of ballot structure and political corruption.

Figure 4.8. Ballot Structure and Political Corruption: Two Distinct Causal Pathways

228
 The reader will note that the distinct causal pathways presented in figure 4.8 suggest

that ballot structure may have an ambiguous overall eect on political corruption. As noted

by the traditional paradigm, electoral systems which encourage intraparty competition ex-

ert considerable demand side pressure in favor of greater political corruption. However, as

shown by the model, these systems are also characterized by more severe agency problems

between rent hungry politicians and bureaucrats, and thus the supply of political corruption

may be depressed relative to party-centric systems. Conversely, electoral systems which

concentrate power in the hands of party elites and proscribe intraparty competition will

likely have modest aggregate demand for illicit resources relative to candidate-centric sys-

tems. Nevertheless, the potential for extraction in such systems is large relative to their

candidate-centric counterparts. The consequence of the existence of countervailing causal

pathways by which ballot structure aects political corruption is that its overall eect may

not be one of frequency or volume, but rather one of type. That is to say, ballot structure

may play a crucial role in defining the nature of political corruption (candidate-directed,

decentralized forms of corruption vs. party-directed, centralized forms of corruption) but,

taken by itself, it may not be a good predictor of the level of political corruption.21

To summarize: electoral rules have distinct demand and supply side consequences for

corruption. Aspects of electoral rules which magnify demand may diminish supply and

vice-versa. Starting with first principles, this chapter has developed an original account

of the supply side of political corruption, one that rests on the rational actions of bureau-

crats, party leaders and the characteristics of relevant political institutions. In this way,

the chapter has conceptualized political corruption as a problem whose dynamics are best

21
The emphasis here on how electoral institutions determine the types of political corruption is similar to
a point made by Rose-Ackerman (2001), who argued that party-centered presidential democracies were more
prone to "legislative corruption" and candidate-centered presidential democracies more prone to "presidential
corruption."

229
elucidated through general equilibrium analysis, understood as the examination of political

and bureaucratic choice in a framework attentive to the fact that the returns available from

dierent courses of action in one sphere are conditioned on the choices and institutions

which characterize the other.

Given the findings of the model, what can we say about the potential impact of electoral

reform in Latin America and elsewhere? Most importantly, I would caution reformers not

to think of ballot reform as a comprehensive solution to the problem of political corruption.

By ameliorating one manifestation of the problem, ballot reform may intensify another.

Nevertheless, in order to merit acceptance by policymakers and the larger scholarly com-

munity as an important consideration in reform proposals, the chapters hyptheses about

the supply side consequences of ballot structure must first stand up to empirical scrutiny. In

the following chapter, the monograph takes that step. Therein, I test the hypotheses about

the supply side determinants of political corruption using extensive survey data produced

by research in Bolivia, Brazil and Chile. It is to this task that I now turn.

7 Appendix

Proposition 1 The maximum level of revenue extracted from the state is a concave in-

creasing function of the expected probability of obtaining elected oce.

Proof. Using the envelope theorem and the fact that is increasing in pe , we have: pe =

pe 0. Similarly, we have concavity because is a concave function of pe : pe pe =

pe pe 0.QED.

230
 Chapter 5
Political Career Paths in the Bureaucracy and the Use
of Institutional Resources in Bolivia, Brazil and Chile

231
1 Introduction

If Brazil or any another country contemplating major electoral reforms were to trade in

its current configuration of electoral rules for another configuration, what would the conse-

quences be in terms of the quality of governance? Would the resulting benefits of such a

transformation exceed its attendant costs?

Providing answers to such questions, even a furtive ones, is no mean feat. Consider a re-

form in ballot structure, the principal focus of this monograph. There are many dimensions

of representation for which a reform in ballot structure might matterparty discipline, the

nature of constituency service, political corruptionjust to name a few. So in order to get a

handle on the impact of a reform in ballot structure, considerations of time and comparative

advantage would demand that analysts narrow the scope of analysis to a small subset of

these dimensions. In this monograph, I have tried to do just that, dedicating my attention

exclusively to political corruption.

And yet the challenges to inference go beyond establishing a suciently narrow aca-

demic focus. One must also be forthcoming about the limitations of contemporary political

analysis. The study of macro-level institutional change in the social sciences oers us no

randomized trials, and even if it did, there is no guarantee that the average eect of insti-

tutional change identified in a given sample would correspond to the outcomes which would

emerge in Brazil or any other country contemplating reform. Therefore, an exact answer

to the question of what would happen in a particular country if it adopted a particular

political reform is unavailable to us. Potentially relevant components of political history,

national culture and social organization are so numerous and their eects so subtle that any

attempt to assign a particular point estimate to the eect of reform is untenable.

This important caveat notwithstanding, the tools of modern social science are well suited

232
to illustrate the types of tradeos which face reformers and to provide an account of the

mechanisms by which institutional change exerts an impact on the quality of representation.

The benefits therefrom are considerable. By alerting reformers to the principle shifts in be-

havior engendered by a reconfiguration of institutions, policymakers know what qualitative

changes to expect from reform ahead of time, even if priors about the precise impact of re-

form are diuse. Moreover, by illustrating the channels through which a particular reform

exerts its impact, reformers may be given the option of implementing ancillary reforms to

mitigate unsavory byproducts associated with a given reform proposal.

The last chapter and the present one are designed to provide a grounded analysis along

these lines for helping us to think about the impact of ballot structure. In the previous

chapter, I utilized a formal model based on the principles of non-cooperative game theory

to examine the supply side consequences of ballot structure on political corruption. In this

chapter, I evaluate empirically the key intuitions of the model, utilizing the results of the

public employees survey that I conducted in Bolivia, Brazil and Chile.

The motivation underlying the formal model was the need to correct for a patent

demand side bias in the literature on ballot structure and political corruption. In this

regard, I noted that configurations of electoral rules which encourage intraparty competition

and elicit high demand on the part of politicians for campaign resources may at the same

time make the extraction of resources from the state more dicult. Similarly, I noted that

configurations of electoral rules which proscribe intraparty competition and centralize power

in the hands of party leaders may facilitate the extraction of resources, even though the

demand that individual politicians have for resources in such contexts may be more muted

than in others.

If my intuition on this point is correct, then important implications for electoral reform

233
follow. In eecting reform towards closed lists from an open list PR system, the path con-

templated by Brazil, a country may experience the benefits of having fewer politicians cut

deals with local entrepreneurs to finance campaigns or engage in other manifestations of

what could be called free agent forms of political corruption. As this type of behavior is

believed to spring from the financial burden associated with self-promotion in an open list

system, closed lists should reduce its incidence. On the other hand, such a transformation

in electoral structure may engender greater political corruption of a dierent stripewhat

we might call machine politics variants of political corruptionby which is meant the

organized pilfering of public resources by networks of party militants in the public admin-

istration.

In the pages that follow, I draw upon the survey data to examine the validity of the

monographs original contribution to the debate on electoral rules and corruption: the no-

tion that electoral rules have important supply side consequences for political corruption. I

seek to examine what evidence, if any, exists in favor of the purported causal linkage between

electoral institutions which provide party leaders with centralized control over political ca-

reers and the ecacy of party corruption networks within the public administration. To the

extent that I discover a positive findingevidence of such a linkI will have established a

rationale for analysts to tread cautiously in extolling the virtues of moving from one format

of PR to the other. Before Brazil or any other country invests scarce resources in a radical

reorganization of its political institutions, the pertinent tradeos associated with doing so

must be understood and appreciated.

234
2 From Models to Evidence

2.1 Overview

As we transition from the relatively pristine realm of analytical politics to the more complex

realm of real world politics, it is worthwhile to elaborate the main intuitions of the analytical

model and clarify the relevant propositions to be tested. I do so here.

My point of departure for examining the impact of ballot structure consisted of an

application of the principal agent model with hidden knowledge. In the model, a party

leader elicits tribute from party militants in the public administration in exchange for

the opportunity to run for legislative oce under the party label. Electoral institutions

and party characteristics determine the value of the nomination given to would be co-

conspirators, which in turn determines the scope of participation in the corruption scheme

and the total amount of resources extracted from the state.

This framework produced a set of relatively sharp expectations about the impact of

ballot structure and district magnitude on political corruption. Closed list PR systems

were held to lend themselves more readily to the illicit use of the state as a means of

financing political activity than open list PR systems. Both types of list PR systems, in

turn, were held to dier in this regard according to the level of district magnitude, with high

magnitude systems being expected to display a greater penchant for supply side political

corruption than low magnitude systems.

In the model, the tendency of closed list systems towards supply side political corruption

stems from the heightened political career control such systems invest in party leaders.

The central idea is that closed list systems facilitate corrupt bargains between the party

leadership and politically ambitious militants within the public administration. Closed lists

provide a technology that enables party leaders to advance the political career prospects

235
of militants who put themselves at risk for the party, and in so doing, encourage greater

involvement in corrupt activity.1

The models implications about the impact of district magnitude stem from the op-

portunities that high magnitude systems present to party leaders to pick winners among

listmates. In a closed list system, high magnitude districts give party leaders substantial

leeway to allocate attractive positions on lists to loyal militants. A position that might be

good in a low magnitude system will be great in a high magnitude one, in the sense that

the proportion of competitors on the list with a lower ranking will be significantly larger

in the second scenario than the first. If co-consipirators are awarded with good positions

on lists, then high magnitude systems will encourage significant participation in corruption

schemes, with substantial resources flowing to the party and nominations being particularly

valuable.2

In the open list case, the causal mechanisms elaborated by the model were somewhat

dierent. Here I considered an arrangement I dubbed targeted support. In this scenario,

party leaders encourage participation in corruption schemes by selectively allocating elec-

toral support to the campaigns of compliant militants, thus favorably dierentiating these

individuals from their fellow listmates. Funneling resources into intraparty competition in

this way allows party leaders to make the oer of a nomination more attractive than it

1
Although I modeled this relationship as a direct one, with a party leader transacting directly with a
large number of militants in the public administration, I recognize that in many contexts these relationships
may be considerably more complex, perhaps characterized by party leaders negotiating with political brokers
who in turn negotiate with militants lower down in the administrative hierarchy. Be that as it may, such
chains of principal agent relations ultimately depend on mechanisms of control at the elite level of the party,
and without them party corruption networks are likely to break down.
2
For a variety of reasons, leaders may choose not to give militants involved in corruption schemes the very
top positions on the list: such positions have high visibility and assigning them to cronies risks sullying the
partys image, debts that the party has incurred to labor unions or other civil society organizations forces
leaders to reserve certain list positions for individuals outside the ranks of the rank and file, and internal
splits may force the party leadership to allocate such posts to faction leaders in order to prevent a break up
of the party. That being said, leaders can still give pretty good positions to whom ever they want, and as
magnitude increases, getting a pretty good position can become almost a guarantee of getting a seat.

236
would otherwise be, thus eliciting greater participation in the scheme and more resources

for the party. The greater district magnitude, the greater the impact of such arrangements,

and thus the more extensive participation in the scheme.

2.2 Propositions to be Tested

One of the virtues of elaborating formal models in the social sciences is that they lend

themselves to hypothesis testing at multiple levels (Morton 1999). Firstly, like all well

articulated social science theory, formal models often produce precise comparative statics

predictions, i.e. hypotheses about how variation in exogenously given parameters induces

change in endogenously determined outcomes. Secondly, as the core assumptions of formal

models are necessarily explicit, these, too, can receive empirical scrutiny. Such scrutiny

is perhaps especially important in models, like the monographs, which derive predictions

about aggregate behavior based on a well-defined theory of individual choice. To the extent

that one can use empirical evidence to establish the plausibility of the postulated micro-level

mechanisms, the aggregate level findings may receive additional support. Finally, formal

models can be tested by evaluating their logical implications. This may entail testing

propositions which do not emerge directly from the core assumptions of the model but

which follow from the models general mode of reasoning. In this case, we may evaluate a

model by applying its core logic to choices over a decision set not originally contemplated

by the formal model.

In what follows, I evaluate the formal model by conducting hypothesis tests that cor-

respond to each of the strategies mentioned above. Let us consider each proposition in

turn.

237
2.2.1 Direct Implications of the Model

Electoral rules and the scope of participation in political corruption My first

hypothesis concerns the impact of specific political institutions on , the proportion of

party militants in the public administration who engage in illicit behavior for the benefit

of their party. As the reader will recall, a transformation in institutions which allows party

l
leaders to make better oers to militants (i.e. produce a lower E[ k+1 ]) leads to a higher

value of p , the equilibrium value of attaining a seat if oered a nomination. A shift from

open to closed lists or from low district magnitude to high district magnitude (holding

ballot structure constant) would each constitute such a transformation. The increase in p

resulting from such changes, in turn, engenders an increase in the proportion of militants

willing to extract, as can be seen by dierentiating with respect to p :

q()f (.)
p = >0 (5.1)
p2 [1
q()]

Importantly, equation (4.2) states the proportion of militants willing to extract is not

only a function of p , and by extension, electoral institutions, but also a function of q(.) and

, the probability of being caught engaging in extraction and the punishment associated

with it, respectively. Thus, in generating predictions about the impact of electoral rules on

, the model requires that we hold q(.) and constant. With this in mind, the comparative

statics predictions are the following:

H1a: Holding the elements which define the riskiness and potential cost of corrupt

action constant, a transformation from open to closed lists will increase the proportion of

party militants engaging in party-directed corruption.

H1b: Holding ballot structure as well as the elements which define the riskiness and

potential cost of corrupt action constant, a transformation from a low magnitude electoral

238
system to a high magnitude electoral system will increase the proportion of party militants

engaging in party-directed corruption.

Notice that the hypotheses are conditional statements. Conditional on the factors which

determine the riskiness and potential cost of corrupt action, certain features of electoral

rules are hypothesized to have a certain eects on political corruption. We can think of the

first set of factors, those which determine the risk of corrupt action, as the characteristics

of the public bureaucracy which define its monitoring capacity. Here I have in mind the

quality of control mechanisms provided by internal auditing organs, the ecacy of the

politys supreme auditing institution as well as the informal mechanisms of vigilance and

control that exist within each public institution. The second set of factors, those which

determine the potential cost of corrupt action, can be thought of as the characteristics of

the legal system which determine the likelihood and severity of punishment in the case that

corrupt action is detected. Here I have in mind the existence of various features of the legal

environment which may subvert the rule of law: the abuse of injunctions to delay trial,

the political penetration of the judicial branch, and most obviously, corruption within the

judiciary itself.

Electoral rules and the impact of political ambition The second hypothesis concerns

the impact of political ambition on the propensity of public employees to use institutional

resources for partisan ends. It stands to reason that public employees interested in a political

career will be more inclined to use institutional resources for partisan purposes than those

without such an interest. That being said, the model leads us to expect that the magnitude

of the impact of political ambition will be mediated in a significant way by the format of

electoral institutions.

To see this, let us reconsider inequality (4.1). Suppose that there is an element of

239
uncertainty involved in militant is choice, such that (4.1) can be rewritten as:

[1 q()]p ri + q() i > 0 (5.2)

where i has a continuous cdf G. In this formulation, i represents an unobservable

aspect of is psychological disposition which modifies the expected utility he receives from

extracting resources. We might think of i as representing random events which exert an

influence on is moral bearings, such as a death in the family, sudden financial diculties

and the like. Were we privy to ri , we could characterize the probability that militant i

extracts resources from his institution as:

G([1 q()]p ri + q()) (5.3)

We consider the variable ri , militant is personal ego rent for holding elected oce,

to be synonymous with is level of political ambition. Thus, ri = 0 denotes that i has no

political ambition and ri > 0 denotes that i has some positive level of political ambition. To

characterize the marginal impact of increasing political ambition from 0 on the probability

that a militant extracts resources, we take the partial derivative of G(.) with respect to ri

for ri = 0. This gives us:

G(.)
= g(q())[1 q()]p > 0 (5.4)
ri (0)

The above states that politically ambitious bureaucrats are always more likely to extract

resources than politically unambitious bureaucrats, irrespective of the electoral system in

force. Note that the magnitude of this eect, however, is dependent on the configuration

of political institutions. We can see this by examining the cross partial derivative:

240
 2 G(.)
= g(q())[1 q()] > 0 (5.5)
ri (0) p

Thus, the micro-level model of bureaucratic choice implies that the impact of political

ambition is a positive function of the level of p , which is endogenous to the electoral system

in the manner stated above. As was the case with our previous hypothesis, my expectations

about the impact of electoral institutions must be stated in conditional form, as q(.) and

play an important role in determining the impact of political ambition on the propensity

to use resources for partisan ends. They are as follows:

H2a: Holding the elements which define the riskiness and potential cost of corrupt action

constant, a transformation from open to closed lists will increase the impact of political

ambition on the propensity to use institutional resources for a party.

H2b: Holding ballot structure as well as the elements which define the riskiness and

potential cost of corrupt action constant, a transformation from a low magnitude electoral

system to a high magnitude electoral system will increase the impact of political ambition

on the propensity to use institutional resources for a party.

2.2.2 Indirect Implications of the Model

Electoral rules and job allocation If the above hypothesis about the impact of elec-

toral structure is valid, then it produces a clear (and rather sinister) implication about

job allocation in the public administration. In contexts in which the configuration of elec-

toral institutions creates a high propensity for politically ambitious bureaucrats to engage

in party-directed corruption, party leaders will do their best to distribute posts in a way

that allows them to take advantage of this state of aairs. In order to produce a maximal

impact from the influence they exert, leaders will be particularly inclined to place party

241
loyalists in the type of posts known to have significant bribe-generating potential. Thus, we

would expect political loyalty to be an important litmus test for access to jobs that have the

potential to generate resources from bribes. In electoral systems where leverage is weaker,

the appeal of politicized hiring for such posts is less clear cut and meritocratic factors may

play a larger role. Stated formally, my expectations are the following:

H3a: All else equal, politically ambitious bureaucrats will be more likely to hold posts

with significant bribe-generating potential in closed list systems than in open list systems.

H3b: Holding ballot structure constant, politically ambitious bureaucrats will be more

likely to hold posts with significant bribe-generating potential in high magnitude systems than

low magnitude systems.

2.3 Electoral Structure and the Proclivity for Supply Side Political Cor-

ruption in Bolivia, Brazil and Chile

The above are the central propositions that I bring to the data. Although they are het-

erogeneous in terms of the specific outcomes they purport to explain, they all point in a

similar direction, namely, that of closed list and high magnitude electoral systems produc-

ing a tendency towards supply side political corruption. In order to use the administrative

experience of our countries to evaluate these hypotheses, I must first specify the manner

in which the analytical model maps into expectations about observable outcomes in the

cases. I take that step here. I briefly describe the key characteristics of electoral rules and

party organization in the three countries and apply the framework of the previous chapter

to each. By design, the cases exhibit substantial variation in terms of the degree of cen-

tralized leadership control over political careers and other relevant aspects of the electoral

system. As a result, I am able to generate relatively sharp expectations about patterns of

bureaucratic behavior in the three countries.

242
 As the reader will recall, my expectation about a countrys proclivity for supply side

political corruption concerns the degree to which formal electoral rules and informal party

organization concentrate control over political careers in the hands of party leaders. In

this regard, Bolivia is clearly the case we would expect to exhibit the most supply side

political corruption. Bolivia can be categorized as having a hyper-centralized electoral

system (Crisp 2002). This country has a closed list system of proportional representation

which dates back to 1956, originally introduced by the Movimiento Nacional Revolucionario

(MNR) party as a gambit to produce centralized leadership control in what was to be

a new, one party state (Kaufman Purcell 1972, Gamarra and Malloy 1995). Although

the one party state never materialized (instead replaced by a succession of alternatively

populist, conservative and venal authoritarian governments from 1964 to 1982), closed list

proportional representation became a lasting feature of the political environment (Gamarra

1997). As competitive elections reemerged in the early 1980s, the old practices pioneered by

the MNR leadership, among which included the assiduous selection and ranking of favored

candidates on closed lists in accordance with services provided to the party and party

leaders, were generalized across the party system.

As mentioned earlier, large district magnitude increases the value of high ranking nom-

inations made to the rank and file. In this regard, district magnitude in Bolivia has had

considerable potential to solidify corrupt contracts. In Bolivia, legislative candidates run for

oce in one of nine departments. Until 1994, average district magnitude for contests in the

Chamber of Deputies was fairly large (slightly over 14), although the reforms of that year

reduced magnitude for candidates running on closed lists to about 7. For those candidates

running in the more populated departments (La Paz, Santa Cruz, Cochabamba, Potos),

having a good list position remains of pivotal importance, as magnitude therein varies from

243
7 to 15.

Other aspects of Bolivian electoral law have conspired to produce a highly centralized,

party-centric system. Until 1994, Bolivian elections were characterized by a fused ballot,

which meant that voters cast a single party list vote for the President and representatives

in the Chamber of Deputies and the Senate. In modified form, this system remains in force

to this day, although currently half the members of the Chamber of Deputies are elected in

single member districts (the other half elected on closed lists with the fused ballot). The

inclusion of a fused ballot assures the alignment of incentives between legislative candidates

and the party leadership. To the extent that party leaders are strongly wedded to their

presidential candidate (or in the case that the presidential candidate is the party president

an empirical regularity in Bolivia), anything legislative candidates might do to improve their

own electoral prospects necessarily improves the lot of the leadership and vice-versa.

By in large, the organizational features of Bolivias parties have served to reinforce the

party-centrism of the electoral system. Although Bolivian parties are marked by noticeable

dierences in certain formal aspects of party organization, such as procedures and rules

outlined in party statutes, in practice they have been dominated by the whims of a strong

central personality and a small coterie of advisors (Teresa Zegada 1996). This has been

the case for Bolivias so-called traditional parties, the MNR, the MIR (Movimiento de la

Izquierda Revolucionaria) and the ADN (Accin Democrtica Nacionalista), as well as for

the populist parties that styled themselves as an alternative to them and even the indigenous

parties that fancy themselves as an alternative to Bolivias reigning political class (Costa

Benavides 2004).

The Brazilian case inhabits an intermediate position in terms of its propensity towards

supply side political corruption. As noted in the previous chapter, in this country can-

244
didates for the federal Chamber of Deputies as well as state and municipal assemblies are

elected according to open list proportion representation. The origin of open list proportional

representation in Brazil is to be found in first period of mass democratic competition, the

so-called Old Republic (1945-1964). Although the military government which took power

in 1964 completely restructured Brazils party system, the electoral system remained in-

tact, with patterns of political representation in the New Republic (1985-present) mirroring

those of the Old Republic in a host of important ways (cf. Ames 1987, 1995, Hagopian

1990, Amorim Neto 2000, Amorim Neto and Santos 2001).

According to my framework, the potential for open list systems to sustain machine

politics is partially determined by district magnitude, with high magnitude systems lending

themselves to more eective targeted electoral support provided by party leaders to favored

militants. In this regard, Brazils uncommonly large magnitude districts certainly create

the potential for successful targeted support.

Average district magnitude is approximately 19, with about 40% of federal deputies

elected in megadistricts where magnitude varies from 39 to 70 (Nicolau 1996). To give an

idea of the implications that high magnitude may entail for the ability of party leaders to

selectively target resources in order to pick winners among listmates, consider the situation

of a candidate belonging to the PTB party running federal deputy in the state of So Paolo

in 1998. This individual seeks one of 70 seats in his district, along with 615 candidates

from competing parties and 45 from his own. Of his many listmates, only five will get seats.

Of these five, the dierence between the proportion of votes captured by the two lowest

vote getters that attain seats, relative to total individual (nominal) votes captured by the

list, is remarkably small1.9 %. Thus, for our hypothetical candidate, any financial or

logistical support he can obtain from the state party leadership, even if it has a relatively

245
small impact on his within list vote share, may have a big impact on the probability that

he lands a seat. In contexts where remarkably small swings in intralist vote share can mean

the dierence between getting a seat and not getting one, a little favoritism can go a long

way.

In terms of actually targeting electoral resources towards favored candidates, however,

it appears that many Brazilian parties have by in large foregone the opportunity presented

to them by high magnitude districts. Legislative candidates in most parties are responsible

for raising funds on their own (Samuels 2004). Party elites do regularly campaign on

the behalf of favored candidates, although the impact of these eorts is hard to gauge.

Nevertheless, the targeting of electoral resources towards favored candidates has been an

important practice in at least one party, which also happens to be the party with greatest

representation in the Chamber of Deputies and the party which currently controls the

presidency: the Partido dos Trabalhadores (PT). Relative to other Brazilian parties, the

PT exhibits strong and stable organization. It possesses a well-defined hierarchy linking its

bases units, the ncleos de base, all the way up to the central party leadership (Meneguello

1989, Keck 1992). Moreover, recent research documents the fact that the PT leadership

has been able to eectively allocate a prized resource, legally fixed television time, in a

way that essentially picks winners among listmates (Samuels 1999). Unfolding revelations

of massive illegal campaign financing in this party, apparently organized and controlled by

senior party ocials, point to the distinct possibility that illicit resources, too, may have

been used to pick winners.3 Given both Brazils configuration of electoral rules and the

importance of the PT at the time I conducted the survey, and given the fact that five of the

3
In this regard, Delbio Suarez, former treasurer of the party and allegedly a mastermind of the scheme,
stated in a sworn deposition that all PT candidates received clandestine funds. He later corrected himself
to exempt a single figure in the party, Brazils president Luiz Incio Lula da Silva (Folha de So Paolo, July
16, 2005).

246
ten institutions I surveyed were controlled by that party, the Brazilian case is categorized

as exhibiting an intermediate/high penchant for supply side political corruption.

Of the three countries included in the survey, the Chilean case exhibits the weakest

propensity towards supply side political corruption. Like Brazil, Chile has an open list

version of proportion representation, with seats allocated to lists according to the DHondt

formula. These features of the electoral system have a long historical legacy in Chile, being

inherited from the vibrant multiparty democracy that existed for forty years prior to the

onset of military rule in 1973. Unlike the other cases, however, military rule did have an

important impact on the basic contours of the electoral system. Whereas Chiles pre-1973

electoral system contained 28 multimember districts with an average district magnitude

of 5.4, the system bequeathed by the military government contains sixty districts, each of

which have a magnitude of 2. The military leadership hoped that the introduction of two

member districts would reduce the fragmentation of the party system and provide a boon

for parties of the right (Siavelis 2000). Only the latter seems to have occurred (Siavelis

1997, Valenzuela and Scully 1997).

In terms of restructuring the incentives which dictate levels of political corruption, the

binomial system, as it is called, has arguably had a substantial impact. The two member

district system makes the establishment of corrupt contracts the most dicult among our

cases. The reason for this rests with the combination of two member districts with another

legacy bequeathed by the military government: an enduring and intense ideological cleavage

between sympathizers and opponents of the military regime.4

An important result of this combination of factors has been the fact that the member

4
To this day, a coalition has never formed between parties who find themselves on opposite sides of this
division. In fact, the political coalitions that emerged to contest the 1988 plebiscite on the military govern-
ments continuation in power have remained unchanged to the present date, despite substantial heterogeneity
exhibited in terms of preferences towards social and religious issues (Londregan 2000).

247
parties of the Concertacin, which represent the left and center left of the political spectrum,

always run their legislative candidates in each district on a single list, as do the member

parties of the Alianza, which represent the right and center right. Because the coalitions

can present only two candidates in each district, the nomination of candidates for particular

districts is a matter of intense and protracted negotiation among the elites of the member

parties (Navia 2004, Siavelis 2002). The need to negotiate nominations with fellow coalition

members, in turn, makes the type of corrupt contracts envisioned by the formal model very

dicult to achieve. A promise of a nomination in exchange for some service provided for

the party, for instance, is a priori incredible unless it has been approved by the leadership

of other coalition parties. To the extent that this service consists of corrupt activities,

the process of interparty negotiation over nominations may threaten to make such delicate

intraparty exchanges in question public, thereby making them unsustainable in the first

place.

A second consequence of the two member district system is that it mitigates the impact

of any targeted support-type arrangements that might emerge. As the reader will recall, the

impact of such support depends on district magnitude. With only two seats per district and

two candidates per list, the marginal impact of electoral resources on the probability that

a favored candidate will get a seat is essentially nil unless one of three scenarios obtain: 1)

the favored candidates vote share is close to that of his listmate (for a list which is expected

to get at least one third of the vote but less than two thirds); 2) the favored candidate is

substantially behind his listmate in the polls but the list in question is close to doubling the

vote share of the opposing coalition; 3) the favored candidate is ahead of his listmate in the

polls and the list in question is close to attaining one third of the vote. Since all of these

occurrences are relatively rare, targeted support is not likely to be an equilibrium behavior

248
in the Chilean political system.

In sum, I expect Bolivias closed list and moderate magnitude proportional representa-

tion system to exhibit the greatest penchant for supply side political corruption, Brazils

open list and high magnitude system to exhibit an intermediate proclivity for the same and

Chiles open list and low magnitude system to exhibit the weakest proclivity towards supply

side political corruption. Table 5.1 maps out my general expectations for the cases based

on the relevant components of electoral rules.

Table 5.1: Electoral Structure and Expectations for the Cases

3 Testing the Model

3.1 Electoral rules and the scope of participation in political corruption

I begin the evaluation of the theoretical framework with an examination of the relationship

between electoral rules and the scope of partisan resource use. Here I use the survey data to

test the two components of hypothesis 1. According to H1a, I expect a greater proportion

of party militants to participate in political corruption schemes in closed list systems than

open list systems. In terms of the cases, this means that if we could produce an estimate of

for each country, I would expect b

BOL > b
BRA and b
BOL > b
CHL , where subscripts follow

249
the convention: BOL = Bolivia, BRA = Brazil and CHL = Chile. In words, I expect

that the proportion of party members in the public administration who use the resources

at their disposal for the benefit of a party will be greater in the closed list case than the

two open list cases.

According to H1b, I expect a greater proportion of party militants to participate in

political corruption schemes in high district magnitude systems than low magnitude systems,

holding ballot structure constant. Thus, for my cases I would expect: b

BRA > b
CHL . Again,

in words, I expect that the proportion of party members in the public administration who

use the resources at their disposal for the benefit of a party will be greater in the high

magnitude open list case than the low magnitude open list case.

In order to evaluate these claims, I must provide an estimate of for each case. How

can I estimate b
BOL , b
BRA and b
CHL using the survey data? I did so for each country

using the randomized response methodology described in chapter 2. One of the randomized

response questions queried respondents as to whether or not they had ever, even once, used

the resources of their institution for the benefit of a political party (see fn15 in Chapter 2 for

exact wording). Another, non-sensitive, question included in the survey asked respondents

whether or not they belonged to a political party. In order to estimate b

, I simply calculated

the proportion of party members which had used resources by applying formula (2.11). Table

5.2 presents the results.

As expected, the estimate of the proportion of party militants that had used resources

in the Bolivian case is larger than the estimates produced in the Brazilian and Chilean

cases. In fact, it is much larger than the estimates produced for the latter two cases. I

estimate that almost half (47.4%) of the Bolivian party members in the sample had used the

resources of their institution for the benefit of a party. This compares to 16.0% estimated

250
to have used resources among the Brazilian party members and 10.7% among the Chilean

party members.

Table 5.2: Proportion of party militants that have used resources for a party (RR estimates)

In order to check if the dierences in estimated proportions across the cases satisfy

standard thresholds of statistical significance, I calculated a two sample t-test for each pair-

wise comparison among the cases. The first two columns of table 5.2 present the empirical

hypothesis tests which correspond to H1a; namely, that the dierence between the Boli-

vian estimate and the Brazilian and Chilean estimates, respectively, should be positive and

greater than 0. The test statistic produced for the Bolivia-Brazil comparison was 28.1 while

the statistic produced for the Bolivia-Chile comparison was 3.70. Thus, in both instances

we can reject the null hypothesis of no dierence between the estimates at well above the

99% confidence level. The expectation of a proclivity of closed list systems towards supply

251
side political corruption seems to find prima facie support in the data.

The third column of table 5.2 presents the empirical hypothesis test which corresponds

to H1b; namely, that the dierence between the Brazilian and Chilean estimates should be

positive and greater than 0. In this instance, the test statistic I calculated was only .53,

which did not allow me to reject the null hypothesis of no dierence in estimates. Thus,

even though the dierence estimate was of the correct sign, based on the comparison of

proportions alone one cannot claim that the data support the thesis of a greater proclivity

of high magnitude systems towards supply side political corruption.

At this point in the presentation, the attentive reader will recall that hypotheses H1a

and H1b were originally stated in conditional form. That is to say, I desire to test the

proposition that political institutions have distinctive eects on supply side political cor-

ruption, holding the real world factors which correspond to the model parameters q(.) and

constant. Naturally, the question arises as to how we might use the survey data from three

distinct country cases in order to hold these factors constant. If the country cases have

distinctive reputations for factors such as the rule of law and accountability mechanisms in

the bureaucracy (which they do), is it then an impossible task to have these elements held

constant in our empirical analysis?

The short answer is no. As the reader will recall, chapter 3 demonstrated that the dataset

exhibits substantial variation in terms of institutional conditions and individual perspectives

within the public administration of each country. Thus, although in the aggregate the

country cases may seem quite dierent, at the individual level it is not too dicult to find

bureaucrats in the dierent countries who match up extremely closely (and in many cases,

exactly) in terms of their perspectives regarding the institutions in which they work. By

comparing only the responses of bureaucrats who match up in terms of the indicators of

252
q(.) and , I can in eect hold constant these features of the bureaucratic environment.

The task of holding constant these potentially confounding variables, in turn, is one that

can be accomplished using the matching techniques discussed in chapter 2.

The survey included a number of questions which capture aspects of the bureaucratic

and legal environment represented by the model parameters q(.) and . Table 5.3 displays

the questions which correspond to each. In terms of measuring the riskiness of corruption

action, respondents were presented with questions designed to gauge perceptions about the

quality of formal and informal control in their institutions. One of these questions asked

respondents to evaluate the likelihood that an internal or external audit or another type of

investigation would occur as a result of a misappropriation of resources in their institution.

Another asked respondents whether or not they believed that the majority of the employees

in their institution would file a complaint if they were aware of a misappropriation of

resources. A dierent survey question was keyed to the potential cost of corrupt action.

Placed immediately after the question on formal control, this question asked respondents

to evaluate the likely fate of the individual who had misappropriated the resources. The

question tapped into perceptions about accountability within institutions as well as the

ecacy of the legal process.

253
 Table 5.3: Covariates of political corruption

Using the questions presented in table 5.3, I conducted two types of matching analyses:

exact and nearest neighbor. Table 5.4 presents the estimates corresponding to each. Let us

consider first the estimates in the column of the table labeled "Exact", which consist of the

results for the exact matching algorithm. These estimates were calculated as follows. For

each country dyad (e.g. Bolivia-Brazil), I created a dataset of matched pairs in which the

members of each pair were party militants from two dierent countries who were otherwise

identical in terms of responses to the questions concerning the riskiness and potential cost

of corrupt action. In this way, I was able to hold these potentially confounding factors

constant.

254
 Using the exact matching strategy, the dierence between the Bolivian case and the

Brazilian and Chilean cases remains large and statistically significant. Thus, we can con-

clude that the dierences in the behavior of party militants for these pairings is not due to

the factors which determine the riskiness and potential cost of corrupt action. This finding

is clearly consistent with the expectations generated by the formal model. On the other

hand, the dierence between the Brazilian and Chilean cases remains statistically insignif-

icant, although the magnitude of the dierence is substantially larger than that estimated

without controlling for confounding variables.

Naturally, I do not believe that the formal model presented in chapter 4 necessarily

provides a complete account of all the reasons why a bureaucrat may use the resources to

his institution for the benefit of a party. In the vernacular of formal political theory, I view

the model as a partial data generating process, not a complete data generating process.

Thus, in producing conditional estimates of the dierence in the propensity of militants

to use resources across the cases, it is important to control for potentially relevant factors

besides those which fall under the rubric of the q(.) and terms. In what follows, I also

introduce controls measuring the adequacy of ones salary to cover basic needs, perceptions

about the likelihood of losing ones post in the near future and a proxy for personal ethics.

The exact question wording corresponding to these controls is presented in the lower half

of table 5.3.

The first two measures can be thought of as indicators of the opportunity cost of cor-

ruption. The higher ones legal wage, for instance, the greater the foregone income should

one lose ones job. Since losing ones job may be a consequence of being caught engaging

in illicit behavior, a higher legal income means that the potential cost of corruption is also

255
higher.5 The logic for including a measure of perceived job stability is similar. The greater

the perceived stability of ones post, the greater the long run income stream associated with

it, and thus the greater the cost of actions which risk the termination of employment.6 The

rationale for including a measure of personal ethics should be self-evident in the context of

governmental corruption.

The estimates in the column labeled "Nearest Neighbor", the nearest neighbor matching

estimates, control for all the covariates presented in table 5.3. Rather than creating a

dataset of matched pairs in which the two individuals in each pair are identical in terms of

the covariates (which is impossible as the number of covariates becomes large), the nearest

neighbor matching algorithm selects as members of each matched pair individuals who are

as close as possible, where closeness is determined by a distance metric discussed in the

appendix of chapter 2.

5
In this regard, a large literature on "eciency wages" has made a compelling theoretical and empirical
case for there being a negative relationship between the wage rate and corruption (cf Becker and Stigler
1974, Van Rijckeghem and Weder 2001, Di Tella and Schargrodsky 2003).
6
On the other hand, too much perceived stability may have the opposite eect. Legal regulations which
make the termination of public employees particularly onerous may empower them to engage in illicit be-
havior with impunity.

256
Table 5.4: Matching Estimates, proportion of party militants that have used resources for a party

Like the previous estimation, the nearest neighbor results confirm the more extensive

scope of partisan resource use among Bolivian party militants than their Brazilian and

Chilean counterparts. The estimated dierences in this regard all met standard thresholds

of statistical significance. The dierence between the Brazilian and Chilean cases, although

of the correct sign and substantively large, was not significant in the nearest neighbor

estimation.

Another way of conducting inference for the same hypothesis is to use all the survey

data in a parametric regression analysis. In so doing, one may examine the influence of

257
party membership in the three cases on partisan resource use, holding constant the factors

in table 5.3. In essence, this strategy consists of fitting a modified logistic regression to

the data, one which makes the necessary corrections for the RR data generating process

(see chapter 2). The advantage of this approach is that one can examine the eect of the

confounding variables, of interest in their own right, as well as conduct some basic tests of

model fitness. The drawback of the approach is that the parameter estimates are contingent

on the functional form specification of the model (e.g., logistic) and may be incorrect in the

presence of unknown interaction eects among the righthandside variables.

258
Table 5.5: RR-modified Logistic Regression of Party Militancy and Partisan Resource Use

Table 5.5 presents the results of this analysis. In addition to the covariates listed in table

5.3, I also included country dummies to control for any country-specific fixed eects. The

estimation shows that the selected regressors were clearly relevant predictors of partisan

resource use. The likelihood ratio statistic for the null hypothesis that the covariates have

no explanatory power, =0 , was such that we could strongly reject this hypothesis.

259
 Of interest are the coecients on the first three variables presented in the table, which

consist of interaction terms denoting that a respondent was a party member in Bolivia,

Brazil and Chile respectively (the omitted category is non-party member). According to

H1, we would expect the first coecient to be greater than the second, which should, in turn,

be greater than the third. Of the three interaction terms, only the coecient for the Bolivian

case was significantly dierent from zero at conventional levels. As expected, the coecient

estimated for the Bolivian case was larger than that estimated for the Brazilian case, which,

in turn, was larger than that estimated for the Chilean case. However, likelihood ratio tests

for dierences in the aforementioned coecients (see bottom righthandside of table 5.5)

would not permit me to reject the null hypothesis of no dierence across the cases.

Nevertheless, the overall dierences in the behavior of party militants in the three cases

were quite stark, as can be seen in figure 5.1. The figure shows the (distributions of)

the predicted probabilities that an average bureaucrat who is a party militant has used

resources in the three cases (all covariates are held at their mean values for the sample).

Like the matching analyses, the figure illustrates the dierences across the cases, holding

constant other potential determinants of partisan resource use. As is obvious upon visual

inspection, there is a marked dierence between the Bolivian case and the other two cases,

while there seems to be little dierence among the Brazilian and Chilean cases.

Also of note in the parametric analysis was the influence of perceptions about the ecacy

of control mechanisms and the adequacy of ones salary. As expected, the better bureaucrats

perceived formal control mechanisms to be in their institution, the less likely they were to

have engaged in an instance of partisan resource use. This eect was strong and statistically

significant. Informal control mechanisms and personal ethics did not seem to play a role in

this process. Perceptions about ones wage did exert substantial influence on resource use.

260
Individuals who felt that their wage covered the costs of basic necessities were significantly

less likely to have engaged in an instance of partisan resource than those who felt otherwise.

This result adds more micro-level evidence to the claim that corruption is negatively related

to the wage rate (cf Van Rijckeghem and Weder 2001, Di Tella and Schargrodsky 2003).

Figure 5.1: Predicted probability of partisan resource use among militants

.In sum, the empirical analysis of the relationship between electoral rules and the scope of

participation in partisan resource use gives prima facie support to the notion that closed list

systems may exhibit greater supply side political corruption. In this regard, the expectations

of the model have met with empirical validation. The empirical analysis has also provided

some evidence in favor of the thesis that high magnitude systems exhibit greater supply

261
side political corruption than low magnitude systems, although here the evidence is much

weaker.

3.2 Electoral rules and the impact of political ambition

I now turn my attention to the relationship between political ambition and partisan resource

use, and the role that electoral institutions play in mediating this relationship. According

to hypothesis 2, I expect that the impact of political ambition will be greatest in the closed

list case, Bolivia, that it will take an intermediate value in the open list and high magnitude

case, Brazil, and that it will have the smallest impact in the open list and low magnitude

case, Chile.

Let us begin the analysis with a visual inspection of the distribution of the key actors

in the formal model: politically-oriented bureaucrats. As figure 5.2 shows, there are sub-

stantially greater numbers of these individuals in the Bolivian public administration than

the Brazilian or Chilean public administrations. For the three cases, the figure shows the

proportion of employees that belong to a political party (total length of the horizontal bar)

and the proportion of employees that were categorized as having a political career (length

of light gray portion of bar).7 Overall, 97 respondents, or 9.7% of the total sample, were

categorized as having political careers in Bolivia, 44 respondents, or 3.7% of the total sam-

ple, were categorized as having political careers in Brazil and only 10 respondents, or 1.7%

of the total, were categorized as having political careers in Chile. As a first look at the

data, this finding jibes well with my expectations.

7
For the Bolivian and Chilean data, an individual was classified as having a political career if he responded
that he was currently a member of a party and was either actively involved in the aairs of the party or had
political aspirations for the future. In the Brazilian case, an additional set of questions allowed us the adopt
a slightly more expansive definition. For this country, an individual was classified as having a political career
if the above condition held or, additionally, if he was currently a party member and had run for elected oce
in the past or was a party member and expressed interest in doing so in the future.

262
 Figure 5.2: Party membership and political careers in the cases

My ultimate concern, however, is with establishing what politically oriented bu-

reaucrats do (and do not do) for the party while in the public administration. In order to

explore this issue empirically, I examined the responses of bureaucrats with respect to the

randomized response question discussed in the previous section, namely, that designed to

measure an individuals own participation in instances of partisan resource use. Due to a

lack of observations on Chilean bureaucrats with political careers (and the fact that RR is

quite data hungry), the Chilean case was dropped from the analysis that follows.

The first step in the analysis was simply to examine the dierence in estimated propor-

tions. For the Bolivian and Brazilian cases, I broke the data into two groups: bureaucrats

classified as having a political career and those who were not. The results are displayed in

the top row of table 5.6. In the Bolivian case, as expected, the dierence between those bu-

reaucrats who had political careers and those who did not was impressive: the former were

.23 more likely to use the resources of their institution for the benefit of a party than the

latter (approximately twice as likely). In the Brazilian case, the dierence between those

who had political careers and those who did not was less pronounced but still substantial:

263
the former were .15 more likely to use the resources of their institution for the benefit of

a party than the latter (approximately 3 times as likely). In both cases, the dierence in

estimated proportions was statistically significant at conventional standards.

Table 5.6: The Eect of Political Careerism in Bolivia and Brazil

As in the earlier analysis, I proceeded to use matching methods in order to control for

the eect of potentially confounding variables. The results of these estimations can be seen

in the middle and bottom rows of the same table. In the exact matching analysis, pairs of

bureaucrats from the same country were selected as matches if one individual was classified

264
as having a political career, the other not, and the two individuals were otherwise identical

in terms of their responses to the questions touching on the riskiness and potential cost

of corrupt action. The nearest neighbor analysis proceeded similarly, with the dierence

being that all the covariates in table 5.3 were controlled for, albeit not necessarily matched

perfectly.

The exact matching results by in large conformed to expectations. The eect of having

a political career in the Bolivian case was large and statistically significant. The estimated

eect of having a political career for a randomly selected Bolivian bureaucrat was to increase

his probability of using the resources of his institution for the benefit of a party by .21 (the

ATE). The estimated eect of political careerism for the average Brazilian bureaucrat was

markedly smaller, .06, and did not reach statistical significance. The overall dierence in

the careerism eect between the two cases was a noteworthy .15.

The nearest neighbor matching estimates also confirmed the greater importance of ca-

reerism in the Bolivian case. In this analysis, the careerism eect in Bolivia was estimated

to be .17, compared to a careerism eect of only .02 in the Brazilian case. The dierence

between the impact of careerism in the two cases was again large at .16, although for neither

case did the estimated careerism eect satisfy standard thresholds of statistical significance.

The final attempt at inference regarding dierences in the eect of careerism across the

cases consisted of a parametric regression analysis, conducted as in the previous section.

I ran the same regression as in Table 5.5, this time limiting the data to the Bolivian and

Brazilian cases. I also substituted the indicator variable for party membership with an

indicator variable for political careers. Table 5.7 displays the results of the analysis.

As expected, the coecient on the interaction term between the Bolivia dummy and

the indicator variable denoting that a bureaucrat had a political career was positive and

265
statistically significant. The coecient on the Brazil and political career interaction term

was also positive, and it was substantively large, but it failed to reach statistical significance.

Interestingly, the latter coecient was actually larger than that estimated for the Bolivian

case. Does this mean that the parametric analysis is suggesting that the marginal eect of

having a political career is greater for the average Brazilian bureaucrat than it is for the

average Bolivian one?

Table 5.7: RR-modified logistic regression of political careers and partisan resource use

Importantly, the answer is no. This can be seen by examining the graph of first dier-

ences shown in Figure 5.3. Using the results of the parametric analysis, I calculated the

266
predicted probability that the average Bolivian bureaucrat would have used the resources

of his institution for the benefit of a party if he had a political career as well as the corre-

sponding probability if he did not have a political career, holding all the other covariates

at their mean values (for Bolivia). I then conducted the same procedure for the Brazilian

case, again holding all other covariates at their mean values (for Brazil). As can be seen in

the graph, the careerism eect in the Bolivian case is substantively large and statistically

significant (.209, t-stat=2.30). The eect in the Brazilian case is smaller, and as can be

seen in the graph, estimated less precisely than in the Bolivian case (.155, t-stat=1.21).

Figure 5.3: Graph of first dierences from the maximum likelihood model

In sum, it seems clear that the eect of political careerism in the Bolivian case is larger

267
than in Brazil, although how much larger depends to a certain extent on the particular

estimation technique utilized. Moreover, it should be underlined that overall the frequency

of partisan resource use is substantially greater among politically-oriented bureaucrats in

the Bolivian public administration than in the Brazilian administration. This fact may have

important implications for how jobs are distributed among bureaucrats, which brings us to

the core concern of my final empirical test.

3.3 Electoral rules and job allocation

If it is in fact true that politically-minded bureaucrats are more inclined to use their ocial

positions to further the interests of their party in polities whose electoral and party systems

establish strong mechanisms of political career control, then other implications follow from

this. The most obvious concerns job allocation. I would expect that in such polities, party

leaders would do their best to see to it that loyalists are placed in posts which grant them

substantial control over resources and numerous opportunities to collect bribes. In this way,

political career control could be parlayed into substantial access to resources for the party.

In polities where political career control is weaker, the incentive to place loyalists in such

posts is less pronounced, since party leaders are less able to convince these individuals to

engage in risky behavior for the party. Moreover, even if loyalists do accept bribes, with

weak political career control leaders are less able to force them to share any ill gotten gains

with the party. In terms of the cases, I expect political careerism to have the strongest

(positive) eect for Bolivia, followed by that for Brazil. I would expect the weakest eect

to be encountered in the Chilean case.

I proceeded to test my hypothesis about job allocation in the following way. One of

the questions included in the survey presented respondents with a hypothetical situation:

"Imagine that an individual of your same rank within your institution accepted bribes

268
regularly without being caught. How much would the annual earnings of this individual

increase as a result?" Respondents were presented with five potential replies: 1) less than

30%; 2) 30% to 50%; 3) 50% to 100%; 4) 100% to 200%; and 5) more than 200%. In essence,

the question measures the potential bribe value of the individuals post. The response to this

question was adopted as the dependent variable in an ordered logistic regression analysis of

the relationship between political careerism and the potential bribe value associated with

ones post.

As in the regression analysis presented in the previous section of the chapter, I estimated

interaction eects between country dummies and the political career indicator variable, thus

capturing the eect of having a political career on the potential bribe value of ones post

in a particular county. Controls included years served in the public sector, years served

in the institution, measures of educational attainment and indicator variables denoting the

type of institution in which a particular bureaucrat worked (i.e., a central bank, a ministry,

an agrarian reform institution or a revenue-generating institution such as a tax service or

customs service).

The ordered logistic regression analysis was performed twice: first on the full sample and

then solely on those institutions which were categorized as being highly politicized according

to bureaucrats responses about political influence in hiring, dismissals and promotions in

their institutions.8 By splitting the sample in this way, I hoped to verify if patterns of

behavior observed in the full sample were being driven by activities practiced solely in the

institutions heavily under partisan control. Table 5.8 presents the results.

8
Measured in this way, the most politicized institutions in Bolivia were the Productive Social Investment
Fund, the Ministry of Agriculture, the Ministry of Economic Development and the Ministry of Sustainable
Development. In Brazil, these institutions were National Department of Infrastructure and Transport, the
National Social Security Service, the National Institute of Colonization and Agrarian Reform and the Min-
istry of Social Development. In Chile, these institutions were the National Institute of Agrarian Development
and the Customs Service.

269
 Table 5.8: Ordered logistic regression of the potential bribe value of post

The findings by in large confirmed expectations. In Bolivia, having a political career was

an important predictor of the type of post one was likely to have. When the analysis was

limited to the most politicized institutions, the estimated eect proved to be even stronger.

270
In neither the Brazilian nor Chilean cases did having a political career seem to make much

dierence in terms of the potential bribe value of ones post.

The estimated eect of the covariates also generally emerged in expected directions.

Happily, more educated bureaucrats were found to be significantly more likely to hold posts

which had the potential to generate significant revenue from bribes, suggesting that control

over matters such as licenses and contracts is generally limited to public servants of a certain

level of educational attainment. Not surprisingly, public servants in the revenue-generating

institutions were also found to be more likely to have the types of jobs with the potential

to generate significant revenue from bribes.

How important was the political careerism eect on potential bribe revenue in Bolivias

more politicized institutions? Figure 5.4 shows the distributions of the predicted probabili-

ties of holding a post in each of the five categories outlined above. The distributions of the

predicted probabilities are shown for a Bolivian bureaucrat that has a political career and

works in one of the politicized institutions and a Bolivian bureaucrat that does not have a

political career and also works in one of said institutions. Throughout, all covariates are

held at their mean values for this subset of institutions.

271
 Figure 5.4: The eect of political careerism in politicized institutions, Bolivia

As can be seen in the graph, having a political career leads to a substantial shift down-

wards in terms of the probability of having a post with low potential bribe value (upper

lefthandside panel of Figure 5.4). Similarly, having a political career leads to a substantial

shift upwards in terms of the probability of having a post in one of the high potential bribe

value categories (final 3 panels of Figure 5.4). In terms of numbers, having a political career

increases the probably of having a post with a potential bribe value of at least 100% ones

annual wage from .10 to .24 and the probability of having a post with a potential bribe

value of at least 50% ones annual wage from .24 to .47. These are not paltry dierences.

272
4 Conclusion

In previous chapter, I developed the argument that electoral rules have important conse-

quences for the ability of political actors to use the state to produce political funds. Electoral

systems which concentrate control over political careers in the hands of party leaders may

facilitate the emergence of corrupt bargains between party elites and politically-minded civil

servants, thus facilitating the use of the state as a source of resources. Similarly, electoral

systems which make political career control weak or diuse make such contracts dicult

to establish, thus militating against the use of the state as a source of resources. Based on

this general framework, I generated specific predictions about the impact of ballot structure

(open versus closed lists) and district magnitude.

In this chapter, I examined empirically the factors which contribute to the use of the

state as a source of resources for political activity. In order to accomplish this task, I

scrutinized the results of a large scale public employees survey that I conducted in Bolivia,

Brazil and Chile. The empirical findings generated from the bureaucratic experience of the

three countries corroborate key aspects of the theoretical framework while providing weaker

evidence for others.

The strongest finding regards the dierence between closed and open list systems. In

Bolivias closed list system of PR, there appears to have developed a pernicious informal

institution whereby partisan bureaucrats could be relied upon to utilize public resources for

their patrons and their patrons, in turn, would see to it that they were placed in precisely the

type of posts which would give them favored access to such resources. No such discernible

pattern emerged for our open list PR cases. In terms of the postulated dierences between

open list systems, the findings are much weaker. Although responses to certain survey

questions indicated that Brazilian bureaucrats generally perceive there to be more political

273
interference in bureaucratic tasks than their Chilean counterparts, in the individual level

analyses I conducted here the dierences between the two cases in terms of the patterns of

supply side corruption were relatively small and not statistically significant.

As a whole, the empirical analysis conducted here provides support for the concern

voiced at the end of the preceeding chapter; namely, that by ameliorating one manifesta-

tion of political corruption, ballot reform may intensify another. In this regard, it is perhaps

somewhat ironic that many analysts and commentators have called on Brazil to make the

switch to closed lists in the midst of that countrys current corruption crisis. The crisis

contains two components, each of which is conceptually distinct from the other. One com-

ponent consists of the scandal created by a vote buying scheme in the National Congress,

in which elites of the governing PT party helped organize monthly payments to selected

legislators in order to guarantee support for the presidents legislative agenda (the mensalo

scandal). The other component consists of revelations of a large scale party slush fund pur-

portedly organized by elites in the uppermost echelons of the PT hierarchy, as well as similar

practices exposed among allied parties (the caixa dois scandal(s)). The irony underlying

the call for a move to closed party lists is that while such a change might reduce the type

of behavior characterizing the first component of the crisis (since closed lists impose party

discipline and reduce the need for individual negotiations with legislators), it may increase

the incidence of behavior characterizing the second (for the reasons I have considered in

this paper). In the long run, a move from open to closed lists might mean trading o less

mensalo against more caixa dois.

Of greater utility may be a reduction in district magnitude. Such a change would have

the advantage of reducing incentives for political corruption on both sides of the equation:

the demand side and the supply side. On the demand side, lower magnitude would entail

274
less intense intraparty competition and thus less need to cultivate a personal vote through

costly campaign spending (cf Carey and Shugart 1995). On the supply side, lower magnitude

would provide less scope for profit sharing and reduce the ability of party leaders to select

winners through the targeting of resources, thus generating less corrupt bargains.

This observation notwithstanding, perhaps the most valuable piece of advice emerging

from present analysis would be that reformers would be well served to invest their energies

in institutional reforms other than those related to the electoral system. In this regard, the

empirical analysis conducted here suggests that initiating bread-and-butter type reforms

aimed at addressing the capacity of control organisms found within and outside public

institutions, along with increases in public employee salaries, would be a fruitful path to

tread in order to increase probity in the public sector. To illustrate this point, nearest

neighbor matching analysis shows that, controlling for a number of confounding variables,

bureacrats who felt that an investigation would be a "likely" or "highly likely" outcome

after a misappropriation of resorces had taken place were .140 less likely to have used

the resources of their institution of the benefit of a party than those who felt otherwise.9

Similarly, such matching also established that bureaucrats who felt their salaries covered

the costs of basic necessities were .113 less likely to use institutional resources in this way.10

By magnifying the risks associated with corrupt behavior, reformers may be able to make

a serious dent in its incidence.

9
In calculating this figure, the control variables utilized were: country dummies, a dummy for political
careers, result.1, result.2, denuncia, ethics.1, ethics.3, salary, stability.1, stability.2, stability.3.
10
Control variables utilized were: country dummies, a dummy for political careers, invest.1, invest.2,
invest.3, result.1, result.2, denuncia, ethics.1, ethics.3, stability.1, stability.2, stability.3.

275
 Chapter 6
Taking Stock of the Findings

276
6.1 What has been learned?

This monograph began by advancing a simple proposition: the structure of

institutions in the political arena has important consequences for the outcomes that obtain in

both the political arena and the bureaucratic arena. Given the mutual interdependency of

choices made by actors in these two realms of the polity, the monograph has argued that

general equilibrium analysis is the appropriate approach to adopt in evaluating the

consequences of institutional reform.

The logic of general equilibrium was applied to the problem of the impact of

electoral system design, specifically, the impact of having an open list versus closed list

proportional representation system on levels of political corruption. In this regard, I argued

that the transformation of an electoral system from OLPR to CLPR, or vice-versa, can be

expected to produce two countervailing effects on the level of political corruption. Firstly,

drawing on an extensive literature on intraparty competition and campaign spending, I

argued that a move from OLPR to CLPR can be expected to lessen legislative candidates

demand for campaign resources and thus exert a dampening influence on the level of

political corruption. Secondly, having constructed an original principal-agent model of party-

directed corruption in public bureaucracies, I argued that a move from OLPR to CLPR can

be expected to increase the leverage of party leaders over militants in the public

administration and thus exert a positive influence on the level of political corruption. The

existence of these two countervailing causal pathways linking ballot structure to levels of

political corruption suggests that the overall effect of ballot structure reform on levels of

political corruption is ambiguous. Although the distinction between OLPR and CLPR has

277
demonstrably important effects on political corruption, these effects are more of form than

of frequency or volume.

The empirical research presented in this monograph has concentrated on

establishing the validity of the second causal pathway just mentionedthat linking CLPR to

more pervasive and effective party corruption networks in the bureaucracy. Drawing on the

public employees survey conducted in Bolivia, Brazil and Chile, the study produced a

number of empirical findings consistent with the notion that CLPR facilitates the use of the

state as a source of electoral resources. Firstly, I found that party militants in the closed list

case (Bolivia) were more likely to have used the resources of their institutions for the benefit

of a party than militants in either of the two open list cases (Brazil and Chile), holding other

factors which contribute to the incidence of illicit resource use constant. Secondly, I found

that the marginal impact of having a political career on the likelihood of using institutional

resources for the benefit of a party was greater in Bolivia than Brazil, again, holding all else

constant. Finally, the Bolivian case was found to be the only one in which having a political

career was a statistically significant predictor of the likelihood that a bureaucrat would

receive the type of post which had the potential to generate significant amounts of revenue

from bribes. Taken together, these findings paint a picture of relatively more developed and

ingrained party corruption networks in the closed list case than the two open list cases.

Needless to say, with only three countries included in the analysis, the findings

presented here are far from an empirical proof that CLPR tends to encourage political

corruption within the bureaucracy in the ways explored in this monograph. Many more

similarly organized surveys would be required before analysts could conclude one way or

another that CLPR systematically produces such outcomes. Nevertheless, this study is the

first to examine the consequences of electoral system design on bureaucratic behavior at the

278
individual level. As such, it should be seen as an opening salvo in a debate about the

bureaucratic effects of electoral reform and how those effects can be measured.

Distilled to its essence, the core claim of the monograph is simple: ballot structure

reform is not a magic bullet for reducing political corruption because its consequences for

the behavior of politicians on the campaign trail and party militants in the bureaucracy pull

in opposite directions. As an instrument of policy, therefore, the monograph serves as a

forewarning to well-intentioned reformers that, to the degree that a reduction in political

corruption is the ultimate goal, institutional reforms of this stripe may prove disappointing.

Although I have concentrated my energies in making an argument about what doesnt

work in reducing corruption rather than what does, this monograph should not be read as a

council of despair. Both the cross-institutional and individual-level statistical analyses suggest

that old-fashion Weberian reform may be able to effect a considerable reduction in political

corruption. The cross-institutional analysis makes clear that one key to reducing the partisan

use of institutional resources consists of vigorous and sustained civil service reform designed

to weaken the influence of political criteria in determining who gets hired, who gets

dismissed and who rises up the ranks in public institutions. The individual-level analysis

shows that two other components of bureaucratic reform also hold great promise for

reducing the incidence of political corruption: the strengthening of accountability

mechanisms within public bureaucracies and increases in public sector salaries.

6.2: Recent trends in the Cases

Given the studys findings about the non-electoral rule based determinants of

corruption, what prospects might the future hold for the cases?

279
 In terms of civil service reform, the trends in the three cases are mixed but do

provide some grounds for optimism.

Chile has made the greatest strides in civil service reform in recent years. A string of

corruption scandals in this country pushed the government and opposition to the bargaining

table on this issue in late January 2003. The result was the so-called Political-Legislative

Accord for the Modernization of the State, Transparency and the Promotion of Growth.

The Accord was made possible due to the bundling of two distinct types of reform: civil

service reform and campaign finance reform (Fuentes 2004). The Concertacin, the

government coalition belonging to the left, wanted to obtain campaign expenditure limits

and was willing to trade off some of its control over the public bureaucracy in order to attain

this goal. The Alianza, the coalition of the right, wanted to reduce the influence of the state

in electoral campaigns and was willing to accept expenditure ceilings as the price for this

benefit. The civil service reforms contained in the Accord created a national civil service

commission (which came into being in January 2004), reduced the number of appointive

posts in the public administration and created a council which will provide the president

with the names of admissible candidates for a number of high level posts in the public

administration. Although it is too early to assess the impact of these changes, they represent

movement in the right direction.

Bolivia, too, has made major efforts in legislating civil service reform, although as

can be seen from the survey data examined in chapter 3, the distance between parchment

and bureaucratic reality in this regard is considerable. The myriad attempts over the last 15

years to implant a more meritocratic civil service include the Civil Service Program and Civil

Service and Administrative Reform project adopted in the early 1990s, as well as the more

ambitious National Integrity Plan (PNI) adopted in 1999, which, in addition to

280
institutionalizing the Customs, Roads and Tax Services, gave birth to Bolivias Statute of the

Public Functionary and the Superintendency of the Civil Service. The degree of actual

improvement entailed by these reforms was certainly meager at the outset, as incoming

administrations routinely dismissed supposed career functionaries hired by their

predecessors. However, the PNI reforms do seem to have had some lasting effects, and the

survey data presented in this monograph suggest that the three services mentioned above

once considered the worst performing institutions in the Bolivian public administration

have exhibited real improvements. Whether these advances can be maintained remains an

open question. In any case, the survey data are clear that much work remains to be done in

laying the groundwork for stable and productive bureaucratic careers within the Bolivian

public service.

The experience with civil service reform in Brazil has also been fairly inconsistent (cf

Rezende 2002). The most recent reforms in this country, begun in the mid 1990s, focused

on introducing greater efficiency in bureaucratic tasks through the adoption of public

management techniques along the lines of the New Zealand model. However, the ministry

created to oversee the public management revolution in the Brazilian public sector was

dismantled within three years of its creation and subsequent administrations have shown

little interest in resurrecting the discontinued reforms. In any case, these reforms did not

address the politicization of the public administration. As noted in chapter 3, the formal

distinction between career civil servants and political appointees is well respected in the

Brazilian public administration but the total mass of appointive posts remains enormous

on the order of 22,000. The litany of political corruption scandals that hit the Workers Party

governmentnearly all of which involved strategically located appointeeshave led to

281
some grumblings about legislation designed to reduce the number of such posts. Only time

will tell if such reforms come to fruition.

In terms of the reinforcement of accountability mechanisms designed to detect

wrongdoing in the public administration, recent trends in the cases are encouraging.

Beginning in the early 1990s, Bolivia gradually introduced a new system of financial

management in the public sector (SAFCO) designed to regularize the inspection of

government accounts by internal auditors and Bolivias Supreme Auditing Institution

(Comptroller General of the Republic). Although the implementation of this system was

painfully slow, modest improvements in transparency seem to have been attained in recent

years. A Unit of Financial Investigations was also created in 1997 in order to detect money

laundering operations related to corruption, drug trafficking, or organized crime. Along the

same lines, a national anti-corruption institution, the Secretariat of the Fight Against

Corruption, was created in August 2002 and has since been upgraded to the status of

Presidential Delegation for Transparency and Public Integrity. Brazil has introduced a similar

set of innovations. The Constitution of 1988 gave considerable powers to the Office of

Public Prosecution (Ministerio Pblico), which has been actively involved in corruption

investigations at all levels of government (Sadek and Cavalcanti 2003). A national anti-

corruption institution, the Comptroller General of the Union, was created in April 2001. It

has played an important role in reinforcing internal control within Brazils public agencies

and has made a major effort to systematically audit the accounts of many of this countrys

municipalities. Brazils Federal Police has also taken on the anti-corruption mantle, giving

corruption investigations greater priority than had been the case in the past. In Chile, recent

reforms of this nature have included the implementation of a new financial management

system and the adoption of standards requiring greater transparency in public procurement.

282
 The situation with public sector salaries is worthy of some concern. Although the

ratio of the public wage rate to GDP per capita is high in the three cases by Latin American

standards (see chapter 3), it appears there might be some serious backsliding in Bolivia. One

of the early moves of the administration of Evo Morales, which came to power in January

2006, was to promulgate an executive decree which drastically cut the salaries of high-level

public servants.1 Although the action was ostensibly driven by noble motives, namely, to

redirect public revenue towards social programs which benefit the poor, such efforts may

prove counterproductive. Clearly, such programs will only bear fruit if there are competent

and honest public servants willing to administer them. If the reduction in the salaries of top-

level administrators in the public sector falls too far below the salaries available for similar

work in the private sector, then many of the most qualified individuals may leave the public

sector. Moreover, those that stay may be tempted to supplement their income through illicit

means. The upshot is that the resulting efficiency loses may, given time, overwhelm the short

term revenue gain produced by the initial reduction in public sector salaries.

6.3: What Still Needs to be Learned?

This study concludes with many important questions hanging in the balance. Perhaps

the most important is the following: just how much of the money that works its way into

political campaigns and party operating expenses in Latin America is off the books and

comes from illicit sources? The consensus of experts on political financing in the region is

that the proportion of illicit money which makes its way into political activity substantial (cf

Zovatto 2004), but it is currently impossible to place a point estimate on this amount.

1
It should be noted that Morales made the move in the most self-sacrificing way possible, cutting his own
salary by more than half.

283
Efforts are currently underway by a team of specialists at the Atlanta-based Carter Center to

get a better sense of the proportion of political funding that goes unreported in a variety of

countries in the Americas. Moreover, the public disclosure of parties declared finances has

been mandated by legislation in eight countries in the region (albeit with different degrees of

compliance and detail in party accounts). Thus, with the passage of time, our ability to gauge

the importance of illicit money in the electoral process may very well improve.

And yet, even if we knew the proportion of campaign finance that was off the books,

important questions would remain. Most relevant from the standpoint of studies of

corruption would be the composition of said financing. Specifically, it would be useful to get a

rough idea of the proportion of money that is injected into the political system from the

indirect or direct use of the state as I have examined it here, as well as the proportions that

are accounted for by other illicit activities such as organized crime and drug trafficking. In

this regard, well publicized scandals in countries such as Bolivia, Colombia, Ecuador and

Peru suggest that for a number of countries in the region at various points in their histories,

the influx of money from nacrotrafficking may have had a considerable role in bankrolling

political campaigns.

Besides questions about the precise weight of political corruption as a source of

electoral resources, another outstanding issue worth exploring is how party organization,

apart from electoral rules, impacts on levels of political corruption. Holding electoral

institutions constant, it would be interesting to examine how party structures might influence

the ability of political actors to use the state as a source of resources. In this regard, there

already exists related work which shows that the electoral impact of patronage may differ

across parties located in the same country (Murillo and Calvo 2004). The recent experience

of the Workers Party in Brazil, with its high internal discipline, considerable centralization

284
and its unique ability (in that country) to collect tithes (on the order of 20% to 30% of

monthly salaries) from its public sector workers, suggests that tracing out the consequences

of party organization for the use of the state as a source of resources might produce

interesting findings as well.

6.4: The Need for (and Challenges of) Institutional Engineering

A great hope characterizes much social science research today. An entire generation

of scholars has pinned its professional prestige on the notion that by implementing changes

in the configuration of formal institutions, policymakers can make real and lasting

improvements in human welfare. Perhaps nowhere have such arguments had more impact

than among students of politics in newly democratizing republics, where the eclipse of

military government has given way to vigorous debates about the appropriate design of the

economy and polity.

For many years, debates about the appropriate design of the economy dominated

this discussion. As many of the worlds new democracies made the transition to competitive

politics in a context of severe economic crisis, it was only natural that questions about the

design of market institutions would initially take precedence. In this vein, the combined

intellectual efforts of academics, bureaucrats and economic experts in the employ of

international lending institutions converged to produce a blueprint for economic reform.

The blueprint, often referred to as the Washington consensus, was embraced with greater

or lesser enthusiasm by developing country governments around the world. However, its

central componentsamong them the privatization of state owned enterprises, trade and

285
financial liberalization and measures to encourage fiscal disciplinewere pretty much

implemented everywhere.

Two decades after the prescriptions of the Washington consensus were put into

place, citizens throughout the developing world have had the opportunity to evaluate the

performance of their young democracies in light of these momentous changes in economic

organization. Many are clearly underwhelmed with what they see. Poverty rates have not

declined to the extent that had been hoped, and inequality remains devilishly resilient.

Pathologies of governance, most notably corruption, have abruptly cut short numerous

governments and left party systems in shambles. Not surprisingly, confidence in the key

organizations and actors of democratic governmentthe Congress and political parties

among themhas reached an all time low (cf PNUD 2004). In certain polities, it is not an

exaggeration to say that conducting the business of government has lost the aura of a

legitimate enterprise.

In light of these developments, many of those voices who in the past championed

policies designed to get prices right came to the realization that such policies were

insufficient in and of themselves (Burki and Perry 1998, IADB 2000). Proper governmental

and social institutions are now widely recognized as forming the bedrock for market

mechanisms that promote growth and prosperity (North 1990). Naturally, this realization

piqued an interest in understanding how the architecture of governance can be fashioned to

best serve the needs of the public. Issues of democratic institutional engineering quickly

leapt to the forefront of much political economy research, with political scientists and

economists both providing important contributions (cf Carey and Shugart 1995, Mainwaring

and Shugart 1997, Norris 2004, Persson and Tabellini 2003, Sartori 1994).

286
 This monograph demonstrates just how challenging such institutional engineering

can be. It has shown how one institutional reform widely thought to produce a reduction in

political corruption may instead produce ambiguous effects. Given the complex ways in

which democratic institutional design governs outcomes in the electoral and bureaucratic

arenas, the impact of institutional reform may be remarkably subtle. The high degree of

subtlety which characterizes the task of tracing out the relationship between explanans and

explanandum, however, is no reason to abandon the important enterprise of restructuring

democratic institutional design in order to improve social welfare. The burning social

problems which have driven reformers and scholars alike to the institutional drafting board

will not disappear on their own. Neither should the efforts of would-be institutional

engineers. Hopefully, this monograph has convinced the reader that we can improve future

attempts at institutional engineering by adopting a more holistic view of institutional reform,

one that appreciates the significant overlap between the bureaucratic and electoral arenas in

many developing countries.

287
 REFERENCES

Abadie, A., Drukker, D, JL Herr and GW Imbens. 2004. Implementing Matching

Estimators for Average Treatment Effects in Stata. The Stata Journal 4 (3): 290-311.

Aberbach, JD and BA Rockman. 1987. Comparative Administration: Methods, Muddles,

and Models. Administration and Society 18: 473-506.

Abranches, SH. 1988. Presidencialismo de Coalizo: O Dilemma Institucional

Brasileiro. DADOS 31(1): 5-33.

Ades, A. and R. Di Tella. 1999. Rents, Competition and Corruption. American Economic
Review 89 (4): 982-993.

Adeser, A, C Boix and M Payne. 2003. Are You Being Served? Political Accountability
and Quality of Government. Journal of Law, Economics and Organization 19 (2): 445-490.

Aguiar, R. 1994. The Cost of Election Campaigns in Brazil. In HE Alexander and R

Shiratori, eds., Comparative Political Finance Among the Democracies. Boulder: Westview,
77-84.

Alston, LJ and B Mueller. 2006. Policy for Pork: Executive and Legislative Exchange in
Brazil. Journal of Law, Economics and Organization 22 (1): 87-114.

Alt, J and D Dreyer Lassen. 2003. The Political Economy of Institutions and Corruption
in American States. Journal of Theoretical Politics 15 (3): 341-365.

Alvarez, A. 1997. Los Dineros de la Poltica. Caracas: Universidad Central de

Venezuela.

Ameringer, C. 2000. The Cuban Democratic Experience: The Autntico Years 1944-
1952. Gainesville: University Press of Florida.

Ames, B. 1987. The Congressional Connection: The Structure of Politics and the
Distribution of Public Expenditures in Brazil's Competitive Period. Comparative
Politics 19 (2): 147-171.

_______. 1995. Electoral Strategy under Open-list Proportional Representation.

American Journal of Political Science 39 (2): 406-433.

_______. 2001. The Deadlock of Democracy in Brazil. Ann Arbor: University of

Michigan Press.

Amorim Neto, O. 2000. Gabinetes Presidenciais, Ciclos Eleitorais e Disciplina

Legislativa no Brasil. Dados 43 (3): 479-519.

________. and F. Santos. 2001. The Executive Connection: Presidentially Defined

Factions and Party Discipline in Brazil. Party Politics 7 (2): 213-234.

288
Anderson, CJ and YV Tverdova. 2003. Corruption, Political Allegiances and Attitudes
Towards Government in Contemporary Democracies. American Journal of Political
Science 47, 1: 91-109.

Andvig, JC and KO Moene. 1990. How Corruption May Corrupt. Journal of Economic
Behavior and Organization 13: 63-76.

Ansolabehere, S, R Behr, and S Iyengar. 1993. The Media Game. New York: Macmillan.

Archer, RP and MS Shugart. 1997. The Unrealized Potential of Presidential Dominance

in Colombia. In S Mainwaring and MS Shugart, eds., Presidentialism and
Democracy in Latin America. Cambridge: Cambridge University Press, 110-159.

Arellano Gault, D. and JP. Guerrero Amparn. 2003. Stalled Administrative Reforms of
the Mexican State. In BR Schneider and B Heredia, eds., Reinventing Leviathan: The
Politics of Administrative Reform in Developing Countries. Miami: North-South Center
Press, 151-180.

Aristotle. The Politics, 3.7, trans. W.Ellis.

Austen-Smith, D and J Banks. 1989. Electoral Accountability and Incumbency. In PC

Ordeshook, ed., Models of Strategic Choice in Politics. Ann Arbor: University of Michigan
Press.

Barro, R. 1973. The Control of Politicians: An Economic Model. Public Choice 14 (1):
19-42.

Bartels, L. 1993. Messages Received: The Political Impact of Media Exposure.

American Political Science Review 87, 267-285.

Becker, GS and GJ Stigler. 1974. Law Enforcement, Malfeasance, and Compensation of

Enforcers. The Journal of Legal Studies 3(1): 1-18.

Bertrand, M., S. Djankov, R. Hanna and S. Mullainathan. 2006. Does Corruption Produce
Unsafe Drivers? Working paper. Available at:
http://post.economics.harvard.edu/faculty/mullainathan/papers/corruption.pdf

Bezerra, MO. 1999. Em Nome das Bases: Poltica, Favor e Dependncia Pessoal. Rio
de Janeiro: Dumar.

Bouissou, J-M. 1997. Gifts, Networks and Clienteles: Corruption in Japan as a

Distributive System. In Della Porta and Mny, eds., Democracy and Corruption in Europe,
London: Pinter, 132-147.

_______. 1999. Organizing Ones Support Base under the SNTV: The Case of the
Japanese Koenkai. In Grofman et al., eds., Elections in Japan, Korea, and Taiwan under the
Single Non-Transferable Vote. Ann Arbor: University of

289
 Michigan Press, 87-120.

Burki, SJ and GE Perry. 1998. Beyond the Washington Consensus: Institutions Matter.
Washington DC: World Bank.

Brescia, M. 2001. Manos Limpias: La Corrupcin en las Empresas Pblicas (de Todos
los Chilenos), 1980-2001. Santiago: Editorial Mare Nostrum.

Bresser-Pereira, LC. 1998. Reforma do Estado para a Cidadania: a Reforma Gerencial

Brasileira na Perspectiva Internacional. Braslia: ENAP.

Bueno de Mesquita, B, A Smith, RM Siverson and JD Morrow. 2003. The Logic of

Political Survival. Cambridge: MIT Press.

Cadot, O. 1987. Corruption as a Gamble. Journal of Public Economics 33 (2): 223-44.

Cain, B., J. Ferejohn and M. Fiorina. 1987. The Personal Vote: Constituency Service and
Electoral Independence. Cambridge: Harvard University Press.

Canache, D and M Allison. 2003. Corrupted Perceptions: The Effect of Corruption on

Political Support in Latin American Democracies. Paper delivered at the XXIV
International Congress of the Latin American Studies Association, Dallas, TX,
March 27-29.

Carey, JM. 2000. Parchment, Equilibria and Institutions. Comparative Political Studies
33 (6/7): 735-761.

________. 1996. Term Limits and Legislative Representation. Cambridge: Cambridge

University Press.

________ and MS Shugart. 1998. Executive Decree Authority. Cambridge: Cambridge

University Press.

________ and MS. Shugart. 1995. Incentives to Cultivate a Personal Vote: A Rank
Ordering of Electoral Formulas. Electoral Studies 14 (4): 417-439.

________ and P. Siavelis. 2005. Insurance for Good Losers and the Survival of Chile's
Concertacin. Latin American Politics and Society 47 (2): 1-22.

Carvalhosa, M., coord. 1995. O Livro Negro da Corrupo. Rio de Janeiro: Paz e Terra.

Casella, G. and RL. Berger. 1990. Statistical Inference. Belmont, CA: Duxbury Press.

Cassese, S. 1993. Hypotheses on the Italian Administrative System. West European

Politics 16: 316-328.

Chalmers, DA. 1977. The Politicized State in Latin America. In JM Malloy, ed.,

290
 Authoritarianism and Corporatism in Latin America. Pittsburgh: University of Pittsburgh
Press, 23-46.

Chang, ECC. 2005. Electoral Incentives for Political Corruption under Open-list
Proportional Representation. Journal of Politics 67 (3): 716-730.

Chaudhuri, A. and R. Mukerjee. 1985. Optionally Randomized Response Techniques.

Calcutta Statistical Association Bulletin 34: 225-229.

________. 1988. Randomized Response: Theory and Techniques. New York: M.

Dekker.

Chehabi, H, and J Linz., eds. 1998. Sultanistic Regimes. Baltimore: John Hopkins
University Press.

Cheng, T-J and S. Haggard. 2001. Democracy and Deficits in Taiwan: The Politics of
Fiscal Policy. In S. Haggard and MD McCubbins, eds., Presidents, Parliaments and
Policy. Cambridge, Cambridge University Press, 183-228.

Chubb, J. 1982. Patronage, Power and Poverty in Southern Italy: A Tale of Two Cities.
Cambridge: Cambridge University Press.

Coppedge, M. 1994. Strong Parties and Lame Ducks: Presidential Partyarchy and
Factionalism in Venezuela. Stanford: Stanford University Press.

Costa Benevides, Jimena. 2004. Os Partidos Neopopulistas na Bolvia (1989-2004). In

Lodola et al., Neopopulismo na Amrica Latina. Rio de Janeiro: Konrad Adenauer.

Cox, GW. 1990. Multicandidate Spatial Competition. In JM Enelow and MJ Hinich, eds.,
Advances in the Spatial Theory of Voting. Cambridge, Cambridge University Press.

_______. 1997. Making Votes Count: Strategic Coordination in the Worlds Electoral
Systems. Cambridge, Cambridge University Press.

_______ and MF Thies. 1998. The Cost of Intraparty Competition: The Single
Nontransferable Vote and Money Politics in Japan. Comparative Political Studies 31, 3:
267-291.

Crisp, BF. 2002. Andean Legislatures. mimeo.

_______. 1997. Presidential Behavior in a System with Strong Parties: Venezuela, 1958-
1995. In S Mainwaring and MS Shugart, eds., Presidentialism and
Democracy in Latin America. Cambridge: Cambridge University Press, 160-198.

_______ and MC Escobar-Lemmon. 2001. Democracy in Latin America: Individuals in

Institutional Contexts. Latin American Research Review 36 (2): 175-192.

_______ and JC Rey. 2001. The Sources of Electoral Reform in Venezuela. In MS

291
 Shugart and MP Wattenberg, eds., Mixed Member Electoral Systems. Oxford: Oxford
University Press, 173-193.

Curtis, GL. 1988. The Japanese Way of Doing Politics. New York: Columbia University
Press.

Daland, RT. 1981. Exploring Brazilian Bureaucracy: Performance and Pathology.

Washington, D.C.: University Press of America.

De Luca, M, MP. Jones and MI Tula. 2002. Back Rooms or Ballot Boxes?: Candidate
Nomination in Argentina. Comparative Political Studies 35: 413-436.

De Soto, H. 1989. The Other Path: The Invisible Revolution in the Third World. New
York: Harper and Row.

della Porta, D. 2000. Social Capital, Beliefs in Government, and Political Corruption. In
SJ Pharr and RJ Putnam, eds., Disaffected Democracies: Whats Troubling the Trilateral
Countries? Princeton: Princeton University Press.

_______ and A Vannucci. 1999. Corrupt Exchanges: Actors, Resources, and

Mechanisms of Political Corruption. New York: Aldine de Gruyter.

Desposato, SW. 2006. Parties for Rent? Ambition, Ideology, and Party Switching in
Brazil's Chamber of Deputies. American Journal of Political Science 51 (1): 62-80.

Di Tella, R and E Schargrodsky. 2003. The Role of Wages and Auditing during a
Crackdown on Corruption in the City of Buenos Aires. Journal of Law and Economics
46: 269-292.

Di Tella, R. and WD Savedoff. 2002,eds. Diagnstico Corrupcin. Washington:

Inter-American Development Bank.

Dias, M. 2004. A Mentira das Urnas. Rio de Janeiro: Editora Record.

Diccionario de la Corrupcion en Venezuela. Vol 1. 1989. Caracas: Consorcio de
Ediciones Capriles C.A.

Diniz, E. 1982. Voto e Mquina Poltica. Rio de Janeiro: Paz e Terra.

Dobel, JP. 1978. Corruption of a State. American Political Science Review 72 (3): 958-
973.

Domingo, P. 2003. Conclusions: Revolution and the Unfinished Business of Nation- and
State-Building. In P Domingo and MS Grindle, eds., Proclaiming Revolution: Bolivia in
Comparative Perspective. Cambridge, Mass: Harvard University Press.

Domnguez, JI., ed. 1997. Technopols. University Park: University of Pennsylvania

Press.

292
_______, with JK Giraldo. 1998. Constructing Democracies: Parties, Institutions,
Market Reforms, the Military and the International Environment. In JI Domnguez,
ed., Democratic Politics in Latin America and the Caribbean. Baltimore: Johns Hopkins
University Press.

Eckstein, H. 1975. Case Study and Theory in Political Science. In FI Greenstein and NW
Polsby, eds. Handbook of Political Science: Strategies of Inquiry. Reading, Mass.: Addison-
Wesley, 79-137.

Efron, B and RJ Tibshirani. 1993. An Introduction to the Bootstrap. Boca Raton:

Chapman & Hall.

Evans, P. 1995. Embedded Autonomy: States and Industrial Transformation. Princeton:

Princeton University Press.

_______. and JE Rauch. 1999. Bureaucracy and Growth: A Cross-National Analysis of

the Effects of Weberian State Structures on Economic Growth. American
Sociological Review 64: 748-765.

Fearon, JD. 1999. Electoral Accountability and the Control of Politicians: Selecting Good
Types versus Sanctioning. In A. Przeworski, SC Stokes and B Manin, eds., Democracy,
Accountability and Representation. Cambridge: Cambridge University Press.

Ferejohn, J. 1986. Incumbent Performance and Electoral Control. Public Choice 50 (1):
5-25.

Figueiredo, AC. and F. Limongi. 1999. Executivo e Legislativo na Nova Ordem

Constitucional. Rio de Janeiro: Editora FGV.

Fleisher, D. 1997. Political Corruption in Brazil: The Delicate Connection with

Campaign Finance. Crime, Law and Social Change 25: 297-321.
_______. 1999. The Difficult Task of Evaluating Corruption in Brazil: the
Corruption Score Board Project. mimeo.

Folha de So Paulo., 9/19/05. Parlementares defendem voto em lista fechada, A-14.

Fox, JA and PE Tracy. 1986. Randomized Response : A Method for Sensitive Surveys.
Beverly Hills: Sage.

Friendly, M. 1994. Mosaic Displays for Multiway Contingency Tables. Journal of the
American Statistical Association 89 (425): 190-200.

Fuentes, C. 2004. El Costo de la Democracia. Santiago de Chile: FLACSO-Chile.

Fukui, H and SN Fukai. 1999. Campaigning for the Japanese Diet. In Grofman et al.,
eds., Elections in Japan, Korea, and Taiwan under the Single Non-Transferable Vote. Ann
Arbor: University of Michigan Press, 121-152.

293
Gamarra, EA. 2003. Political Parties Since 1964: The Construction of Bolivia's
Multiparty System. In P Domingo and MS Grindle, eds., Proclaiming Revolution: Bolivia
in Comparative Perspective. Cambridge, Mass: Harvard University Press.

_______. 1997. Hybrid Presidentialism and Democratization: The Case of Bolivia.

In S Mainwaring and MS Shugart, eds., Presidentialism and
Democracy in Latin America. Cambridge: Cambridge University Press, 363-393.

_______ and JM Malloy. 1995. The Patrimonial Dynamics of Party Politics in

Bolivia. In SP Mainwaring and TR Scully, eds., Building Democratic Institutions: Party
Systems in Latin America. Stanford: Stanford University Press, 399-433.

_______. 1987. Political Stability, Democratization and the Bolivian National

Congress. PhD thesis, University of Pittsburgh.

Garretn, MA and Gonzalo Cceres. 2003. From the Disarticulation of the State to the
Modernization of Public Management in Chile: Administrative Reform Without a
State Project. In BR Schneider and B Heredia, eds., Reinventing Leviathan: Politics of
Administrative Reform in Developing Countries. Coral Gables: North-South Center Press,
113-149.

Gasiorowski, MJ and TJ Power. 1998. The Structural Determinants of Democratic

Consolidation: Evidence from the Third World. Comparative Political Studies 31 (6):
740-771.

Gay, R. 1994. Popular Organization and Democracy in Rio de Janeiro. Philadelphia:

Temple University Press.

Geddes, B. 1994. Politicians Dilemma: Building State Capacity in Latin America.

Berkeley: University of California Press.

_______. 1990. How the Cases You Choose Affect the Answers You Get: Selection
Bias in Comparative Politics. Political Analysis 2, 131: 152.

_______. 1991. A Game Theoretic Model of Reform in Latin American Democracies.

American Political Science Review 85 (2): 371-392.

_______. and AR Neto. 1999. Institutional Sources of Corruption in Brazil. In KS

Rosenn and R Downes, eds., Corruption and Political Reform in Brazil. Miami: North-
South Center Press.

Gerber, AS and DP Green. 2000. The Effects of Personal Canvassing, Telephone Calls,
and Direct Mail on Voter Turnout: A Field Experiment. American Political Science
Review 94 (3): 653-64.

Goel, RK and MA Nelson. 1998. Corruption and Government Size: A Disaggregated

Analysis. Public Choice 97: 107-120.

294
Golden, MA. 2003. Electoral Connections: The Effects of the Personal Vote on Political
Patronage, Bureaucracy and Legislation in Postwar Italy. British Journal of Political
Science 33: 189-212.

_______. and L Picci. 2005. Proposal for a New Measure of Corruption, Illustrated
with Italian Data. Economics and Politics 17 (1): 37-75.

_______. and ECC Chang. 2001. Competitive Corruption: Factional Conflict and
Political Malfeasance in Postwar Italian Christian Democracy. World Politics 53: 588-
622.

Gouva, GP. 1994. Burocracia e Elites Burocrticas no Brasil. So Paulo: Editora

Paulicia.

Graham, LS. 1965. The Clash Between Formalism and Reality in the Brazilian Civil
Service. PhD thesis, University of Florida.

Graham, R. 1990. Patronage and Politics in 19th Century Brazil. Stanford, Stanford
University Press.

Greenberg, BG, A Abul-Ela, WR Simmons, DG Horvitz. 1969. The Unrelated Question

Randomized Response Model: Theoretical Framework. Journal of the American
Statistical Association 64: 520-539.

Greene, WH. 2000. Econometric Analysis. Upper Saddle River, NJ : Prentice Hall.

Grindle, MS. 1977. Patrons and Clients in the Bureaucracy: Career Networks in Mexico.
Latin American Research Review 12 (1): 37-66.
Grofman, B, S-C Lee, EA Winckler and B Woodal. 1999. Elections in Japan, Korea, and
Taiwan under the Single Non-Transferable Vote. Ann Arbor: University of Michigan Press.

Groisman, E. and E. Lerner. 2000. Responsabilizacin por los Controles Clsicos. In

Consejo Cientfico del CLAD, coord., La Responsabilizacin en la Nueva Gestin Pblica
Latinoamericana. Buenos Aires: CLAD, 75-115.

Gupta, S., Davoodi, HR and E. Tiongson. 2000. Corruption and the Provision of Health
Care and Education Services. IMF Working Paper No. 00/116.

Habib, M and L Zurawicki. 2002. Corruption and Foreign Direct Investment. Journal of
International Business Studies 33 (2).

Hagopian, F. 1990. Democracy by Undemocratic Means: Elites, Political Pacts and

Regime Transition in Brazil. Comparative Political Studies 23 (2): 147-170.

Hartigan, JA. and B. Kleiner. 1981. Mosaics for Contingency Tables. In WF Eddy, ed.,
Computer Science and Statistics. New York: Springer-Verlag, 268-273.

_______. 1984. A Mosaic of Television Ratings. The American Statistician 38, 32-35.

295
Hedstrm, P and R Swedberg, eds. 1998. Social Mechanisms: An Analytical Approach to
Social Theory. Cambridge: Cambridge University Press.

Hempel, CG. 1965. Aspects of Scientific Explanation. New York: Free Press.

Hochman, C. 1992. Os Cardeais da Previdncia Social: Gnese e Consolidao de uma

Elite Burocrtica. DADOS 35 (3): 371-401.

Hodess, R. 2004. Introduction. In R. Hodess, T. Inowlocki, D. Rodriguez and T. Wolfe,

eds., Global Corruption Report 2004. London: Pluto Press, 11-18.

Horcasitas, JM and JA Weldon. 2001. Reforming Electoral Systems in Mexico. In MS

Shugart and MP Wattenberg, eds., Mixed Member Electoral Systems. Oxford: Oxford
University Press, 209-230.

Hrebenar, RJ. 2000. Japans New Party System. Boulder: Westview Press.

Huntington, SP. 1968. Political Order in Changing Societies. New Haven: Yale
University Press.

Huber, E and M Dion. 2002. Revolution or Contribution? Rational Choice Approaches in

the Study of Latin American Politics. Latin American Politics and Society 44 (3):
1-28.

Inter-American Development Bank (IADB). 2000. Development Beyond Economics:

Economic and Social Progress in Latin America. Washington, DC: IADB.

International Bank for Reconstruction and Development (IBRD). 1951. Report on Cuba.
Baltimore: Johns Hopkins University Press.

International Institute for Democracy and Electoral Assistance (IDEA). 1997. Electoral System
Design. Stockholm: IDEA.

Iyengar, S and DR Kinder. 1987. News that Matters. Chicago: University of Chicago
Press.
Johnson, C. 1982. MITI and the Japanese Miracle. Stanford: Stanford University Press.

Johnson, RN and GD Libecap. 1994. The Federal Civil Service and the Problem of
Bureaucracy: The Economics and Politics of Institutional Change. Chicago: University of
Chicago Press.

Jones, MP. 2002. Explaining the High Level of Discipline in the Argentine Congress. In
S Morgenstern and B Nacif, eds., Legislative Politics in Latin America. Cambridge:
Cambridge University Press, 147-184.

296
_________. 1997. Evaluating Argentinas Presidential Democracy: 1983-1995. In
S Mainwaring and MS Shugart, eds., Presidentialism and Democracy in Latin America.
Cambridge: Cambridge University Press, 259-299.

_________, S Saiegh, PT Spiller and M Tommasi. 2002. Amateur Legislators-

Professional Politicians: The Consequences of Party-Centered Electoral Rules in a
Federal System. American Journal of Political Science 46 (3): 656-669.

Kang, D. 2002. Crony Capitalism: Corruption and Development in South Korea and the
Philippines. Cambridge: Cambridge University Press.

Katz, R. 1985. Preference Voting in Italy: Votes of Opinion, Belonging or Exchange.

Comparative Political Studies 18 (2): 229-249.

Kaufmann, D A Kraay and P Zoido-Lobatn. 1999. Governance Matters. World Bank.

mimeo.

Kaufman Purcell, S. 1972. The Impact of Revolution: A Comparative Analysis of Bolivia

and Mexico. London: Sage Publications.

Keck, ME. 1992. The Workers Party and Democratization in Brazil. New Haven: Yale
University Press.

King, G. 1998. Unifying Political Methodology. Ann Arbor: University of Michigan

Press.

______, RO Keohane and S Verba. 1994. Designing Social Inquiry. Princeton: Princeton
University Press.

Kinzo, MDG. 1998. Funding Parties and Elections in Brazil. In P Burnell and A Ware,
eds., Funding Democratization. Manchester: Manchester University Press.

Kohli, A. 2004. State-Directed Development : Political Power and Industrialization in

the Global Periphery. Cambridge : Cambridge University Press.

Knack, S. and P. Keefer. 1995. Institutions and Economic Performance: Cross-Country

Tests Using Alternative Institutional Measures. Economics and Politics (7) 3: 207-27.

Krieger, G, F Reodrigues and EC Bonassa. 1994. Os Donos do Congresso. So Paulo:

Editora tica.

Krosnick, J. and D Kinder. 1990. Altering the Foundations of Presidential Support

Through Priming. American Political Science Review 84: 497-512.

Kuk, AYC. 1990. Asking Sensitive Questions Indirectly. Biometrika 77, 436-438.

Kumaraswamy, P. 1980. A Generalized Probability Density Function for Double-

Bounded Random Processes. Journal of Hydrology 46 (1/2): 79-88.

297
Kunicova, J and S Rose-Ackerman. 2002. Electoral Rules as Constraints on Corruption.
mimeo.

La Porta, R., F Lopez-de-Silanes, A Shleifer and R Vishny. 1999. The Quality of

Government. Journal of Law, Economics, and Organization 15 (1): 222-279.

La Prensa, 5/18/2004, Scioli recibo proyectos de una ONG, p.3.

Lambsdorff, JG. 2003. How Corruption Affects Persistent Capital Flows. Economics of
Governance 4 (3).

Lanyi, A, W Guevara and S Bell. 2000. Bolivian Customs Reform: A Case Study of
Consolidating Democratic Institutions. Working Paper No. 247. Center for
Institutional Reform and the Informal Sector, University of Maryland.

Leite, C and J Weidmann. 1999. Does Mother Nature Corrupt? IMF Working Paper.

Lensvelt-Mulders, GJL., JJ. Hox, PGM. van der Heijden, CJM Maas. 2005. Meta-
Analysis of Randomized Response Research: Thirty Years of Validation. Sociological
Methods and Research 33 (3): 319-348.

Levitsky, S. 2003. Transforming Labor-Based Parties in Latin America. Cambridge:

Cambridge University Press.
_______ and L Way. 2002. The Rise of Competitive Authoritarianism. Journal of
Democracy 13 (2): 5165.

Leyton, A. 1994. Administrative Reform Program: The Case of Bolivia. In SA.

Chaudhry, GJ Reid and WH Malik, eds. Civil Service Reform in Latin America and the
Caribbean: Proceedings of a Conference. Washington, DC: World Bank, 158-165.

Linz, JJ. 1994. Presidential or Parliamentary Democracy: Does It Make a Difference? In

JJ Linz and A Valenzuela, eds., The Failure of Presidential Democracy: Comparative
Perspectives (Vol. 1). Baltimore: Johns Hopkins University Press, 3-87.

Londregan, JB. 2000. Legislative Institutions and Ideology in Chile. Cambridge:

Cambridge University Press.

Lpez Echague, H. 1996. El Otro: Una Biografia de Eduardo Duhalde. Buenos Aires:
Planeta.

Madalla, GS. 1983. Limited Dependent and Qualitative Variables in Econometrics. New
York: Cambridge University Press.

Mainwaring, SP. 1991. Politicians, Parties and Electoral Systems: Brazil in Comparative
Perspective. Comparative Politics 24 (1): 21-43.

________. 1993. Presidentialism, Multipartism and Democracy: The Difficult

298
 Combination. Comparative Political Studies 26 (2): 198-228.

________. 1995. Brazil: Weak Parties, Feckless Democracy. In S Mainwaring and

TR Scully, eds., Building Democratic Institutions: Party Systems in Latin America. Stanford:
Stanford University Press.

________. 1999. Rethinking Party Systems in the Third Wave of Democratization.

Stanford: Stanford University Press.

________ and MS Shugart, eds. 1997. Presidentialism and Democracy in Latin

America. Cambridge: Cambridge University Press.

________ and A. Prez-Lin. 2005. Latin American Democratization since 1978:

Democratic Transitions, Breakdowns and Erosions. In F. Hagopian and S.P.
Mainwaring, eds., The Third Wave of Democratization in Latin America: Advances and
Setbacks. Cambridge: Cambridge University Press.

________ and A. Prez-Lin. 1997. Party Discipline in the Brazilian Constitutional

Convention. Legislative Studies Quarterly 22 (4): 453-483.

Malamud, C and E Posada-Carb, eds. 2005. The Financing of Politics: Latin American
and European Perspectives. London: Institute for the Study of the Americas.

Marcel, Mario. 1999. Effectiveness of the State and Development Lessons from the
Chilean Experience. In G. Perry and DM Leipziger, eds., Chile: Recent Policy Lessons
and Emerging Challenges, Washington, DC: World Bank, pp. 265-325.

Mauro, P. 1995. Corruption and Growth. Quarterly Journal of Economics 110 (3): 681-
712.

_______. 1997. The Effects of Corruption on Growth, Investment and Government

Expenditure: A Cross-Country Analysis. In KA Elliott, ed. Corruption in the Global
Economy. Washington: Institute for International Economics.

Mayorga, RA. 1998. El Financiamiento de los Partidos Polticos en Bolivia. In P del

Castillo and D Zovatto, eds., La Financiacin de la Poltica en Iberoamrica. San Jos:
Instituto Interamericano de Derechos Humanos.

McCullagh, P. and JA. Nelder. 1983. Generalized Linear Models. London: Chapman
and Hall.

Meier, KJ and TM Holbrook. 1992. I Seen My Opportunities and I Took Em: Political
Corruption in the American States. Journal of Politics 54(1): 135-155.

Meneguello, R. 1995. Electoral Behavior in Brazil: The 1994 Presidential Elections.

International Social Science Journal 47 (4):627-.

________. 1989. PT: A Formao de um Partido. Rio de Janeiro: Paz e Terra.

299
Meza, V, J Castellanos, O Avila, R Romero and L Salomn. 2002. Corrupcin y
Transparencia en Honduras. Tegucigalpa: CEDOH.

Ministerio de Planejamento, Oramento e Gesto (Brazil). 2004. Boletim

Estatstico de Pessoal 99 (July). Available at:
http://www.servidor.gov.br/publicacao/boletim_estatistico/bol_estatistico.htm

Miyake, I. 1999. Candidate Evaluation and Voting Choice under the Japanese Electoral
System. In Grofman et al., eds., Elections in Japan, Korea, and Taiwan under the Single
Non-Transferable Vote. Ann Arbor: University of Michigan Press, 153-180.

Montinola, GR and RW Jackman. 2002. Sources of Corruption: A Cross-National Study.

British Journal of Political Science 32: 147-170.

Morales, JA. 2003. The National Revolution and its Legacy. In P Domingo and MS
Grindle, eds., Proclaiming Revolution: Bolivia in Comparative Perspective. Cambridge, Mass:
Harvard University Press.

Morgenstern, S. 2004. Patterns of Legislative Politics. Cambridge, Cambridge University

Press.

Morris, S. 1991. Corruption and Politics in Contemporary Mexico. Tuscaloosa:

University of Alabama Press.

Morton, RG. 1999. Methods and Models: A Guide to the Empirical Analysis of Formal
Models in Political Science. Cambridge: Cambridge University Press.

Murillo, MV and E Calvo. 2004. Who Delivers? Partisan Clients in the Argentine
Electoral Market. American Journal of Political Science Vol. 48 (4): 742-757.

Nam, M. 1995. The Corruption Eruption. Brown Journal of World Affairs 2, 2: 245-261.

National Commission for the Protection of Human Subjects of Biomedical Research.

1978. The Belmont Report : Ethical Principles and Guidelines for the Protection of Human
Subjects of Research. Washington, DC: US Government Printing Office.

Navia, P. 2004. Legislative Candidate Selection in Chile. Paper prepared for conference
Pathways to Power: Political Recruitment and Candidate Selection in Latin
America.

Newell, J. 2000. Party Finance and Corruption: Italy. In R. Williams, ed., Party Finance
and Political Corruption. New York: St. Martins Press.

Nice, DC. 1983. Political Corruption in the American States. American Politics Quarterly
11 (4): 507-517.

Nicolau, JM. 1996. Multipartidarismo e Democracia. Rio de Janeiro: FGV.

300
Njaim, H. 1981. El Financiamiento de la Maquinaria Partidista en Venezuela. In JC Rey
et al., El Financiamiento de los Partidos Politicos y la Democracia en Venezuela. Caracas:
Editorial Ateneo, 25-80.

Norris, P. 2004. Electoral Engineering: Voting Rules and Political Behavior. Cambridge:
Cambridge University Press.

North, DC. 1990. Institutions, Institutional Change, and Economic Performance.

Cambridge: Cambridge University Press.

Nunes, E. 1997. A Gramtica Poltica do Brasil : Clientelismo e Insulamento

Burocrtico. Rio de Janeiro : J. Zahar Editor.
Nunes Leal, V. 1975. Coronelismo, Enxada e Voto. So Paulo: Alfa mega.

Nye, JS. 1967. Corruption and Political Development: A Cost-Benefit Analysis.

American Political Science Review 61 (2): 417-427.

O Donnell, G. 1999. Horizontal Accountability in New Democracies. In A. Schedler, L.

Diamond and MF Plattner, eds., The Self-Restraining State: Power and Accountability in
New Democracies. Boulder: Lynne Rienner.

Orellana Vargas, P. 2004. Probidad y Corrupcin en Chile: El Punto de Quiebre.

Polis, Revista On-line de la Universidad Bolivariana 3 (8).

Osborne, D. and T. Gaebler. 1992. Reinventing Government: How the Entrepreneurial

Spirit is Transforming the Public Sector. Reading: Addison-Wesley.

Otero, D. 1997. El Entorno: La Trama ntima del Aparato Duhaldista y sus Punteros.
Buenos Aires: Nuevohacer.

Persson, T and G. Tabellini. 2003. The Economic Effects of Constitutions. Cambridge:

MIT Press.

_______ and F. Trebbi. 2003. Electoral Rules and Corruption. Journal of

the European Economic Association 1 (4): 958-989.

Persson, T, G Roland and G Tabellini. 1997. Separation of Powers and Political

Accountability. Quarterly Journal of Economics 112 (4): 1163-1202.

Petras, J. 1969. Politics and Social Forces in Chilean Development. Berkeley: University
of California Press.

Petrocik, J. 1996. Issue Ownership and Presidential Elections. American Journal of

Political Science 54: 825-850.

Pharr, SJ. 2000. Officials Misconduct and Public Distrust: Japan and the Trilateral
Democracies. In SJ Pharr and RD Putnam, eds. Whats Troubling the Trilateral

301
 Countries. Princeton: Princeton University Press.

Prado, L .1997. Transportes e Corrupo: um Desafio Cidadania. Rio de

Janeiro: Topbooks.

Programa de las Naciones Unidas para el Desarrollo (PNUD). 2004. La Democracia en

Amrica Latina: Hacia una Democracia de Cuidadanos y Cuidadanas. New
York: PNUD.

Przeworski, A, ME Alvarez, JA Cheibub and F Limongi. 2000. Democracy and

Development: Political Institutions and Well-being in the World, 1950-1990. Cambridge:
Cambridge University Press.

Putnam, R. with R. Leonardi and RY. Nanetti. 1993. Making Democracy Work: Civic
Traditions in Modern Italy. Princeton: Princeton University Press.

Ramseyer, JM and FM. Rosenbluth 1993. Japans Political Marketplace.

Cambridge: Harvard University Press.

Rangel, DA. 1973. Los Mercaderes del Voto. Valencia: Editores Distribuidores Vadell
Hermanos.

Rauch, JE. and P. Evans. 2000. Bureaucratic Structure and Bureaucratic Performance in
Less Developed Countries. Journal of Public Economics 75: 49-71.

Reed, SR. 1996. Democracy and the Personal Vote: A Cautionary Tale from Japan.
Electoral Studies 13 (1): 17-28.

Rehren, A. 2002. Clientelismo Poltico, Corrupcin y Reforma del Estado en Chile. In

Valds Prieto, ed., Reforma del Estado, Vol. II: Direccion Publica y Compras Publicas.
Santiago de Chile: CEP.

Rezende, Flvio da Cunha. 2002. Por Que Reformas Administrativas Falham? Revista
Brasileira de Cincias Socias 17 (50): 123-184.

Rhodes, M. 1997. Financing Party Politics in Italy: A Case of Systemic Corruption. West
European Politics 20 (1): 54-80.

Riggs, FW. 1963. Bureaucrats and Political Development: A Paradoxical View. In

J. La Palombara, ed., Bureaucracy and Political Development. Princeton:
Princeton University Press, pp. 120-167.

Rose Ackerman, S. 2001. Political Corruption and Democratic Structures. In AK Jain,

ed., The Political Economy of Corruption. London: Routledge.

Rosenbaum, PR. 2002. Observational Studies. New York: Springer-Verlag.

________. and DB. Rubin. 1983. The Central Role of the Propensity Score in

302
 Observational Studies for Causal Effects. Biometrika. 70:41-55.

Sadek, MT and RB Cavalcanti. 2003. The New Brazilian Public Prosecution: An Agent
of Accountability. In SP Mainwaring and C Welna, eds., Democratic
Accountability in Latin America. Oxford: Oxford University Press, 201-227.

Salomn, L. 2002. Corrupcin y Transparencia. In Meza et al., Corrupcin y

Transparencia en Honduras. Tegucigalpa: CEDOH, 143-169.
______. 2004. Democracia y Partidos Polticos en Honduras. Tegucigalpa: CEDOH.

Samuels, DJ. 1999. Incentives to Cultivate a Party Vote in a Candidate-Centric Electoral

System. Comparative Political Studies 32 (4): 487-518.

______. 2001. When Does Every Penny Count? Intra-party Competition and Campaign
Finance in Brazil. Party Politics 7 (1): 89-102.

______. 2002. Pork Barreling is not Credit Claiming or Advertising: Campaign Finance
and the Sources of the Personal Vote in Brazil. Journal of Politics 64 (3): 845-863.

______. 2004. Political Ambition, Candidate Recruitment and Legislative Politics in

Brazil. Paper prepared for conference Pathways to Power: Political Recruitment and
Candidate Selection in Latin America.

Samuelson, PA. 1976. Economics. New York: McGraw-Hill.

Sartori, G. 1994. Comparative Constitutional Engineering: An Inquiry into Structures,

Incentives, and Outcomes. Basingstoke: Macmillan.

Scheers, NJ and CM Dayton. 1988. Covariate Randomized Response Models. Journal of

the American Statistical Association 83: 969-974.

Schiavo-Campo, S., G. de Tommaso and A. Mukherjee. 1997. An International Statistical

Survey of Government Employment and Wages. World Bank Policy Research
Working Paper 1806.

Schick, A. 1998. Why Most Developing Countries Should Not Try New Zealand's
Reforms. World Bank Research Observer, vol. 13 (1): 123-131.

Schneider, BR. 1999. The Desarrollista State in Brazil and Mexico. In Meredith Woo-
Cumings, ed., The Developmental State. Ithaca: Cornell University Press, 276-305.

_______. 1991. Politics within the State: Elite Bureaucrats and Industrial Policy in
Authoritarian Brazil. Pittsburgh: University of Pittsburgh Press.

_______. and B Heredia, eds., Reinventing Leviathan: The Politics of Administrative

Reform in Developing Countries. Miami: North-South Center Press.

Scott, JC. 1969a. Analysis of Corruption in Developing Nations. Comparative Studies in

303
 Society and History 11 (3): 315-341.

______. 1969b. Corruption, Machine Politics and Political Change. American Political
Science Review 63 (4): 1142-1158.

Seligson, M. 2002. The Impact of Corruption on Regime Legitimacy: A Comparative

Study of Four Latin American Countries. Journal of Politics 64 (2): 408-433.

Shefter, M. 1977. Party and Patronage: Germany, England and Italy. Politics and Society
7: 403-452.

Shin, DC. 1999. Mass Politics and Culture in Democratizing Korea. Cambridge:
Cambridge University Press.

Shugart, MS and JM Carey. 1992. Presidents and Assemblies: Constitutional Design and
Electoral Dynamics. Cambridge: Cambridge University Press.

Siavelis, P. 1997. Continuity and Change in the Chilean Party System. Comparative
Political Studies 30: 651-74.

______. 2000a. Disconnected Fire Alarms and Ineffective Police Patrols: Legislative
Oversight in Post-Authoritarian Chile. Journal of Interamerican Studies and World Affairs
42 (1): 71-98.

______. 2000b. The President and Congress in Postauthoritarian Chile. University

Park: Pennsylvania State University Press.

______. 2002. The Hidden Logic of Candidate Selection for Chilean Parliamentary
Elections. Comparative Politics 34 (2): 419-438.

Siegel, GB. 1964. The Vicissitudes of Governmental Reform in Brazil: A Study of the
DASP. PhD thesis, University of Pittsburgh.

Silveira, FE. 1998. A Deciso do Voto No Brasil. Porto Alegre: EDIPUCRS.

Simon, A. 2002. The Winning Message: Candidate Behavior, Campaign Discourse, and
Democracy. Cambridge: Cambridge University Press.

Sinha, A. 2003. Rethinking the Developmental State Model: Divided Leviathan and
Subnational Comparisons in India. Comparative Politics 35 (4):.

Smith, PH. 1979. Labyrinths of Power: Political Recruitment in Twentieth Century

Mexico. Princeton: Princeton University Press.

Snyder, R. 2001. Scaling Down: The Subnational Comparative Method. Studies in

Comparative International Development 36 (1): 93-110.

Speck, BW. 2000. Mensurando a Corrupo: Uma Reviso de Dados Provenientes de

304
 Pesquisas Empricas. In Speck et al., Os Custos da Corrupo. So Paulo: Konrad
Adenauer.
Spink, P. 1999. Possibilities and Political Imperatives: Seventy Years of Administrative
Reform in Latin America. In LC Bresser Pereira and P Spink, eds., Reforming the State:
Managerial Public Administration in Latin America. Boulder: Lynne Reiner, 91-114.

Stepan, A and C Skach. 1993. Constitutional Frameworks and Democratic Consolidation:

Parliamentarism versus Presidentialism. World Politics 46 (1): 1-22.

Stokes, WS. 1949. The "Cuban Revolution" and the Presidential Elections of 1948. The
Hispanic American Historical Review 31: 37-79.

Suleiman, E. 2003. Dismantling Democratic States. Princeton: Princeton University

Press.

Taylor, MM. 1996. When Electoral and Party Institutions Interact to Produce Caudillo
Politics: The Case of Honduras. Electoral Studies 15 (3): 327-337.

Tendler, J. 1997. Good Government in the Tropics. Baltimore : Johns Hopkins University
Press.

Teresa Zegada, M. 1996. Democratizacion Interna de los Partidos Polticos em Bolvia.

La Paz: ILDIS.

Torres, MDF. 2004. Estado, Democracia e Administrao Publica no Brasil. Rio de

Janeiro: Editora FGV.

Treisman, D. 2000. The Causes of Corruption: A Cross-National Study. Journal of Public

Economics 76: 399-457.

Valenzuela, A. 1994. Party Politics and the Crisis of Presidentialism in Chile: A Proposal
for a Parliamentary Form of Government. In JJ Linz and A Valenzuela, eds., The
Failure of Presidential Democracy: Comparative Perspectives (Vol. 1). Baltimore: Johns
Hopkins University Press, 91-150.

________. 1984. Parties, Politics and the State in Chile: The Higher Civil Service. In
EN Suleiman, ed., Bureaucrats and Policy Making: A Comparative Overview. New York,
NY: Holmes and Meier, pp. 242-279.

________. 1978. The Breakdown of Democratic Regimes: Chile. Baltimore : Johns

Hopkins University Press.

________. and T. Scully. 1997. Electoral Choices and the Party System in Chile.
Comparative Politics 29: 511-527.

van der Heijden, PGM and G van Gils. 1996. Some Logistic Regression Models for
Randomized Response Data. In A Forcina, GM Marchetti, R Hatzinger and G

305
 Galmatti, eds. Statistical Modelling: Proceedings of the 11th International Workshop on
Statistical Modelling. Perugia, Italy: University of Perugia.

________. 2000. A Comparison of Randomized Response, Computer Assisted Self

Interview and Face-to-Face Direct Questioning: Eliciting Sensitive Information in
the Context of Welfare and Unemployment Benefit Fraud. Sociological Methods &
Research 28, 505-537.

Van Rijckeghem, C and B Weder. 2001. Bureaucratic Corruption and the Rate of
Temptation: Do Wages in the Civil Service Affect Corruption, and by How
Much? Journal of Development Economics 65: 307-331.

Vargas, OR. 2000. Crculos del Infierno: Corrupcin, Dinero y Poder. Managua: Foro
Democrtico.

Veja, Corrupo: possvel ganhar a guerra contra ela. No. 1906, May 18, 2005.

Wantchekon, L. 2003. Clientelism and Voting Behavior: Evidence from a Field

Experiment in Benin. World Politics 55: 399-422.

Warner, SL. 1965. Randomized Response: A Survey Technique for Eliminating Evasive
Answer Bias. Journal of the American Statistical Association 60: 63-69.

Waterbury, J. 1992. The Heart of the Matter? Public Enterprise and the Adjustment
Process. In S Haggard and RF Kaufman, ed., The Politics of Economic Adjustment.
Princeton: Princeton University Press, 182-220.

Weber, M. 1946. Bureaucracy. In HH Gerth and CW Mills, trans. From Max Weber:
Essays in Sociology. New York: Oxford University Press.

Weyland, K. 2002. Limitations of Rational-Choice Institutionalism for the Study of Latin

American Politics. Studies in Comparative International Development 37 (1): 57-85.

________. 2000. The Brazilian State in the New Democracy. In PR Kingstone and
TJ Power, eds., Democratic Brazil: Actors, Institutions and Processes. Pittsburgh: University
of Pittsburgh Press, 36-57.

Williams, R. ed. 2000. Party Finance and Political Corruption. New York: St. Martins
Press.

Willis, E. 1995. Explaining Bureaucratic Independence in Brazil: The Experience of the

National Economic Development Bank. Journal of Latin American Studies 27 (3): 625-
661.

Wittman, DA. 1989. Why Democracies Produce Efficient Results. Journal of Political
Economy 97 (6): 1395-1424.

306
_______. 1995. The Myth of Democratic Failure: Why Political Institutions Are
Efficient. Chicago: University of Chicago Press.

Woodall, B. 1999. Japan under Construction. Berkeley: University of California Press.

World Bank. 1997. The State in a Changing World. Washington, D.C.: World Bank.

_______. 2000a. Bolivia: From Patronage to a Professional State, 2 vols. Report No.
20115-BO.

_______. 2000b. Cambodia Governance and Corruption Diagnostic: Evidence from

Citizen, Enterprise and Public Official Surveys. Available at:
http://www.worldbank.org/wbi/governance/capacitybuild/d-surveys.html#reports

World Economic Forum (WEF). 2004. Global Competitiveness Report. Geneva: World
Economic Forum.

Zovatto, Daniel G. 2004. Dinero y Poltica en Amrica Latina: Una Visin Comparada.
San Jos: FLACSO.

307