Sunteți pe pagina 1din 31

Proportional Representation and Female Parliamentarians Author(s): Rob Salmond Reviewed work(s): Source: Legislative Studies Quarterly, Vol.

31, No. 2 (May, 2006), pp. 175-204 Published by: Comparative Legislative Research Center Stable URL: http://www.jstor.org/stable/40263381 . Accessed: 11/01/2013 01:25
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Comparative Legislative Research Center is collaborating with JSTOR to digitize, preserve and extend access to Legislative Studies Quarterly.

http://www.jstor.org

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

ROB SALMOND University ofCalifornia-Los Angeles

Proportional Representation and Female Parliamentarians


This article does thechoiceof a nation's electoral asks,"Whateffect system of its I haveon thegender over time?" find that theelectoral composition parliament toplay, has an important butprevious work has overstated, of system part byfactors howmuch ofa difference between twoandthree, an electoral can make.This system an updated offemale article contributes nonlinear an improved theory representation, acrossspaceandtime, andmore dataset onwomen's modern statistical representation on thisquestion. than used in research techniques previously

This articleis about the politicalrepresentation of women. havefound that researchers Successive leftism, highlevelsof female andproportional labor force (PR) electoral representation participation, on thepercentage ofa nation's ruleshavea causalimpact parliament andactivists haverelied onthis research whoarewomen. Policymakers "If about possible electoralreforms: to help guide theirthinking B is better forwomenthanelectoral electoral A, then system system refrain. Theliterature B should is a common liberal be adopted" system to policymakers as itcouldbe, is notas useful as itstands, however, from somepersistent andithassuffered methodological shortcomings. on thequestion of female all previous studies Almost represenin in parliament haveasked,"Whatfactors tation explaindifferences intimeTThisquestion, at a given levelsoffemale point representation suited to thepolicymaker's whileinteresting, is notperfectly inquiry, inthis factor have effect woulda change which is,"What independent overtimeT' This article aimsto on levelsofwomen'srepresentation demands and to closerto thepolicycommunity's provideanswers This that earlier research hasmissed. the goalrequires dynamics capture of analysisused in previous research be thatmanyof themethods In particular, theprevious of usingpurely crosspractice rethought. This article makesuse of a panel needsupdating. datasets sectional 2 1 advanced industrialized democracies overa period dataset covering withdata from 281 elections. Earlierestimation of about50 years, linear are also which reliedon purely models, regression procedures, LEGISLATIVE STUDIESQUARTERLY, XXXI, 2,May2006 175

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

176

Rob Salmond

ofthegrowth updatedhere.FollowingMatland's (1993) in-depth study in Norway,I propose a logit-esque S-curve of female representation in parliament. model of the growth of women's representation The analysis in this article focuses on one known correlateof female legislativerepresentation: electoralrules. Policymakersoften debate unilaterally electoral rules in orderto achieve certain changing of the enhanced outcomes, representation policy including descriptive over similar debates do hold not, however, community. Policymakers manaof the workforce the female (or proportion unilaterally changing gerial workforce), alteringthe level of culturaltoleranceforfemale leadersin thecountry, or shifting thenation'sdominant religiontoward one more liberal on genderissues. Such shiftsare beyond the direct controlof the legislature. The electoralrules are almostthe only way can legislators directly "engineer"increasesin thenumberof women in parliament. Thus, this article concentrates on examining how is likelyto be. successfulsuch electoralengineering This narrowfocuson theimpactof electoralruleson representation allows for methodological innovationsin the measurement of, and accountingfor, other(oftencultural)causes of women's representation.I propose a solution to the problem of "measuringculture": it.In thisarticle, control for without actuallymeasuring politicalculture I also introducemore-sophisticated capturing ways of quantitatively the essence of electoral systemsthanhave been used in most of the on thistopic. previous literature My resultsshow thatthe choice of electoralsystemdoes matter as in parliaments but not as substantively forwomen's representation a UnitedStates/ A shift from has suggested. muchoftheearlierresearch United Kingdom-style single-member-district (SMD) system to a PR district (low system leads to an magnitude) Norwegian-style women of increasein the descriptive by between 1.5% representation and one-halfthe size of and 7%. This increase is between one-third earlier estimatesof the impact of such a reform.Shiftsto a Dutchstyle (high districtmagnitude) PR systemor a mixed systemmay to an Irish-style single transferproduce a greaterimpact,but a shift able vote systemis less beneficial. The remainderof this article is divided into four sections. In on electoral systemsand Section I, I review the existing literature on thosestudiesthathave tried women's representation, concentrating In Section II, I outlinethenonlinear to testtheirtheoriesempirically. thedata used and present of changes in women's representation theory And theresults. In Section III, I present and interpret to testthetheory. in Section IV, I offer conclusions.

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Proportional Representation and Female Parliamentarians I. ElectoralSystems

177

The relationship between electoral and therepresentasystems been thesubjectof a smallmountain of tionof womenhas already theempirical in research. Before we review relation academic findings itis useful to setoutthecausal logicthrough which to thisquestion, to benefit theelectoral cause ofwomen. PR is hypothesized PR Be Better for Women? Why Might ofproportional The keydifference between systems representalies in their electoralarrangements "district tionand majoritarian DM The measures how are,on (DM). manypoliticians magnitude" In most each electoral district. elected systems, by majoritarian average, a single that electoral elects eachconstituency system politician, giving are knownas single-member-district a DM of one. These systems is In all PR systems, on theother hand,each constituency systems. than one more for politician. responsible electing that arerelevant tothe effects on elections The DM has multiple that women's of (1999,555) argues Reynolds representation. question an for bosses to stand lowest"creates incentive SMD party system any ingeographical these candidates common-denominator districts; rarely A related claimis that SMD elecorminorities." turn outtobe women than district-level toproduce more-adversarial tions tend campaigning in win a a candiorder to local SMD PR elections. election, Indeed, PR arelosers. Not so under that all other candidates ensure datemust or at least more accustomed to are better some rules. Men, argue, than them battle of thisgladiatorial women, (all making type political inSMD systems candidates for attractive elsebeing parties equal)more 1955;Rule 1981). (Duverger DM systems tobetter Another high representation argument linking such for women lies in theincreased for provide opportunities systems at In the district level. SMD ticket socialengineering balancing through cannot be balanced slateofcandidates a localdistrict's bygensystems, There is onecandidate, and or location. sexual orientation, race, der, only InPR systems with lowDM, some ticket orheisn't. either sheisa woman thedemands of multiple is possible,butaccommodating balancing DM become easier as increases Such acts is difficult. balancing groups rural for of demands andthe further women, communities, representation to each other. and ethnic groupsare less likelyto be in opposition DM as rises(Darcy, becomes ticket Thus, increasingly easy balancing Lakeman Matland Clark and 1994,115; 1976,161; Welch, 1993).

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

178

Rob Salmond

Higher district magnitudes are also associated with higher numbers of parties in the political system (Duverger 1954). This increasein thenumber ofpartiescan have two effects. Reynolds( 1999, that in with 555) argues highlyproportional systems highDM "small an incentive to are able to have and parties gain representation parties broadentheiroverall electoralappeal by makingtheircandidatelists of small as diverse as possible." Second, the presence in parliament a point liberal-left who sometimes make parties, genderegalitarianism of difference is between themselvesand a center-left party, likelyto aid operation ofthe"contagionfrom theleft"theory of growing female forward Matland and Studlar (1996). representation put by Taken together, a small catalog of these mechanismsrepresent reasonswhyPR might better thanother be expectedto perform systems in termsof genderegalitarianism. There has, in fact,been verylittle should prove to be dispute in political science thatPR theoretically better forwomen thanSMD electoralsystems. But have theneattheoreticalmodels translated well intothe messy world of real politics? Is PR Better for Women? The State of theLiterature on theState of the World There have been two strandsto the empirical literature:one looking only at advanced industrialized democracies, the other in theworld.This articlespeaks directly includingalmostall countries - at present, to the narrower literature only panel data going back 50 - but nonethelessit is years are not available across the entireworld that of thisresearch. we considerthefindings of bothstrands important Studies of Advanced IndustrializedDemocracies. The assertion thatPR is better fortherepresentation of women dates to theworkof Castles (1981), Rule (1981, 1987), andNorris(1985, 1987). BothNorris least squares (OLS) (1985) and Rule (1987) employedlinearordinary methods to examine cross-sectional data frombetween 19 and 24 westerncountries.Both found thatthe electoral systemis the most of the level of female legislativerepresentation. significant predictor level of this Neither, however,providedan estimateof thesubstantive effect, reporting only standardizedcoefficients. A latergeneration of researchers, includingMatland (1998) and Kenworthyand Malami (1999), explored this question again. They also employed OLS methodsand cross-sectionalsamples (although data taken Matland does employa pooled cross-sectional design from at threetimeperiods). Both the Matland and Kenworthy and Malami

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Representation Proportional

179

PR and a strong and significant studies found between relationship of PR at the between levelsofrepresentation, 11% estimating impact world.These are verylargeeffects when and 16% in thedeveloped acrossthedeveloped world themeanlevelofwomen's representation theperiod1990-95. was only18% during studies here other Thefour controlled, highlighted among things, and education levels.All female laborforce for participation average also controlled fordifferent levels of leftism, exceptNorris'sstudy seats for women.LaterI source of another parliamentary purported variables areadequate orreliable. ornotthese control whether discuss in thisarea thathas been used only One avenueforresearch Matland is that ofthecase study. (1993) lookedattheimpact sparingly and party on in Norwegian district of changes magnitude magnitude have examined Canada and some scholars women'srepresentation, andMatland andStudlar 1996;Studlar 1994)andCostaRica (Matland Laterin as Matland and 2004; Taylor1997) testing grounds. (Jones in western countries as I use recent electoralreforms thisarticle, of the illustrate the to "natural comparative performance experiments" literature. thosein theexisting hereagainst modelsdeveloped levels offemale to factors believed other the explain partly Among the are cultural variables,especially comparatively representation andNorthern EurointheScandinavian toward drive "equity" strong Norris and Norris 1 countries 2003; 995; Inglehart (Bystydzienski pean of left-wing parties(Duverger1955; Lakeman 1993); the strength ofa multiparty (Lakeman 1976)- although system 1970);the presence is tells us thata multiparty 's system Duverger (1954) hypothesis and of the caused PR; demographic presence by probably andlow levelsofCasuchas highlevelsofeducation characteristics, tholicism (Rule 1981;Welch1978). variable as theonly have advancedthecultural Some scholars are the public Not in the variable only system. fullyindependent of to concerns favorable attitudes equityhighly general strongly butthey are also correfemale with correlated representation, higher latedwiththepresenceof a PR system (Norris1987, 120). Some from that have researchers suggested, therefore, a causalarrow points PR of and the "fairness" cultures 2004, 183). (Norris adoption valuing that this is showed Rokkan byBoix (1999),however, (1970) followed in northern that conservative instead the not case,establishing parties PR as a protection destroyed against being completely Europe adopted PR extended. was progressively at theballot box as thefranchise Thus, not for reasons of inclusiveness. These for was adopted gain, partisan PR is notendogenously causedby comparatively showthat findings

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

180

RobSalmond

inclusive cultures. We maytherefore proceedwithanalysis political ofthetwovariables separately. withtheexisting literature serveto cast doubt Threeproblems had as few as 19 of these studies over their First, many findings. Withsamplesthis variables. observations butup to six independent results mustbe questioned. of any statistical small,the reliability reliable results. the N make for more would Dramatically increasing for used to control thevariables The secondproblem concerns in main the which culture, is, explanation competing political many ways, that in representation for differences levels.It is doubtful cross-national of element thevariables the relevant culture, political actually capture and in positions of leadership attitudes toward women namely, having measures of used are common control variables Two of the most power. of levels. Both female education women's laborforce and participation for reasons tovariation these measures areblunt instruments andaresubject on these ofwomenas leaders.1 notrelated to theacceptance Relying work. in small-N measures can lead to unstable estimates, especially researchers is that with theexisting literature The third problem ofthedata on OLS models. almost Simpleinspection exclusively rely overtime.In New Zealand,it confirms that theprocessis notlinear for to runforparliament womenwerepermitted took60 yearsafter that levelsto reach5%, butitonlytook20 yearsafter representation after World 25 years for torisecloseto30%. In thefirst representation inNorway increased WarII, thelevelofrepresentation only5%, but inthenext itincreased by25%. Thesearenotisolated 25-year period, inorder toaccount is necessary modeling examples. More-sophisticated OLS models canalso leadtoimpossible for suchpatterns. predictions. arenot. at 0 and 100; linear arebounded predictions Percentages in thisliterature, for theshortcomings To correct anynewstudy on countries observations should: on this (1) Gather repeated question of the timeframe; under overa lengthy (2) Model theimpact study whoarefemale oflegislators variables on thepercentage independent and ofimpossible insucha wayas toavoidthepossibility predictions; that do not culture forcontrolling forpolitical (3) Develop methods ofculture. "measurement" relyon thestatistical overtime studies haveexamined To date, twopublished changes the small-Nproblem. in women'srepresentation, thusovercoming overtheperiod collected dataon 20 countries McAllister and Studlar forwomen that listPR does about10% better 1950-2000and found andMcAllister than andStudlar does SMD (McAllister 2002; Studlar theirstudiesalso have important however, 2002). Unfortunately, flaws.2 methodological

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Proportional Representation

181

The first is that, likethecross-sectional methodological problem and work McAllister Studlar's relied on OLS analysis. Secstudies, the authors inflated their number of cases and thereond, artificially - bycounting inthe results fore their confidence eachyear as a separate hadbeenan election ornot.3 case whether there McAllister and Third, for intheir madenoattempt tocontrol autocorrelation Studlar regression that is akinto saying thelevelofrepresentation model. Thisomission in election at time/is dependent variables onlyon theindependent at all on thelevel of representation at t-L This and notdependent if we consider, forexample,incumbency is unrealistic assumption it inflate and serves to theestimated size of artificially advantages, the electoral further other inflated coefficients, including system. They the impactof theelectoral (fordata reasons) system by excluding ofwomen's female another known correlate representation: participationin thelaborforce. cast significant doubton the These methodological problems itis worth McAllister and Studlar's Nonetheless of findings. reliability inflation of electoral the effects due that, system despite likely noting size of their the increased dataset to thesemethodological problems, to thecrossestimated leads to a smaller impactof PR, compared studies. sectional stream ofresearch with GlobalScope.A second Studies addressing link and between electoral thepurported systems women's representaitself totheadvanced industrialized tion hasnotconfined democracies, In under the number of cases investigation. expanding thereby greatly that thechoiceofelectoral there is nota consensus this lineofresearch, Neither Oakes and forwomen'srepresentation. is important system nor andNorris's of 73 nations (2003) (1993) study Inglehart Almquist's between found 62-nation anystatistically significant relationship study otherlarge-N rulesand womenin parliament. electoral Moreover, Malami's of and (1999) and parts Kenworthy analyses including over 140 Norris's(2004) work,bothof whichreliedon data from - found of PR that weresignificant butmuchsmaller effects nations intheOECD studies. found than those (1999) also Reynolds typically and found listPR fared of over 160 countries an analysis conducted thansimpleplurality rule.All of thesestudies better significantly in addition to other forlevels of economicdevelopment controlled variables.4 andpartisan cultural inthisarticle's is helpful The globalliterature endeavor, despite these studies showPR to have a of cases. thedifferent First, range than that onfemale smaller legislative representation suggested impact

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

182

RobSalmond

or Kenworthy and Malami.Thisdifference byMatland mayarise,as Matland the role because (1998, 120) suggests, playedbyinstitutions in fostering women's representation is dwarfed by the role that economic has to these development play.Alternatively, widerstudies on a very the results one can obtain whenrelying mayhighlight poor small the these studies some of methods Second, global employ sample. are improvements on thoseused in earlier studies, rich-country-only and thesebroader investigations guidesome of themethodological I describe in thefollowing decisions section. II. Theoryand Data A Nonlinear Theory ofFemaleLegislative Representation In a 1993 article, Matlanddiscussedthechanging Richard of in His research identified female pattern representation Norway. in four the rise of from to 36% as women's 0% stages representation of1989(Matland "Giants these 1993,746-50). He titled among stages which ofparliament were"elected members Men,"during anyfemale in spiteofbeingwomen"; a single which involved "One Is Enough," ineachdistrict; No female candidate on eachparty's ticket "Tokenism which sawthefirst use ofgender and "Gender More," Equality quotas; in whichwomenbecame"secondamong Just aroundtheCorner," equals." A similar can be toldaboutNew Zealand,thefirst country story to grant women theright ofthetwentieth tovote.Through themiddle 5% by themidcentury, representation reaching grewveryslowly, 1970s.At that theValuesParty, a smallleft-wing movetime, protest as The of its ment, promoted gender equity part platform. party gained substantial votersupport, butNew Zealand's SMD electoral system that ensured thegroupgainedno seats.Following thedemiseof the ValuesParty, of its activists moved into the LabourParty many larger with of thedemonstrated for their and,armed publicsupport message for Labourintoselecting morefemale candidates equality, pressured winnable seats(Hill andRoberts 1990;McLeay2000). Femalereprein Labour'scaucus rose quickly. sentation (some years Responding themainright-wing followed suit, later)to thisdevelopment, party andby 1999,there was a female a female minister, opposition prime and 30% ofthemembers ofparliament leader, (MPs) werewomen. in Denmark,Finland,Germany, Female representation and Swedenalso seemsto followthepattern of a periodof zero growth riseto slow growth, followed giving by a secondperiodof dramatic

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Proportional Representation FIGURE 1 An S-Curve Model ofWomen's Representation

183

All ofthese with theexception inrepresentation levels. shifts countries, in the1990s.5 a slowing rateofgrowth also experienced ofGermany, between time It appears, that thegeneral therefore, relationship is notlinear butS-shaped. andthepercentage ofMPs whoarewomen inNorway, research is supported Thiscontention byMatland's in-depth data from contextual evidencefrom New Zealand,and quantitative mentioned. relevant countries theseand other already inFigure1,which The general S-shapeofthiscurveis sketched from no representation to intofivebroad"stages"shifting is divided in of no sense These rigid equitable representation. stagesare, course, Thefirst four arelabeled arejustheuristic devices. orfixed; stages they whilethefifth thesame as thoseMatlandfoundin Norway, stage, has not been achieved "True Mathematically, anywhere. Equality," yet the "log odds" of the such a model can be derivedby calculating in held women: of seats parliament by percentage
, , , f [constraint] ] / Logodds= \n\M {[percentage] J (\)

In equation(1), constraint an ex ante researcherrepresents the valueofthe limit onthe leveltowhich percentage imposed predicted

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

184

Rob Salmond

canrise. Notethat this limit is different from onLogodds. anyconstraint Forthisarticle, thevalueoftheconstraint is setat 50%, becausethis levelof representation in thelegislature, approximates gender parity thealmostuniversally in this area of thefeminist agreed-upon goal The results in no on thischoice are, however, project. waydependent ofconstraint level.Detailsofrobustness in checks onthis issueappear the results section. TheS-shaped model as outlined aboveis only useful ina cross-national ifoneassumes that all countries inthesample study started women at the same time. This giving representation assumption is clearly of years thenumber false,so I used a variableindicating sincesuffrage was first to a variant a found on variable women, granted Studlar and McAllister to control for the time at important by (2002), which womenstarted to winpolitical rights.6 One consequence of thetransformation is thatit changesthe "natural" for each inthesystem. If of the variables signs independent we expecteda variableto be associated withan increase in the oflegislators whoarewomen, that then variable wouldbe percentage with a decreaseinthelog oddsofthepercentage. associated In order to solvethisproblem, I performed a simplelinear transformation on thelog odds.Thistransformation restores the"natural" for each sign oftheindependent variables' estimated coefficients. It does notaffect thecoefficients inanyother theconstant) (except way.The dependent variable for theregressions is therefore:
= X-\\J[[cOmtraint] DepVar -X^\ L {[percentage] )\ CollectingData to Testthe Theory (2)

Thissection outlines theother variables usedinmy independent The literature review that four broad classes of analysis. suggested variables inparliament: affect thepercentage ofwomen instipolitical of and cultural socioeconomic tutions, factors, patterns partisan support, factors. This article one uses two measures of politicalinstitutions, the salient of the electoral and the other meacapturing aspects system the(logged)size oftheassembly. I measured causesof suring partisan women's in the of representationparliament using percentage seatsin awarded to leftist to CastlesandMair's parliament parties [according I in thisarticle, socioeconomic measures (1984) framework]. ignore becausethe case selection ensures that countries areunder wealthy only I discusstheproblem consideration. in a later of quantifying culture subsection.

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Representation Proportional

185

21 countries relieson a dataset from 1950 Thisarticle covering intheanalysis inthe areamong thewealthiest to 2001. The countries ofdemocratic elections the haveall hada number world, during period, mentioned. The set dataon thevariables andhavereliable previously willbe familiar to regular readers on thissubject, of countries reprethe "usual a list of Australia, Austria, suspects": Belgium, senting Ireland, Finland, France, Greece, Canada,Denmark, (West)Germany, New Zealand, theNetherlands, Norway, Portugal, Spain, Italy, Japan, United andthe United States. Each the Switzerland, Sweden, Kingdom, and democratic electionis forthe lowerhouse of the legislature, in where womenwerenotallowedto vote(onlya problem elections a N These observations total of not included. were give Switzerland) elections. 281 separate in somedetailhow thisarticle The restof thissection explains on How from traditional approaches twokeydataquestions: departs culture? and electoral one quantify should (1) systems (2) political Previous workon thistopichas ElectoralSystems. Quantifying electoral forcategorizing of methods a wide array offered systems. variables.Matland(1998), Most analyseshave reliedon dummy andNorris and Studlar McAllister (2003) use a (2002), andInglehart of PR in theelectoral is no element variable coded as 0 whenthere that andNorris andas 1 otherwise. (2003, 141) admit Inglehart system electoral versusproportional this"simplemeasureof majoritarian suchas district other variations, important mayfailtocapture systems the level of or disproportionality." magnitude a three-point andMalami( 1999) propose scale,where Kenworthy such diverse ofmixed SMD is codedas 0, anyform system (including member mixed the as German-style proporvery proportional systems Irishsingletransferable tional vote, proportional (MMP), moderately is a Italian andtokenly systems) supplementary-member proportional in two addition has PR is a 2. This and 1, problems approach party-list contains themixed mentioned. tothebluntness First, category already in more to its name of electoral a diverse livingup systems, range the Malami considered. and than Second, coding Kenworthy ways of list-based that theestimated theresult mechanism impact imposes of mixed twicethesize of theestimated PR will be exactly impact this reason to is no theoretical there even expect result. systems, though a use series of variables and McAllister Studlar (2002) dummy of electoral list fourbroadfamilies to represent systems: PR, other The first andthird and other PR, first-past-the-post plurality. plurality, but the second and fourth are categories categories self-explanatory,

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

186

RobSalmond

of non-purebins"intowhichall forms appearto act as "remainder electoral are thrown. The "other PR," for type systems category and the is home for the low-DM Irish high-DM example, system German The "other mayhave actedas a system. plurality" category for French two-round oritcouldhavealso elections, dummy majority 1994 Italian and post1996 Japanesesystems, includedthe postonhowthe authors chosetotreat mixed member majoritarian depending is The of unlike (MMM) systems. problem lumping together systems andNoms technotas bad inthisschema as intheMatland/Inglehart instrument. butitis stilla blunt earlier, niquediscussed a relies on coding scheme Reynolds (1999) ten-category that he and Ben Reilly variables) (operationalized using dummy This is certainly and (Reynolds Reilly1997). coding system developed all an improvement, that arenarrow to ensure as thecategories enough within and there are no researchereach are similar systems category is This scheme between imposed relationships anyof thecategories. inthepresent number notuseful because of the small however, study, of countries underconsideration here.AlthoughI examine 281 If I adoptedthe there are only2 1 countries elections, represented. and Reillyapproach, thenthecategory Reynolds singletransferable vote(STV) wouldoperate as an Ireland country-dummy, singlenonMMP would and transferable vote(SNTV) wouldbe a Japan dummy, this scheme almost be a Germany Thus, using dummy. anyregressions to and it wouldbe impossible wouldsuffer from overdetermination, of thatelectoral tell whether the resulton STV was an indication ofIreland. or a measurement oftheidiosyncrasies system's impact is to present a codinginstrument The challenge forthisarticle forelectoralsystems thatuses few variables, ensuresthatsimilar different elecand electoral are qualitatively systems grouped together PR toral arenot, andaccurately measures thelogicbywhich systems A measure is hypothesized share oflegislative seats. toaffect women's that all thesethings is district Manyof the magnitude. accomplishes women'srepresentawhichPR is supposed to affect effects through also allowsus to tionoperate thesystems' DM. Thismeasure through variable this ina single variable. the electoral Moreover, capture system PR category that shifts within the broad allowsestimation oftheimpact electoral haveonwomen's UsingDM toquantify might representation. is nothing new in politicalscience,havingbeen used in systems and Shugart worksby Taagepera (1994), (1989), Lijphart scholarly andCox (1997). variablefor DM as an explanatory Matland(1993) considers it in favorof party levels beforerejecting women'srepresentation

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Proportional Representation

187

ofseatsa party hasina particular district. (PM), thenumber magnitude PM is a variable to because most of DM's DM, Certainly superior level.Nonetheless, DM and aresupposed to occurat theparty effects If DM is 1, then PM are likely to be at leastmoderately correlated: PM cannot be more than1; ifDM is 100,then atleasttwo eachparty's all other PMs willbe above 1. PMs willbe largeandalmost I use DM inplaceofPM inthis becauseofthedifficulty of study 1993 PM data. For his of Matland collected study Norway, obtaining in every inNorway ofeachparty district for dataon theperformance forone 10 elections.This is a verytaxingdata collectioneffort - indeed, Matland went toliveinNorway whilecollecting his country PM for onecountry was very then data.Ifcollecting taxing, collecting and thatis certainly a task wouldbe Herculean, it for21 countries the of this article. beyond scope DM is a goodproxy for that To ensure PM, I rana simple regresin themost PM of all the successful sionpredicting average parties of the21 countries under election recent study, using parliamentary The regression over52% of theDM as thesole predictor.7 explained in was significant of averagePM, and theDM variable thevariance at a of 0.001. This test shows direction the expected that, /?-value thebestvariable for theappropriPM is certainly capturing although DM is stilla verygood ate partof an electoral system's operation, in of the information contained that much a variable captures variable, without and that can be collected thePM variable learning multiple months on theroad! andspending many languages itcan is required before A little oftheDM measure fine-tuning 1 a in DM from to 5 is to have much A shift be useful. likely larger than a shift from 96 to 100.Forthis onwomen's representation impact thefinal electoral variable. DM is logged togenerate the reason, system PR electoral are and list-based SMD systems single-tier Purely For more like values of complicated systems, log(DM). easytoassign themeasurement two-tier listPR ormixed however, systems, problem whichDM translates into The mainmechanism is harder. through All is theparty list. else beingequal, systems female representation to be electedfrom lowerplaces on DM enablewomen with a higher withlower listthanwouldbe electedin districts thedistrict's party PM and therefore theparty's is higher district (Matland magnitudes, thequestion for these other is, 1993;Noms 1996).Therefore systems to electMPs? listdo yougo in order How fardowna party's withtwotiersof districting For all electoral (whether systems I used PR or someform ofmixed member two-tiered system), simple oftheDMs at thelowerandhigher tiers.8 a weighted average

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

188

RobSalmond

"Measuring" Political Culture.As noted earlier,available forpolitical statistical limits theways one can control information not available culture. that thestatistics are Once we recognize simply time we a to directly measure culture over frame, lengthy political needto consider other approaches. the Theapproach is different andnew.Theideabehind usedhere in one culture method is to reliably "undercount" theroleofpolitical it in another thus thenreliably "overcount" regression, regression, If assumes a within.9 one robust confidence interval to work giving levels that noroleindetermining different national cultures playalmost of all othervariables of femalerepresentation, thenthe estimates willbe biasedupwards. On theother theelectoral (including system) culture national hand,if one assumesthat political playsalmostthe will be variables sole determinative of all other role,thenestimates ofthebiasesin biaseddownwards. The trick is to knowthedirection the"measures" ofculture andtopaira control biasedinone direction with a control biasedin theother. In thefirst theyearintothe "undercontrol" one enters method, from theassumption for culture. This follows to equation proxy step for women that a tolerance that there is a single western culture develops leadersovertime.The method does notallow anybetween-country inpositions of inthedegree towhich societies women variance accept in culture variation actual cross-national Therefore, political power. any will show up in theregression as stochastic noise,severely system effect found the the effect found on culture and inflating dampening is mitiwithculture. Thisundercontrol on other variables correlated of theyears-since-suffrage by theinclusion gatedto a smallextent a to say that it is a big stretch variable. Even with thismodification, nation's culture by (1) a singleinternapolitical todayis determined inthe1920sor 1930s.The tional culture and(2) a legislative decision little. variable the bias inclusion ofthesecond Regressions very negates because these cultural undercontrols are they suppress wrong, using in favorof other variables. These regressions theeffects of culture oftheelectoral overstate theimpact system. of seats The cultural "overcontrol" method uses thepercentage a multitude of for heldbywomen after theprevious election tocontrol the sometimes-unidentifiable culturalfactorsthatcan influence in I compared all thoseelections of femalerepresentation. progress which was X% and noted which theprevious levelof representation I also addedto theovercontrol movedmorequickly regresupwards. variable and thelaggeddependent sionsan interaction term between thenumber of yearsthathave elapsed since theprevious election,

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Representation Proportional

189

is elections between oftime amounts becausethepassageofdifferent its to is related variable the how affect dependent closely likelyto laggedvalue. variablealso servesto of the laggeddependent The presence the that results. the bias Usingit,oneassumes regression significantly to cultural due is at level of representation(t-1) entirely processes; anda very ofculture's thisis thetheoretical highone impact, ceiling and electoral the that assumes One at that. system, assemblysize, at no had of makeup parliament absolutely impact (t-1), and partisan areevident variables other ofthese ornoteffects whether one then tests leads This method restrictive this in of at (0 spite harsh, assumption. dominate to tends variable cultural the where a to regression unfairly of othertruly and artificially the regression depress the effects andRao Maddala Griliches variables 1961; (Achen2000; explanatory overconcultural the the For using completeness, regressions 1973). variablesused in the cultural trolalso includeall of the cultural overcontrol undercontrol usingthiscultural Regressions regressions. statistical or a substantive bias becausethey arewrong, unfairly against the understate These variables. other on all impact regressions finding oftheelectoral system. usedhereareunfair, variables control biased, All ofthecultural alone taken this for and and wrong, gives a reason,anyregression identical otherwise in Taken however, regressions pairs, poorestimate. confidence a veryrobust controls cultural use different that provide lie. must variables other all of true the which within interval impact III. Data Analysis forthe support good initial analysisprovides Simplebivariate whole the Across women's for matters DM the ideathat representation. the female and DM between correlation the percentlogged sample, is 0.364, significant at/?< 0.001. Even with age of thelegislature with theanalysis, from SMD cases excluded loggedDM's correlation < . 0.001 at remains women's 0.285,againsignificant at/? representation the electoral about claims reliable and more-concrete make to In order we require analysis. however, more-sophisticated impact, system's Methods Statistical I use panelovertime, countries Becausethedatacovermultiple 1995 and 1996 Katz and Beck errors standard corrected (PCSEs). (See allow which of the of discussion fora full desirability PCSEs,

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

190

Rob Salmond

desirable to be madeaboutthedata.)10 Two statistically assumptions in are useful here. The first, assumptions particular heteroskedasticity acrosspanels,modelstheidea that thecombination of independent variables be better at levels ofwomen's may systematically predicting in in one than are another. The second representation country they desirable of is that for PCSEs allow contemporaneous property they correlation that cross acrosspanels, thus theideathat events modeling - forexample, national boundaries culture temporary pop emphasis ongender orinternational events likethe United Nations' issues, Beijing - can affect conference on thestatus of women levelsof representationacrossdifferent countries at the same time.11 These additional in results also serve to the overconfidence assumptions guard against that theearlier least estimation (GLS) generalized squares technique tended to produce (Beck andKatz 1995). Mostpaneldatasets a statistical ofthis alsoemploy correction type in for first-order autocorrelation described (AR1), buttheregressions this article do not.In theovercontrol an AR1 correction is regressions, not becauseofthe of the variable, necessary presence lagged dependent In theundercontrol which thesamefunction performs (onlystronger). an AR1 the addition of term is for tworeasons. undesirable regressions, theAR1 technique usedwith PCSE paneldatamakesuse ofthe First, variable. theAR1 term to theundercontrol laggeddependent Adding in would therefore result both the underandovercontrol regressions in them, a term which variable regressions having laggeddependent AR1 woulddefeat their the exclusion of Second, any jointpurpose.12 term that of from theundercontrol ensures the estimates regressions AR1 DM's impact an are notdepressed cultural controls by through that ofDM's thus extra confidence theestimates correction, providing in in these too are, fact, impact regressions high. Results Regression Tables 1 and 2 report theregression The mainanalysis results. thatthisarticlerelieson is provided in Table 1. Table 2 provides substantive ofthe Table1 regressions. When interpretation interpreting itis very theresults to that the should remember important regressions be readinpairs: oneregression for overcontrol for culture anda second but forthe exclusionof the lagged (with identicalspecification for cultural undercontrol. variable) dependent The first theelectoral to notice about Table 1 is that system thing in variable is and the direction (theloggedDM) significant expected in bothundercontrol and in one of thetwo overcontrol regressions

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Proportional Representation TABLE 1 Causes of Female Legislative Representation


All Cases Undercontrol Overcontrol (1) Constant p-value ln(DM) p-value YearssinceSuffrage p-value Year p-value Variable Lagged Dependent p-value Variablex LaggedDependent Elections Yearsbetween p-value size) ln(Assembly p-value LeftStrength -0.050 (0.094) 0.595 0.011*** (0.002) -39.830*** (9.414) 0.000 0.300*** (0.023) 0.000 0.027*** (0.004) 0.000 0.019*** (0.005) 0.000 (2) -13.501*** (4.184) 0.001 0.050** (0.016) 0.002 0.006** (0.002) 0.006 0.007** (0.002) 0.002 0.886*** (0.068) 0.000 -0.010 (0.017) 0.580 -0.018 (0.042) 0.663 0.001 (0.001) -0.052** (0.114) 0.648 0.019*** (0.004) PR Cases Only

191

Undercontrol Overcontrol (3) -46.741*** (11.634) 0.000 0.268*** (0.039) 0.000 0.024*** (0.005) 0.000 0.022*** (0.006) 0.001 (4) -12.411* (4.922) 0.012 0.031 (0.025) 0.211 0.005* (0.002) 0.038 0.006* (0.003) 0.013 0.866*** (0.086) 0.000 0.007 (0.024) 0.762 -0.061 (0.055) 0.262 0.002 (0.002)

p-value R2 N

0.000 0.699 280

0.321 0.924 272

0.000 0.626 166

0.340 0.922 160

standard errors are used throughout and appearin parentheses. Note: Panel-corrected < 0.001; all testsare two-tailed. *p < 0.05; **p < 0.01; ***/?

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

192

Rob Salmond

The onlytime loses signifitheelectoral variable regressions. system canceis whenthecultural overcontrols areinplace and thesampleis limited to PR cases only. inthe Giventhiscombination ofconstraints and the null we should not be this sample regression, surprised by I which literature. does not a to the extant finding, representchallenge 21 robustness checks on each of models and (2) (which performed (1) utilize thefull toverify that no onecountry's observations were sample) the In electoral to all of these effect statistical driving system significance. theresult survived unscathed.13 Thisdataset therefore checks, generally in the consensus that matters. the literature the electoral supports system in Table 1 behavegenerally Other variables More as expected. womentendto getintoparliament as timegoes on {Year)and when the battle for recedes into the sinceSuffrage). suffrage rights past{Years Notsurprisingly, an thelagged variable is excellent predictor dependent of women'srepresentation. in leftism tendto increase the Increases levelofrepresentation, in this effect the cultural although disappears overcontrol models. The assembly size appears notto matter. I reran themainregressions, of models(1) and (2), a number times tomakesurethat thechoiceofconstraint levelon thedependent I reran variable was notdriving theresults. themodels Specifically, with theconstraint setat30%,40%, 60%, 70%, 80%,90%, and 100%. Thesemodelstested fortheimpact of different constraint levelsand for different levelsofthedependent variable atwhich theindependent In a logit-type variables have maximum modelwitha 50% impact. thepoint for variable ofmaximum constraint, impact anyindependent is whenthe dependent variableis 25% by assumption. Thus, the if"setting" robustness checksalso determine thispoint of maximum at 15%,20%,andso onmakes a difference totheestimates.14 flexibility In both under-and overcontrol witha models, regressions constraint lower set lowerthan50% have an R2 value substantially than the50% model(R2for for the30% undercontrol model, example, models the50% model's0.699). Undercontrol dropsto 0.604 from with a constraint sethigher than 50% havevery marginally improved R2 values,rising the regressions, by up to 0.005. In theovercontrol andovercontrol, R2.In all models, under50% modelhas thehighest thesignificance of theelectoral variable remains remarkably system similar. inother areusually valuesonthe variable models exactly The/?thesameas inthe50% model, valueis andthemaximum change in/?for 0.004. The 50% modelshowsa slightly substantive effect higher electoral than do modelswith constraints setabove50% and systems a lowersubstantive thanthebelow-50%model.The overall effect ofthesesubstantive estimates arenarrow, from 0.252 to 0.341 ranges

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Proportional Representation
N? vo N? 00 sO Tt

193

'B

(j

<L) c

DoQ

a. o

rj

*> "

-7 ^
08

h d

"O

j OQ

3 o o

o^

\O

^ ri

"1 on

o^

o^

so

^ vd

vO

yO

vO

w .2

00

<N

bo g -^ <D >

on
>, 2 o. g -33

O
^^

Se10

o
|^^<n S S <u

5 S 2

HS

S. |

13

g^^^

g |

jh

I
-C O, o ^~

5 *

00

> I

S |

S| U

g g g ? ? ?

| ggg^g S J "S2
S g S s

e I

tel lit S g 1I I

g S

! l! l'I jiLiti oj<^5>^a: !

a 2

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

194

Rob Salmond

intheundercontrol models andfrom 0,044to0.069 intheovercontrol models. Taken together, this evidence shows thatthe choice of does notdrive constraint levelin thedependent variable calculations thesubstantive results. Whatis notyetclear from Table 1 is howmuchtheelectoral into 2 matter. Table transforms these coefficients systemmight in thepercentage estimated shifts of seatsheld by womenundera from theMatland number ofdifferent scenarios. estimates Comparable and and Malami articles are also (1998) (1999) Kenworthy provided. The estimates than thosefrom either aregenerally moreconservative theMatland or Kenworthy and Malamiarticles. of thedifferent modelsto Table 2 also showsthesensitivities in PR's advanthe mechanism which much of DM, changes through become to women is to This article's estimates tage supposed operate. on theother as DM rises.Matland'sestimate, monotonically larger in his in DM; all that matters is entirely insensitive to changes hand, is sharedwith estimation is thata system uses PR. This property and Malami's McAllister and Studlar's (2002) estimates. Kenworthy with those of and McAllister Studlar estimates, (2002), suggest along DM and representaa non-monotonicity in therelationship between that a shift from SMD to low-DM listPR tionlevels;bothestimate thana shift to a high-DMmixedsystem. wouldbe morebeneficial than Theseestimates aremore driven instruments bythe bythecoding dataandrun of DM women's counter tothesupposed on repreimpact sentation. With better dataandbetter measures oftheelectoral system andculture, this that ofPR's effect article estimates suggests previous on thenumber offemale to some overwere, extent, parliamentarians blown.Note also thatthetheory-driven interval confidence (CI) in this article is generally no wider than thestatistically derived 95% CIs from theother about culture twoarticles. Thetwoextreme assumptions do notproduce a CI that includes absolutely everything. theEstimates on theRoad: SomeNatural Taking Experiments It nowremains intothefield for us to takethevarious estimates and look at theability of each estimate to predict changein cases of electoral reform within a singlecountry. inlevelsofrepresentation Table3 presents thechange following fiverecent within cases ofmajorelectoral reform from thesample.15 Francechangedfrom a double-ballot to PR in system majoritarian time for the1986 election andbackagainintime for1988.Italy shifted andin from a listPR system to a mixedsystem for the1994election,

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Representation Proportional
-S 3 /-s

195
U
;o

I
S

s I -8
^ ^, as

o "3 s

~.

06

"

^ -

'I

^
JJ C ^ OO

jgri
W c

od

<N

vo

<L>

1_
sp

vp

no

sO

^_
^p

|a

So

^t

Ti

VO

(N

CO

<N

<N

Op

Op

Sw
q> g

|,

^s

il

!|

fe-o S

S^

^2 8- *i l^ "" ?

awin ill S
5
*3

? ? ?1
^
'
\P

'

?
*
+
N

w> S

^r +

W S.1 D m +
^^9
"^OO-J 2nJQQ+

^
'
N

^*
+
NU

reT!>-''ONP

**\
e^i Q
C ^

^ W
T

vo I

i
O +

^^ (N't*
+

S>|Di

vq

<N

'

CO

On

"^

+ s
^-o
g *n /- s so

U
-6 D--

II

C 0> g

N o^ w-j

N? 0sen

?
00 00 On -

?
<N On On -

N? o^ O

N o^ O)

sO ^s. ON

.VO eoOO .WON >.-

^O On On -

VO^U O^ On ^ On *"*O3H

2s s
cd

I f&

i I

, 2

I r i M I $e

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

196

Rob Salmond

1996both New ZealandandJapan a majoritarian movedfrom system to a mixedPR system. Boththerawchangeinrepresentation andthe to the electoralsystem are listedin Table 3. I changeattributable calculated theattributable an S-curve changefigures by estimating modelsimilar to theone in Figure1 on each country's levelofrepresentation inFrance's data(prior to either reform usingthepre-reform an of what estimate the level of case). Thismethod gives representationlikely wouldhave been sans electoral reform, modeling growth due to cultural and partisan factors. The difference betweenthis is predicted changeandtheactualchangein women'srepresentation taken tobe thechange the attributable totheelectoral Thus, system.16 estimated levelofrepresentation sansreform relieson representation - in general, overthewholepre-reform 10 to 15 elecpatterns period tions. Whilethismethod not the most measuremay provide precise ment itis goodenough that follows. to makethepoint instrument, None of themodelson offer in Table 3 come away smelling of roses.All theconfidence failto makea correct intervals entirely more than half thetime. inthisarticle ButtheCI advanced prediction does a goodjob of explaining thechange(boththerawand attributable change,in fact)in New Zealand,and it comes within1% of inrepresentation theshift andinthesecond levelsinJapan predicting In every French andwithin reform. case French 3% ofthefirst reform, butone,theCI generated a better of this does predicting by analysis job infemale than orKenworthy either theMatland changes representation andMalamiCIs. Theoneexception which is the NewZealandcase,for and Malamialso correctly the amount of Kenworthy change. predict inthisarticle inboth themodeladvanced makeserrors direcFurther, at times the and at other times tions, overestimating underestimating since oftheelectoral Thisdisparity is to be expected, impact system. themodels'estimates are of theaverageeffect acrossall countries whilethey case. arebeingtested actualchangein a particular against in the less creditably Estimates from earlierstudiesperform TheKenworthy andMalamimodeldoesa good experimental settings. theresult inNew Zealandandcomeswithin 1.3% of job ofpredicting thechange inJapan, 3.5 standard butitdoes notgetwithin predicting of predicting Matland'sestimate does deviations theother changes. notgetwithin theimpact ofany 3.5 standard deviations ofpredicting of thewithin-country shifts. On theeightoccasionsthattheearlier the models failtopredict thelevelofchange, their errors overestimate Thispattern is further evidence electoral time. system's impact every that advanced here thelower estimates oftheelectoral system's impact aremoreaccurate than earlier estimates.17

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Proportional Representation IV. Conclusion

197

to theliterature has madefour contributions on the Thisarticle and ithas linkbetween electoral First, egalitarianism. systems gender PR and levels of that theargument female shown linking higher reprebutalso that theconceptual and sentation correct, is, in broadterms, in earlier studies led those studies to present methodological problems Matland's(1998) the impactof the electoralsystem. overestimate PR causes women'srepresentation to be 15.6% higher that estimate wouldbe bearslittle resemblance to theexperiences than itotherwise that have made majorchangesto their of anyof thefourcountries and Malami faresonlya rules.Researchfrom electoral Kenworthy inthis makesmorearticle Themodeladvanced little better. generally models and than either of the earlier does not accurate predictions in one direction. makeitserrors consistently addressed thequestion ofchange this article hasdirectly Second, and itsrelationship to electoral overtimein women'srepresentation asksof thepolicycommunity Thisis thequestion generally systems. in this and of the research most thescientific previous community, itdirectly. areahas notbeenable to address a significant the datasetused in thisarticlerepresents Third, are based on Thisarticle's estimates efforts. advanceon mostearlier than research the more observations times ten Norris, Matland, byRule, and Studlar studies and Malami.The twoMcAllister and Kenworthy their overcount datatothis that use similar N, causing severely study's to be overstated. results their usedtostudy the research methods this article hasupdated Fourth, in It treats women's a number of fronts. on this growth repreproblem an intuitively idea butone that had as nonlinear, sentation appealing for electoral Itusesa continuous measure beentried before. not systems, howtheelectoral theories about that fits wellwith an approach system now.This measure allows tested before matter butone rarely might kind of electoral ofalmost estimates tobe madeabouttheimpact any measures couldnotdo. reform, something previous whichconcerns The last of the methodological innovations, unmeasurable for almost for statistical concepts techniques controlling on womenand outsidethe literature has application like culture, be directly observed ofpolitical culture cannot Someaspects politics. is tomeaThe solution tothis to not ina reliable fashion. problem try at all. Instead of one culture surepolitical directly making extremely what culture should about unreliable is,researchers assumption political a each to the maketworeliably extreme assumptions, counterweight

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

198

Rob Salmond

itprovides a very robust other. The advantage ofthisapproach is that CI that can still be narrow This to be useful. requires enough approach little datacollection thedependent which is especially variable, beyond world useful when into toward ofthe anydistance history, parts looking that do not where datasources areunreliable, or at aspectsofculture havedirect andobservable effects. of itsown, The article, of course, is notwithout imperfections It wouldbe nice ifa method and somehavebeenraisedin thetext. on the couldbe devisedto letthedatadecidewhattheendconstraint level of representation will be, rather thanhavingit imposed by the researcher.18 It wouldbe especially nice ifthedataweresymmetric, with in the40%-50% range as in the0%-10% as many observations if this research this scenario ever occurs, question range although and willprobably At present I cannot thesethings, be moot. provide therefore are left thosechallenges for others. Rob Salmondis a Ph.D. candidatein politicalscience,4289 CA Bunche Los Angeles, Hall, University ofCalifornia-Los Angeles, 90095<rsalmond@ucla.edu> anda Pre-Doctoral Fellow, Department Ann Political 6634 Haven Science, Hall, University of Michigan, of MI 48109-1045. Arbor,

NOTES
Clare I thank StanMarsh, Kathleen Bawn,Jeffrey Lewis,MichaelLewis-Beck, staff andanonymous MikeThies, Jana vonStein, andtheeditorial reviewers, Salmond, thisarticle's all of whomprovided and suggestions during veryusefulcomments I also thank theInterparliamentary Union,Duane Swank,andArend development. version ofthis for thisarticle is based.A previous thedataon which Lijphart sharing PoliticalScience article was presented at the2003 annualmeeting of theAmerican in Philadelphia, Association Pennsylvania. 1. Femalelaborforce a good example.Raw participaparticipation provides oratwhatlevels. tionlevelstellus onlythat women areworking, notinwhatsectors with women Countries ortourism industries employed mayhavemany largeservice with inthelowerechelons an otherwise ofthosesectors, whereas country equivalent because women as "employed," a large sector notcount agricultural may many working levels conclude from suchparticipation a family farm. Can we really they helptorun thesecond?Recogof female leaders than that thefirst is morepermissive country ofwomen's haveincluded a measure thisdifficulty, someresearchers particinizing this and Malami 1999). Although (Kenworthy pationin professional occupations the it suffers from on the is an measure, previous adjustment certainly improvement culture a nation's same problem. Researchers need to knowto whatextent accepts theproposition thatwomencan lead just as well as mencan, notsimply perform

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Proportional Representation

199

or accountancy tasksas well. The statistics management onlydeal withthelatter. Educational attainment statistics suffer from similar deficiencies. 2. It is also unfortunate that McAllister and Studlar's tablesdo not regression allowthereader to construct a 95% confidence interval around their estimates point for anyvariable. 3. To understand is inappropriate, consider theBritish case whythis technique in 1980 and 1981.Members wereelectedto theBritish in 1979,andthe parliament electionwas notheld until1983. Thus,theeventthatdetermined the subsequent in 1980 was the 1979 election, variable and thesameeventalso entirely dependent in 1981.The observations thelevelofrepresentation in Britain determined for1980 and 1981 aretwoobservations of thesamecase, nottwoseparate cases. McAllister treat and Studlar, theseas twoseparate however, (and,by thestandard assumptions cases. The averageparliamentary term of theOLS model,independent) acrosstheir thiscounting leads to an inflation in Therefore, years. strategy sampleis aboutfour inturn ofcases bya factor offour, to artificially smallstandard their number leading in theresults. errors and substantial overconfidence 4. In addition, Matland(1998) also examined16 developing countries and was theelectoral not for women's found that system important representation. 5. The period1988-98saw representation whereas growby4.3% in Sweden, In Finland, 12 yearshad seen 15.5% growth. theprevious representation marginally in theperiod1991-99,after declined growing by 13% overtheprevious12 years. theperiodsince 1990 has seen representation And in Denmark, grow5%, whereas inNorway hasremained almost theperiod1977-90saw 13% growth. Representation 20 years. static sincethelate 1980s,after rising byabout25% intheprevious Reprealso after a 20-year sentation levelsinNew Zealandhavesimilarly stagnated, period to notethat thePR countries do appearto have It is interesting ofexplosive growth. New Zealand(which is nowPR levelthan either stabilized at a higher representation to be seenwhether butwas previously SMD) or Canada(whichis SMD). It remains PR countries. will stabilize at lowerlevelsthan or notother SMD countries that an authoritarian interlude after women 6. Forthosecountries experienced I use the date when was first won suffrage (Greece,Portugal, Spain), suffrage first fora period does becausebothwomenandmenlostpolitical rights Simply granted. women started a position ofpolitical not mean redemocratization, that, againfrom upon to men. compared powerlessness variable thenumber of seatswon by thedependent 7. I calculated by taking inthecountry. ofelectoral districts that valuebythenumber anddividing eachparty in theregression, becausethere is good reasonto term I did notincludea constant areno seatson offer, andI usedrobust PM equals0 whenthere a party's believethat in thedata,buttheresult is notdepento combat standard errors heteroskedasticity in decisions. Some of theresidual variance of thesemethodological denton either in somehigh-DMPR counforrepresentation PM is likely due to legal thresholds is 127. The N in theregression tries. with a DM of6, and is elected at a lower tier 8. Thus,if75% oftheparliament with a DM of 100,then theoverall at a higher, national tier theother 25% areelected theidea that itis possible is codedas 29.5. Thismethod DM ofthesystem captures it listandstillgetelectedinthisform ofmixedsystem than to be loweron a party's PR system with a DM ofsix,butprospective candidates must is ina simple generally

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

200

Rob Salmond

be placedhigher thanin systems witha 100-seat One might national district. argue that thiscounting is unfair to MM systems andto MMP inparticular. People strategy in SMDs inMMP systems elected andNew Zealandareoften likethoseinGermany on their thePR tier theparty listbefore thePR seatsare listfor andcomeoff party's then once allocated. If theSMD winners are all menbutthelistis gender balanced, are all themencomeoff whenthelistallocations theparty listit looksveryfemale in a gender-balanced the(PR tier)listcompoTherefore done,resulting parliament. and theDM as it sition is thekeydeterminant of gender balancein theparliament, of of DM in thecontext relates to thenational listmight be theappropriate measure with this different thisresearch To I reran test this the question. theory, regressions theMM systems' DM. Therewas no changeinthestatistical significance codingfor or substantive of thevariable. importance 9. Another is toemploy a fixed-effects for culture goodwaytoproxy political which measures each country's culture model, using entirely independently political howdummies. Thatmethod is notappropriate forthisresearch country question, becausethere is virtually nowithin-country ontheindependent variable variation ever, in estimates oftheefof interest This approach wouldresult (theelectoral system). from fect ofelectoral ofthem France, systems beingdriven by six maincases,three would be and one each from and New Zealand. Thus,theestimates Italy, Japan, unstable and notgeneralizable. highly with a 10. The regressions incombination use panel-corrected standard errors correction forcommon-pool first-order autocorrelation (AR1), as recommended by Beck and Katz (1995). Beck and Katz (1996) do express somecaution aboutusing I in theregressions. variableis present PCSEs and AR1 whena laggeddependent differandfound retested theaffected modelswithout theAR1 correction negligible and ence in the substantive or statistical of any variable.Kristensen significance can be madeoverthePCSE methWawro to further that (2003) point improvements on inpaneldatawith variables. Theseimprovements rely odology laggeddependent effects a fixed-effects tousingfixed model.Becausethedataherearenotwellsuited PCSEs in thisanalysis. and Wawro themselves (as Kristensen note),I retained correla11. Although thePCSE technique does control forcontemporaneous one I estimated twodummy tionacrosspanels, a version ofthemodelwith variables, after theBeijingdeclaration hadbeenmadeandanother (onlycoded indicating years thestart of the theyearsafter as a 1 forEuropean Unionmember states)indicating Neither of these Union'sinitiatives to getmorewomenintoparliaments. European and the variables of thedependent variable, dummy proveda significant predictor alter estimates. inclusion ofthesevariables did notmaterially anyoftheother to an anonymous reviewer forsuggestions 12. I am verygrateful regarding AR1 corrections. I reestimated models(1) and(2) ofTable 1, 13. Forfurther robustness checks, variableand thenusinga simple first (linear)dependent usingtheuntransformed variableforelectoral the same as thevariableused by Matland dummy systems, I ranthese inplace oftheloggedDM variable. simply (1998) andothers, regressions ofmethodin thisarticle arenotdriven to showthat thegeneral results byanyform which Theresults thesameinthese remain regressions, broadly ologicalskullduggery. mix aspectsof thenew modelwithaspectsof oldermodelson thistopic.These as previous butnotas much also indicate that theelectoral matters, system regressions

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Proportional Representation

201

Fullresults ofthese research has suggested. areavailablefrom theauthor regressions uponrequest. checksperformed hereshouldinspire in the 14. The robustness confidence butthevery bestwayto modelthisdatawithout at all, either results, anyconstraints interms ofmaximum levelsorinterms ofpoints ofsymmetry, wouldbe to adaptthe variables used by Nagler(1994) to workin the scobitmodelforbinary dependent Such a methodological advanceis, however, context ofpaneldataand percentages. thescope ofthisarticle. well beyond in thistest 15. I do not includeanalysisof France's 1958 electoral reform theMatlandnorKenworthy and Malami articles even claimedto because neither levelsin the 1950s. of changein representation explainpatterns hereunderstates 16. One might that themeasurement thetrue argue employed in representation becausepastincreases areassumed oftheelectoral system, impact levels in 1993 had gone In New Zealand,forexample, to continue. representation at that time. theseincreases SMD system thanin anyother Whyassumethat higher of womenstagnated whenCanada's representation at just would have continued, inthe1990s?There theproporaretwoanswers. above20% inan SMD system First, rosein parallelwithparliamentionof womenelectedto New Zealandmayoralties in the 1980sand early1990sand continued to rise,without any tary representation is anybias therest ofthe1990s.Second,ifthere shift totheelectoral through system, itamounts to less than1% in mostof thesenatural in thiscase, then experiments drawnfrom the difference to theconclusions notenoughto makeany substantive analysis. to earlier researchers. 17. Some readers arguethatthistestis unfair might itwanted to answer a different that started After all, thisarticle question by noting scholars havestudied women's Other workon thesubject: askedin earlier that from between electoral environments by lookingat thestaticdifferences representation differences acrossspaceand aimis to lookat dynamic acrossspace,butthisarticle's arenaand research outof itsstatic to taketheearlier is a bitunfair time.It probably what time. Thisis,however, tochanges over estimates the precisely policymakers apply in how in a givenplace are interested do withthisresearch. exactly Policymakers - and untilnow overtime rulesmight changerepresentation changesto electoral that didnotdirectly answer that havehadno choicebuttouse research question they now exists an alternative is to show that there of Table 3 The as their purpose guide. theearlier and thisestimate of theelectoral estimate effect, outperforms system's that concerns in thecontext research policymakers. butthey do that allow suchestimation, somemethods 18. Thereare,in fact, ofwomen ofthisresearch notworkin thecontext project. Usingthelog oftheratio inwhich thedependent variable allowsa regression variable tomenas thedependent a constant as time as timegoes to zeroand approaches infinity negative approaches in of a be the which would That constant, implied intercept infinity. approaches variable, taking1 overall time-based independent usingthisdependent regression is theconstraint variables and leavingother variables untransformed, independent in myformulations of thisregression. Two problems, I have imposed that however, in thisstudy. timeis nowhere near"approaching arisein usingthisregression First, oftheconstant term arevery that estimates Second,the imprecise. infinity," meaning measures of all independent relieson unbiased estimation of theconstant unbiased

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

202

Rob Salmond

variables and thatconditionis not met in the researchdesign used here. I reranthese estimations thespecifications thatundercontrolled forculture anyway,and predictably, gave estimates of the constant of 70-71%, with standard errorsof about 8, while those specificationsovercontrolling for culturegave estimates in the mid-30s, with standarderrorsof about 6. The factthatthe likelyceiling forwomen's representation lies somewherebetween30% and 70% will notcome as news to anybody.The patterns of substantiveand statisticalsignificanceon the electoral systemvariables remained thatI performed unchanged in these regressionspecifications.It should be reiterated robustnesschecks on the regressionsreportedin this articleusing different levels of I theconstraint, and I foundno significant in coefficients. am the verygrateful changes to Kathleen Bawn and an anonymous reviewer for suggestionsregardingtechnique in this area.

REFERENCES
H. 2000. "Why Lagged Dependent Variables Can Suppress the Achen, Christopher ExplanatoryPower of Other IndependentVariables." PolmethWorkingPaper. Katz. 1995. "What To Do (and Not To Do) withTimeBeck, Nathaniel,and Jonathan Series Cross-Section Data." AmericanPolitical Science Review 89: 634-47. Katz. 1996. "Nuisance vs Substance: Specifyingand Beck, Nathaniel, and Jonathan EstimatingTime-Series Cross-Section Models." Political Analysis 6: 1-36. Boix, Carles. 1999. "Settingthe Rules of the Game: The Choice of Electoral Systems in Advanced Democracies." AmericanPolitical Science Review 93: 609-24. in Electoral Politics: Lessons fromNorway. LonJillM. 1995. Women Bystydzienski, don: Praegar. and theElectoralSystem." Castles,FrancisG 198 1. "Female LegislativeRepresentation Politics 1: 21-27. Political Scales: Some 'Expert' Castles, Francis G, and PeterMain 1984. "Left-Right Judgments." European Journal of Political Research 12: 73-88. Cox, Gary. 1997. Making VotesCount: StrategicCoordinationin the WorldsElectoral Press. Systems.Cambridge: Cambridge University Elections, and RepresenDarcy,Robert,Susan Welch, and JanetClark. 1994. Women, tation.New York: Longman. Duverger,Maurice. 1954. Political Parties. New York: Wiley. Duverger,Maurice. 1955. The Political Role of Women.Paris: UNESCO. Griliches,Zvi. 1961. "A Note on Serial CorrelationBias in Estimatesof Distributed Lags." Econometrica 29: 65-73. Hill, Roberta,and Nigel S. Roberts. 1990. "Success, Swing, and Gender: The Perforin New Zealand." Politics25: 62-80. mance of WomenCandidatesforParliament Ronald and Norris. 2003. F., Inglehart, Pippa Rising Tide: Gender Equality and Cultural Change. Cambridge: Cambridge University Press. Jones,Mark P. 2004. "Quota Legislation and the Election of Women: Learning from the Costa Rican Experience." Journal of Politics 66: 1203-23. Lane, and Melissa Malami. 1999. "Inequalityin Political Representation: Kenworthy, A WorldwideComparativeAnalysis." Social Forces 78: 235-68. ofMichigan 2d ed. AnnArbor:University Kmenta,Jan.1997. ElementsofEconometrics, Press.

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

Proportional Representation

203

J.Wawro. Ida Pagter, andGregory 2003. "Lagging theDog? The RobustKristensen, inthePresence Standard Errors ofSerial Correlation and nessofPanel-Corrected Effects." Observation Columbia University. Specific Unpublished manuscript. Vote: A Study Enid.1970.HowDemocracies London: Lakeman, ofElectoral Systems. FaberandFaber. inParliament." andWomen Parliamentarian Enid.1976."Electoral Lakeman, Systems 56: 159-62. and Party Systems. Arend. 1994. Electoral Systems Oxford:Oxford Lijphart, Press. University inRegression for SerialCorrelation Models G.S.,andA.S. Rao. 1973."Tests Maddala, and Correlated Errors." Econometrica with Variables Serially LaggedDependent 47: 761-74. FemaleRepresentation E. 1993."Institutional Variables Richard Matland, Affecting TheCase ofNorway." Journal 55: 737-55. inNational ofPolitics Legislatures: inNational DevelRichard E. 1998."Women's Matland, Legislatures: Representation 23: 109-25. Countries." Studies Quarterly Legislative oped and Developing of Women Candi1996. "The Contagion Richard E., and DonleyStudlar. Matland, Canadaand Districts andProportional datesin Single-Member Representation: Journal Norway." ofPolitics58: 707-33. on Effects Richard E., and MichelleA. Taylor.1997. "ElectoralSystem Matland, Evidence from Costa Rica." Theoretical and Women's Arguments Representation: 30: 186-210. PoliticalStudies Comparative and Women's 2002. "Electoral McAllister, Ian,and DonleyStudlar. RepreSystems 39: 3-14. A Long-Term sentation: Perspective." Representation In LeftTurn:TheNew Zealand 2000. "The New Parliament." McLeay,Elizabeth. Victoria Bostonet al. Wellington: GeneralElectionof 1999, ed. Jonathan Press. University Estimator to Logit and Probit." 1994. "Scobit: An Alternative Nagler,Jonathan. American Journal ofPoliticalScience38: 230-55. in Western Noms, Pippa. 1985. "Women'sLegislative Europe."West Participation Politics 8: 90-101. European TheComparative Position andSexualEquality: Norris, ofWomen Pippa.1987.Politics Books. CO: Wheatsheaf Democracies. in Western Boulder, In Genderand Party Recruitment." Norris, Legislative Pippa. 1993. "Comparing London:Sage Publications. Lovenduski and PippaNorris. ed. Joni Politics, Elections In Comparing Democracies: Recruitment." Norris, Pippa.1996."Legislative G Niemi, and ed. Lawrence in GlobalPerspective, and Voting LeDuc, Richard Publications. Norris. London: Sage Pippa Behavior. RulesandPolitical New Norris, Engineering: Voting Pippa.2004.Electoral Press. York:Cambridge University A CrossinNational 1993."Women andElizabeth Ann, Oakes, Legislatures: Almquist. Theories." Research Gender TestofMacrostructural National Population Policy and Review12: 71-81. in theLegislatures oftheWorld." 1999."Women and Executives Andrew. Reynolds, Politics51: 547-72. World IDEA Handbook eds. 1997. TheInternational and Ben Reilly, Andrew, of Reynolds, IDEA. Electoral Design.Stockholm: System

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

204

Rob Salmond

to theComparative Stein. 1970. Citizens, Parties:Approaches Elections, Rokkan, Oslo: Universitetsforlaget. Study oftheProcessofDevelopment. Factorsin Contextual Rule,Wilma.1981. "WhyWomenDon't Run: The Critical 34: 60-77. Women's Recruitment." Western PoliticalQuarterly Legislative Contextual and Women's Factors, Rule,Wilma.1987."Electoral OpportuSystems, Political inTwenty-Three Democracies." Western for Election toParliament nity 40: 477-98. Quarterly In Rule,Wilma.1994."Parliaments of,by,and forthePeople:ExceptforWomen?" F. Electoral in Comparative ed. WilmaRule andJoseph Systems Perspective, Zimmerman. London:Greenwood. In Electoral Matthew S. 1994."Minorities andUnrepresented." Shugart, Represented inComparative RuleandJoseph F.Zimmerman. ed.Wilma Systems Perspective, London:Greenwood. E. Matland. 1994."TheGrowth ofWomen's andRichard Studlar, RepresentaDonley, and theElection of 1984:A Reaptionin theCanadianHouse of Commons CanadianJournal praisal." ofPoliticalScience27: 53-79. A ComparaMass Exist? and Ian McAllister. 2002. "Does a Critical Studlar, Donley, since 1950."European tiveAnalysis of Women's Legislative Representation Journal 41: 233-53. ofPoliticalResearch InElectoral Rein. 1994. the Attrition." Systems Taagepera, "Beating Law ofMinority London: F.Zimmerman. inComparative ed.Wilma RuleandJoseph Perspective, Greenwood. and 1989. Seats and Votes:TheEffects S. Shugart. Rein,and Matthew Taagepera, Press. Determinants New Haven,CT: Yale University ofElectoral Systems. of Womento Public Office:A Discriminant Welch,Susan. 1978. "Recruitment Western 31: 372-80. Political Analysis." Quarterly

This content downloaded on Fri, 11 Jan 2013 01:25:26 AM All use subject to JSTOR Terms and Conditions

S-ar putea să vă placă și