World Development Vol. 38, No. 1, pp.

37–47, 2010
Ó 2009 Elsevier Ltd. All rights reserved
0305-750X/$ - see front matter
www.elsevier.com/locate/worlddev
doi:10.1016/j.worlddev.2009.06.001

Homogenization and Specialization Effects of International Trade:

Are Cultural Goods Exceptional?
JESSE CHU-SHORE *
Massachusetts Institute of Technology, Cambridge, MA, USA
Summary. — In contrast to the logic that international trade leads to greater specialization and differentiation of products, cultural
industries are often still protected from imports, in part, because of the worry that trade will lead instead to homogenization. Is this true
for cultural goods and if so, is this different from other goods? I consider the effects of homogenization on industrial development, pro-
pose a network-based method of identifying homogenization in global trade patterns, and test a range of industries. I find evidence of
homogenization in many industries, calling into question a major justification for free trade.
Ó 2009 Elsevier Ltd. All rights reserved.

Key words — global, world, international trade, cultural industries, homogenization, specialization

1. INTRODUCTION individual psychic and social life could be unduly shaped by

Free trade, by providing greater markets and greater compe-
tition, is supposed to engender specialization in those indus-
tries or products in which a country has a particular
endowment or competitive advantage. Free trade is thus
thought to lead to a greater variety and higher quality of
goods available to consumers across the globe. Despite this,
there is much debate as to whether cultural goods, such as
movies, books, and music, ought to be given special exemption
from international trade agreements, in part because of the
fear of homogenization.
Since the Uruguay Round of trade talks that led to the cre-
ation of the World Trade Organization (WTO), for example,
‘‘audiovisual” products and services that ‘‘typically reflect
the social and cultural characteristics of nations and their peo-
ples” (WTO Council for Trade in Services, 1998) have been
substantially omitted (even if not officially exempted) from
the standard free trade-oriented commitments made by
WTO member countries. The United Nations Educational,
Scientific and Cultural Organization (UNESCO) has adopted
a Convention on the Protection and Promotion of the Diver-
sity of Cultural Expressions that enshrines in an official docu-
ment the rights of governments to protect domestic culture,
cultural expressions, and significantly, cultural goods indus-
tries. Although this convention has no mechanism for enforce-
ment, it lends legitimacy to the protection of culture from the
influence of foreign trade. Individual countries therefore have
considerably more de facto leeway to practice direct protection
or provide subsidies to cultural industries than they do with
other goods.
This special treatment, known as the ‘‘cultural exception” is
a contentious issue. Countries that are already significant
exporters of cultural goods, such as the United States, have ar-
gued that special treatment is unwarranted. Supporting special
treatment are countries such as France and Canada, which
have expressed concern over the likely loss of their domestic
cultures and cultural products due to the influence of imported
goods.
An important theme in this debate is this concern that inter-
national trade would lead to homogenization of local cultures.
For individual countries, the meanings, expressions, and val-
ues that are conveyed through cultural goods and that shape
37
imports, resulting in a loss of local culture. From an economic
perspective, influence from imported goods could threaten the
unique endowment that a distinct local culture provides to lo-
cal cultural industries that could be developed and exported
for economic gain. The UNESCO convention also appeals
to the sense that loss of cultural diversity would be a serious
concern at the global scale, as well as for individual countries.
This argument—that trade leads to homogenization in cul-
tural industries—appears to contradict the most basic logic
of international trade in other industries, that trade engenders
specialization. Specialization should support the production of
local cultural goods – products in which a given country has a
comparative advantage – at least as a first approximation
based on the standard logic of trade.
My aims in this paper are twofold. First, I wish to evaluate
this notion of the homogenizing effects of trade in the cultural
industries. Cultural industries are not only socially important,
but represent real economic development opportunity, and it
is essential to get trade and industrial policy right. In 1998,
for example, global sales in recorded music were worth $38.6
billion, and world trade in cultural products in general was
estimated at $388 billion, an impressive sum, even taking def-
initional and measurement difficulties into account (UNE-
SCO, 2006). International trade in cultural goods is
important to developed countries that have thriving domestic
industries, but it is likely to be even more important to devel-
oping countries, for which cultural industries represent a po-
tential leading export sector (e.g., Bourne & Allgrove, 1996;
Kozul-Wright & Stanbury, 1998; Pratt, 2004; Scott, 2004).
Many developing countries that have weak manufacturing sec-
tors and little prospect of becoming competitive in other rap-
idly growing industrial sectors such as high tech goods
* Sincere thanks are due to Alice H. Amsden, Michael J. Piore, Frank Levy,
Pol Antras, and Christopher Wheat for many helpful comments on earlier
drafts of this paper. Many thanks also to Tom Snijders for his help with
implementation of the statistical method. Special thanks to four
anonymous reviewers for substantial insights and suggestions which greatly
improved the paper. All errors and inaccuracies are my own. This research
was completed while the author was on a National Science Foundation
Graduate Research Fellowship. Final revision accepted: June 8, 2009.
38 WORLD DEVELOPMENT
nonetheless have thriving musical and literary cultures, with
abundant human capital and productive ‘‘technology.” It
would therefore be a lost opportunity of immense proportions
to suppress rather than support nascent cultural industries in
developing countries because of wrong-headed trade policy.
My second aim is to evaluate this notion that the cultural
industries are exceptional. If trade has a homogenizing effect
on cultural goods, it may have a homogenizing effect on other
goods as well. Specialization implies that everyone gains from
trade. A finding of homogenization would not have such an
implication and raises some serious concerns about the effects
of trade on the development of industry in developing coun-
tries. If trade has a homogenizing effect on domestic cultural
industries, there are strong arguments to be made for support
or protection of domestic producers and products. The welfare
of the world’s poorest has been entrusted to this notion of spe-
cialization; if there is another side of the effects of trade, it
must be understood so that it can be brought into policy deci-
sions to the benefit of all.
Although the approach of this paper involves an analysis of
the pattern of trade, it is less a trade study than an inquiry into
the premises upon which arguments about development are
built. I am not seeking to understand the determinants of
who exports which products or how much is exported, but
rather an aspect of the complementary issue of the effects of
trade on domestic industries. Finally, although the questions
addressed in this paper are inspired by aspects of the ‘‘cultural
exception,” this paper is not directly concerned with the full
scope of that debate. Rather, I take it as a starting point for
an investigation into the possible homogenizing effects of trade
on industrial development more generally.
The plan of the rest of the paper is as follows. In Section 2 I
review literature on and related to the idea of a homogenizing
effect of trade, and how it might affect the development of
domestic industries. In Section 3, I propose a method of distin-
guishing between homogenization and specialization effects by
looking at the overall pattern of world trade in individual
goods as networks and comparing their characteristic topolo-
gies. I argue that certain topological characteristics in cross
section are likely to have resulted from a process of homoge-
nization, while others are likely to have resulted from a pro-
cess of specialization. I evaluate two cultural goods and two
uniform commodities and compare the results, finding that
the two cultural goods show strong evidence of homogeniza-
tion effects from trade, while the two uniform commodities
show strong evidence of specialization. I go on to test a range
of manufactures that are neither uniform commodities nor
goods that are exempted from trade agreements on ‘‘cultural”
grounds. The results from analyzing trade patterns for these
goods describe a spectrum between the extremes represented
by the two cultural goods and the two undifferentiated com-
modities, with most showing significant evidence of trade lead-
ing to homogenization. Section 4 examines alternative
explanations for the patterns of trade found in Section 3. Fi-
nally, Section 5 discusses the implications for policy and fur-
ther research.
2. DOES TRADE LEAD TO HOMOGENIZATION?
(a) Homogenization in the cultural industries
It has been observed that in cultural industries, such as tele-
vision (Lee, 1980) and feature films (Fu, 2006), exports dispro-
portionately originate in a few countries. And in the film
industry at least, there is a trend toward increasing concentra-
tion. Fu (2006) shows that in the feature film industry, the
United States and the United Kingdom have been increasing
their shares of international trade in movies, while other coun-
tries have been steadily declining. Inquiries into such global
market dominance could be classified into three broad catego-
ries: imperialistic, purely economic, and diffusion based.
Explanations based on media imperialism (e.g., Mattelart &
Polan, 1978; Schiller, 1976) emphasize the active role that eco-
nomic actors in developed countries play in creating and main-
taining international trade arrangements that benefit
themselves at the expense of those in developing countries.
American firms have lobbied for media infrastructure and reg-
ulation in trade partner countries that benefits or is compatible
with the type of output that they produce, for example (Lee,
1980). In these accounts, homogenization is a negative and
intrinsic consequence of international hegemony.
Purely economic explanations (e.g., Dupagne & Waterman,
1998; Lee & Waterman, 2007; Wildman & Siwek, 1988) are
based on the size of an exporter’s home market. A large home
market is an advantage for exporters of cultural goods in part
because of economies of scale in distribution of these goods
and services. For example, in the production of feature films,
most of the cost of production occurs up front in the process
of creating the original cut of the film: actor’s and crew mem-
ber’s wages, equipment, location rentals, editing, scoring, mas-
tering, and so on. Once the original has been produced, unit
reproduction costs (i.e., the cost of creating a playable copy
from that original) are very low. The more sunk costs of pro-
duction can be spread over many units, the cheaper each unit
can be sold or leased for. Furthermore, the larger the home
market, the more specific steps in the production process can
be carried out by specialized external contractors, increasing
the overall efficiency of the production process. Therefore,
the prices of cultural goods originating in large home markets
are likely to be lower than the prices of goods from smaller
markets. This implies an international division of labor in
which, ceteris paribus, the countries with the largest home mar-
kets (roughly the large, rich countries) specialize in the pro-
duction and export of cultural goods.
The role of cultural homogenization is more central in diffu-
sion-based accounts of trade in cultural goods. In this tradi-
tion, the influence of imported goods on local culture is
assumed, but the authors emphasize the agency and indepen-
dence of consumers in importing countries. People select,
adapt, and give new contexts and meanings to the cultural
goods they import. The foreign culture does not necessarily
overwhelm or displace the local culture as in the imperialistic
or home-market based accounts, but rather enriches and in-
forms it, as homogenization co-exists with a process of differ-
entiation. These authors (e.g., Falkenheim, 2000; Pool, 1977),
see the emergence of non-US subcenters of television produc-
tion and export (Telenovelas from Mexico, for example) as
evidence supporting this perspective.
(b) Homogenization and innovation
Moving beyond accounts of trade in cultural goods, the lit-
erature on innovation more generally is consistent with a pro-
cess of homogenization due to trade. Most product
innovations occur as incremental changes to an existing set
of knowledge. This is to say that innovations in new products
depend on older products, or that technological change (or
change in the characteristics of goods) is path dependent
(e.g., Cohen & Levinthal, 1990). If imports serve as reference
points for innovation, it would imply a convergence of domes-
tic and global technological trajectories.
 HOMOGENIZATION AND SPECIALIZATION EFFECTS OF INTERNATIONAL TRADE
Likewise for consumers, it could be said that taste for cur-
rent products is shaped in large part by what products each
consumer has been exposed to in the past. The diffusion of
innovations literature shows that the more similar new prod-
ucts are to what is already known, the more rapidly and exten-
sively they are adopted in the aggregate (Di Benedetto,
Calantone, & Zhang, 2003; Haveman, 1993; Rogers, 1995,
and references therein). Consumers who have common expe-
riences in consumption of a given type of industrial product
therefore have a common set of reference points to which fu-
ture products can be related. Each individual’s positive or neg-
ative impression of any given reference point is of course a
personal matter and not necessarily illuminating for analysis
at the social level. The sharing of common reference points
within an aggregate population of consumers is enough to sug-
gest that new products related to those reference points have
an increased probability of diffusing widely through that pop-
ulation, that is, that there is a large potential market for such
related goods.
Trade between two countries in a given industry generates
common experience in relation to which new products are cre-
ated and consumed. This is my interpretation of the finding of
the homogenizing effect of trade. Homogenization does not
mean that exactly the same products are produced in multiple
places, but rather that the product reference points that shape
future rounds of production converge over time for countries
engaged in trade with each other. Homogenization also im-
plies that not only does the extent of the market affect the pos-
sibilities for trade, but trade itself affects the extent of the
market.
Specialization, on the other hand, could only meaningfully
be said to occur among countries that already share product
reference points to some degree, in other words, among mar-
kets with consistent definitions of what is demanded. Special-
ization implies that given a set of reference points, the
products of one country relate to those reference points in a
particular way that is of value to domestic and foreign con-
sumers and is different from products exported by other coun-
tries. Specialization is sustainable when the specializing
country has some particular advantage in producing that vari-
ety of product.
(c) Is homogenization good or bad?
In effect, the idea of free trade assumes a homogenous world
in which the potential export market for any given good is the
set of countries with consumers able to afford that good. In
this assumption, there is no theoretical place for a process of
homogenization. However, the results of this study provide
evidence of such a process, which must therefore be accounted
for in industrial policy. For industries such as books and mu-
sic recordings, there are obvious arguments to be made about
why countries may wish to protect local industries by limiting
the imports of foreign products. However, even for industries
which are not obviously ‘‘cultural” there are nonetheless rea-
sons to think that homogenization may not necessarily be
good.
There have been many challenges to the idea that free trade
is universally beneficial. The traditions of dependency and
world systems theory argue that trade forces industries in
developing countries to assume peripheral and dependent roles
relative to the markets of developed countries. (Amin, 1974;
Wallerstein, 1975). This peripheral role strictly limits the de-
gree to which such countries can capture the gains from trade.
An alternative view, suggests that the integration of domestic
industries with the demands of foreign markets was essential
39
to the export led-growth of the East Asian Economies (e.g.,
Amsden, 1979; Amsden, 2001). Among the major themes of
this literature is that technological knowledge is not easily
gained, even if consumers can adapt and come to like new
products easily. If trade changes what is demanded by con-
sumers faster than domestic industry can accumulate new pro-
ductive knowledge, then trade in new products puts foreign
producers (who already produce these products) ahead of
the producers in the importing country. Therefore, developing
countries would be wise to implement policies to protect un-
ique goods or capabilities in the context of foreign trade.
There is also the possibility that homogenization could pre-
vent countries from developing their own, unique products
which they could go on to export for economic gain, without
having to play industrial catch-up. Differences between the
products of one nation and those of another could occur by
specializing in relation to shared reference points, or it could
occur due to a difference in reference points in the first place.
Homogenization implies that the latter type of differentiation
would not occur, and thus would prevent one way in which
a country could develop a novel endowment in a given indus-
try.
Admittedly for many industries, another word for homoge-
nization could be ‘‘standardization.” One could also think of
homogenization as an analog to standardization in the domain
of tacit, rather than explicit knowledge. The impact of a for-
eign standard on domestic development is complex. On the
one hand, the arrival of a foreign standard requires the domes-
tic industry to catch up with the technological capabilities of
the country which exports the goods to which the standard ap-
plies. Investments in technologies which do not conform to the
new standard may be rendered valueless in the transition
(Farrell & Saloner, 1986; Katz & Shapiro, 1992). However,
the flip side of the necessity to catch up and keep pace with
foreign standards is the increased opportunity for technological
learning from abroad. Standards can allow for easier interna-
tional technological transfer and partnerships. The impact of
foreign standards on developing countries depends therefore
on the competitive assets of domestic firms relative to the glo-
bal market (Ratanawaraha, 2006). While explicit standards
can be negotiated with foreign actors by firms and govern-
ments to maximize developmental benefits (learning) and min-
imize costs (loss of non-conforming sunk investments), tacit
homogenization of technology and products is more difficult.
The implication is that for countries with significant techno-
logical assets, for example, in cultural or other consumer
goods, homogenization could put domestic industries into an
inferior position relative to global competitors.
At the global scale, standardization could have a beneficial
effect on economic growth if it allows innovations in different
countries to be complementary to each other, rather than
redundant or mutually irrelevant. However, standardization
taken to the extreme could theoretically lead to technological,
organizational, or institutional ‘‘lock-in,” in which innovation
is limited by overinvestment in existing standards and/or
homogeneity of perspectives and experience on the part of
innovators and firms (Olson, 1982). Limiting innovation
would almost certainly limit possibilities for future economic
growth. This is to say that whether or not standardization
or homogenization is good for any particular country or set
of countries, it could be a negative for all countries if it pro-
gresses too far at the global scale.
A process of homogenization implies a tension between po-
sitive and negative economic consequences: homogenization
could prevent a country from developing unique technology
or products, and yet could open up potential export markets
40 WORLD DEVELOPMENT
by aligning domestic products with what is demanded in for-
eign markets. As these consequences rely on technological
change over time, they would invite dynamic, rather than sta-
tic policy responses. In existing accounts of trade, however,
homogenization is either assumed or assumed away. The sta-
tus of this assumption dictates to a large extent the type of pol-
icy conclusions that can be drawn from the research. To my
knowledge, the literature lacks a broad-based empirical test
of this critical assumption.
3. METHODS, DATA, AND TESTS
(a) Testing for homogenization and specialization effects
(i) Testing for homogenization effects
To test the hypothesis that trade has a homogenizing effect,
it must first be made operational. To that end, I begin by ask-
ing what we would expect to observe if trade had a homoge-
nizing effect. If imports influence domestic technology and/
or taste, then the result would be that countries that traded
with each other would become more similar in these dimen-
sions for the good being traded.
I assume for now that for a trade tie to exist, what is pro-
duced and exported by the exporting country (roughly, the
technology) must be somehow compatible with what is ex-
pected or demanded by the importing country (the taste). It
cannot be inferred from the pattern of trade ties what is being
produced, of course, so the meaning of ‘‘similar” here is only
that there is some correspondence between product and mar-
ket across the countries in question. Two countries engaged
in bilateral trade are in this broad sense ‘‘similar,” but could
be producing very different goods that are nonetheless com-
patible with each other’s markets. A lack of a trade tie does
not necessarily imply dissimilarity, however, unless other bar-
riers to trade such as geographic distance are accounted for.
Looking beyond bilateral trade to larger patterns suggests a
method of testing for homogenizing effects of trade. Let i, j,
and k represent three arbitrary countries selected from the
set of countries that engages in international trade in a given
good. If i exports to j, then i and j must be somewhat similar.
If j also exports to k, then j and k must be similar. If i is similar
to j, which is similar in turn to k, then i ought to be similar to
k. I therefore expect to observe a greater than random preva-
lence of links (i, k), given the existence of links (i, j) and (j, k).
In other words, I expect to observe transitive triads such as the
left side of Figure 1.
Similarities in other dimensions than technology and taste
could also lead to a finding of transitive triads in trade pat-
terns. In the presence of transportation costs, geographically
similar (i.e., proximate) countries are more likely to trade with
each other and less likely to trade with distant countries. At
the network level, this could result in trade flows among near-
Figure 1. Structures hypothesized to occur at a greater than random rate in
trade in cultural goods (left) and ordinary goods (right).
by countries forming a number of transitive triads. It is there-
fore necessary to control for exogenous sources of similarity
before making inferences about the effects of trade itself.
Taken alone, of course, a finding of a significant transitivity
effect in trade patterns cannot be attributed to similarity in
technology and/or taste, however plausible that seems. A more
credible test would involve a comparison between industries in
which we a priori believe trade to be shaped by similarities and
differences in technology and taste, that is, cultural industries,
against industries in which we firmly believe technology and
taste are irrelevant to trade, such as uniform commodities. If
trade patterns for cultural goods exhibit a significant tendency
to transitivity after controlling for exogenous factors, whereas
uniform commodities do not show this tendency, we can pro-
visionally see evidence of similarity in technology and/or taste
as the factor which leads to this outcome.
I will therefore examine patterns of trade for two cultural
goods, books and sound recordings, and two uniform com-
modities, anthracite and unmilled corn. The testable versions
of the first hypothesis are therefore:
H1a: World trade patterns in books and sound recordings
exhibit a transitivity effect.
H1b: World trade patterns in anthracite and unmilled corn
exhibit a transitivity effect.
I expect to find support for H1a, and not for H1b.
(ii) Testing for specialization effects
Unlike cultural goods, which could be described heuristi-
cally as highly varied and constantly changing, ordinary goods
are well-defined commodities, in which any country might
trade under standard market assumptions. I therefore expect
to see exports of ordinary goods originating in a relatively
small number of countries that specialize in production of
each good because of some sort of advantage or endowment.
In terms of trade network structure, this would be described
by the variable representing ‘‘out-stars” such as the right-hand
side of Figure 1. The estimated parameter for the out-star var-
iable captures the degree to which the origins of export flows
are concentrated in relatively few countries. An insignificant
out-star parameter would indicate that there is no such ten-
dency beyond that explained by other variables.
H1c: World trade patterns in books and sound recordings
exhibit an ‘‘out-star” effect.
H1d: World trade patterns in anthracite and unmilled corn
exhibit an ‘‘out-star” effect.
I expect to find support for H1d, and not for H1c.
(iii) Testing for homogenization in other goods
There were a priori reasons to believe that trade flows of
obviously cultural goods, such as music, would exhibit transi-
tivity. But which other goods would this kind of logic apply
to? So far I have described cultural goods as having technol-
ogy and taste characteristics which vary across countries and
which cannot be evaluated with objective quality measures.
Contrasting with this is the purely utilitarian, readily measur-
able quality characteristics and global market viability of or-
dinary goods such as, say, coal. But between these extremes
there must be goods with a mix of cultural and utilitarian
qualities. Allen Scott, a veteran analyst of cultural industries,
describes an ‘‘unbroken continuum of sectors” from the cul-
tural to the utilitarian (Scott, 2004). I therefore expect to reject
the hypothesis implicit in free trade agreements:
H1e: Other goods do not exhibit transitivity effects.
I test hypothesis H1e in the analyses of the networks defined
by international trade flows in furniture, sound recordings,
sewing machines, cutlery, passenger vehicles, shoes (leather,
 HOMOGENIZATION AND SPECIALIZATION EFFECTS OF INTERNATIONAL TRADE
rubber, or plastic sole), wine, color TVs, butter, unmilled corn,
gas generators, and parts thereof, anthracite, cargo ships
(excluding tankers), and iron and steel bars and ingots.
Examining differentiated goods raises the issue of intra-
industry trade, which is described by the ‘‘new” trade theory
(Krugman, 1980). Both old and new trade theories should pre-
dict a finding of out-stars characterizing network structure. In
old trade theory, countries specialize in producing and export-
ing those goods in which they have a comparative advantage
or endowment, and other countries import them. In new trade
theory, countries might specialize in producing an individual
variety of a good. In this case, a given country might be both
an exporter and an importer of goods in a certain industry.
But each exporting country would likely export its specialized
output to a number of other countries. At the network level,
this would amount to a structural tendency to out-stars as
well, with the only difference being that these exporting coun-
tries also have incoming trade flows.
(b) Exponential random graph model (ERGM) analysis in
StOCNET
(i) The model
The hypothesis tests that I propose are not so much about
the level of trade or even which countries are exporters so
much as the structure of the larger patterns of trade. That
is, the hypotheses describe network-level effects and thus must
be tested with network-based tools, rather than the standard
approach to studying bilateral international trade, which is
the gravity model. The gravity model’s history of empirical
success does provide some guidance in terms of explanatory
variables to include in a network-based investigation, how-
ever. The standard version of the model relates the level of
bilateral trade to the combined economic ‘‘masses” (GDPs)
of a pair of countries, and a term representing ‘‘gravity” by
the inverse of the great circle distances separating them. It is
also customary to include dummies for common language,
colonial past, and contiguity of borders.
Because the observations are hypotheses concern larger
structures than bilateral relationships, candidate models are
essentially limited to those from drawn from the network anal-
ysis literature. Among these, the most appropriate set of tools
in network analysis for testing hypotheses about structural ef-
fects are the exponential random graph models (ERGMs),
also known as p* models. These have been developed in the
context of social network analysis to estimate structural statis-
tics for networks without the independence assumption that is
a principal feature of standard statistical approaches (Frank &
Strauss, 1986; Pattison & Wasserman, 1999; Snijders, 2002;
Snijders, Pattison, Robins, & Handcock, 2006; Wasserman
& Pattison, 1996). Indeed the hypotheses I wish to test are pre-
mised on the idea that individual observations (the presence or
absence of a trade tie) are actually dependent in some way or
ways on certain other observed ties.
This method grew out of the literature on the social networks
of individuals, and most empirical applications have been in re-
lated settings. For example, Thurner and Binder (2008) apply it
to an analysis of the influence of informal networks on formal
networks in the context of international relationships among
political actors. However, its use is indicated in any setting in
which the units of analysis are relationships among actors
(such as the trade flows between countries), when there is rea-
son to believe that the relationships are dependent on each
other. As a statistical method, it has been applied beyond so-
cial networks as well Guo, Fu, and Xing (2007) study networks
of gene expressions in fruit flies, for example.
41
The papers by Anderson, Wasserman, and Crouch (1999)
and Robins, Pattison, Kalish, and Lusher (2007) are good
introductions to this analytic method, and the following over-
view is largely a summary of the salient points from these two
sources that are important for the present context. The basic
premise of the approach is that the observed network (of trade
ties, in this case) is one realization of a stochastic process,
which is to be modeled. The same stochastic process could
have generated a different pattern of ties, but any such in-
stance of the generating process would have structural charac-
teristics in common, even if the exact pattern of ties differed.
This set of possible networks is described as a probability dis-
tribution, with the exact probability of a given configuration
of ties depending on the parameters in the model that de-
scribes the stochastic process. The maximum likelihood ap-
proach to estimating the parameters of the stochastic process
results in a model in which the observed network has the high-
est probability of occurring.
These models consist of a family of probability functions,
P h fY ¼ yg ¼ expðh0 uðyÞ  wðhÞÞ ð1Þ
where y is the observed set of ties defining an adjacency matrix
representing the network being modeled, h is a vector of
parameters, u(y) is a vector of sufficient statistics of the graph
or digraph, and w(h) is a normalizing constant, ensuring that
the probabilities of all possible graph configurations sum to
one (Snijders et al., 2006).
Estimation of structural parameters—those that involve
dependencies between one random variable (one tie between
two nodes, or in terms of the present paper, a trade flow from
one country to another) and the state of other random
variables (trade flows) involves complications unfamiliar in or-
dinary regression models. Early approaches (Frank & Strauss,
1986; Wasserman & Pattison, 1996) were based on a pseudo-
loglikelihood approach. Although models constructed in this
fashion could be practically estimated using any logistic
regression software after preparing the data appropriately,
the statistical properties of the resulting estimator are not well
understood for random graph models, and the value of its
standard errors is questionable (Snijders, 2002).
Snijders (2002) therefore elaborated a Monte Carlo simula-
tion method of arriving at maximum likelihood parameter
estimates, with the corresponding well-understood t-statistics.
Snijders’ method has been implemented in the computer pro-
gram SIENA, which is a component of the network analysis
platform StOCNET (Snijders, Steglich, Schweinberger, &
Huisman, 2007). In addition to the structural parameters, SIE-
NA can accommodate independent variables in the form of
attributes of individual nodes and dyadic covariates.
A technical difficulty with the estimation of ERGMs is that
for many specifications, much of the parameter space can be
degenerate—that is, producing simulated networks that are
either empty or maximally connected. This can prevent con-
vergence of the estimation algorithm. This is a particular prob-
lem in estimating a parameter for transitive triads, which we
are especially interested in.
For example, in a simulated network, if the presence of a gi-
ven tie would lead to the completion of three transitive triads,
then the log-odds of that tie being present increases by three
times the estimated parameter. The presence of this tie further
increases the number of transitive triads that would be com-
pleted by other ties as the simulation process continues, leading
to a large increase in the likelihood of their inclusion in the sim-
ulated network, and so on until all possible ties are predicted.
One tactic is to estimate models conditional on the total
42 WORLD DEVELOPMENT
number of ties in the observed network. This can lead to great-
er success in parameter estimation, but there are still many data
sets for which satisfactory estimates cannot be obtained. There-
fore Snijders et al. (2006) have developed alternative statistics
for estimating these models. The basic idea is that geometri-
cally decreasing weights are put on larger numbers of triads
that would be completed by a given tie. This has the effect of
attenuating the tendency to cascade into a maximally con-
nected network and reduce the degenerate areas of the param-
eter space. Several structural effects are better estimated with
this type of specification. All these are denoted (for reasons ex-
plained in Snijders et al. (2006)) by the prefix ‘‘Alternating.”
(ii) Model parameters
The principal structural parameters of interest are those for
transitivity, denoted by ‘‘alternating k-triangles” and the com-
peting tendency for exports to originate in a few countries, de-
noted by ‘‘alternating out-stars.” Three other effects were
added to the model as structural control variables. The reci-
procity effect captures the increased tendency for i to export
to j conditional on j exporting to i. The ‘‘alternating in-star”
effect represents the tendency for certain countries to import
from a large number of other countries. The ‘‘alternating
two-path” effect represents a pattern of exporting from one
country to a second, and the second to the third. As such it
could be thought of as a ‘‘precondition” to transitivity (Snij-
ders et al., 2007), or a tendency to a directed, hierarchical net-
work structure.
The parameter estimate for reciprocity may seem like an-
other variable of interest for studying homogenizing effects
in trade. However, it depends on reciprocal market demand
and the presence of a developed industry capable of exporting
in both countries. It may even in certain circumstances capture
information about the different status characteristics of both
countries’ exports. In other words, a deeper substantive inter-
pretation of the reciprocity parameter in this context is prob-
ably unwarranted without further study.
Independent covariates included in the model are the prod-
uct of GDPs, inverse of great circle distances between national
capitals, the presence of a common language spoken by great-
er than 9% of both countries’ populations, a present or past
colonial relation, and contiguity of borders. To control for
non-homothetic preferences, I also included effects for GDP
per capita in the exporting and importing countries, but be-
cause they did not substantively affect outcomes, they were
ultimately excluded from the final model.
(c) Data and data preparation
I use reports of the values of exports in the year 2005 from
the United Nations Commodity Trade Database (COM-
TRADE). For these data, the value of an export flow is re-
ported by each exporting country in United States dollars in
free on board (FOB) terms, which includes costs of the ex-
ported goods and the costs incurred in getting them to the
country’s border, but excludes shipping and insurance costs in-
curred in transporting them to the destination country. Re-ex-
ports, which are goods that merely pass through a country’s
jurisdiction without having being produced there, are typically
reported as part of exports. However, because my aim is to
study domestic production and technology, I have subtracted
re-exports from exports wherever they have been reported to
arrive at a figure for net exports, which includes only goods
originally shipped from the origin country. Other authors have
tended to prefer to use imports, rather than exports as the
source of data on trade flows for reasons of accuracy, but this
approach would not allow the subtraction of re-exports, which
I deemed to be the more significant distortion of the data for
the present study. Weight data are not reported by some coun-
tries and were therefore not an option for this analysis.
I define the presence of a directed trade flow dichoto-
mously—a flow is coded with a one if it is present or a zero
if it is absent or extremely small. Trade flows from exporting
countries are added to the network and coded as existing in
the descending order of total value. Starting with the largest
reported flow, trade flows are added to the network until
95% of the global total of reported trade has been represented
in the network. The smallest trade flows, which are large in
number, but together total less than five percent of total re-
ported world trade in that commodity are omitted from the
network. This has the analytical value of omitting the ‘‘long
tail” of the distribution of trade flow magnitudes and focuses
on those flows that are of significant size.
Dichotomous coding strips the dependent variable of some
information. However, for goods that are traded in part in
the form of designs or licenses, rather than physical products,
the reported level of trade may be quite inaccurate in any case.
Dichotomous coding thus allows for easier comparisons
across goods.
To further achieve comparability across various commodi-
ties, I used the same reference set of 156 countries for each
trade network. This has the advantage of defining a popula-
tion of countries under study that may or may not be engaged
in trade in a given commodity. However, for most goods, there
are a large number of countries that do not engage in trade.
Including non-trading countries introduces many zeros into
the data matrix representing the trade network, and results
in much greater computing time for each network under
study—anywhere from a few hours to a few days per network.
Goods for which the top 95% of trade is concentrated in an
especially small number of trade flows could not be analyzed
because the iterative estimation algorithm could not converge
on stable parameter estimates. With a smaller reference group
of countries, such as the industrialized countries, these omitted
goods could be studied in the same manner.
Non-structural independent variables are implemented as
dyadic covariates, with one value for each pair of countries.
For historical reasons, the StOCNET platform only accepts
integer values between 0 and 255 for dyadic covariates. The
original data were thus normalized to a maximum value of
255, and rounded to the nearest integer to accommodate this
limitation. Because the dependent variable is dichotomous,
however, it is extremely unlikely that this transformation af-
fected the analysis in any significant way. Finally, dyadic
covariates are centered by subtracting the mean.
(d) Results
The estimated magnitudes of the parameters for the transi-
tivity effect and the out-star effect are plotted in Figure 2. A
full table of estimated parameters, t-statistics, and effect mag-
nitudes for all model effects is reported in Appendix. Effect
magnitudes represent the increased likelihood of observing a
tie in the presence of a one unit increase in dummy and struc-
tural variables, or a two-standard deviation increase for con-
tinuous explanatory variables.
As expected, there are very large transitivity effects for the
two cultural goods and very small ones for the two commod-
ities. I therefore reject H1b. Conversely, there are very large
out-star effects for the two commodities and very small ones
for the two cultural industries. I therefore reject H1c. For
the range of other goods, we see a continuum between the high
 HOMOGENIZATION AND SPECIALIZATION EFFECTS OF INTERNATIONAL TRADE
Figure 2. Trade network Structure, various commodities.
specialization uniform commodities and the high-transitivity
cultural goods. In other words, although trade policy treats
manufactures in the same way as uniform commodities, their
patterns of trade are not necessarily similar. I reject H1e and
find evidence a mismatch between policy and empirical reality
for trade in a wide range of ordinary goods: trade leads to
homogenization.
4. IS TRANSITIVITY ENDOGENOUS TO THE TRADE
NETWORK?
(a) Hypotheses
To support the idea that trade leads to homogenization of
technology and taste, it remains to be shown that the transitiv-
ity effect of section II is endogenous to the network and not
the result of some other process. In other words, we must dis-
tinguish between the following hypotheses:
Exogenous hypothesis:
H2a: the observed tendency to transitivity is a result of sim-
ilarities between countries that are exogenous to the process of
trade.
Endogenous hypothesis:
H2b: the observed tendency to transitivity is a result of sim-
ilarities between countries that are caused by trade itself.
In this section, I provide several exhibits that support an
endogenous explanation of transitivity in trade networks.
(b) Specification test
The variables in the original model which derive from the
standard specification for gravity models capture directly or
are proxies for the most obvious explanations for exogenous
similarities that could produce transitive trade patterns. The
variables for colonial ties and common language capture signif-
icant historical sources of cultural similarities, as well as the
resultant present day ease or frequency of international ex-
change. The gravity and contiguity variables capture spatial
similarity, and the product of GDPs captures, to an extent, eco-
nomic similarity. It is conceivable, therefore, that the transitiv-
ity parameter is spuriously picking up transitivity that should
have been explained by the exogenous explanatory variables.
I re-ran the estimation process with the transitivity parameter
fixed to zero and subsequently compared the resulting model
to the observed network to evaluate if this was the case.
Table 1 reports the result of modeling trade in music with-
out the alternating k-triangles parameter. Observed values
are compared with a distribution of simulated networks for
43
each model. Although the other variables do produce a con-
siderable incidental transitivity effect, it is much lower than
the actual trade patterns exhibit.
(c) Controlling for omitted exogenous sources of similarity
Having found that there is more transitivity in patterns of
trade than can be accounted for by the independent variables
in the model, it nonetheless remains possible that this excess
transitivity could be due to some omitted exogenous explana-
tory variable. Although it is impractical and inadvisable to test
a large number of explanatory variables for which we have no
prior reason to include in the model, we can still seek evidence
of omitted variables in general. If there is some important exog-
enous source of similarity in technology and taste, then it ought
to affect trade in more than one narrowly defined industry. If
this is the case, there should be correspondence between spe-
cific ties that form transitive triads in multiple industries. If
there is no such correspondence, then we must conclude that
either each specific has a unique, unobserved, exogenous
source of similarity (not a very plausible story), or that the
source of similarity is endogenous to the process of trade.
Although questionable as a true model of trade, we could use
one trade network as a control variable in a model of another
trade network. In practice, this means including an indepen-
dent dyadic covariate matrix, coded as ‘‘1” for all ties present
in the ‘‘control” network and ‘‘0” otherwise. For example, in
an analysis of trade in sound recordings, we could include
the trade network defined by the sum of trade in all goods in
the COMTRADE database as a test of dimensions of similarity
not captured by the control variables already in the model. We
could also posit that an unobserved exogenous socio-cultural
similarity between countries is more likely to affect trade in cul-
tural goods than in other goods. To test this possibility, we
could include the trade network of one cultural good as a con-
trol variable to proxy for such an unobserved similarity in the
analysis of another. Returning to music, we could include trade
in books as such a control variable (Table 2).
In both cases, the parameter for transitivity is moderately
lower than that in the original model, more so for the imple-
mentation of books as a proxy for exogenous difference than
for the use of the total trade network. This suggests that there
is some unobserved factor which is correlated with the transi-
tivity parameter. Still, even after controlling for unobserved
variables, the magnitude of the transitivity effect is statistically
significant and very high compared to, say, the original esti-
mates for coal, corn, or cargo ships. I therefore maintain a
high level of suspicion that transitivity is in large part due to
an endogenous process of trade itself.
(d) An exceptional case: antiques
Intuitively, I would expect antiques to fall on the cultural
goods side of the spectrum, because they are extremely differ-
entiated and their value does not stem from universally held
functional criteria. However, they are an interesting test case
because the technology used to produce them could not be
influenced by present-day imports because the producers are
all dead. They are thus unconcerned by changes in fashion
and also immune from subconscious influences. Still, if the
exogenous hypothesis for explaining transitivity in trade pat-
terns were correct, we would expect trade in antiques to
exhibit transitivity strongly. It does not. In fact, this is the
only good I analyzed, other than iron and steel puddled bars
and ingots, that had an insignificant estimate for the
transitivity parameter. Interestingly, trade in antiques also
44 WORLD DEVELOPMENT

Table 1. Effect of fixing transitivity parameter to zero

Structural measure Observed data Original model Fixing transitivity parameter to zero
Mean of sim. networks t-stat. for fit Mean of sim. networks t-stat. for fit
Number of mutuals 131 129.16 0.285 131.95 0.129
Alternating Out Stars 893.2 894.45 0.154 891.16 0.277
Alternating In Stars 780.02 779.14 0.073 779.07 0.088
Alternating k-triangles 904.89 905.02 0.009 784.12 4.1568
Alternating 2-paths 3696.57 3677.66 0.123 3703.63 0.0528
Number of ties  gravity 4345.25 4131.23 0.327 4299.48 0.075
Number of ties  colony 46.46 46.57 0.026 46.23 0.044
Number of ties  common language 50.98 49.32 0.098 53.67 0.21
Number of ties  contiguous 57.81 56.09 0.23 57.75 0.009
Number of ties  GDP product 4640.99 4631.85 0.077 4626.67 0.107

Table 2. Using other trade networks to control for omitted exogenous

sources of transitivity on trade in music plained by the dominant perspective that trade leads to
Parameter Estimate S.E. Estimate S.E.
economic gains through specialization, but it is consistent with
the idea that trade leads to homogenization of technology
Trade in books 2.0702 0.1533 and/or taste. For countries which produce distinctive or un-
All trade 2.2321 0.2109 ique goods, but are not already international leaders in export-
Reciprocity 0.7826 0.2178 0.7864 0.2087 ing these goods, this finding supports the idea that policies
A. Out stars 0.9865 0.2023 0.7991 0.2012 must be enacted to counter the homogenizing effect of trade.
A. In stars 0.1081 0.1909 0.431 0.197 What form these policies should take is a matter for future re-
A. k-triangles 1.1718 0.1656 1.3423 0.1723 search. The other major empirical finding of the present study
A. 2-paths 0.002 0.0097 0.011 0.0097 is that patterns of trade for individual industries range dra-
Gravity 0.0166 0.0037 0.022 0.0033 matically from extremes defined by cultural goods on the
Colony 0.1789 0.2849 0.4849 0.244
one hand to uniform commodities on the other hand. The
Language 0.3202 0.1489 0.8725 0.1289
broader implication of these findings is that international
Contiguous 0.0562 0.2605 0.0253 0.2409
trade policy is based on an incomplete theory of the effects
Product of GDPs 0.0867 0.0158 0.0556 0.0133
of trade. In this final section, I discuss the findings and specu-
late on their implications, with the intention of suggesting
exhibits the strongest tendency to out-stars of any good propositions to be examined in future research.
analyzed (Figure 3).
In sum, while it is not possible to conclusively reject the (b) Transitivity is a feature of trade networks
exogenous hypothesis for the origin of transitivity in trade net-
works, I have endeavored to show in this section with these Higher degrees of transitivity characterize those goods that
three tests that the most plausible story is the endogenous one. we a priori consider to be ‘‘cultural.” But every good exam-
ined, except for antiques (a special case) and iron and steel
bars and ingots, exhibited a significant tendency to transitivity.
5. DISCUSSION AND IMPLICATIONS FOR FURTHER The interpretation of this is not so much that cultural goods
RESEARCH are like ordinary goods, but rather that we should systemati-
cally revise our picture of all industries to include effects we
(a) Summary only expected in cultural goods. To take a small example,
the gravity model is the empirical benchmark for explaining
The principal empirical finding of this paper is that transitiv- trade, but it does not account for structures of interdepen-
ity is a feature of international trade patterns. This is not ex- dency within the trade network. Rather, it makes the standard
assumption that all variables with explanatory power are
exogenous to the trade network. The results presented here
suggest that endogenous effects—probably through channels
of technology and taste—may in fact be extremely important
to modeling trade—not only in cultural goods industries, but
in most other industries as well.

(c) The cultural industries as a model

The cultural industries have been considered difficult to ana-

Figure 3. An exceptional case: network structure for trade in antiques.
lyze with traditional economic methods. Perhaps this is be-
cause, with their high semiotic content, they are the furthest
away from the classical origins of economic analysis: uniform,
mass produced goods. Indeed, this paper began by asking if
the cultural industries were an exception to the basic theory
of international trade and finished by using them as an
example for the types of questions that could be asked
more generally. Cultural industries bring to the fore issues of
 HOMOGENIZATION AND SPECIALIZATION EFFECTS OF INTERNATIONAL TRADE
consumption, prior reference points, definitions of, and con-
ventions structuring what is demanded that vary across social
units of analysis. To better understand them it becomes neces-
sary to consider large-scale patterns of human interaction
explicitly, an analytical approach that is likely to serve us well
in studying other phenomena in today’s increasingly con-
nected world. I believe that in this era of knowledge- and
information-based economic development rather than being
considered the most irregular, the cultural industries could in-
stead become a new best example and an alternative pole to
uniform commodities in development studies, which could
illuminate many obscure and complex mysteries of the devel-
opment of economies in general.
(d) Conclusion
I have found strong evidence that trade has a homogenizing
effect for cultural goods and therefore empirical justification
REFERENCES
Anderson, C. J., Wasserman, S., & Crouch, B. (1999). A p* primer: Logit
models for social networks. Social Networks, 21(1), 37–66.
Amin, S. (1974). Accumulation and development: A theoretical model.
Review of African Political Economy, 1(1), 9–26.
Amsden, A. H. (1979). Taiwan’s Economic History: A case of etatisme
and a challenge to dependency theory. Modern China, 5(3), 341–379.
Amsden, A. H. (2001). The rise of the rest: Challenges to the west from late-
industrializing economies. New York: Oxford University Press.
Bourne, C., & Allgrove, S. M. (1996). Prospects for exports of entertain-
ment services for the English-speaking Caribbean: The case of music.
World Bank Informatics and Telecommunications Division, Working
Paper.
Cohen, W. M., & Levinthal, D. A. (1990). Absorptive capacity: A new
perspective on learning and innovation. Administrative Science Quar-
terly, 35(1), 128–152.
Di Benedetto, C., Calantone, R. J., & Zhang, C. (2003). International
technology transfer: Model and exploratory study in the People’s
Republic of China. International Marketing Review, 20(4), 446–462.
Dupagne, Michael, & Waterman, David (1998). Determinants of U.S.
television fiction imports in Western Europe. Journal of Broadcasting
and Electronic Media, 42(2), 208–220.
Falkenheim, B. J. C. (2000). Asymmetries reconfigured: South American
television flows in the 1990s. Canadian Journal of Communication,
25(2).
Farrell, J., & Saloner, G. (1986). Installed base and compatibility:
Innovation, product preannouncements, and predation. The American
Economic Review, 76(5), 940–955.
Frank, O., & Strauss, D. (1986). Markov graphs. Journal of the American
Statistical Association, 81(395), 832–842.
Fu, W. W. (2006). Concentration and homogenization of international
movie sources: Examining foreign film import profiles. Journal of
Communication, 56(4), 813–835.
Guo, F., Hanneke, S., Wenjie Fu, Xing, E. P. (2007). Recovering
temporally rewiring networks: A model-based approach. In Proceed-
ings of the 24th international conference on machine learning http://
doi.acm.org/10.1145/1273496.1273537.
Haveman, Heather A. (1993). Follow the leader: Mimetic isomorphism
and entry into new markets. Administrative Science Quarterly, 38(4),
593–627.
Katz, M. L., & Shapiro, C. (1992). Product introduction with network
externalities. Journal of Industrial Economics, 40(1), 55–83.
Kozul-Wright, Z., & Stanbury, L. (1998). Becoming a globally competitive
player: The case of the music industry in Jamaica. UCTAD discussion
paper no. 138. http://www.unctad.org/en/docs/dp_138.en.pdf.
Krugman, P. (1980). Scale economies, product differentiation, and the
pattern of trade. American Economic Review, 70(5), 950–959.
Lee, Chin-Chuan (1980). Media imperialism reconsidered: The homogeniz-
ing of television culture. Beverly Hills: Sage.
Lee, Sang-Woo, & Waterman, D. (2007). Theatrical feature film trade in
the United States, Europe, and Japan since the 1950s: An empirical
45
for exempting them from free trade commitments. Perhaps
more fundamentally, I found that cultural goods are not
exceptional in this way, and that many other ordinary goods
reveal this sort of ‘‘cultural” pattern of trade by demonstrating
evidence of a strong homogenization effect. The presence of
such an effect implies that trade liberalization is not necessarily
good for everyone, and that there are winners and losers.
Admittedly, the findings here beg for much additional research
before we can fully understand the implications of homogeni-
zation from trade on development. For now, however, I will
say that at the least, if products and technologies that are un-
ique to a given country are of any value, then that country
should seek to support and protect them from the homogeniz-
ing influence of trade, even where there is no ‘‘infant industry”
justification for such protection.
study of the home market effect. Journal of Media Economics, 20(3),
167–188.
Mattelart, A., & Polan, D. B. (1978). The nature of communications
practice in a dependent society. Latin American Perspectives, 5(1),
13–34.
Olson, Mancur (1982). The rise and decline of nations: Economic growth,
stagflation, and social rigidities. New Haven: Yale University Press.
Pattison, P., & Wasserman, S. (1999). Logit models and logistic
regressions for social networks: II. Multivariate relations. British
Journal of Mathematical and Statistical Psychology, 52(2), 169–193.
Pool, I. de Sola (1977). The changing flow of television. Journal of
Communication, 27(2), 139–149.
Pratt, A. (2004). The music industry in Senegal: The potential for economic
development. UNCTAD.
Ratanawaraha, A. (2006). Late standardization and technological catch-up.
Unpublished dissertation, MIT Department of Urban Studies and
Planning <http://dspace.mit.edu/handle/1721.1/37674>.
Robins, G., Pattison, P., Kalish, Y., & Lusher, D. (2007). An introduction
to exponential random graph (p*) models for social networks. Social
Networks, 29(2), 173–191.
Rogers, EverettM (1995). The diffusion of innovations. The Free Press: New
York.
Schiller, H. I. (1976). Communication and cultural domination white plains.
International Arts and Sciences Press.
Scott, A. J. (2004). Cultural-products industries and urban economic
development: Prospects for growth and market contestation in global
context. Urban Affairs Review, 39(4), 461–490.
Snijders, T. A. B. (2002). Markov Chain Monte Carlo estimation of
random graph models. Journal of Social Structure, 3(2).
Snijders, T. A. B., Pattison, P. E., Robins, G. L., & Handcock, M. S.
(2006). New specifications for exponential random graph models.
Sociological Methodology, 36(1), 99–153.
Snijders, T. A. B., Steglich, C. E.G., Schweinberger, M., & Huisman, M.
(2007). Manual for SIENA version 3. University of Groningen, ICS,
Groningen. University of Oxford, Department of Statistics, Oxford.
<http://stat.gamma.rug.nl/stocnet>.
Thurner, P. W., & Binder, M. (2008). European Union transgovernmental
networks: The emergence of a new political space beyond the nation-
state?. European Journal of Political Research, 48(1), 80–106.
UNESCO. (2006) http://www.unesco.org/culture/industries/trade/
html_eng/question.shtml Accessed 10.25.06.
Wallerstein, I. (Ed.) (1975). World Inequality. Quebec: Black Rose.
Wasserman, S., & Pattison, P. (1996). Logit models and logistic
regressions for social networks: I. An introduction to Markov random
graphs and p*. Psychometrika, 61(3), 401–426.
Wildman, S. S., & Siwek, S. (1988). International trade in films and
television programs. Cambridge, MA: Ballinger.
WTO Council for Trade in Services. (1998) AUDIOVISUAL SERVICES;
Background note by the Secretariat. WTO document 98-2437
www.wto.org/english/tratop_e/serv_e/w40.doc.
46 WORLD DEVELOPMENT

APPENDIX A
ESTIMATED PARAMETERS FOR ERGM TRANSITIVITY ANALYSIS—VARIOUS GOODS

Reciprocity A. Out A. In A. A. Gravity Colony Language Contiguous Product

stars stars k-triangles 2-paths of GDPs
Books
Parameter estimates 1.0843 1.0122 0.6058 1.7624 0.0005 0.0177 1.9752 0.6261 1.2174 0.1322
Standard errors 0.1931 0.1690 0.1376 0.1331 0.0051 0.0026 0.1872 0.0606 0.1608 0.0162
t ratios 5.6152 5.9893 4.4026 13.2412 0.0980 6.8077 10.5513 10.3317 7.5709 8.1605
Size of effect 2.9574 2.7516 0.5456 5.8264 1.3043 7.2081 1.8703 3.3784 2.8233
Furniture
Parameter estimates 0.2347 0.7962 0.0837 1.7430 0.0050 0.0148 0.5793 0.2476 1.1388 0.0650
Standard errors 0.1863 0.1715 0.1537 0.1535 0.0100 0.0031 0.2054 0.1464 0.1799 0.0095
t ratios 1.2598 4.6426 0.5446 11.3550 0.5000 4.7742 2.8204 1.6913 6.3302 6.8421
Size of effect 2.2171 5.7145 1.2487 1.7848 3.1230 1.6658
Sound recordings
Parameter estimates 0.9572 0.9672 0.1846 1.6535 0.0048 0.0189 0.8173 0.4775 0.6195 0.1154
Standard Errors 0.1955 0.1862 0.1663 0.1624 0.0086 0.0026 0.2292 0.0853 0.2360 0.0144
t ratios 4.8962 5.1944 1.1100 10.1817 0.5581 7.2692 3.5659 5.5979 2.6250 8.0139
Size of effect 2.6044 2.6306 5.2252 1.3280 2.2644 1.6120 1.8580 2.4744
Sewing machines
Parameter estimates 1.0266 1.0908 0.3897 1.5885 0.0234 0.0101 0.3479 0.2998 1.3704 0.0399
Standard errors 0.2815 0.1518 0.1237 0.1234 0.0121 0.0030 0.2923 0.1167 0.2125 0.0071
t ratios 3.6469 7.1858 3.1504 12.8728 1.9339 3.3667 1.1902 2.5690 6.4489 5.6197
Size of effect 0.3582 2.9767 0.6773 4.8964 1.1637 1.3496 3.9369 1.3679
Cutlery
Parameter estimates 0.0624 1.1691 0.3414 1.5470 0.0028 0.0109 0.4102 0.3648 1.5511 0.0938
Standard errors 0.2105 0.1692 0.1328 0.1233 0.0080 0.0025 0.2088 0.0860 0.1832 0.0131
t ratios 0.2964 6.9096 2.5708 12.5466 0.3500 4.3600 1.9646 4.2419 8.4667 7.1603
Size of effect 3.2191 0.7108 4.6974 1.1777 1.4402 4.7167 2.0885
Passenger vehicles
Parameter estimates 0.6300 1.2888 0.0338 1.4557 0.0147 0.0095 0.2260 0.0685 0.8566 0.0393
Standard errors 0.2232 0.1991 0.1676 0.1618 0.0165 0.0028 0.3155 0.1375 0.1927 0.0067
t ratios 2.8226 6.4731 0.2017 8.9969 0.8909 3.3929 0.7163 0.4982 4.4453 5.8657
Size of effect 1.8776 3.6284 4.2875 1.1533 2.3551 1.3614
Shoes-leather, rubber or plastic sole
Parameter estimates 0.1486 0.9843 0.1808 1.3999 0.0118 0.0079 0.3048 0.2895 1.3185 0.0226
Standard errors 0.2109 0.1631 0.1522 0.1343 0.0149 0.0024 0.2522 0.0767 0.1694 0.0049
t ratios 0.7046 6.0349 1.1879 10.4237 0.7919 3.2917 1.2086 3.7744 7.7834 4.6122
Size of effect 2.6759 4.0548 1.1259 1.3358 3.7378 1.1941
Women’s dresses all fabrics
Parameter estimates 0.0067 0.9189 0.3408 1.2094 0.0287 0.0102 0.8024 0.2581 0.8987 0.0404
Standard errors 0.2094 0.1393 0.1315 0.1118 0.0124 0.0025 0.2357 0.0863 0.1892 0.0072
t ratios 0.0320 6.5966 2.5916 10.8175 2.3145 4.0800 3.4043 2.9907 4.7500 5.6111
Size of effect 2.5065 1.4061 3.3515 0.9717 1.1654 2.2309 1.2945 2.4564 1.3733
Wine
Parameter estimates 0.8317 1.5596 0.5402 1.0614 0.0730 0.0060 1.0429 0.3818 1.5609 0.0242
Standard errors 0.3133 0.1730 0.1469 0.1142 0.0178 0.0026 0.2788 0.0769 0.2154 0.0052
t ratios 2.6546 9.0150 3.6773 9.2942 4.1011 2.3077 3.7407 4.9649 7.2465 4.6538
Size of effect 0.4353 4.7569 1.7164 2.8904 0.9296 1.0942 2.8374 1.4649 4.7631 1.2092
Color TVs
Parameter estimates 0.5083 1.7848 0.1842 1.0479 0.1574 0.0069 0.1590 0.1378 1.5311 0.0305
Standard errors 0.2146 0.1643 0.1290 0.0962 0.0133 0.0029 0.3398 0.1377 0.2219 0.0062
t ratios 2.3686 10.8631 1.4279 10.8929 11.8346 2.3793 0.4679 1.0007 6.9000 4.9194
Size of effect 1.6625 5.9584 2.8517 0.8544 1.1091 4.6233 1.2706
 HOMOGENIZATION AND SPECIALIZATION EFFECTS OF INTERNATIONAL TRADE 47

APPENDIX A—Continued
Reciprocity A. Out A. In A. A. Gravity Colony Language Contiguous Product
stars stars k-triangles 2-paths of GDPs
Books
Butter
Parameter estimates 0.3740 1.6104 0.2979 1.0307 0.0894 0.0059 0.7483 0.2126 1.1023 0.0079
Standard errors 0.2702 0.1441 0.1182 0.0798 0.0129 0.0023 0.2066 0.0842 0.1770 0.0026
t ratios 1.3842 11.1756 2.5203 12.9160 6.9302 2.5652 3.6220 2.5249 6.2277 3.0385
Size of effect 5.0048 1.3470 2.8030 0.9145 1.0926 2.1134 1.2369 3.0111 1.0640
Corn, unmilled
Parameter estimates 0.2374 2.0981 0.5355 0.6693 -0.0776 0.0045 -0.2584 0.1551 1.9361 0.0157
Standard errors 0.5700 0.1637 0.1569 0.1206 0.0271 0.0039 0.4169 0.1402 0.2812 0.0030
t ratios 0.4165 12.8167 3.4130 5.5498 -2.8635 1.1538 -0.6198 1.1063 6.8851 5.2333
Size of effect 8.1507 1.7083 1.9529 0.9253 6.9317 1.1312
Gas generators and parts thereof
Parameter estimates 0.1938 1.7512 0.5054 0.6344 0.0350 0.0084 0.9333 0.2406 1.0062 0.0320
Standard errors 0.3209 0.1628 0.1527 0.1049 0.0149 0.0032 0.2301 0.1115 0.2398 0.0075
t ratios 0.6039 10.7568 3.3098 6.0477 2.3490 2.6250 4.0561 2.1578 4.1960 4.2667
Size of effect 5.7615 1.6576 1.8859 1.0356 1.1344 2.5429 1.2720 2.7352 1.2856
Anthracite
Parameter estimates -1.0620 2.4325 1.2643 0.6663 -0.1183 0.0057 0.5576 0.2112 1.9251 0.0148
Standard errors 1.2104 0.2291 0.2276 0.2641 0.0729 0.0067 0.5194 0.2659 0.4018 0.0046
t ratios -0.8774 10.6176 5.5549 2.5229 -1.6228 0.8507 1.0735 0.7943 4.7912 3.2174
Size of effect 11.3873 3.5406 1.9470 6.8558 1.1232
Cargo ships (excl. tankers)
Parameter estimates 0.0879 1.9417 1.3910 0.5857 0.0715 0.0047 0.1185 0.2637 0.6668 0.0085
Standard errors 0.6330 0.1756 0.1633 0.1310 0.0333 0.0038 0.4496 0.2711 0.3752 0.0027
t ratios 0.1389 11.0575 8.5181 4.4710 2.1471 1.2368 0.2636 0.9727 1.7772 3.1481
Size of effect 6.9706 4.0189 1.7962 0.9310 1.0690
Iron and steel bars and ingots
Parameter estimates 1.3974 1.5624 1.0941 0.2563 0.0184 0.0061 0.3685 0.1981 1.5954 0.0221
Standard errors 0.4621 0.1702 0.1727 0.1334 0.0332 0.0037 0.3193 0.1621 0.2688 0.0047
t ratios 3.0240 9.1798 6.3353 1.9213 0.5542 1.6486 1.1541 1.2221 5.9353 4.7021
Size of effect 4.0447 4.7703 2.9865 4.9303 1.1895
Antiques
Parameter estimates 1.0971 2.7326 2.0313 0.0129 0.1675 0.0142 0.9492 0.5956 1.0775 0.0299
Standard errors 0.5419 0.6187 0.6017 0.3754 0.0465 0.0078 0.4023 0.2577 0.4556 0.0076
t ratios 2.0245 4.4166 3.3759 0.0343 3.6021 1.8205 2.3594 2.3112 2.3650 3.9342
Size of effect 2.9954 15.3728 7.6239 1.1823 2.5836 1.8141 2.9373 1.2645
Note: effect magnitudes are only reported for parameters with t ratios greater than 2.
Effect magnitudes are calculated for an increase of two standard deviations in the independent variable for gravity and GDP product, and for a one unit
change for dummy and structural variables.

Available online at www.sciencedirect.com