7/27/2019 Dissertation Submitted Augmented http://slidepdf.com/reader/full/dissertation-submitted-augmented 1/274 INSIGHTS FROM BOOK TRANSLATIONS ON THE INTERNATIONAL DIFFUSION OF KNOWLEDGE A DISSERTATION SUBMITTED TO THE DEPARTMENT OF ECONOMICS AND THE COMMITTEE ON GRADUATE STUDIES OF STANFORD UNIVERSITY IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY Isabelle Yin Fong Sin May 2011
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Scientific, and Cultural Organization. From this bibliography, I compile a data set of
over 2 million translations published in 80 countries since the 1949, including detailedinformation on each title translated. I then document the main patterns of translation
flows.
In chapter 4, I employ a gravity framework to study how distance affects
translation flows between countries. This sheds light both on the barriers to
international idea diffusion and on the underlying causes of the negative relationship
between distance and trade. Translations differ from trade in that they have zero
transportation costs, but they are subject to similar search and information costs
and costs of forming contracts. I estimate a gravity model where bilateral translation
flows vary with the sizes of the countries and the distance between them, and find the
elasticity of translations with respect to distance to be between -0.3 and -0.5 for the
1990s; these values are significantly smaller than the equivalent elasticity for trade
found in the literature, suggesting a significant role for transportation costs in the
distance effect on trade. In addition, I present several pieces of evidence that suggest
supply-side frictions play a larger role in the distance effect on translations than doconsumer preferences. For instance, the speed with which titles are translated, which
is likely to largely capture supply frictions as opposed to demand factors, decreases
significantly with distance.
Finally, in joint work with Ran Abramitzky (chapter 5), I study how the
collapse of the Communist regime in Eastern Europe at the close of the 1980s
affected the international diffusion of ideas. We show that while translations
between Communist languages decreased by two thirds with the collapse, Western-
to-Communist translations increased by a factor of seven and reached Western levels.
Convergence was full in economically-beneficial fields such as sciences and only partial
in culturally-beneficial fields such as history. The effects were larger for more Western-
oriented countries. These findings help us understand how institutions shape the
but flows of disembodied ideas are more interesting because these are the flows
that generate spillovers. In contrast, previous empirical research has largely usedproxies for idea flows such as trade, foreign direct investment (FDI), and migration.
Although this literature offers many important insights, these proxies measure idea
flows embodied in something else, and thus the idea flows they capture may not cause
spillovers.
In this dissertation I propose a new measure of idea flows, namely book
translations, and use it to study the factors that affect the international diffusion
of ideas. The purest measure of idea flows used previously is patent citations.1 Book
translations offer a complementary measure of knowledge diffusion that captures a
broader definition of knowledge. Although much of the endogenous growth literature
has implicitly or explicitly restricted its attention to technological ideas narrowly
defined (e.g. Romer, 1990), a much wider range of ideas are likely to be important
for economic and intellectual development. Book translations encompass a broader
set of ideas than purely technological knowledge. They are also generated by a quite
different process to patent citations, meaning the two measures capture differentaspects of knowledge flows.
As a measure of international idea flows, book translations have a number of
appealing features. Translations, as distinct from the physical books that are
translated, are a measure of pure flows of disembodied ideas. The impetus behind
translations is the desire to make the ideas contained in the books accessible to
1An extensive literature uses patent citations to study questions such as whether flows such astrade, migration, and foreign direct investment (FDI) increase knowledge spillovers (e.g., MacGarvie,2006, Branstetter, 2006, Agrawal, Kapur and McHale, 2008, and the relationships between geographyand knowledge spillovers (e.g., Jaffe, Trajtenberg and Henderson, 1993, Bottazzi and Peri, 2003,Thompson and Fox-Kean, 2005, MacGarvie, 2005, Griffith, Lee and van Reenen, 2007, Criscuolo andVerspagen, 2008, Singh and Marx, 2011). The answers to such questions could inform domestic policythat encourages or impedes research, trade, immigration, or FDI. They may also have implicationsfor the theoretical modeling of knowledge spillovers and for predictions about income convergencebetween countries.
speakers of other languages, thus translations do not primarily occur as the by-product
of some other process such as trade or migration. Importantly, translations are flowsof non-rival ideas: a country can translate a book and use its ideas without affecting
the use of the ideas by other countries. Conveniently, book translations are readily
quantifiable and classifiable by the type of idea they contain, and they capture a broad
range of technological, organizational, cultural and social ideas that are important
for economic growth and intellectual advancement.
However, book translations face some limitations as a measure of idea flows.
Because the process of translating and publishing a book is time-consuming, book
translations do not capture very new ideas. They only capture flows of ideas between
speakers of different languages, and they only capture codifiable ideas, meaning they
exclude tacit knowledge. Finally, some people are multilingual, so have access to the
ideas in books before the books are translated into their native language.
In order to study book translations empirically, I construct a new data set
from an international bibliography of translations collected by the United Nations
Educational, Scientific and Cultural Organization (Unesco). The data set contains
information on every translation published between 1980 and 2000 for up to 80
countries each year, including information on the country and year in which the
translation was published, the subject of the book, the original and target languages,
the length of the book, the author, and the original and translated titles of the book.
This amounts to over two million translations. For every fifth year from 1949 to 1979,
I also digitized from hard copy a representative sample of the translations published
annually in each country, including information on the country and year of translation,
the subject of the book, the original and target languages, and the length of the book.
This amounts to approximately 100,000 translations. For specific sub-samples I also
collected additional information such as the year in which the original book was first
published, the sub-field of the book, and the author’s political views. These data
allow me to draw a detailed picture of the changing translation flows between a wide
range of countries for the period 1949 to the present.
In Chapter 2, I motivate my study of international idea flows with an overview
of the economics of ideas and their diffusion between countries. I also discuss
in more detail the qualities of ideas that the theoretical literature suggests to be
important for growth, and motivate the use of book translations as an empirical
measure of international idea flows. Finally, I provide a framework for thinking at
the microeconomic level about when translations are likely to occur.
In Chapter 3, I describe the data collection, and document a range of patterns of
the international flow of translations and their historical trends. I find the most
translated language is English, which accounted for over 46% of all translations
between 1949 and 2000, rising from 31% in 1959 to 61% in 1999. The most prolific
translating country in 1959 was the Soviet Union, which translated nearly 5,000 titles
that year. By 1999, after the collapse of communism in Eastern Europe and the
disintegration of the USSR, Germany had risen to be the top translating country
with nearly 10,000 translations annually; it was followed by Spain, France, and Japan.European countries translate more on average than countries on other continents, and
Muslim countries (such as Albania, Egypt, and Turkey) translate 73% fewer titles than
Roman Catholic countries (such as Peru, France, and Slovakia), even after controlling
for population, income and openness defined as trade as a fraction of GDP.
I also find that in the years 1998-2000, the bulk of translated books were translated
within 10 years of publication, though the speed with which titles are translated varies
considerably by original language and translating country. English and Italian books
are translated faster than French and German books on average; richer countries and
countries that trade more translate faster.
In Chapter 4, I study how distance affects translation flows. By studying the
relationship between measures of distance between countries and translation flows, I
shed light on an important type of impediment to the free international diffusion of
ideas. I also shed light on the factors underlying the negative relationship betweendistance and trade in goods. Translations have many commonalities with trade in
goods, but differ in that they are exempt from all costs related to physical relocation.
The relationship between distance and translations is thus informative about the
aspect of the distance effect in trade that is not driven by transportation costs.
Translations (and similarly trade) may decrease with distance for both supply and
demand reasons. Supply frictions such as distance-varying search and information
costs and costs of negotiating contracts could cause translations to decrease withdistance. Translations (or trade) may also fall off with distance because distance
is correlated with tastes, meaning closer countries cater better than more distant
countries to local tastes in books (or products).
I first estimate a gravity model of translation flows, in which the flow between
two countries depends on the sizes of the two countries and on the distance between
them. I estimate the elasticity of translations with respect to distance to be -0.3 to
-0.5 during the 1990s, which is considerably smaller than the equivalent elasticity fortrade found in the literature, which usually ranges from -1.08 to -1.24.2 The difference
between these estimates suggests transportation costs play a substantial role in the
distance effect on trade.
I next conduct several tests to study the relative roles of supply-side frictions and
consumer tastes in the distance effect on translations. Results suggest an important
role for search and information costs and a lesser role for demand factors. First, the
distance effect decreased between 1949 and 1999. This is consistent with informationcosts, which declined substantially over this period, being an important limiting factor
for translations; this result stands in contrast with the puzzling result in the trade
literature that distance did not become less important over this period. Second, the
distance effect is larger in the fields of natural and applied science, where tastes are
less important, than in the fields of arts, literature and philosophy, which have ahigher cultural component. If geographically correlated tastes were the major driving
factor, the opposite would be true. Third, cultural distance between countries does
inhibit translation flows, but accounts for relatively little of the overall distance effect,
suggesting non-cultural factors play a large role. Finally, the speed with which titles
are translated, which likely captures supply frictions as opposed to demand factors,
also decreases significantly with distance.
My results have important implications for the international diffusion of ideas.They suggest that, despite the fact ideas have no transportation costs, idea flows
are hindered both by geographic distance and cultural distance between countries.
Furthermore, idea flows into less developed countries are hindered more by distance
than idea flows into more developed countries. This relationship works against income
convergence between rich and poor countries: the countries that can benefit most from
catch-up growth by adopting foreign ideas seem to face greater frictions in accessing
these ideas. However, the inhibiting effect of distance has decreased over time, whichsuggests that even the barriers surrounding less developed countries may be lower in
the future.
In Chapter 5, I use the translation data (with Ran Abramitzky3) to investigate
how institutions shaped the international diffusion of ideas in one of the largest events
in modern history, the collapse of Communism in Eastern Europe. We study how the
collapse affected translation flows within Communist Europe and between Western
and Communist Europe, how the effects varied across countries, and how they variedacross book fields.
The collapse of Communism was an important historical event and is thus worth
studying for its own sake, but our research also sheds light on a range of broader
3Both coauthors contributed equally to every aspect of the work involved in this paper.
issues. The collapse was a large shock that swiftly moved countries from nearly
complete isolation from Western ideas to full openness. Because our measure of ideaflows captures a broad range of ideas, the paper sheds light on the type of ideas
most likely to be affected by policy changes that reduce information restrictions. In
particular, we can examine whether the collapse of Communism had a stronger effect
on ideas that contain more “useful knowledge” for economic development than on
“less-useful” knowledge with more cultural content.
To shed further light on the role of preferences in the flow of ideas, first we compare
translation patterns in the Soviet countries with patterns in the more western-orientedSatellite countries. Second, we test the degree of convergence in translation flows
between Eastern and Western Europe post collapse.
We find that the collapse of Communism resulted in a sevenfold increase in
translations of Western European titles in the Satellite countries, suggesting a huge
increase in the inflow of Western ideas, and a threefold decrease in translations
of Communist titles, suggesting a decline in the flow of ideas between Communist
countries. These patterns also imply a substitution of Satellite countries away fromCommunist ideas and towards Western ideas.
Furthermore, we find evidence consistent with a surprising degree of cultural
convergence of Satellite countries and Western Europe. Given censorship was lifted
with the collapse of Communism, remaining differences in translation patterns likely
reflect differences in tastes for certain ideas between Eastern and Western Europe.
Since the end of Communism in Eastern Europe, the traditionally more Western-
looking Satellite countries have increased their translations of Western European titles
to Western levels. We interpret this convergence to the West as evidence that Satellite
preferences were either similar to Western ones or became like them quickly following
the collapse.
We find both an increase in Satellites’ translations of older titles and a jump
in translations of newer titles. These findings are consistent with both catching up
on the stock of ideas that were missed out on under Communism and a convergencebetween Satellite countries and Western Europe in the diffusion of new Western ideas.
In contrast, we find that the collapse of Communism had little effect on Western
translations in Soviet countries, suggesting the diffusion of Western ideas into these
countries was much more limited. The difference between the effect for the Soviet
and the Satellite countries suggests preferences play an important role in determining
the ideas that diffuse into a country.
We also find the effects of the collapse of Communist translations of Western titlesvaried significantly by field. The effects were larger and convergence to the West was
greater in fields that can be considered more “economically useful”, such as applied
science and economics, more ideological fields, such as philosophy and religion, and
for more threatening titles. They were smaller for less “economically useful” fields,
such as arts, and more “objective” fields, such as the exact sciences. Finally, we study
the translation patterns of a sample of particularly influential titles. We show most of
these titles were not translated anywhere in Communist Europe prior to the collapseof Communism, and with the collapse their translation increased dramatically in
Communist but not Western Europe. This affirms that the collapse of Communism
resulted in a genuine increase in access to important Western ideas within Communist
Europe.
A key lesson from our study is that incentives play a major role in shaping the
international flow of knowledge. Distortion of these incentives by institutions can
have long-lasting effects that can only be remedied by institutional change.
To summarize, in this dissertation I introduce a new measure of the flow of ideas
between countries, namely book translations. I assemble a novel data set of over two
million translations in 80 countries over the period 1949-2000. I use this measure
to study questions such as how physical and cultural distances between countries
shape the international diffusion of ideas, and what the role of institutions is in
shaping the international diffusion of knowledge. I find that, even today, translationflows decrease significantly with both physical and cultural distance, although this
relationship was stronger in earlier decades. Physical distance matters more for
less developed translating countries, and inhibits both the quantity and speed of
translations. Overall, my results are consistent with search and information costs
being a major driving force behind translation flows. Furthermore, I find (with
Ran Abramitzky) that the Communist regime in Eastern Europe severely distorted
international idea flows, artificially inflating flows between Communist countries, andsuppressing flows of ideas from the West into the Communist Bloc. The strong inflow
of economically-useful Western ideas into the Satellite countries, but not the Soviet
countries, upon the collapse of Communism suggests preferences play a strong role in
determining the types of ideas that diffuse into a country. Finally, this study of book
translations generates insights into the economic consequences of linguistic divisions
2.1. THE IMPORTANCE OF IDEAS IN HISTORICAL PERSPECTIVE 11
Revolution and the enlightenment movement that preceded it. Mokyr (2002) explains
much of the Industrial Revolution as resulting from a change in the feedbackmechanism between two distinct types of useful knowledge. The first type of useful
knowledge is propositional knowledge about natural phenomena and regularities; the
second type is instructional or prescriptive knowledge, or techniques. Both types of
knowledge may reside in the minds of people or in storage devices such as books
from which they can be retrieved. Learning or diffusion of knowledge involves the
transmission of existing knowledge from one person or storage device to another. The
union of all propositional knowledge held by people or devices in a society is referredto as omega; the union of all techniques is referred to as lambda. A new discovery
adds to the set lambda or omega.
For any technique or element of lambda to exist, someone (though not necessarily
the person implementing the technique) must know enough about the elements of
omega upon which the technique is based to make the technique possible. More
broadly, each set of omega-knowledge possessed by a society makes possible many
sets of techniques, or lambdas. Which lambda is realized will depend on factors suchas the culture and institutions of the society, which affect the preferences and priorities
of agents and the rewards and penalties associated with suggesting new techniques.
Prior to 1800, serendipity played the dominant role in technological progress.
Techniques had narrow bases in omega, and thus the flow-on effects of new discoveries
ran into diminishing returns and petered out without ever leading to sustained
increases in technological progress. Furthermore, inventions based on limited
knowledge of the natural phenomena behind them were often treated with suspicion
by the public, making the spread of their use difficult.
Mokyr (2002) argues that the reason the Industrial Revolution didn’t die out
after 1820 in the manner of all previous episodes of growth is that the scientific
revolution of the seventeenth century and the Industrial Enlightenment of the
eighteenth century had broadened the epistemic base in omega of techniques in
lambda, allowing the feedback mechanism between the two types of useful knowledgeto switch from negative to positive. The Industrial Enlightenment affected the two
types of knowledge and their interactions in several ways. First, it reduced the
cost of accessing best-practice artisanal techniques. Second, by the generalization of
techniques it improved understanding of why they worked, thus broadening their bases
in omega. Third, it improved communication and interactions between the people who
understood the propositional knowledge and those who used the techniques. The time
of the Industrial Revolution was also the time of a knowledge revolution, in which theorganization, storability, accessibility and communicability of information in omega
advanced greatly. Much knowledge that was previously unwritten was codified in
books, and many scientific and technical works switched from being written in Latin
to being written in the vernacular, thus becoming accessible to the users of lambda
knowledge. The diffusion of knowledge between European countries was facilitated
by rapid translations of key works, and considerable movement of skilled individuals
between countries. This historical perspective on the role of knowledge in growthhighlights, among other things, the importance for further technological advancement
of how knowledge is stored and diffused both within and between societies.
2.2 The economic theory of ideas
Macroeconomic models such as Solow’s classic growth model, in which technological
change is absent or exogenous, explain growth through the accumulation of physical
and human capital. However, history reveals a long-term trend of accelerating growth
that cannot be explained by similarly increasing levels of capital (e.g., Jones and
Romer, 2010). An increasing rate of technological change is therefore required to
explain this accelerating growth rate. An exogenous and exogenously accelerating rate
of technological change is obviously an unsatisfactory explanation for this historical
fact, so the economics profession has delved deeper into the nature of ideas and thedrivers of technological progress within the economic machine to look for an answer.
A likely candidate explanation for the acceleration in growth has emerged in the non-
rival nature of knowledge and the positive feedback between population and ideas.
2.2.1 One-country models
The central advance of endogenous growth models over the previous, exogenous
growth models was to recognize that technological change is determined within
the economic system and occurs as the result of actions taken by individuals and
firms, rather than being independent of economic activity. The first macroeconomic
models to include endogenous growth abstracted from the complication of interactions
between countries in their formal modeling. Such models can therefore be considered
to capture economies that are totally closed off from the rest of the world, or multiple-
country worlds in which countries are perfectly integrated. Although neither scenario
is the most interesting case, such models nevertheless provide important insights into
the role of ideas in economic growth.
In a seminal paper, Romer (1990) lays out clearly aspects of how to think
about ideas in a growth context that became widespread in later literature. Romer
essentially adds endogenous technological change to a neoclassical growth model,
where technological change is formulated as the invention of new ways to combine
raw materials to make producer durables. Once invented, ideas are excludable in
production, which provides firms with an incentive divert resources away from final
goods production and towards innovation. However, they are non-rival and are not
excludable in their use to produce further ideas, which produces a spillover effect.
Importantly, the research sector has increasing returns to scale in its two inputs,
human capital and the existing knowledge stock. The transmission of ideas within
the economy is assumed to be complete and free. In this model, economic growth is
ultimately driven by the accumulation of ideas. A further prediction of the modelis that a larger stock of human capital, not just more people, translates into more
research and thus faster growth.
An important debate that has arisen in choosing characteristics macroeconomic
models of growth ought to have is that of scale effects. Models such as Romer (1990),
Aghion and Howitt (1992) and Grossman and Helpman (1991) exhibit “strong” scale
effects, meaning the long-run growth rate of the economy increases with the scale
of the economy. In contrast, in models such as Jones (1995), Kortum (1997) and
Segerstrom (1998) exhibit only “weak” scale effects. That is, the long run level of per
capita income increases with scale or, equivalently, the economy has increasing returns
to scale. Jones (2005) argues that the predictions of strong scale effects have been
shown empirically untrue. For instance, strong scale effects are difficult to reconcile
with the empirical facts that research effort has grown over time, but US growth rates
have remained relatively stable for over a century. Weak scale effects, however, which
are largely synonymous with idea-based growth models, seem more plausible thantheir alternatives.
Several key properties of knowledge have emerged in the growth theory literature:
it is non-rival, may be excludable, is disembodied, and its accumulation responds to
incentives. (See, for example, work by Romer, Jones and Helpman.) Non-rivalry
is arguably the most important property of ideas for endogenous growth. It is the
technological characteristic that the use of an idea by one party in no way limits its
simultaneous use by others. This property generates spillovers and increasing returns
by the duplicability argument. That is, consider a production process that uses as
inputs both traditional rival inputs such as physical capital and labor, and knowledge,
which is non-rival. Then, supposing all rival inputs can be perfectly duplicated, a
doubling of the rival inputs will double output because the same knowledge can be
used in both identical plants at the same time. If, in addition, knowledge is doubled,
then output will more than double, hence the production function exhibits increasingreturns in all its inputs. Another consequence of the non-rivalry of knowledge is that
the total value to society of an idea is increasing in the size of the society, because
it can be used by more people or firms simultaneously (Jones and Romer, 2010).
This means there are advantages to integrating populations into as large groups as
possible.
Knowledge may or may not be excludable. It may be able to be protected
through secrecy, or through the patent system in an appropriate institutional setting.How quickly and fully ideas diffuse depends on the incentives inventors have to
exclude others from their inventions, which depend on institutions. In choosing their
institutions, societies face a trade-off between the stronger incentives for innovation
provided by high excludability, and the efficient utilization of existing ideas provided
by low excludability.
Knowledge itself is disembodied. A knowledge flow that is embodied in a capital
good may not produce a spillover; and if it does, it might be a pricing or pecuniary
externality rather than a technological one (Jaffe and Trajtenberg, 1999). Human
capital is ideas embodied in people. Unlike the ideas themselves, human capital is rival
because a person working on one project is limited in her ability to simultaneously
work on another.
Finally, knowledge accumulation responds to incentives. Research and devel-
opment, for example, is the purposeful generation of new knowledge. It comes
at an opportunity cost, and thus responds to incentives (Helpman, 2004). The
adoption of foreign technologies depends on institutions and the incentives they create
(Romer, 2010). Even learning-by-doing, by which an additional output, knowledge,
is generated by a production process, responds to the incentives that reward this
Sixth, across countries, high investment rates are a far better predictor of high levels
of income than of high growth rates of income.
Although this evidence is far from definitive, it does suggest economists ought
to take the international diffusion of ideas seriously. Indeed, considerable theoretical
work has been done in this area, such as Nelson and Phelps (1966), Parente and
Prescott (1994), Romer (1994), Howitt (2000), Lucas (2009), Eaton and Kortum
(1996, 1997, 1999), Kortum (1997), Keller (2004) and Romer (2010). Klenow and
Rodrıguez-Clare (2005) calibrate a model of the international diffusion of knowledge
that builds off previous models, particularly Eaton and Kortum (1999). In order to
be consistent with the evidence, they focus on a model in which, in steady state,
all countries grow at the same rate because of international technology spillovers,
and policy difference across countries result in TFP level differences, not growth
rate differences. In a simple version of their model, there exists a world frontier
in technology, growth in which is determined by a weighted aggregate of worldwide
research activity. The research efforts of individual countries determine how close to
that frontier they will get. “Research”, which encompasses both R&D and efforts toadopt foreign technology, is more effective at increasing productivity the further is the
country from the world technology frontier. However, even in the absence of research
effort, some technology adoption from abroad occurs. They assume international
technology spillovers decrease in distance, and they capture additional barriers to
technology adoption in a country-specific “R&D tax”.
Some useful insights are derived from their model and calibration. For one thing,
large cross-differences in institutional or policy barriers to technology adoption are
required to make the model fit the data. In addition, the calculated benefit from the
world being connected is huge: the calibrated model suggests world GDP would be
a mere 6% of its current value if international idea diffusion were absent. Because
knowledge diffusion is costly, differences in levels of investment in knowledge creation
traffic. The two types of ideas share many similarities. Both are non-rival and offer
potential benefits from being shared internationally. The adoption of either is costlyand is affected by incentives. In Romer’s model, which includes idea diffusion between
countries, productivity in a country depends both on the local stock of technological
ideas and on local rules. In addition, rules affect how likely foreign technological
ideas are to be adopted locally, because they determine the degree of excludability
the inventors hold, and the incentives local firms have to adopt the ideas. Rules also
have the potential to be adopted from overseas, though this may be difficult if it
requires the agreement of large numbers of people.
One potential reason for the relative neglect of non-technological ideas in the
literature may be that, compared with technological ideas, they remain a vague
concept and one that is even more difficult to measure empirically. Theoretical
discussions have long acknowledged the importance of institutions for economic
performance (e.g., North, 1990, among others), but only more recently has the
empirical literature begun to provide evidence along these same lines. A notable
example is Acemoglu, Johnson and Robinson (2001), who use differences in the
mortality rates faced by European colonists, which affected the colonial institutions
they established, and thus current institutions through institutional persistence, and
estimate large effects of institutions on income per capita. Although the institutions of
a country are heavily path-dependent and knowledge of an improvement over existing
institutions by no means guarantees its adoption, such knowledge is a necessary
condition for its adoption. As a result, the diffusion of ideas about institutions has
an important role in institutional change and development.
levels of conventional trade flows may actually indicate inefficiently low international
diffusion of ideas.
The second potential channel for technology diffusion is exports. The mechanism
here is less immediately obvious than in the case of imports, but may operate
because firms benefit from dealing with international customers that possess more
advanced knowledge. For instance, the international customer may require higher
quality standards than domestic customers, and may provide the domestic firm with
information on how to meet those standards. Although case studies find support for
such a role of exports, the econometric evidence is consistent with exports having noeffect on domestic TFP (Keller, 2004). These two channels suggest that policies that
affect trade, especially imports, may have implications for the rate at which a country
adopts foreign technology.
The third channel is foreign direct investment. Multinational firms tend to be
the most R&D-intensive (e.g., Criscuolo, Haskel and Slaughter, 2010), hence they
offer a large potential for technology diffusion between the countries where the
parent company and the subsidiaries operate. The parent company may share firm-specific technology with its subsidiaries in other countries, high-quality inputs may
be provided to the subsidiaries, or labor training may generate learning externalities.
Empirical evidence for such effects is mixed, though some studies find significant
effects that are economically important in magnitude. Furthermore, spillovers seem
to be larger in high-tech industries (Keller, 2004).
Knowledge may also diffuse between countries when carried by migrants and
temporary visitors. In addition to codifiable knowledge, people may carry tacit
knowledge, which may be strongly complementary to physical technology. This
channel suggests a strong geographic element to knowledge diffusion, because the
cost of moving people is highly correlated with distance.
In line with much of the empirical literature that he summarizes, Keller (2004)
doesn’t draw an explicit distinction between “pure” ideas and ideas embodied in
capital goods. This distinction is important, because embodied ideas may not providethe technological spillovers that are essential in theoretical models of endogenous
growth (e.g. Jaffe and Trajtenberg, 1999).
2.4.2 Pure idea flows: Patent citations and translations
The primary measure of pure idea flows in the literature is patent citations. The
literature on patent citations considers a patent to indicate a piece of technological
knowledge, and a patent citation indicates “a given bit of knowledge being useful in
the development of a descendant bit” (Jaffe and Trajtenberg, 1999). That is, patent
citations proxy for spillovers from R&D. Patent citations can therefore be used to
measure how geographic and other factors affect knowledge diffusion. The answers
such studies can provide have important implications for how technological change
and growth ought to be modeled in theoretical papers, and may guide the formation
of appropriate policy in the areas of science and technology (Jaffe and Trajtenberg,
1999).
A series of papers by Jaffe, Trajtenberg and Henderson (1993, 1996, 1999) find
a number of regularities of patent citations. Citations are localized both within and
between countries. That is, patents are much more likely to be cited by other patents
from the same region of the country than from other regions of it, and by patents from
the same country than from foreign countries. However, this localization effect fades
over time and any local advantage is eventually eliminated. This pattern of fading
localization over time can be explained by the two competing forces of knowledge
diffusion and obsolescence. Knowledge diffuses geographically over time, hence it
reaches closer locations more quickly, but at the same time at any location that
might use it, it passes through a natural life cycle from discovery, to usefulness, to
obsolescence. There are also some country-specific patterns to patent citations. For
example, the Japanese tend to cite a new patent quickly. Citation flows also tend to
be bidirectional. That is, if there is a large citation flow from country A to countryB, then there is likely to also be a large citation flow from country B to country A.
More recently, MacGarvie (2005) discusses some of the correlates of international
patent citation flows. Geographic distance, which MacGarvie interprets as a proxy
for difficulty of communication, inhibits citation flows, though this negative effect
decreased over the period 1980 to 1995. Lack of a common language inhibits citation
flows; import flows and FDI are positively correlated with citation flows.
Patent citations have many desirable qualities as a measure of idea flows, but they
also have some well-known limitations. First, the overall fraction of research output
that is ever patented is small. The decision to patent an invention is a strategic one;
in many circumstances, secrecy is viewed as a preferable means to protect a discovery,
or the discovery may not be considered worth protecting at all. Second, as stated by
Jaffe, Trajtenberg and Henderson (1993), “Ex post, the vast majority of patents are
seen to generate negligible private (and probably social) returns.” Third, incentives to
patent differ by time and place, and with the type of invention, which implies caution
must be used in interpreting comparisons between quantities of patents or patent
citations in different countries or periods, or the presence or absence of citations to
any particular patent. Finally, patent citations are limited to measuring the spread
of scientific and technical knowledge.
Book translations, the measure of the diffusion of ideas that I introduce in this
dissertation, have several attractive qualities for this purpose. They capture a broad
range of types of ideas, including technical, social, and organizational; in the parlance
of Romer (2010), they include both “technological ideas” and “rules”. Translations
themselves are non-rival and disembodied, thus they have the potential to create the
externalities that are key to growth in the theoretical literature. Besides not being
embodied, translations do not occur as a side effect of some other activity: a book
is translated in order for people who speak the target language to gain access to
the ideas contained in the book, thus occurs as the direct result of the desire forknowledge to spread. Unlike patents, there are no strategic considerations involved
in the translation of books. Finally, book translations have a natural quantification,
and they are classifiable by type.
Naturally, book translations also have some limitations as a measure of idea flows.
They only capture idea flows between languages, so cannot measure, for instance, idea
flows between the US and Britain. By their nature, the ideas they can transmit must
be codifiable, thus they miss flows of tacit knowledge. Because of the length of timeinvolved in writing and publishing a book, they tend not to capture very new ideas.
Finally, some people are multi-lingual, thus have access to ideas that are published
in languages other than their own before the books are translated.
2.5 When will translations occur?
A microeconomic perspective
The diffusion of ideas between the groups that create them is critical because it
enables a much greater rate of idea accumulation, and makes certain ideas available
to societies that lack the ability to generate them domestically. At the same time, it
decreases the duplication of effort involved in the creation of ideas.
However, language barriers hinder the diffusion of ideas between linguistically
distinct groups. This is especially true for ideas captured in books, which can
only be read by people who speak the language in which the books are written.
Despite the limitations inherent in using written language to capture ideas, books
are an important means for storing and transmitting many types of ideas between
individuals separated by space or time: a book may detail a technology itself or
provide information that allows a technology to be adapted to new circumstances or
used most fully or efficiently; it may capture ideas on how to organize a society, aneconomic or political system, or a firm, or explain how an agent can exploit such an
existing system; it may contain, in the parlance of Mokyr (2002), “propositional
knowledge” about natural phenomena and regularities, which forms a basis for
the discovery of new technologies; it may contain ideas such as literature that are
consumption goods in themselves, or suggest activities such as arts, sports or hobbies
to consumers that improve their mapping from material consumption of goods and
services to utility; or it may contain many other sorts of ideas besides.
2.5.1 Translations versus bilingualism:
The individual’s viewpoint
Although language barriers are an impediment to the spread of ideas captured in
books, they can be overcome either through translation or through bilingualism.
Consider the choice faced by an individual endowed with some native language who
is interested in a particular idea described in a book. A book capturing the idea may
have been published in his native language, in which case he will read the original
with no difficulty. However, it may be the case that the idea was written in a foreign
language, and no adequate substitute was originally written in his native language. If
the foreign book has not been translated into the individual’s native language, he has
the options of learning the language in which the book was written and reading it in
the original, or obtaining the idea in some other form (such as from a bilingual person
who has read the original), from which he may receive it in an incomplete, diluted
or distorted version. If, on the other hand, the foreign book has been translated into
his native language, he may choose to read the translation, which is likely to be less
costly for him to read and fully understand, but which may lack some information
may experience greater demand for translations from each other because their
preferences favor each other’s ideas. At the same time, such similarities may meanthey tend to generate original titles that contain ideas that are close substitutes, thus
making translations redundant. Thus it is conceptually unclear whether translations
will increase or decrease with similarity between the countries.
In addition to these demand-side considerations, supply-side considerations are
likely to play a role in determining translations. Importantly, a publisher can only
translate a title once it knows the title exists. This might occur serendipitously, or
may be the result of costly search. The chance of a serendipitous discovery occurringis increasing in the degree of interactions between the populations of the original and
target languages. At the extreme, where there two languages coexist in the same
country, and especially where they are geographically mixed, the chance is very high.
In the case where costly search is involved, it seems likely search efforts of a publisher
will be directed towards languages that publish a lot of original titles, and that are
able to test titles in large domestic markets.
A second supply-side consideration is transaction costs. A publisher’s calculationof whether translating a title is likely to be profitable will also account for the costs of
completing the transaction, which may vary with the culture or country of the original
title, and also with differences in culture or business practice between the original and
translating countries. Greater physical communication costs between the countries,
such as those caused by being in different time zones, may also decrease the likelihood
of translation.
2.5.3 Implications for translation flows
The mechanisms outlined above suggest the size of the target language population,
both in terms of numbers and in terms of wealth and education, should matter
for translation flows for several reasons. A larger population has less reason to be
bilingual and thus may translate more (bilingualism effect). They also offer a bigger
potential market for copies of each title translated, thus may attract translations of awider range of titles (market size effect). However, they are likely to also produce more
original titles that act as substitutes for potential translations and that may crowd
translations out of the market (substitution effect). At small sizes, it seems likely the
bilingualism and market size effects will dominate and translations will increase with
population. However, the substitution effect may become relatively more important
at larger sizes, and there may actually be a size beyond which the substitution effect
dominates and translations decrease with the size of the target language population.The size of the original language population must also be important for transla-
tions. An economically small linguistic group will not produce a large number of titles
that have the potential to be translated, so outward translations are likely to increase
with the size of the language group, at least up to a point. However, foreigners also
have less incentive to learn a small language, so if they want to read titles written
in such a language they must first translate them, whereas they may instead learn a
larger language to read it in the originals.
When we consider differences across fields in propensities to translate, the
predictions are again ambiguous. Fields in which the viewpoints of original titles
differ more across languages may be translated more because domestic substitutes
for foreign titles are less frequently available, or may be translated less because
the interests of potential readers differ more across language groups. However, in
fields where internationally books tend to be concentrated in a lingua franca (such as
English in many academic disciplines) we unambiguously expect fewer translations
relative to original publications.
Various types of similarities or close relationships between linguistic groups may
increase or decrease translation flows. Close relationships (such geographic proximity,
trading relationships, mutual membership of international treaties, etc) between two
linguistic groups foster bilingualism between them, making translations less necessary.
They also tend to align the types of original titles produced by the two groups, whichcould reduce translations through substitution, or increase them through enhancing
relevance. Additionally, these relationships could generate demand for the ideas of the
other group by increasing awareness of the existence of the ideas. Such relationships
also mean publishers are more likely to be aware of titles to potentially translate, and
face relatively low communication and transaction costs for doing so.
Along similar lines, cultural and other similarities between groups decrease the
costs of translation, and may either increase translations because of greater mutualinterests or decrease them because of the availability of domestic substitutes.
Finally, if two languages are sufficiently similar that the cost of learning one for a
person who speaks the other is very low, the prevalence of bilingualism may reduce
book published in the country and intended for circulation.1
Titles are categorized into fields according to the nine main categories of theUniversal Decimal Classification (UDC) system: General; Philosophy (including Psy-
chology); Religion and Theology; Law, Social Sciences, Education; Natural and Exact
Sciences; Applied Sciences; Arts, Games, Sports; Literature (including books for
children)2; History Geography, Biography (including memoirs and autobiographies).
The bibliographic entry for each translation includes information on the country,
city, and year in the which the translation was published, the language of the original
title and the target language into which it was translated, the field (UDC class) of the title, the number of pages or volumes of the title, the author, and the original
and translated titles of the book.3
Digital translation data: 1979 to 2000
For approximately the period 1979 to 2000, I acquired the IT from Unesco in digital
format. Prior to 1979, these data do not exist in digital form. Beyond 2000, there
are still translations reported for some countries, but in many cases reporting of the
translations published in these years is clearly still incomplete. I do not use data from
countries in years where translations are incompletely reported. The digital record for
each translation includes the full bibliographic record for the translation, and usually
bibliographic details of the original title.
1Note that although there may be a delay of several years between the national depository of acountry receiving a translation and Unesco listing the translation in the IT, the IT reports the yearin which such translations were published, not just the year in which they were reported. I attributethem to the former and disregard the latter.
2Philology and Linguistics were a separate (and very small) category prior to 1970, and thenwere combined with Literature. I group them with Literature for all years for consistency.
3In a few instances, the IT reports that a title was translated from its original language via anintermediate language. In these cases, I consider the idea flow to be from the original language tothe final language, with the intermediate language just part of the mechanism. I thus count theseas translations from the original language to the target language, and disregard the intermediatelanguage.
In the regressions that aim to capture contemporary translation patterns, I use
translation data from two points in time, the first being the annual average for 1993to 1995, and second being the annual average for 1998 to 2000. This period is
short enough to likely have a relatively constant relationship between translations
and distances throughout. In addition, the two points in time fit into the pattern of
every fifth year that I use for examining historical trends in translations, as described
below. Finally, including two periods as opposed to just one allows more precise
estimates of the relationships of interest. The averaging process reduces noise in
the data, while limiting the number of time-varying fixed effects required, which isnecessary to be able to feasibly estimate the PML model I use, as described in Section
4.2 of chapter 4.4 At the same time, this maximizes the number of countries in the
sample: if data are available for a country for only one or two years in either of the
three-year periods, I use the average translations for those one or two years.
Hand-collected translation data: 1949 to 1979
For translations prior to 1979, the IT exists only in hard copy format. The total
number of translations listed in the IT in one year is often thirty or forty thousand;
to make digitization manageable, I restrict my digitization effort to every fifth year
from 1949 to 1979. I choose to begin my sample period with 1949 because Unesco
only began systematic data collection in 1948. Specifically, Unesco did not compile
translations for the period 1941-47, and the pre-war data (1932 to 1940) were collected
by a different institution and are not entirely comparable. Because some countries
do not report their translations to Unesco every year, and in order to maximize
the geographical coverage of my historical translation data, where the exact year of
interest was not available for a country but the preceding or following year was, I
4Using just the years 1994 and 1999, instead of the averages as described here, does notsubstantially alter the results, though it increases the standard errors on the estimates.
substitute that year instead.5 The years for which I have translation data for each
country are listed in Appendix Table A.2.Within each country, year and field, I take a 100, 50, 20, 10 or 5 percent sample
of entries.6 This amounts to approximately 100,000 records in total. I choose the
percentage to give me approximately 100 titles (or collect data on all translations
where the total number is fewer than 100) in total for each country-year-field group. In
all subsequent work I weight observations according to the inverse of their probability
of being sampled. For each entry I sample, I record the reporting country, original
and target languages, UDC category, year of publication and number of pages of thebook.
For the historical translation series, I combine these newly-digitized data with
digitized data provided by Unesco for every fifth year from 1979 to 1999.
Speed of translation data
The bibliographic entries in the IT in general do not include the year in which the
original title was first published. In order to investigate the factors driving the speed
with which titles are translated, I randomly sampled 20 non-fiction titles translated
from each of English, French, German and Italian in each country (or took all titles
where the total number of translations in the country from the original language was
fewer than 20) in the period 1998-2000. This amounts to a total of approximately
2,900 titles. For each sampled title I used online sources such as Worldcat and the
5Where data exist for consecutive years, they are very highly correlated, so this approximationis unlikely to have a significant effect on the results.
6My sample within each country-year-category group is pseudo-random in the following sense.Entries in the IT are identified by an entry number which starts at one each volume and eithercounts up throughout the whole volume, or restarts from one at the start of each new country entry.If I am taking a one in n sample, I sample every title whose identification number is a multiple of n . I do this instead of taking a genuinely random sample for speed of data entry, and because theordering within each group of titles alphabetically by author means this method is unlikely to biasmy sample with respect the original or target language, the main dimensions of interest that varywithin such a group.
has the same religion as a person from the other country with missing religion (results
not presented). The two distance measures are highly correlated and regression resultsare unaffected. As a second alternative measure, I use an indicator variable for the
most widespread religion in the two countries being the same. Because this last
measure uses less of the variation in the data, regression results using it tend to be
weaker statistically, but point in the same direction.
Linguistic distance
My primary measure of linguistic distance is based on the linguistic tree measure usedby Fearon (2003)9. This measure of linguistic distance is intended to capture how
long ago the two languages split from each other, which proxies for both the degree
of dissimilarity of the languages, and the cultural distance that has evolved between
the speakers of the languages.
My primary distance measure is generated as follows. First, each language is
classified as in the 16th edition of Ethnologue. For example, Spanish is classified as
follows:
- Indo-European
- Italic
- Romance
- Italo-Western
- Western
- Gallo-Iberian
- Ibero-Romance
- West-Iberian
- Castilian
- Spanish
9Whom I thank for kindly sharing his data with me.
The significance of 15 here is that this is the maximum number of nodes any one
language has in Ethnologue’s classification scheme. The main difference between thetwo measures is that related languages with relatively few nodes in the language tree,
such as Czech and Slovak, are considered relatively close according to my primary
measure (0.2 for Czech and Slovak), but less close according to Fearon’s measure
(0.86 for Czech and Slovak). According to both measures, 80 percent of language
pairs worldwide are distance 1 from each other. The two measures yield similar
results in the regressions.
Genetic distance
I use Spolaore and Wacziarg’s (2009) measure of genetic distance. This distance
is defined at the country-pair level and captures the time elapsed since the two
populations’ last common ancestors. Where the population of a country consists
of more than one genetically distinct group, the population-weighted average over
the different groups is used.
Hofstede’s (1980, 2001) cultural distance measure
My first survey-based measure of cultural distance is the variance-adjusted average
of Hofstede’s (1979, 1980, 1982, 1983, 2001) four cultural dimension measures:
power distance, uncertainty avoidance, individualism, and masculinity.10 These four
dimensions were generated from surveys of 88,000 IBM employees in 53 different
countries. They relate especially to values in the workplace, but are closely tied in to
basic anthropological and societal issues (Hofstede and Bond, 1984).
The first dimension is “power distance”, defined as “the extent to which less
powerful members of institutions and organizations accept that power is distributed
10This method of combining Hofstede’s dimensions was used previously by studies including Kogutand Singh (1988) and Ng, Lee and Soutar (2007).
unequally.” The second dimension is “uncertainty avoidance”, or “the extent to
which people feel threatened by ambiguous situations, and have created beliefs andinstitutions that try to avoid these.” The third dimension is a continuum that ranges
from “individualism”, or “a situation in which people are supposed to look after
themselves and their immediate family only,” to “collectivism”, or “a situation
in which people belong to in-groups or collectivities which are supposed to look
after them in exchange for loyalty.” The fourth dimension is a continuum between
“masculinity”, or “a situation in which the dominant values in society are success,
money, and things,” and “femininity”, or “a situation in which the dominant valuesin society are caring for others and the quality of life.”
My measure of cultural distance based on Hofstede’s cultural dimensions is given
by
HofstedeDistij =1
4
4k=1
(I ki − I k j )2
V ark
(3.5)
where i and j denote countries, k denotes the dimension, I ki is country i ’s value for
dimension k , and V ark is the variance across countries of the index for dimension k .
Differences between countries in these dimensions reflect differences in values,
priorities, and accepted norms. Such differences may hinder translation flows from
the supply side. Furthermore, they may mean original titles written in the countries
are likely to encompass more different world views, which may make them more
demanded in translation because they have no domestic substitutes, or less demanded
because the ideas they contain are less acceptable.
Schwartz’s (1994, 1999) cultural distance measure
My second survey-based measure of cultural distance is based on Schwartz’s (1994,
1999) seven cultural value dimensions. Schwartz’s framework is theory-driven, with
elements derived from earlier work in the social sciences. The first of Schwartz’s
dimensions is “conservatism”, defined as “a cultural emphasis on maintenanceof the status quo, propriety, and restraint of actions or inclinations that might
disrupt the solidary group or the traditional.” Conservatism stands in opposition
to two types of autonomy: autonomy in ideas and thought, called “intellectual
autonomy”, and autonomy in feelings and emotions, called “affective autonomy”.
Intellectual autonomy is defined as “a cultural emphasis on the desirability of
individuals independently pursuing their own ideas and intellectual directions.”
Affective autonomy is “a cultural emphasis on the desirability of individualsindependently pursuing affectively positive experience,” such as pleasure, or an
exciting or varied life.
The next dimension is “hierarchy”, or “a cultural emphasis on the legitimacy of
an unequal distribution of power, roles and resources,” which has clear commonalities
with Hofstede’s power distance dimension. Hierarchy stands in opposition to
“egalitarianism”, or “a cultural emphasis on transcendence of selfish interests in favour
of voluntary commitment to promoting the welfare of others.”
The next is “mastery”, meaning “a cultural emphasis on getting ahead through
active self-assertion,” which opposes “harmony”, defined as “a cultural emphasis on
fitting harmoniously into the environment.”
I follow the approach of Ng, Lee and Soutar (2007), and construct an average
distance on Schwartz’s dimensions analogously to my average using Hofstede’s
dimensions, but where the sum is instead over the seven dimensions.
steady over the period 1949-1999. Applied science is the next most translated field,
with 10.6% of translations overall, and an increase in importance from 6.2% in 1949to 13.9% in 1999. Social science is next largest at 10.3%. The other fields range from
4.9% to 7.4%.
Table 3.3 compares the distribution across fields of translations with that of all
books published. Overall, literature and philosophy titles are most over-translated
relative to original publications; social science is particularly under-translated,
and natural science and history are somewhat under-translated. When restricting
attention to non-fiction titles, applied science is also translated more than expectedgiven the total publication of titles in the field.
The importance of translations in a field relative to total books published is
likely to depend on a number of factors. Translations will occur less the greater
are the foreign language ability of the audience for the field, and the willingness of
the audience to read in foreign languages. For instance, scientific audiences may be
better at reading English than audiences for popular literature, suggesting popular
literature will be translated relatively more than scientific titles; indeed, Table 3.3suggests this is the case.
Translations will occur more in fields where books written in different languages
are poorer substitutes for each other. For instance, it may be that all organic
chemistry textbooks are relatively close substitutes, so such books are translated
infrequently, whereas philosophy titles written in different languages are derived from
different traditions and thus are very poor substitutes for each other, hence are
translated a lot. Such forces are consistent with the low translation rate of natural
science and the high translation rates of philosophy, literature and religion.
The extent to which the same ideas are relevant in different countries will also
affect the relative importance of translations. For example, history books generally
focus on a specific region of the world, and countries tend to be more interested in
across languages, especially titles with the potential to be of interest internationally.
Economies of scale in idea production (e.g. resulting from a thriving intellectualcommunity) and publication (publishing a book involves a relatively large fixed cost
relative to the cost of printing each copy) may be one reason original titles are more
concentrated across languages than is income. The concentration of originals across
original languages may also be partly because speakers of small languages writing in
areas of international interest learn larger languages and publish in them in order to
reach a wider readership. A similar pattern of concentration across languages appears
in another measure of idea creation, namely research and development spending: thelargest 7 industrialized countries accounted for 84% of world R&D spending in 1995,
which is considerably larger than their 64% share of world GDP.11
Table 3.5 and Figure 3.1 paint further details of the original languages from
which various countries translate. Total translations in most Western European
countries were on the rise over the second half of the twentieth century. With the
exceptions of East Germany, the United Kingdom, Malta and Portugal, each Western
European country translated most from English every year data were available. Thepercentage of translations from English was relatively steady in most cases, though
for some countries it increased mildly over time. French and German were also
heavily translated in most of these countries; Russian was translated very little. The
United Kingdom translated most from French, and Portugal from Spanish, French,
or English.
The source languages of translations published by Central and Eastern European
countries were more mixed: many translated most out of their own languages orRussian, though by 1999 most were translating most out of English. Several of
these countries show two distinct periods when translation patterns changed. First,
after the death of Stalin in 1953 and with the Khrushchev Thaw that followed,
translations from Russian declined somewhat in importance and translations from
English increased. Second, a similar change of much greater magnitude occurredwith the collapse of Communism in 1989. These changes are less evident in Turkey,
Yugoslavia and Albania, which were outside the sphere of Soviet influence for most
of this period.
Most South American countries translated most from English; French was also
translated heavily. Over the period, German replaced French as most translated in
USA. Total translations grew steadily in American countries such as the USA, Canada
and Brazil, but fell in Argentina.
English and French were the languages most translated in Africa. English was
most translated in most of Asia, and grew considerably percentage-wise in the
prominent case of Japan, which also increased its total translations steadily over
the period. East Asia translates mostly from English, and to a modest degree from
French and German. India and Israel translated considerably from Russian until the
collapse of communism, at which point such translations fell off.
Appendix Figure A.1 shows translations by continent and field from each of the
languages English, French, German, and Russian over time. A few main patterns
are evident. The dominance of English as an original language holds across fields,
continents and time. Russian is an important source of natural science titles and, to
a lesser extent, applied science titles, especially in America. However, translation of
other Russian titles is relatively limited outside Eastern Europe in the Communist
era. French and German are relatively similarly translated, though French tends to
be more translated in Western Europe and German in Eastern Europe, especially
in the fields of natural, applied and social science, and philosophy. The most
dramatic change over time is the collapse in translations from Russian and increase in
translations from English, French and German in Eastern Europe upon the collapse
of Communism; these changes occurred right across fields.
good, with the diversity of translated titles demanded by a country increasing more
than proportionately with income.The main country of a language, such as Germany (for German), but not
Switzerland, translates considerably more than the secondary countries of the
language. This could be a result of such countries tending to have larger publishing
industries, and translating titles that are exported to other countries that speak the
same language. For example, many books for sale in Australia and New Zealand were
published in the US or UK.
Columns 3 to 6 of Table 3.6 show that countries with more educated populations,as measured by (i) the proportion of adults aged 25+ with post-school education or
(ii) the average years of schooling of the population aged 25+, translate more, but
this is entirely due to their higher incomes.12
Table 3.7 shows that landlocked countries translate considerably more than
countries with a coastal border. This seems counterintuitive given we usually consider
landlocked countries to be more isolated. However, column 8 of Table 3.7 suggests
this effect is driven by cross-continent differences. Specifically, most of the landlockedcountries in the sample are European, and European countries tend to be heavy
translators, yet landlocked European countries translate no more than European
countries with coastlines. More open countries, as measured by trade as a percentage
of GDP, actually translate less. This effect is statistically significant when using
exports to measure trade, but is smaller in magnitude and insignificant when using
imports. Again, this effect is driven by variation between continents.
Columns 4 and 5 of Table 3.7 suggest that more democratic and less autocraticcountries translate more, though the effects are not statistically significant.13 The
magnitude of the coefficients suggests a country with the highest level of democracy
12Education data are from Cohen and Soto (2007).13The democracy and autocracy variables are from the Polity IV data set; both vary on a scale of
will translate 65% more than a country with the lowest level of democracy.
The number of official languages in a country is not significantly correlated withthe number of titles the country translates. This is perhaps surprising given each title
translated in a country with more languages becomes available to only some fraction
of the population, so more translations are required to generate the same access to
foreign titles. However, this effect may be counterbalanced by the fact each translated
title faces a smaller market in the country, so translation of any individual title is less
worthwhile.
Even after controlling for population, GDP per capita, and openness, Muslimcountries translate less than other countries. For instance, relative to Roman
Catholic countries, they translate 73% fewer titles. European countries translate more
than African, Asian, Middle Eastern and American countries, and Pacific countries
translate less.
3.2.6 Western Europe translates quickly
I next document patterns in the data on age of translated titles. In the years 1998-
2000, the bulk of translated books were translated within 10 years of publication,
with a mode of 1 to 2 years, but some countries systematically translate faster than
others. Figure 3.4 shows speed varies also considerably by original language. On
average, English and Italian titles are translated relatively quickly, and German and
French titles more slowly; these patterns vary substantially across countries, as shown
by Appendix Figure A.2. For each translating country, this figure shows a kernel
density plot of the age of titles at translation for translations from English, French,
German and Italian. Only original languages with at least 15 translations in the
country during the time period of interest are plotted.
Western European countries generally translate these four languages quickly, but
even within Western Europe there is significant variation: Spain is relatively slow,
translate slower (though these results are mostly not statistically significant). A
potential explanation is that countries with larger populations have more ideas(books) produced (published) internally, making foreign books less important.
Columns 1 and 6 of Table 3.9 show that landlocked countries translate significantly
faster than non-landlocked countries. Much of this effect acts through trade: countries
that trade more relative to their GDP, especially import more, translate faster, and
controlling for exports or imports as a fraction of GDP decreases the coefficient on
landlocked and turns it insignificant. Across specifications, a 10 percent increase in
imports as a fraction of GDP corresponds to a 3 to 4 percent decrease in median bookage at translation. The importance of trade for translation speed could suggest that
countries for which trade is more important are more outward-looking, and either
more interested in foreign countries, or more aware of what is currently occurring in
them.
Countries where the dominant religion is an Indian or other religion may translate
somewhat slower than Roman Catholic, atheist, Eastern Orthodox, Muslim or
Protestant countries. Continents do not differ significantly in how quickly theytranslate.
s : Th i s t a b l e pr e s en t s t h e p er c en t a g e of or i gi n al t i t l e s an d t r an s l a t i on s p u b l i s h e d i n1 9 9 9 t h a t f e
l l i n t o e a ch fi el d .Th e
s am p
l ei n cl u d e s d a t af r om al l a v ai l a b
l e c o un t r i e s .Th e d a t a on p u b l i c a t i on s of or i gi n al t i t l e s ar ef r om
French French French German German French German German German
8.9% 7.9% 5.9% 10.0% 13.8% 10.4% 13.0% 6.9% 10.3%
Italian Russian Italian Spanish Spanish Italian Italian Italian Italian
3.1% 2.5% 1.9% 2.4% 5.0% 2.6% 6.7% 2.5% 6.9%
Spanish Italian Swedish Italian Italian Ancient Greek Spanish Spanish Latin
2.5% 1.7% 1.4% 2.4% 3.8% 2.0% 5.7% 2.5% 4.2%
Swedish Danish Spanish Finnish Latin Latin Dutch Swedish Spanish
1.4% 1.7% 1.1% 1.4% 1.5% 1.8% 1.3% 1.8% 2.8%
Russian Estonian Danish Russian Russian Spanish Danish Russian Arabic
1.3% 1.6% 1.0% 1.4% 1.5% 1.1% 1.2% 1.6% 1.7%
Dutch Spanish Finnish Swedish Polish Dutch Catalan Japanese Hebrew
0.9% 1.5% 0.7% 1.1% 1.2% 0.7% 1.1% 1.0% 1.7%
Latin Macedonian Dutch Dutch Dutch Chinese Russian Dutch Ancient Greek
0.8% 1.4% 0.7% 1.1% 1.0% 0.7% 1.0% 1.0% 1.7%Danish Swedish Russian Danish Danish Russian Swedish Norwegian Sanskrit
0.8% 1.4% 0.4% 0.9% 0.9% 0.7% 0.9% 0.7% 1.1%10
6
7
8
9
2
3
4
5
8
9
10
1
4
5
6
7
10
1
2
3
1
2
3
4
5
6
7
8
9
Notes: This table presents the ten most translated original languages and their percentagesof total translations for each field of translations. Panel A presents data in total for everyfifth year from 1949 to 1999; Panel B presents data for 1959; Panel C presents data for1999. The sample includes, for each year, data from all available countries.
Table 3.5: Most translated original language by country
Year
Country Total 1949 1959 1969 1979 1989 1999
Western Europe
Austria English English English English English English English
58.6% 51.5% 58.1% 58.4% 55.8% 68.1% 60.1%
Belgium English English English English
49.6% 57.1% 38.9% 47.7%
Switzerland English English English English
47.3% 45.8% 46.2% 49.8%
East Germany Russian Russian Russian
36.7% 38.0% 30.5%
Germany English English English English English English
60.2% 43.1% 55.3% 59.2% 58.2% 64.8%
Denmark English English English English English English English
57.7% 57.4% 63.3% 58.8% 53.6% 62.1% 62.4%
Spain English English English English English English English
48.8% 54.9% 45.5% 33.7% 46.5% 50.1% 53.2%
Finland English English English
56.0% 51.4% 62.6%France English English English English English English English
60.4% 60.8% 48.5% 52.5% 59.8% 65.8% 60.0%
UK French French French
25.0% 25.0% 27.2%
Greece English English English English
39.2% 31.0% 49.7% 31.3%
Iceland English English English English
55.1% 41.1% 53.2% 65.7%
Italy English English English English English English English
47.4% 45.2% 40.6% 36.4% 49.7% 47.4% 59.6%
Malta English English Italian
40.0% 66.7% 33.3%
Netherlands English English English
67.5% 63.8% 73.8%
Norway English English English English English English English
64.5% 61.9% 68.4% 69.2% 56.0% 62.8% 69.9%
Portugal English Spanish French
37.2% 39.0% 37.9%
Sweden English English English English English
65.1% 55.8% 64.4% 62.5% 67.9%
Central and Eastern Europe
Albania Albanian Russian Albanian Albanian Albanian English
26.6% 52.8% 63.2% 48.9% 35.0% 27.2%
Bulgaria Russian Russian Russian Russian Russian Russian English
32.9% 82.2% 48.7% 37.9% 38.4% 27.6% 68.2%
Czechoslovakia Czech Russian Czech Czech Czech
25.6% 27.4% 28.7% 25.1% 29.4%
Hungary English Russian Russian Hungarian Hungarian English English
27.0% 67.0% 24.6% 41.0% 38.1% 32.5% 55.4%
Poland English Russian English English English English English
42.2% 48.9% 21.9% 28.0% 17.9% 34.8% 65.2%
Romania Romanian Russian Romanian R omanian R omanian English
32.5% 43.9% 40.8% 48.0% 33.9% 44.2%
USSR Russian Russian Russian Russian Russian
50.6% 43.7% 44.1% 54.3% 52.1%
Turkey English English English English
52.9% 42.9% 46.7% 55.7%
Yugoslavia Serbo-Croatian Russian English Serbo-Croatian Serbo-Croatian English
26.4% 38.5% 24.6% 26.6% 24.8% 28.8%
Australasia
Australia English French German Ancient Greek
19.2% 22.2% 33.3% 33.3%
New Zealand English English Tokelauan
59.0% 70.6% 50.0%
Notes: This table presents the most translated original language in each translating countryand its percentage of total translations for the country. Data in the “Total” column are forevery fifth year from 1949 to 1999, where available.
Most translated original language by country continued
Year
Country Total 1949 1959 1969 1979 1989 1999
America
Argentina English English English English English
52.9% 45.1% 57.9% 56.9% 52.9%
Brazil English English English English English English
64.6% 48.7% 56.1% 68.2% 66.8% 66.3%
Canada English English English English
73.7% 75.7% 73.6% 72.6%
Chile English English English
51.5% 65.9% 50.9%
Colombia English English English English
75.5% 79.7% 66.7% 80.9%
Peru Spanish Spanish English English Portuguese Machiguen
39.6% 72.2% 36.4% 39.3% 25.0% 26.3%
Suriname Saramaccan Aukan
20.4% 33.3%
US French French French German German German German
22.6% 31.9% 32.9% 26.8% 22.3% 20.8% 19.3%
Venezuela English English English
40.4% 55.6% 54.5%
Africa
Egypt English English English English
77.1% 80.3% 75.0% 77.0%
Madagascar English French
50.0% 42.9%
Tunisia French French French French
64.9% 62.5% 75.0% 100.0%
Asia
China English English English
70.8% 80.2% 65.3%
India English English English English English
37.3% 44.9% 35.7% 29.9% 35.3%
Israel English English English English English English
53.5% 35.5% 33.5% 53.2% 75.3% 79.2%
Japan English English English English English English
69.1% 46.9% 58.6% 54.1% 69.2% 80.5%
South Korea English English English English English English75.4% 62.6% 50.8% 76.4% 72.5% 75.9%
Kuwait English English English English
36.6% 43.8% 31.2% 60.0%
Sri Lanka English English English English English
60.1% 73.7% 47.4% 45.7% 59.7%
Myanmar English English English English
81.2% 82.1% 36.2% 90.5%
Malaysia English English English English
80.1% 81.8% 78.1% 77.5%
Pakistan English English English Urdu Arabic
56.1% 62.2% 29.9% 50.0% 44.4%
Philippines English English English English
67.7% 58.3% 92.3% 60.0%
Saudi Arabia Arabic English English Arabic
43.8% 100.0% 100.0% 50.0%
Singapore English Malay English English Chinese
50.0% 100.0% 87.5% 71.4% 33.3%
Syria English English English English
41.5% 47.1% 34.4% 51.7%
Thailand English English English English
88.7% 79.8% 91.5% 88.3%
Vietnam French French French
42.7% 56.9% 35.6%
Notes: This table presents the most translated original language in each translating countryand its percentage of total translations for the country. Data in the “Total” column are forevery fifth year from 1949 to 1999, where available.
Notes: This table presents OLS regression results predicting translations flowing into a
country in a year. The years included are every fifth year from 1949 to 1999.Landlocked
is a dummy for the country being landlocked. The religion dummies are for the mostwidespread religion of the country. The variable Country is the main country of their
main language is an indicator for whether the translating country is the primary countryof their most widespread language. For example, it takes the value 1 for Germany, but0 for Switzerland. Standard errors are clustered at the country level. Asterisks denotesignificance at: * p<0.10, ** p<0.05, *** p<0.01.
of p o p ul a t i on wi t h p o s t - s c h o ol e d u c a t i on
-2 . 6 6 7 * * *
- 0 . 3 6 1
0 . 9 5 6 * * *
0 . 3 3 0 * *
( 0 . 8 8 3 )
( 0 . 7 8 6 )
( 0 .2 1 9 )
( 0 .1 4 3 )
A v e r a g e
y e a r s of s c h o ol i n g
- 0 .1 2 5 * * *
- 0 . 0 1 6
0 . 0 4 0 * * *
0 . 0 0 8
( 0 . 0 3 3 )
( 0 . 0 6 2 )
( 0 . 0 0 8 )
( 0 . 0 1 2 )
Or i gi n a l
l a n g u a g e f i x e d e f f e c t s
N o
Y e s
Y e s
Y e s
Y e s
Y e s
N o
Y e s
Y e s
Y e s
Y e s
Y e s
R - S q u ar e d
0 .1 9 2
0 .2 9 9
0 .1 6 5
0 .2 3 1
0 .1 8 1
0 .2 3 0
0 . 3 2 6
0 . 3 8 4
0 .2 9 5
0 . 3 9 5
0 .2 9 3
0 . 3 8 6
O b s er v a
t i o n s
1 6 3
1 6 3
9 7
9 7
9 7
9 7
1 6 4
1 6 4
9 8
9 8
9 8
9 8
C o u n t r i e s
5 0
5 0
2 9
2 9
2 9
2 9
5 0
5 0
2 9
2 9
2 9
2 9
M e d i a n a g e o f t i t l e s a t t i m e o f t r a n s l a t i o n ( l n )
P r o p or t i o n o f t i t l e s t r a n s l a t e d w i t h i n2 y e ar s o f f i r s t p u b l i c a t i o n
N o t e
s : Th e t a b l e pr e s en t s r e s ul t s of O
L S r e gr e s s i on s pr e d i c t i n g t h e s p
e e d of t r an s l a t i on.An o b s er v a t i oni n t h e s er e gr e s s i on s
i s an
or i gi n al l an g u a g ei n a t r an s l a t i n g c o un t r y.Th e or i gi n al l an g u a g e s i n cl u d e d ar eEn gl i s h ,F r en ch , G er m an an d I t al i an.
Onl y
n on-fi c t i on t i t l e s t r an s l a t e d i n t o an offi ci al l an g u a g e of t h e t r an
s l a t i n g c o un t r y ar ei n cl u d e d .Th e e d u c a t i on v ar i a b l e s
ar e d efi n e d f or t h e p o p ul a t i on a g e d
2 5 +. O b s er v a t i on s ar e w ei gh t e
d b y t h en um b er of t r an s l a t i on
s c on t r i b u t i n g t o t h e
m e a s ur e of s p e e d . S t an d ar d er r or s ar
e cl u s t er e d a t t h e c o un t r yl e v el .
A s t er i s k s d en o t e s i gni fi c an c e a t :
s : Th e t a b l e pr e s en t s r e s ul t s of O
L S r e gr e s s i on s pr e d i c t i n g t h e s p
e e d of t r an s l a t i on.An o b s er v a t i oni n t h e s er e gr e s s i on s
i s an
or i gi n al l an g u a g ei n a t r an s l a t i n g c o un t r y.Th e or i gi n al l an g u a g e s i n cl u d e d ar eEn gl i s h ,F r en ch , G er m an an d I t al i an.
Onl y
n on-fi c t i on t i t l e s t r an s l a t e d i n t o an offi ci al l an g u a g e of t h e t r an
s l a t i n g c o un t r y ar ei n cl u d e d .L a
n d l o c k e d i s a d umm y
f or t h e c o un t r y b ei n gl an d l o ck e d .Th
er el i gi on d ummi e s ar ef or t h em o s t wi d e s pr e a d r el i gi on of t h e c o un t r y. O b s er v a t i on s
ar e w ei gh t e d b y t h en um b er of t r an
s l a t i on s c on t r i b u t i n g t o t h em e a s ur e of s p e e d . S t an d ar d er r or s ar e cl u s t er e d a t t h e
c o un
t r yl e v el .A s t er i s k s d en o t e s i gni fi c an c e a t : * p < 0 .1 0 ,* * p < 0 . 0 5 ,
and other types of dissimilarities between countries may inhibit translation flows, for
instance by increasing search and transaction costs. By studying the relationshipbetween measures of distance between countries and translation flows, I shed light on
an important type of impediment to the free international diffusion of ideas.
The second is to shed light on the factors underlying the negative relationship
between distance and trade in goods. This negative relationship is a robust finding
in international economics, but the driving factors behind it remain imperfectly un-
derstood.1 The most obvious contributing factor to the relationship is transportation
costs, but an increasing literature demonstrates that transportation costs account for
only a fraction of the total distance effect.2 For example, in a recent paper, Feyrer
(2011) uses time-varying exogenous variation in effective distance generated by the
temporary closure of the Suez Canal to estimate that only half the elasticity of trade
with respect to distance is driven by transportation costs.
An alternative approach to shedding light on the factors underlying the distance
effect in trade is to consider trade in goods for which specific costs are known to be
absent. Blum and Goldfarb (2006) study how distance affects trade in digital goods
consumed over the internet, which have no transportation, time, or distribution costs.
In this paper I consider another setting in which transportation, time, and much of
the distribution costs are negligible by studying how distance affects the translation
of books. Specifically, only a single copy of the title, in digital or hard copy form,
must travel between the countries in order for a translation to occur. Translations
thus have effectively zero transportation costs, both direct (freight, insurance) and
indirect (e.g. holding cost for the goods in transit). Translations are also largely free
from several other trade costs (discussed in Anderson and van Wincoop (2004): they
avoid border-related costs, policy barriers such as tariffs and quotas, and many legal
1e.g., Disdier and Head (2008), Blum and Goldfarb (2006), Feyrer (2011).2Anderson and van Wincoop (2004).
and regulatory costs. In general, none of the costs related to the physical movement
of goods apply to book translations.
I use data on translations published in a large number of countries for the period
1949 to 2000, which I digitized from Unesco’s Index Translationum. The data are
described in detail in chapter 3 of this dissertation.
I study the effect of physical distance on translation flows within an augmented
gravity framework, where the translation flow between a pair of countries is affected
by characteristics of the two countries (such as GDP) and the distance between them
(section 4.3.1). Even though translations have zero transportation costs, I find they
decrease significantly with distance. However, I estimate the elasticity of translations
with respect to distance to be 0.3 to 0.5 during the 1990s, which is considerably
smaller than the equivalent elasticity for trade found in the literature, which usually
ranges from 1.08 to 1.24.3 Under the assumption that the non-transportation costs
faced by translations vary with distance in the same way as the equivalent costs for
trade, the magnitudes of these coefficients suggest that roughly half to three quarters
of the elasticity of trade with respect to distance is the result of transportation costs;
comparisons with results using a more similar estimation method, from Santos Silva
and Tenreyro (2006), decrease this range to a third to three fifths. Although my
method is very different, these results are comparable to Feyrer’s (2011) estimate
that half the elasticity of trade with respect to distance is the result of transportation
costs.
Translations (and similarly trade) may decrease with distance for both supplyand demand reasons. Supply frictions such as the search and information costs of
identifying titles worth translating are likely to increase with distance, as are the
various costs of negotiating contracts.4 Translations (or trade) may also fall off with
distance because distance is correlated with tastes, meaning closer countries caterbetter than more distant countries to local tastes in books (or products).5
Furthermore, these supply and demand factors may not remain constant over
time. I next examine how the relationship between distance and translations changed
over the latter half of the twentieth century (section 4.3.2). I find it fell significantly
over the period 1949-1999, especially in the last two decades. This result contrasts
with the puzzling finding that the effect of distance on trade in goods did not
decline over this period, instead remaining high.6
However, it is consistent with
the finding that the effect of distance on patent citations has fallen over time;7 both
the translations and patent citation results are easily explicable by the decrease over
time in communication and information costs.
In comparisons across fields of the effect of distance on translations, I find the
distance effect is larger for titles in exact and applied science than for titles in fields
where taste ought to play a larger role, such as philosophy, arts and literature (section
4.3.3). This suggests a greater importance for contracting or search and information
costs relative to consumer tastes in driving the distance effect. These results also
demonstrate that distance matters even for translations of “economically useful”
titles, not just for titles that may be considered largely consumption goods.
To further explore the roles of contracting costs, search and information costs, and
tastes, I add controls for religious distance, linguistic distance, genetic distance, and
4That is, both trade in goods and translations are subject to the costs that are related to forminga contract between parties in different countries. These costs may vary with distance, and includethe time and legal costs of negotiating and enforcing the contract, direct and indirect costs relatedto transacting between currencies, and the costs of overcoming any language barriers that existbetween the parties.
5Blum and Goldfarb (2006) suggest this factor plays a significant role in the distance effect fortaste-dependent digital goods.
6Disdier and Head (2008).7MacGarvie (2005) and Griffith, Lee and van Reenen (2007).
survey measures of differences in cultural values, all of which are expected to capture
some element of cultural differences (section 4.3.5). Cultural differences could affecttranslations through the supply channel by increasing contracting costs, or through
the demand channel by decreasing the similarity of taste for books. Religious and
linguistic distance and Hofstede’s (1980, 2001) survey measure of cultural distance are
negatively related to translations, but physical distance remains important even when
these controls are added. Furthermore, adding these controls reduces the elasticity
of translations with respect to physical distance by at most a quarter. This suggests
that cultural differences contribute to distance-varying contracting costs or to demandthat prefers titles written in nearby countries, but that other distance-varying costs
play a larger role.
I next allow the elasticity of translations with respect to distance to vary by the
level of development of the translating country, as measured by GDP per capita or
urbanization (section 4.3.6). I find a strong differential effect: the effect of distance
is 89% weaker for a translating country on the 75th percentile of GDP per capita
among countries in my data than for a country on the 25th
percentile. The fact thattranslations flowing into poorer countries are more affected by distance may have
important implications for the international diffusion of knowledge. Specifically, it
suggests that countries that are further from the world knowledge frontier, and thus
that can benefit most from adopting ideas that already exist elsewhere, are actually
less able to access these ideas.
Finally, to further explore the role of supply-side frictions relative to taste in the
relationship between distance and translations, I study the speed with which titles
are translated (section 4.4). Time is required for foreign publishers to discover that
a title exists and is worth translating, and to negotiate the rights to translate it; the
greater the supply-side frictions, the longer this will take. However, a title in enough
demand to be worth translating in five years time is likely to also be worth translating
now. The speed of translations thus likely reflects frictions such as information and
contracting costs, rather than tastes. I find that distance substantially decreases thespeed with which translations are published: a 50 per cent increase in the distance
between the countries corresponds to a 3 percentage point decrease in the proportion
of translations published that are of titles originally published no more than two years
previously. Because the speed and quantity of translations are intimately related,
this suggests that information and contracting costs also play a substantial role in
the effect of distance on quantity of translations.
4.2 Empirical strategy
To shed light on the determinants of translation flows and the factors underlying
the negative relationship between distance and trade, in section 4.3 I investigate the
relationship between geographic distance and translations, which are not subject to
any trade costs related to physically moving goods. To decompose the distance effect
I find, I add further controls for various types of distance or dissimilarity between the
countries.
The basic specification is a gravity model in multiplicative form estimated using
the pseudo-maximum-likelihood (PML) estimator recommended by Santos Silva and
Tenreyro (2006) for the gravity equation specifically and constant-elasticity models in
general. Santos Silva and Tenreyro consider a constant elasticity model of the form
T ij = α0Y α1
i
Y α2
j
Dα3
ij
ηij (4.1)
where ηij is a multiplicative error term with E (ηij|Y i, Y j, Dij) = 1 and where ηij
is assumed to be statistically independent of the regressors. They show that if
ηij is heteroskedastic in a manner that depends on the regressors, then lnηij is
not independent of the regressors, and thus linearizing equation (x) and estimating
it by OLS leads to inconsistent estimates. In fact, they demonstrate that this
heteroskedasticity is usually present and substantial in gravity models of trade,so estimating the relationship in multiplicative form using their PML estimator is
preferred. The nature of translation data suggests such heteroskedasticity is also likely
to be present here, so I use their PML estimation technique. A further advantage
of this method is that it has no difficulty with observations where the value of the
dependent variable is zero.
The equation I thus estimate is:
transijt = αdistφijeβiteγ jtijt (4.2)
the more familiar linearized form of which is:
ln(transijt) = α + φln(distij) + β it + γ jt + ν ijt (4.3)
where α ≡ ln(α) and ν ijt ≡ ln(ijt). Here transijt is the number of translations into
language-in-country pair i , from language j , in year t , distij is the geographic distance
between the main country of language j and the country denoted by i , the β s are
time-varying fixed effects for target language-in-country, the γ s are time-varying fixed
effects for original language, and is a error term with mean 1. The coefficient of
interest is φ, which is the elasticity of translation flows with respect to geographic
distance.
I run specifications where I control for the population and GDP per capita of the
original and translating countries instead of including time-varying origin and target
fixed effects. However, Anderson and van Wincoop (2003) show such a specification
is likely to result in a biased estimate of the coefficient on distance because it suffers
omitted variable bias. In the translation context, it does not account for the average
barriers to translation from all possible original languages faced by a country. These
are likely to be correlated with average distance from potential original countries,
and thus with the distance from any one original country, so failing to account forthem introduces bias. Thus my preferred specification includes time-varying origin
and target fixed effects.
I also augment this model by including measures of non-physical distance between
the countries, such as differences in culture.
4.2.1 Original languages and target languages and countries
for gravity model
An observation in the gravity equation I estimate is an original language, a target
language in a country, and a year. Two questions then arise. First, from what set of
original languages should translations be included, and should this vary by translating
country? Second, into which target language or languages in each country should
translations be included?
I do not allow the set of original languages to vary by translating country. That is,
each target language in a translating country in a year contributes the same number of
observations to the regression, one for each language in a set of original languages that
does not vary by country. The advantage of this method is that it does not impose
any priors about which countries will translate from which languages. However, it
does mean many measured translations flows are zero. The set of original languages
out of which I consider translations in my primary specification is the set of the most
widely spoken 100 languages as listed by Ethnologue. From these languages, I drop
those out of which translations are never published.
One option for target languages in translating countries would be to include an
observation for translations into each possible language in every country. However,
most of these flows will be zero. In fact, most would also not signify a relevant transfer
native speakers are in the country, the population of native speakers in the country
relative to worldwide, and whether the language is national or official in the country.Very small countries (e.g. Monaco) are not counted as major unless they are the main
country of the language. The major countries for each original language are listed
in Appendix A. I then set the original country of translations in country C from
language L to be language L’s major country that is physically closest to country C.
4.3 How distances affect translations
In this section, I study how bilateral translation flows are affected by distance between
countries. This is a setting with many commonalities with trade in goods, but in which
transportation, time, and much of the distribution costs are negligible. Specifically,
because only a single copy of the title must travel between the countries in order
for a translation to occur, translations have effectively zero transportation costs,
both direct and indirect. They are also largely free from border-related costs, policy
barriers such as tariffs and quotas, and many legal and regulatory costs. That is,
translations face zero costs related to the physical movement of goods. Translations
are, however, expected to be subject to all the costs of contracting between parties
in different countries, plus search and information costs, that trade in goods face
and that may vary with distance. In addition, both trade and translations may occur
more between closer countries because consumers in these countries have more similar
tastes, thus more demand for each other’s books or products. Studying how distance
affects translations thus sheds light on the factors beyond transportation costs that
contribute to the negative relationship between distance and trade.
My estimation framework is an augmented gravity model, in which (directional)
translations between two countries depend on the economic sizes of the countries, and
the physical distance between them. I assume a constant elasticity functional form,
and estimate the model by PML as described in section 4.2. I add controls such as
the cultural distance between the countries to measure the extent to which countriestranslate more from their neighbors because they are more culturally similar to them.
4.3.1 The negative distance effect: Neighboring countries
translate more from each other than from distant
countries
In a basic gravity specification with physical distance as the only distance measure,
presented in Table 4.1, I find a strong negative correlation between the number of titles
translated and the physical distance between the original and translating countries.
Appendix Table B.1 presents the same specifications, but uses OLS and predicts the
natural log of the number of translations plus 1. The strong negative correlation
is again present, though the coefficients on distance are smaller in magnitude. The
data used in these regressions are a panel of the years 1994 and 1999, a short enough
period that we expect the relationship between distance and translations to have
remained relatively constant. Column 1 presents the basic gravity specification where
the number of translations flowing from one country to another depends on the
populations of the two countries, their GDP per capita, and the distance between
them. For each target language in a translating country, we consider translations
from the same set of original languages, namely those of the 100 languages most
widely spoken in the world that are ever translated. For each of the 56 countries with
translation data in at least one of the two years, we consider translations into each
of the languages that are official in the whole of the country. To generate distance
measures, I assign each original language to its main country as described in section
4.2.2.
As expected, the population and GDP per capita of the translating country are
positively and significantly correlated with translation flows with elasticities of 0.72
and 0.75 respectively. The elasticity of translations with respect to the populationof the original country is 1.1; the elasticity with respect to the GDP per capita of
the original country is 3.3. This strong relationship between wealth of the original
country and translations suggests the creation of ideas with international relevance is
very much concentrated in rich countries, whereas less rich countries tend to consume
ideas created elsewhere. The OLS version of this regression, presented in column 1 of
Appendix Table B.1, shows these basic covariates have moderate explanatory power:
the R-squared in this regression is 0.15.Column 2 of Table 4.1 adds controls for colonization relationships between the
original country and the translating country in either direction. There are relatively
few of these in the data, particularly because the translating country must have at
least one official language that differs from the language of the colonizer in order
for the pair to appear in the data, and neither direction of colonizing relationship is
significantly correlated with translation flows.
In these first two specifications, the elasticity of translations with respect to
geographic distance is -0.9, suggesting a 10% increase in the distance between two
countries corresponds to a 9% decrease in translation flows between them. However,
Anderson and van Wincoop (2003) show such a specification is likely to suffer omitted
variable bias as explained in section 4.2, so in column 3 I add in time-varying fixed
effects for original language and target language-translating country pairs. The
elasticity falls in magnitude to -0.47, but remains significant.
Next I add controls for the original and translating countries being contiguous,
and the original language being widely spoken in the translating country. Both are
associated with significantly higher translations, but their inclusion doesn’t eliminate
the relationship between distance and translations. The interpretation of these two
effects is similar. Sharing a land border with a country suggests the populations will
both interact more, implying lower search and transaction costs of translating from
each other, and have more similar tastes, implying greater demand for translations.Similarly, mixing geographically with a group that speaks a foreign language can be
expected to stimulate translations from both the demand and the supply sides.
Columns 5 to 10 of Table 4.1 run the same specification as column 4, but vary
the sample of original languages and translating countries in a number of ways.
Column 5 restricts the original languages to those in the top 100 that can be
unambiguously attributed to a single country, which eliminates many of the large
original languages such as English, German and Spanish. The elasticity of translations
with respect to distance increases in magnitude to -1.1 in this specification. Column
6 restricts original languages to the four main “research languages”, namely English,
French, German and Japanese. The magnitude of the correlation is similar, though
significance decreases because of the much smaller sample size. Column 7 uses all
of the top 100 original languages, but attributes each to the country in which it is
widely spoken that is geographically closest to the translating country (as explained
in section 4.2.2), instead of to its main country. The coefficient on distance falls
slightly in magnitude, which suggests geographic proximity to a secondary country
of a language may be a less-than-perfect substitute for geographic proximity to the
main country of the language for the purpose of enhancing idea flows. Column 8
differs from column 4 in that it restricts the sample of target languages in translating
countries to those where the translating country is the main country of the target
language. For example, it includes translations into German in Germany, but excludes
translations into German in Switzerland. The coefficient of interest is unaffected.
Column 9 instead includes all the target languages for each translating country
that are (i) official in at least part of the translating country, (ii) spoken natively
by at least 500,000 people in the country, and (iii) spoken by at least 5% of the
country’s population. Results are again largely unaffected. Finally, column 10 looks
at translation flows only within Europe. That is, it includes original languages in
the top 100 that have a European country as their main country, and translatingcountries that are European. The coefficient on distance increases slightly.
Overall, it seems that in the 1990s a 10 percent increase in distance corresponded
to roughly a 3 to 5 percent decrease in translations, despite translations having
zero transportation costs. This suggests there are significant distance-varying costs
involved in translation, which may relate to search and information, or to the costs of
forming contracts. Geographic correlation of tastes that causes demand to decrease
with distance may also contribute to the distance effect.
This elasticity of 0.3 to 0.5 is significantly lower than those found in the literature
on trade in goods, which generally range from 1.08 to 1.24.9 Under the (admittedly
strong) assumption that the non-transportation costs faced by translations vary with
distance in the same way as the equivalent costs for trade, the magnitudes of these
coefficients suggest that roughly half to three quarters of the elasticity of trade with
respect to distance is the result of transportation costs. However, this comparison
may be confounded by the use of PML estimation in this paper. Where Santos Silva
and Tenreyro (2006) use PML as opposed to OLS to estimate the distance effect on
trade in a model with importer and exporter fixed effects, their coefficient falls from
-1.3 to -0.75. This lower elasticity estimate for trade suggests a third to three fifths
of the distance effect in trade is due to transportation costs. These estimates are in
the same range as the value of a half found by Feyrer (2011) using a very different
4.3.2 The negative correlation between physical distance and
translations decreased over time
To gain further insight into the causes of the effect of distance on translations, I next
estimate how this effect changed over time. Figure 4.1 illustrates how the correlation
between physical distance and translations changed over time. These correlations are
coefficients from regressions of translations on geographic distance, origin and target
fixed effects, and controls as in column 4 of Table 4.1, run separately for each fifth year
from 1949 to 1999. The figure presents the 95% confidence interval of the coefficienton translations for two different sets of translating countries: the solid blue lines are
for the consistent set of 9 countries for which data are available every year; the dashed
red lines are for all the countries for which data are available in the particular year.
In each case, the magnitude of the negative correlation decreased significantly over
the period 1949 to 1999, particularly over the last two decades.
This contrasts with the changes seen in the distance effect in trade, which,
according to Disdier and Head’s (2008) meta-analysis of the results from many
papers, rose mid-century and has remained persistently high since. The decrease
in the inhibitory effect of physical distance on translations over time is consistent
with several causal mechanisms. For instance, the ease of international travel and
communication decreased over this period, and their costs fell. This could have
both weakened the relationship between distance and the search, information, and
transaction costs of translation, and stimulated interest in geographically distant
cultures. If search and information costs are higher on average for books than for
goods, this could explain why the distance effect decreased for translations but not
for trade. Note MacGarvie (2005) similarly finds a decrease in the effect of distance
on patent citations over the period 1980-1995, which is also consistent with such a
4.3.3 Translations of different types of books are affected
differently by physical distance
There are a number of reasons to expect translations of titles in different fields to
be affected differently by physical distance. On the demand side, fields differ in the
extent to which their ideas are region-specific. For instance, history titles frequently
focus on a particular region of the world, thus are likely to be of more interest to
countries in that region. Similarly, religion titles tend to relate to a specific religion,
and thus will be of more interest in countries where that religion is widespread, whichtend to be geographically clustered. Conversely, many natural science ideas (such
as ideas in physics and chemistry) are equally relevant anywhere in the world. In
addition, the degree to which titles written in different languages are substitutes for
each other varies by field. In fields with high substitutability, there may be no reason
to translate from very distant languages because nearby languages are sufficient to
meet demand, thus if costs rise with distance translations may fall off quickly with
distance. In fields with low substitutability, a specific idea can only be sourced fromone language, so distance is likely to play a lesser role in determining translation
flows.
Figure 4.2 shows the coefficients and 95% confidence intervals of the coefficient on
physical distance when translations in each field are regressed on physical distance
and other controls as in column 4 of Table 4.1. Physical distance and translations
are negatively correlated for all fields of translation, though the magnitude of the
correlation varies across fields. Perhaps surprisingly, physical distance has the largestinhibitory effect on translations in the fields of natural science and applied science,
and the smallest in philosophy and arts.
These results by field demonstrate that distance matters even for translations of
“economically useful” titles such as titles in natural and applied science, not just for
titles that may be considered largely consumption goods, such as many philosophy,
arts, or fiction titles. Furthermore, the fact distance has a greater effect for titleswith less of a cultural or taste component suggests taste differences that increase
with distance may have a lesser role in driving the distance effect on translations
relative to supply-side frictions.
4.3.4 Countries with similar physical environments translate
more from each other
The negative relationship between translation flows and physical distance could be
driven by several factors, all of which apply to trade in goods to some extent: search
and information costs involved in identifying foreign titles worth translating; costs
of negotiating rights to translate a title; and tastes for ideas that differ more widely
between more distant countries.
One reason tastes for ideas may be more similar in neighboring regions is that
physical environment (such as climate, terrain, the types of plants that will grow
etc) tends to be more similar in neighboring regions, and the physical environment in
which a society lives might affect the types of ideas that are relevant or interesting to
its members. To estimate the importance of this effect, I augment the basic gravity
model with the difference between countries in altitude profile, biome region profile,
and climate region profile. Column 1 of Table 4.2 presents the results from this
regression. Differences in altitude profile and biome region profile significantly inhibit
translation flows, but together these three differences explain only a modest 6.5% of
the negative correlation between physical distance and translations. The coefficient
on altitude profile difference suggests that, relative to two countries with the same
altitude profiles, two countries with altitude profiles that are only 90% similar will
translate 8% less from each other. However, much of this correlation can be shown
and information costs, play a large role in the distance effect.
The process of globalization over the past half century has made the worldsmaller in many ways; international travel has become cheaper and faster, and
global communications have improved beyond measure. The forces that have allowed
distant cultures to mingle more easily may have decreased cultural barriers to the
flow of ideas. Also plausible is that globalization has caused a reactionary increase in
nationalism that may have actually decreased receptiveness to foreign ideas. It is thus
unclear theoretically how the relationship between cultural distances and translation
flows will have changed over time. Appendix Figure B.1 shows the negative correlationbetween religious distance and translations tended to increase between 1949 and 1999,
while the correlation between linguistic distance and translations tended to decrease.10
4.3.6 Translations published in more developed countries
decrease less with physical distance
To yield further insight into the drivers of the negative correlation between trans-
lations and distance, in Table 4.3 I allow the effect of distance to differ by various
characteristics of the original or translating country. In column 1, I allow the effect of
distance to differ by the wealth of the translating country. I find the effect of distance
is 89% weaker for a translating country on the 75th percentile of GDP per capita
10In column 9 of Table 4.2, I add controls for trade flows in each direction between the original andtranslating countries, in order to see descriptively how trade in ideas (translations) are correlatedwith trade in goods. Note the coefficients on these variables in particular should not be interpretedcausally because of reverse causality and unobserved heterogeneity. For instance, trade flowsbetween countries may cause an increase in idea flows and thus an increase in translations, buttranslations may increase understanding and decrease transaction costs, thus increasing trade. Tradein manufactured goods and translations may also be complements. The coefficients on both importsand exports are positive and significant, and are similar in magnitude: a 10% higher flow of tradein either direction corresponds to a 2.4% higher translation flow. The direction of this effect isconsistent with the causality stories running in either direction. One interesting point to note isthat inclusion of these two trade variables eliminates the negative correlation between distance andtranslations. However, as Appendix Tables B.2 and B.3 show, this was not the case prior to the1990s.
among countries in my data relative to the 25th percentile, or 70% weaker for a country
with urbanization rate on the 75th
percentile relative to the 25th
percentile (column 3).More developed countries can differ from less developed countries across a multitude
of dimensions, making it difficult to establish the causal mechanism behind these
results. For instance, communication technologies tend to be more advanced, reliable,
and widespread in richer countries, which could reduce search and information costs.
However, the fact that translations published in poorer countries are more affected
by distance has potentially important implications for the international diffusion of
knowledge. Specifically, it suggests that countries that are further from the worldknowledge frontier, and thus that can benefit most from adopting ideas that already
exist elsewhere, are actually less able to access these ideas.
Similarly, distance is significantly less important for translations of titles origi-
nating in wealthier countries. The effect of distance is 49% weaker for translations
from original countries on the 75th percentile of GDP per capita relative to the 25th
percentile (column 2); it is 52% weaker for a country with urbanization rate on the
75th
percentile relative to the 25th
percentile (column 4).
Distance is also significantly less important for translating countries that are
more democratic, as shown in column 5. The effect of distance is 44% weaker
for a translating country on the 75th percentile of democracy relative to a country
on the 25th percentile. Such a relationship could be observed if more democratic
countries were less threatened by ideas that differed more from their own than were
less democratic countries. The level of democracy in the original country is not
significantly correlated with the strength of the relationship between physical distance
is physically closest to the translating country, as described in section 4.2.2.11 This
correlation is statistically significant when controlling for the populations and GDPsper capita of the original and translating countries (column 1 in both tables), but loses
significance due to lack of power when origin and target fixed effects are included. A
50 per cent increase in the distance between the countries corresponds to a decrease
of 2.5 percentage points in the proportion of translations that were translated within
two years of first being published when original languages are attributed to their main
countries, and a 3 percentage point decrease when original languages are attributed
to their closest major countries. Columns 2 to 9 of these tables suggest linguisticdistance may decrease the speed of translation and distance may affect less the speed
with which richer countries translate.12
The negative effect of distance on speed of translation suggests translations face
significant supply-side frictions such as information and contracting costs that increase
with distance. These frictions likely play a substantial role in the negative relationship
between quantity of translations and distance.
4.5 Conclusions
In this paper, I study how flows of book translations between countries are correlated
with the physical distance between the countries. Understanding this relationship
is both important for understanding the impediments to the international diffusion
of ideas, and informative about the underlying causes of the negative relationship
between distance and trade in goods. Unlike goods, translations do not face any of the
costs related to physical relocation, though they too are subject to frictions such as the
11The most salient difference between the two is that in the first case translations from Englishare attributed to the USA for European translating countries, whereas in the second case they areattributed to the UK for these countries.
12The lack of power in these regressions makes identifying significant effects of interest difficult.
costs related to negotiating contracts and search and information costs. In addition,
both translations and trade may decline with distance because consumer tastes aregeographically correlated. Studying the determinants of translation flows is thus
informative on the drivers beyond transportation costs of the negative relationship
between trade in goods and distance.
I estimate a gravity-type model in which translation flows are affected by
characteristics of the original and translating countries (such as GDP per capita)
and the distance between them. I find an elasticity of translations with respect to
distance of between -0.3 and -0.5 for the 1990s, which is substantially smaller than
the corresponding elasticity for trade estimated in the literature, suggesting a sizeable
fraction of the distance effect in trade is due to transportation costs.
Several pieces of more refined analysis of the relationship between translations and
distance are consistent with an important role for search and information costs and
a lesser role for demand factors in the negative relationship between translations and
distance. First, the distance effect decreased between 1949 and 1999 for translations
but not trade; this is consistent with information costs, which may be higher for
translations than goods because books are more heterogeneous, and which almost
certainly fell over this period, being an important factor for driving translations.
Second, the distance effect is larger in the fields of natural and applied science, where
tastes are less important, than in the fields of arts, literature and philosophy, which
have a higher cultural component. This is the opposite to what we would expect
if geographically correlated tastes were the main driving factor behind the distance
effect. Third, cultural distance between countries does inhibit translation flows, but
accounts for relatively little of the overall distance effect, suggesting non-cultural
factors play a large role. Finally, the speed with which titles are translated, which is
likely to largely capture supply frictions as opposed to demand factors, also decreases
Figure 4.1: The negative correlation between geographic distance andtranslations decreased over time
Notes: This figure shows the 95% confidence interval of the coefficient on geographic distancein regressions of the number of translations (ln) on distance (ln) and other controls as incolumn (4) of Table 4.1, run separately by year. The solid blue line is for the consistentset of 9 countries for which data are available each year; the dashed red line is for all thecountries for which data are available in any one year.
Figure 4.2: The negative correlation between geographic distance andtranslations by field
Notes: This figure shows the point estimate and 95% confidence interval of the coefficienton geographic distance in regressions of the number of translations (ln) on distance (ln) andother controls as in column (4) of Table 4.1, run separately by book field. Data are for the
N o t e s : T h i s t a b l e p r e s e n t s t h e r e s u l t s o f P M L r e g r e
s s i o n s ( a s d e s c r i b e d i n s e c t i o n 4 . 2 ) o f t h e n u m b e r o f t r a n s l a t i o n s f r o m
a n o r i g i n a l l a n g u a g e
t o a t a r g e t l a n g u a g e i n a t r a n s l a t i n g c o u n t r y i n a y e a r . T h e s a
m e o r i g i n a l l a n g u a g e s a r e i n c l u d
e d f o r
e v e r y t a r g e t l a n g u a g e ; z e r o v a l u e s a r e i n c l u d e d i n t
h e e s t i m a t i o n , a s a l l o w e d b y t h e P M L p r o c e d u r e . T h e s e t o f o r i g i n a l
l a n g u a g e s a n d t a r g e t l a n g u a g e / c o u n t r i e s i n c l u d e d v a r y b y c o l u m n . T h e y e a r s i n c l u d
e d a r e 1 9 9 4 a n d 1 9 9 9 .
T h e o r i g i n a l l a n g
u a g e s i n c o l u m n s 1 - 4 , 8 a n d 9 a r e a l l t h o s e l a n g u a g e s i n t h e 1 0 0 m
o s t w i d e l y s p o k e n l a n g u a g e s w o r l d w i d e
t h a t a r e e v e r t r a n s l a t e d ( t o p 1 0 0 l a n g u a g e s ) . T h e o
r i g i n a l l a n g u a g e s i n c o l u m n 4 a r e t h o s e o f t h e t o p 1 0 0 l a n g u a g e s t h a n
c a n u n a m b i g u o u s l y b e a s s i g n e d t o a s i n g l e o r i g i n a l c o
u n t r y . T h e o r i g i n a l l a n g u a g e s i n
c o l u m n 5 a r e t h e f o u r m a j o r “ r e s e a r c h
l a n g u a g e s ” , n a m e l y
E n g l i s h , F r e n c h , G e r m a n a n d J a
p a n e s e . T h e o r i g i n a l l a n g u a g e s i n c o l u m n 7 a r e t h e t o p 1 0 0 l a n g
u a g e s ,
b u t t h e o r i g i n a l c o u
n t r y u s e d f o r e a c h l a n g u a g e i s
t h e g e o g r a p h i c a l l y c l o s e s t c o u n t r y w h e r e t h e l a n g u a g e i s w i d e s p r e a d ,
r a t h e r t h a n t h e m a i n c o u n t r y o f t h e l a n g u a g e . T h e o
r i g i n a l l a n g u a g e s i n c o l u m n 1 0 a r e t h o s e i n t h e t o p 1 0 0 l a n g u a g e
s t h a t
a r e E u r o p e a n .
T h e t a r g e t l a n g u a g e s i n c o l u m n s 1 - 7 a r e t h e l a n g u a g e s t h a t a r e o ffi c i a l i n t h e w h o l e o f t h e t r a n s l a t i n g c o u n t r y
. T h e
t a r g e t l a n g u a g e s i n c o l u m n 8 a r e t h e l a n g u a g e s t h a t a r e o ffi c i a l i n t h e w h o l e o f t h e t r a n s l a t i n g c o u n t r y , a n d f o r w h i c h t h e
t r a n s l a t i n g c o u n t r y
i s t h e m a i n c o u n t r y o f t h e l a n g u
a g e ( e . g . G e r m a n i n G e r m a n y ,
b u t n o t G e r m a n i n S w i t z e r l a n d ) . T h e
t a r g e t l a n g u a g e s i n
c o l u m n 9 a r e t h e l a n g u a g e s t h a
t a r e 1 ) o ffi c i a l i n a t l e a s t p a r t
o f t h e t r a n s l a t i n g c o u n t r y , 2 ) s
p o k e n
n a t i v e l y b y a t l e a s t
5 0 0 , 0 0 0 p e o p l e i n t h e c o u n t r y , a n d 3 ) s p o k e n b y a t l e a s t 5 % o f
t h e c o u n t r y ’ s p o p u l a t i o n . T h e
t a r g e t
l a n g u a g e s i n c o l u m n
1 0 a r e t h e l a n g u a g e s t h a t a r e o ffi c i a l i n t h e w h o l e o f t h e t r a n s l a t i n g c o u n t r y , f o r E u r o p e a n c o u
n t r i e s
o n l y . S t a n d a r d e r r o r s
a r e r o b u s t . A s t e r i s k s d e n o t e s i g n i fi c a n c e a t : * p < 0 . 1 0 , * * p < 0 . 0
e s e n t s t h e r e s u l t s o f P M L r e g r e s s i o n s ( a s d e s c r i b e d i n s e c t i o n 4 . 2 ) o f t h e n u m b e r o f t r a n s l a t i o n s f r
o m a n
o r i g i n a l l a n g u a g e t o
a t a r g e t l a n g u a g e i n a t r a n s l a t i n
g c o u n t r y i n a y e a r . T h e s a m e o r i g i n a l l a n g u a g e s a r e i n c l u d e d f o r
e v e r y
t a r g e t l a n g u a g e ; z e r o v a l u e s a r e i n c l u d e d i n t h e e s t i m
a t i o n , a s a l l o w e d b y t h e P M L p r o c e d u r e . T h e o r i g i n a l l a n g u a g e s
a r e a l l
t h o s e l a n g u a g e s i n t h e m o s t w i d e l y s p o k e n 1 0 0 l a n g u a g e s w o r l d w i d e t h a t a r e e v e r t r a n s l a t e d . T h e t a r g e t l a n g u a g e / c o u
n t r i e s
i n c l u d e d a r e a l l t h e
l a n g u a g e s t h a t a r e o ffi c i a l i n t h
e w h o l e o f t h e t r a n s l a t i n g c o u n
t r y . T h e y e a r s i n c l u d e d a r e 1 9 9
4 a n d
1 9 9 9 . T h e a l t i t u d e p r o fi l e , c l i m a t e r e g i o n p r o fi l e , a n d
b i o m e r e g i o n p r o fi l e d i ff e r e n c e
v a r i a b l e s a r e a l l c o n s t r u c t e d t o v a r y
b e t w e e n 0 ( n o o v e r l a p i n p r o fi l e s ) a n d 1 ( i d e n t i c a l p r o fi l e s ) .
R e
l i g i o u s
d i s t a n c e i s t h e
p r o b a b i l i t y a r a n d o m l y c h o s e n p e r s o n
f r o m t h e t r a n s l a t i n g
c o u n t r y a n d a r a n d o m l y c h o s e
n p e r s o n f r o m t h e o r i g i n a l c o u n
t r y h a v e t h e s a m e r e l i g i o n . S t a
n d a r d
e r r o r s a r e r o b u s t . A
s t e r i s k s d e n o t e s i g n i fi c a n c e a t : *
p < 0 . 1 0 , * * p < 0 . 0 5 , * * * p < 0 . 0 1 .
Notes: This table presents the results of PML regressions (as described in section 2) of the number of translations from an original language to a target language in a translatingcountry in a year. The same original languages are included for every target language;zero values are included in the estimation, as allowed by the PML procedure. The originallanguages are all those languages in the most widely spoken 100 languages worldwide that
are ever translated. The target language/countries included are all the languages that areofficial in the whole of the translating country. The years included are 1994 and 1999.The altitude profile, climate region profile, and biome region profile difference variables areall constructed to vary between 0 (no overlap in profiles) and 1 (identical profiles). Religious
distance is the probability a randomly chosen person from the translating country and arandomly chosen person from the original country have the same religion. Democracy ismeasured on a scale of 0 to 10. Standard errors are robust. Asterisks denote significanceat: * p<0.10, ** p<0.05, *** p<0.01.
4 . 4 : T r a n s l a t i o n s o c c u r f a s t e r b e t w e e n c l o s e r c o u n t r i e s , 1 9 9 8 - 2 0 0 0
D e p e n d e n t v a r i a b l e : p r o p o r t i o n o f t i t l e s t r a n s l a t e d w i t h i n 2 y e a r s o f f i r s t p u b l i c a t i o n
V a r i a b l e
( 1 )
( 2 )
( 3 )
( 4 )
( 5 )
( 6 )
( 7 )
( 8 )
( 9 )
P h y s i c a l d i s t a n c e b e t w e e n o r i g i n a l a n d t r a n s l a t i n g c o u n t r i e s ( l n )
- 0 . 0
5 1
* * *
- 0 . 0
3 1
- 0 . 0
3 1
- 0 . 0
1 8
0 . 0
0 3
- 0 . 9
9 3
- 5 . 3
0 8
- 0 . 3
9 9
- 0
. 3 2 8
( 0 . 0 1
7 )
( 0 . 0
4 9 )
( 0 . 0
5 2 )
( 0 . 0
4 8 )
( 0 . 0
5 0 )
( 0 . 8
5 8 )
( 3 . 5
7 9 )
( 0 . 2
6 3 )
( 0 . 3
0 0 )
P o p u l a t i o n o f t r a n s l a t i n g c o u
n t r y ( l n )
0 . 0 1
5 *
( 0 . 0 0
8 )
G D P p e r c a p i t a o f t r a n s l a t i n g c o u n t r y ( l n )
0 . 1
3 1 * * *
( 0 . 0 1
3 )
P o p u l a t i o n o f o r i g i n a l c o u n t r y ( l n )
- 0 . 2 1
9 * *
( 0 . 1 0
2 )
G D P p e r c a p i t a o f o r i g i n a l c o u n t r y ( l n )
0 . 1 9
2
( 0 . 3 0
3 )
R e l i g i o u s d i s t a n c e
0 . 1
4 2
0 . 1
8 5
0 . 1
9 3
( 0 . 1
3 7 )
( 0 . 1
3 3 )
( 0 . 1
3 5 )
L i n g u i s t i c d i s t a n c e
- 0 . 2
3 2
- 0 . 2 2 5
( 0 . 1
3 9 )
( 0 . 1
4 0 )
G e n e t i c d i s t a n c e
- 0 . 1 8 9
( 0 . 2
5 8 )
P h y s i c a l d i s t a n c e ( l n ) * G D P
p e r c a p i t a o f t r a n s l a t i n g c o u n t r y ( l n )
0 . 1
0 1
( 0 . 0
9 1 )
P h y s i c a l d i s t a n c e ( l n ) * G D P
p e r c a p i t a o f o r i g i n a l c o u n t r y ( l n )
0 . 5
3 6
( 0 . 3
6 5 )
P h y s i c a l d i s t a n c e ( l n ) * u r b a
n i z a t i o n o f t r a n s l a t i n g c o u n t r y ( f r a c t i o n )
0 . 5
1 0
( 0 . 3
6 4 )
P h y s i c a l d i s t a n c e ( l n ) * u r b a
n i z a t i o n o f o r i g i n a l c o u n t r y ( f r a c t i o n )
0 . 3
9 7
( 0 . 4
0 7 )
O r i g i n a l a n d t r a n s l a t i n g c o u n t r i e s a r e c o n t i g u o u s
0 . 0 2
1
0 . 0
9 3
0 . 1
0 9 *
0 . 1
0 0
0 . 1
0 9 *
0 . 1
0 9 *
0 . 0
7 4
0 . 1
0 8 *
0 . 0
7 6
( 0 . 0 4
3 )
( 0 . 0
6 4 )
( 0 . 0
6 1 )
( 0 . 0
6 2 )
( 0 . 0
6 1 )
( 0 . 0
6 4 )
( 0 . 0
6 8 )
( 0 . 0
6 1 )
( 0 . 0
6 6 )
O r i g i n a l l a n g u a g e i s w i d e s p r e a d i n t r a n s l a t i n g c o u n t r y
0 . 2
6 7 * * *
0 . 3
1 4 * * *
0 . 3
2 4 * * *
0 . 3
8 4 * * *
0 . 3 7
6 * * *
0 . 3
1 0 * * *
0 . 3
1 7 * * *
0 . 3
0 9 * * *
0 . 3 1 0 * * *
( 0 . 0 7
6 )
( 0 . 1
0 0 )
( 0 . 0
9 3 )
( 0 . 0
8 5 )
( 0 . 0
8 3 )
( 0 . 0
9 9 )
( 0 . 0
7 4 )
( 0 . 1
0 1 )
( 0 . 0
8 7 )
T a r g e t l a n g u a g e / c o u n t r y f i x e
d e f f e c t s
N o
Y e s
Y e s
Y e s
Y e s
Y e s
Y e s
Y e s
Y e s
O r i g i n a l l a n g u a g e f i x e d e f f e c t s
s : Th e t a b l e pr e s en t s r e s ul t s of O
L S r e gr e s s i on s pr e d i c t i n g t h e s p
e e d of t r an s l a t i on.An o b s er v a t i oni n t h e s er e gr e s s i on s
i s an
or i gi n al l an g u a g e an d a t ar g e t l an g u a g ei n a t r an s l a t i n g c o un t r y.Th e or i gi n al l an g u a g e s i n cl u d e d ar eEn gl i s h ,F r en ch ,
G er m
an an d I t al i an. Onl yn on-fi c t i on t i t l e s t r an s l a t e d i n t o an offi c
i al l an g u a g e of t h e t r an s l a t i n g c o un t r y ar ei n cl u d e d .
L a n d
l o c k e d i s a d umm yf or t h e c o un t
r y b ei n gl an d l o ck e d .Th er el i gi on d ummi e s ar ef or t h em o s t wi d e s pr e a d r el i gi on of t h e
c o un
t r y. O b s er v a t i on s ar e w ei gh t e d b y t h en um b er of t r an s l a t i on s c on t r i b u t i n g t o t h em e a s ur e of s p
e e d . S t an d ar d er r or s
ar e c
l u s t er e d a t t h e c o un t r yl e v el .A s t er i s k s d en o t e s i gni fi c an c e a t : *
l a t i o n s o c c u r f a s t e r b e t w e e n c l o s e r c o u n t r i e s ,
1 9 9 8 - 2 0 0 0 : a s s i g n i n g o r i g i n a l
l a n g u a g e t o c l o s e s t c o u n t r y w h e r e i t i s s p o k e n w i d e l y
D e p e n d e n t v a r i a b l e : p r o p o r t i o n o f t i t l e s t r a n s l a t e d w i t h i n 2 y e a r s o f f i r s t p u b l i c a t i o n
V a r i a b l e
( 1
)
( 2 )
( 3 )
( 4 )
(
5 )
( 6 )
( 7 )
( 8 )
( 9 )
P h y s i c a l d i s t a n c e b e t w e e n o r i g i n a l a n d t r a n s l a t i n g c o u n t r i e s ( l n )
- 0 . 0 6
0 * * *
- 0 . 0
4 3
- 0 . 0
4 2
- 0 . 0
5 1 *
- 0 .
0 4 6
- 0 . 6
8 3 *
- 0 . 0
3 4
- 0 . 2
1 6
- 0 . 0
1 5
( 0 . 0
1 3 )
( 0 . 0
3 3 )
( 0 . 0
3 3 )
( 0 . 0
2 9 )
( 0 . 0 2 8 )
( 0 . 3
8 1 )
( 0 . 1
8 6 )
( 0 . 1
4 3 )
( 0
. 0 4 2 )
P o p u l a t i o n o f t r a n s l a t i n g c o u
n t r y ( l n )
0 . 0
1 5
( 0 . 0
0 9 )
G D P p e r c a p i t a o f t r a n s l a t i n g c o u n t r y ( l n )
0 . 1
1 8
* * *
( 0 . 0
1 5 )
P o p u l a t i o n o f o r i g i n a l c o u n t r y ( l n )
0 . 0
5 4
* * *
( 0 . 0
1 0 )
G D P p e r c a p i t a o f o r i g i n a l c o u n t r y ( l n )
- 0 . 1
6 9 *
( 0 . 0
9 2 )
R e l i g i o u s d i s t a n c e
0 . 1
2 9
0 . 1
7 3
0 . 1 6 2
( 0 . 1
1 2 )
( 0 . 1
1 0 )
( 0 . 1 1 1 )
L i n g u i s t i c d i s t a n c e
- 0 . 2
7 0 * *
- 0 . 3
0 3 * * *
( 0 . 1
2 6 )
( 0 . 1 0 2 )
G e n e t i c d i s t a n c e
0 . 0 5 0
( 0 . 1 1 3 )
P h y s i c a l d i s t a n c e ( l n ) * G D P
p e r c a p i t a o f t r a n s l a t i n g c o u n t r y ( l n )
0 . 0
6 9 *
( 0 . 0
4 0 )
P h y s i c a l d i s t a n c e ( l n ) * G D P
p e r c a p i t a o f o r i g i n a l c o u n t r y ( l n )
- 0 . 0
0 1
( 0 . 0
1 8 )
P h y s i c a l d i s t a n c e ( l n ) * u r b a
n i z a t i o n o f t r a n s l a t i n g c o u n t r y ( f r a c t i o n )
0 . 2
6 2
( 0 . 2
0 0 )
P h y s i c a l d i s t a n c e ( l n ) * u r b a
n i z a t i o n o f o r i g i n a l c o u n t r y ( f r a c t i o n )
- 0 . 0
2 9
( 0
. 0 3 8 )
O r i g i n a l a n d t r a n s l a t i n g c o u n t r i e s a r e c o n t i g u o u s
- 0 . 0
0 1
0 . 0
8 2
0 . 0
9 8 *
0 . 0
9 5 *
0 . 0 9 3
0 . 1
0 9 *
0 . 0
8 2
0 . 0
9 6
0
. 0 8 9
( 0 . 0
4 1 )
( 0 . 0
5 9 )
( 0 . 0
5 4 )
( 0 . 0
5 6 )
( 0 . 0 5 6 )
( 0 . 0
5 6 )
( 0 . 0
5 9 )
( 0 . 0
6 0 )
( 0
. 0 6 1 )
O r i g i n a l l a n g u a g e i s w i d e s p r e a d i n t r a n s l a t i n g c o u n t r y
0 . 2
3 0
* * *
0 . 2
8 8 * * *
0 . 3
1 8 * * *
0 . 3
6 7 * * *
0 . 3 8
1 * * *
0 . 3
7 6 * * *
0 . 2
9 1 * * *
0 . 3
5 6 * * *
0 . 2
9 8 * * *
( 0 . 0
5 0 )
( 0 . 0
8 5 )
( 0 . 0
7 2 )
( 0 . 0
7 4 )
( 0 . 0 6 7 )
( 0 . 0
9 4 )
( 0 . 0
8 4 )
( 0 . 0
8 7 )
( 0
. 0 8 4 )
T a r g e t l a n g u a g e / c o u n t r y f i x e
d e f f e c t s
N o
Y e s
Y e s
Y e s
Y
e s
Y e s
Y e s
Y e s
Y e s
O r i g i n a l l a n g u a g e f i x e d e f f e c t s
Economists and economic historians have long recognized the importance of knowl-
edge and ideas for growth and development.1 Indeed, much of the “new” growth
theory highlights idea accumulation as key to explaining accelerating growth.2
Moreover, the international sharing of ideas plays a huge role: according to one
estimate3, world GDP would be just 6% of its current level if countries did not share
ideas.
Nevertheless, there is little empirical work on the international flows of ideas4 for
1See, for example, Kuznets (1966), Mokyr (2002, 2009, 2010), Romer (1990, 1993), Grossman andHelpman (1991), Jones (2005), Klenow and Rodrıguez-Clare (2005), and Jones and Romer (2010).
2See, for example, Romer (1986, 1990), Helpman (2004), and Jones (2001, 2005).3Klenow and Rodrıguez-Clare (2005)4Paul Romer makes this point forcefully in his 2010 paper.
two main reasons. First, ideas are challenging to measure. Second, it is challenging
to capture the two main properties of ideas, namely non-rivalry and disembodiment.5
Vitally, ideas are non-rival, meaning the use of an idea by one party in no way affects
its simultaneous use by another; this non-rivalry drives technological spillovers.6 Ideas
are also disembodied; an idea that is embodied in a purchased piece of equipment
may not generate a technological spillover.7
We address these challenges by suggesting a new measure of the international
flow of ideas and a setting in which to study how policies and institutions shape
the international diffusion of ideas. Given the importance of ideas for growth, it isimperative to understand how their spread can be affected by policy and institutional
changes.
Specifically, we use book translations as a measure of the international flow of
ideas. Translations are an attractive measure of the diffusion of ideas because they, as
opposed to the physical books that contain them, are both non-rival and disembodied,
and their key purpose is to transmit written ideas, information or knowledge between
speakers of different languages. In the absence of translation, many ideas stored inwords might never leave the language or country in which they were conceived. Of
course, book translations are not the only way societies gain new knowledge8, but they
are an important channel for the flow of ideas between linguistically distinct groups,
and are both quantifiable and classifiable by field and specific content. Moreover, the
types of ideas captured by translations are broad, ranging from technical ideas (such
5Note that measures such as trade, migration, and foreign direct investment are informative inmany ways, but they measure embodied flows of ideas, which are not as such non-rival.
6See, for example, Romer (1986, 1990, 2010), Helpman (2004), and Jones and Romer (2010).7See, for example, Jaffe and Trajtenberg (1999).8An alternative measure is patent citations, which track the diffusion of particular technological
knowledge across disciplines and geographical space (see, for example, Jaffe, Trajtenberg andHenderson, 1993, Jaffe and Trajtenberg, 1999, 2002, and Jaffe, Trajtenberg and Fogarty, 2000).Foreign research and development (R&D) was also suggested as a measure of knowledge spillovers(Coe and Helpman, 1995, Keller, 2002), as well as international trade and foreign direct investment(Keller, 2004, 2009). Book translations are a complementary measure.
as in physics or engineering books), to ideas that are essentially social or cultural
(such as in books on religion, philosophy, or literature). Finally, empirical analysisof translations is possible because systematic data on translations can be generated
from national bibliographies.
The setting we propose is the collapse of Communism in Eastern Europe, which is a
natural place to identify the effect of policy on idea flows. The collapse of Communism
was a large shock that swiftly moved countries from nearly complete isolation from
Western ideas to full openness. Because our measure of idea flows captures a broad
range of ideas, this paper sheds light on the type of ideas most likely to be affectedby policy changes that reduce information restrictions. In particular, we can examine
whether the collapse of Communism had a stronger effect on ideas that contain more
“useful knowledge” (as coined by Mokyr, 2002) for economic development than on
“less-useful” knowledge with more cultural content.
More broadly, we examine how economic incentives shape the international
diffusion of knowledge, which economic historians view as one of the most crucial
economic phenomena of all (see various work by Joel Mokyr). The wider lesson fromour paper is that when these incentives are seriously impaired by institutions, this
can have severe effects that are only remedied as institutional change occurs.
This study of the Communist regime and its collapse in Eastern Europe is not only
a natural context for the study of international idea flows, but it also contributes to
our understanding of this highly important episode in history. First, this is the first
study to assess how Communism affected idea flows.9 Second, while it is known that
9There is a literature that documents and explains the transition of Eastern European countriesfrom Communism into market economies (e.g. Blanchard, 1994, 1996, 1997, Aghion and Blanchard,1994, Frye and Mansfield, 2003). There is also a literature exploring the “natural experiment”created by the collapse of Communism in Eastern Europe and elsewhere to learn about individuals’preferences and behavior (e.g. Munich, Svejnar and Terrell, 2005, Fuchs-Schundeln and Schundeln,2005, Alesina and Fuchs-Schundeln, 2007, Fuchs-Schundeln, 2008, Abramitzky, 2008). However, thispaper is the first to test the effect of the collapse of Communism on the flow of information andideas.
Communist Europe had low inflows of Western knowledge and ideas (e.g. Garton Ash,
1995, Harrison, 2003, 2005), the role of differences in preferences for ideas betweenEast and West has never been clear. Instead, the emphasis is typically on the stronger
censorship of Western ideas in Eastern Europe. Our empirical strategy sheds light
on the role of preferences. To the extent we see convergence in translation rates to
Western levels post collapse, we can conclude that Eastern preferences were either
similar to Western ones or became like them quickly following the collapse. If there
was no convergence despite the end of censorship, then we can conclude that Eastern
European preferences for ideas differ from Western preferences.We begin by comparing translation patterns in former Communist countries before
and after the collapse. To account for possible general changes in translations over the
1980s and 1990s, we also compare translation patterns in Communist countries with
those in Western European countries. To shed further light on the role of preferences
in the flow of ideas, we first compare translation patterns in the Soviet countries
with patterns in the more western-oriented Satellite countries. Second, we test the
degree of convergence in translation flows between Eastern and Western Europe postcollapse. We then test the effect of the collapse and the degree of convergence to the
West of book translations in different fields, to better understand what type of ideas
are more likely to increase once information restrictions are lifted.
We use newly-collected data on almost 800,000 book translations for the period
1980 to 2000. The data were extracted from Unesco’s Index Translationum (IT), an
international bibliography of the translations published annually in a wide range of
countries.
We present four main sets of results. First, we use graphs and regression analysis
and show that when Communism collapsed the overall flow of translations from
Western Europe into the Soviet satellites increased by a factor of seven. At the
same time, we document an offsetting two-third decrease in Communist-to-Satellite
translations. These large magnitudes emphasize just how much the flow of ideas was
affected by the collapse of Communism. In contrast, translations of Western titles intothe former Soviet countries, which had less Western orientation than the Satellites,
barely increased. We further show that these findings are not driven by changes in
the publishing industry that allowed a larger total number of books to be published.
In fact, the total number of books published in Communist countries didn’t increase
with the collapse of Communism, and may have actually declined. Another striking
pattern that emerges is that Western European countries translated very little from
Communist languages, both before and after the collapse of the Eastern Bloc. Second,we show that whereas the Satellite countries converged to Western countries in their
level of translations of Western titles, Soviet countries did not. This suggests that
non-Soviet Eastern Europe has similar preferences for ideas to the West but the
former Soviet Union does not. The Satellite countries not only started to catch up
on translation of older titles (stocks), but they also increased their rate of translation
of current titles (flows) and converged to Western levels of these translations. This
suggests both a convergence in the flow of new ideas, and a catching up on the stock of ideas. Interestingly, even in the Satellite countries, translations of Communist titles
remained higher that in the West.
Third, we show that the effect of Communism’s collapse was larger for the more
“ideological” book fields. Translations of titles in fields such as religion, philosophy,
and the social sciences, were highly suppressed under Communism because they were
perceived as especially threatening to the Communist regime. For instance, religion
was considered an enemy of the Communist regime and was firmly suppressed under it.
Once Communism collapsed, translations of titles in religion increased dramatically,
especially Christian titles. Similarly, translations in philosophy and the social sciences
(especially economics) jumped post collapse. In contrast, the study of exact sciences
was strongly supported by Communist governments, and was important for the
experienced larger increases in translation post collapse than did other influential
titles.
Our findings are consistent with a dramatic increase in the flow of Western ideas
into former Communist countries when Communism collapsed, and with a decline
in the flow of ideas between Communist countries. The effect of the collapse of
Communism on the flow of ideas reflected both high suppression of idea flows during
the Cold War and East/West differences in preferences for ideas. For example, the
higher effect of the collapse on translations in philosophy and economics relative to
exact sciences illustrates the role of severe suppression. The convergence in Western
ideas between the more Western-oriented Satellite countries and Western Europe,
and the lack of convergence between the more Russian-oriented Soviet countries
and Western Europe illustrates differences in tastes for Western ideas. Similarly,
the remaining differences in translations between East and West in some fields, and
between Soviet and the West in all fields, illustrate how cultural differences persisted
even after Communism collapsed.10
This paper proceeds as follows. In Section 5.2 we present the data on book
translations and explain the construction of our measures of idea flows. Section 5.3
briefly outlines the historical context of publishing in Communist Europe and of the
collapse of Communism. Section 5.4 describes our empirical strategy for examining
the effect of the collapse of Communism on book translations. Section 5.5 presents
results on the effect of the collapse of Communism on the total flow of translations.
Section 5.6 presents results breaking translations down by book field. Section 5.7
presents our analysis of the effect of the collapse on influential titles. Section 5.8
discusses further translations as a measure of the diffusion of ideas and concludes.
10This illustration is consistent with the literature showing how history can shape culture (e.g.Greif, 1994, Guiso, Sapienza and Zingales, 2008, Nunn and Wantchekon, forthcoming, and Fletcherand Iyigun, 2010; see also the surveys by Tabellini, 2010 and Nunn, 2009 on the historical origins of culture).
5.2.1 The flow of book translations across countries
The translation data are extracted from Unesco’s Index Translationum (IT), an
international bibliography of the translations published in a wide range of countries.
These data originate at the national level through the law of legal deposit, which
specifies that every book published that is intended for circulation must be submitted
to the national depository. The national depository then compiles a list of the
publications that are translations, and submits this list to Unesco, which standardizes
the entries across countries to form the IT.
Titles in the IT are categorized according to the nine main categories of the
Universal Decimal Classification (UDC) system: General (0.1% of translations
in the period 1980-2000); Philosophy (including Psychology, 5.3%); Religion and
Theology (5.7%); Law, Social Sciences, Education (8.5%); Natural and Exact Sciences
(4.2%); Applied Sciences (11.4%); Arts, Games, Sports (5.2%); Literature (including
books for children, 52.3%)11; History, Geography, Biography (including memoirs and
autobiographies, 6.6%).12
The bibliographic entry for each translation includes information on the country,
city, and year in the which the translation was published, the language of the original
title and the target language into which it was translated, the field (UDC class) of
the title, the number of pages or volumes of the title, the author, and the original
and translated titles of the book.
We use data on the translations in Communist countries (our group of interest)and Western European countries (our comparison group) over the period 1980 to
2000, which comprise approximately 800,000 translations. We limit our Communist
11Literature also includes the very small category Philology and Linguistics.12For a detailed description of the subfields that make up each UDC field, see https://www.
countries to European countries that were part of the Eastern Bloc and that were
Warsaw Pact members in the 1980s, meaning they were under heavy Soviet controlpre-collapse because Soviet troops were permitted to be stationed within their
borders. Our Communist countries are thus seven former Soviet countries (Russia,
Belarus, Estonia, Latvia, Lithuania, Moldova, and the Ukraine), Bulgaria, the Czech
Republic, Hungary, Poland, Romania, and Slovakia.13 The Western European
countries in our sample are: Austria, Belgium, Switzerland, Denmark, Spain, Finland,
France, Iceland, Italy, the Netherlands, Norway, Portugal, and Sweden. Results are
unchanged if we include the USA in the group of Western countries. We include eachcountry only in the years it reported consistently, resulting in an unbalanced panel.
Note that Germany is excluded from the analysis because our data do not allow us to
distinguish whether a translation after unification was in East or West Germany, and
in any case the country post collapse was a single market with a common language.
The UK is also excluded because it stopped reporting its translations to Unesco in
1990. However, we do use translations from all Western and Communist languages
flowing to these countries, including translations from English.Creation of translation series over time for some of these countries is complicated
by the fact they only became separate countries upon the upheaval of interest in
the middle of our period of study. Prior to 1992, the USSR as a whole reported
its translations; prior to 1993, Czechoslovakia as a whole reported its translations.
Our data provide a rare opportunity to nevertheless allocate the idea flows to the
constituent countries. Specifically, we allocate the translations reported by the USSR
and Czechoslovakia to one of their constituent countries based on the city in whicheach translation was published.
We note that the translations reported are only those that were submitted to
13We omit Yugoslavia because it escaped the Soviet sphere in the Tito-Stalin split of 1948, andAlbania because it withdrew from the Warsaw pact in 1968; thus in our period of interest they wereno longer politically aligned with the Soviet Union.
the central depository of the country. In particular, this excludes samizdat , the
illegal books published under the Communist regime. The exclusion of these titlesis unfortunate, but is unlikely to affect our analysis. The number of samizdat
translations produced under Communism is not available, but they were likely only
a small fraction of total translations. These illegal publications were largely political
magazines and bulletins defending human rights and criticizing repression. Although
some were poems and books, both locally written by dissidents and translated from
foreign publications, the large personal risk involved in owning such books meant
their circulation was limited, and the ideas contained therein were not available tothe general populace.
5.2.2 Translation of influential titles
To test the effect of the collapse of Communism on the most influential titles,
we extract from the Index Translationum data the translation patterns of titles
considered important and influential in the West. The titles selected, listed inAppendix C.3, are those given in any one of three lists. The first is the Central
and East European Publishing Project’s (CEEPP) list of the 100 books that have
been most influential in the West since 1945. This list was assembled in 1994, and
appeared in Garton Ash (1995). The second is the Modern Library’s list of the
100 best non-fiction books of the 20th century published in English.14 The third is
National Review’s best 100 non-fiction books of the 20th century.15 A considerable
number of titles appear in more than one of these lists. We include only titles that were
originally published before 1985 (to allow them enough time to have been translated
before the collapse), and we omit all titles that were not translated in any of our
14The “Board’s List”, available at http://www.randomhouse.com/modernlibrary/
when Gorbachev came to power in the USSR. Among the reforms he instituted,
perhaps the most important two were perestroika , restructuring of the economy andpolitical system, and glasnost , openness in the media and culture. Through these
sets of gradual reforms, the Soviet Union began to move in the direction of a market
economy, with a decrease in centralization and the emergence of private firms, and the
increase in the freedom of people to express their views on a range of topics without
fear of retribution.
An important consequence of glasnost was that people could now openly air
their dissatisfaction with the Communist regime. This freedom spread to the Sovietsatellites, and was likely a contributing factor in revolutions that heralded the fall of
the Berlin Wall and the collapse of the Communist regimes in the Satellite countries
in the last few months of 1989.
The Communist USSR held together for nearly a further two years, though the
power of the Soviet Communists was waning and nationalism in the Soviet republics
was on the rise. Late in 1991, a conservative coup in Russia aimed at preventing
the disintegration of the Soviet Union was staged. Its unintended effect was just the
opposite; the USSR was officially dissolved.
The Communist countries had many commonalities, but there was heterogeneity
between them in the extent to which they had a Western orientation. The former
Soviet countries had a more Russian orientation, the preferences of their consumers
favored Western ideas less, and they maintained stronger ties with Russia and
demonstrated less effort or desire to integrate with Western Europe. However, the
three Baltic states of the Soviet Union, Latvia, Lithuania and Estonia, were more
similar to the Satellites than they were to the Soviet nations. Historically, they were
relatively recent additions to the USSR (annexed in 1940), and had always maintained
their more Western feeling. They were the first among the Soviet nations to declare
their independence from the Soviet Union. Furthermore, their independent streak was
highlighted when, upon the collapse of the Soviet Union, they were the only three
Soviet states not to join the Commonwealth of Independent States (CIS), the loosealliance of independent countries that succeeded the USSR. Since the disintegration
of the USSR, the former Communist countries have coalesced into two trading blocs:
the Russia-focused CIS countries in one, and the Western-centered non-CIS countries,
including the Baltic states, in the other. For this reason, our main analysis groups
the three Baltic states with the Satellites, but we note that the results are similar
when excluding them from the analysis or when assigning them to a separate group.
Figure 5.2 is a map showing the Soviet countries, Satellites plus Baltic states, andthe Western European countries in our analysis.
5.3.2 Restricting information flows:
Publishing and censorship under Communism
Prior to Gorbachev’s reforms, book publishing in the Soviet Union16 was a state-
run industry that produced vast numbers of books with little regard for consumer
demand.17 All publishers were owned and operated by the government, and each had
its own subject area or field in which it enjoyed a complete monopoly. Book prices,
like other prices and wages in the publishing industry, were strictly controlled; each
subject had a designated price range, chosen to ensure the subjects the government
intended to be widely read were available at low cost. Selection of the titles published
was centrally coordinated and crafted according to the government’s grand plan.18
Central to the organization of the Soviet publishing system was the conception
of publishing as an ideological activity. Reading was viewed as a way in which the
16We discuss the publishing and censorship system of the Soviet Union, which is the one bestunderstood by Western scholars and observers during the Communist period. The publishingindustries of the other Communist countries varied in their exact details, but were similar in theirprinciples.
social consciousness of individuals was shaped, thus full state control over the material
published and its availability to citizens was vital. Profits and publishing in order tomeet demand were considered less important, though periodically concern surfaced
in Soviet publishing circles about the shortages of books in specific fields.
The process determining the exact titles printed in any year was complex and
centrally planned to a high degree. USSR-level and republic-level authorities decided
on the proportion of total books published in the coming year that would be in
each subject area, and assigned printing capacity, paper, and binding materials to
individual publishers. Working within these bounds and other specifications given tothem, publishers compiled their own lists of planned printings, each item on which
then received an approval, rejection, or other recommendation from a “coordinating”
central authority. Considerations for the coordinating authority were maintaining the
subject monopolies of the printing houses, avoiding duplication of subject matter, and
economy in the use of paper, which was often in short supply.
Additional centralized planning occurred that was related to the publication
of translations.19 Foreign titles were selected for translation by utilizing experts
employed for the purpose at home, representatives located in numerous countries
abroad, and foreign visiting experts such as scientists. The representatives located
abroad reviewed tens of thousands of new books annually. They then bought copies
of the most important titles from local bookshops, and mailed them back to their
publishers in the USSR.20
Censorship of books intended for sale in the USSR was the domain of Glavlit
(occasionally referred to by its full name, the “Chief Administration for the Protection
of State Secrets in the Press attached to the Council of Ministers of the USSR”).
Editors of publishing houses were expected to use their good sense in selecting titles
for publication, but the corrected galley-proofs (granki ) then had to be perused by
Glavlit “both for the mention of prohibited topics and for the observance of politicallines and nuances” (Walker, 1978, page 66) before publication could occur.
Censorship of translations followed a somewhat different, but undoubtedly no less
rigorous, process, explained by Walker (1978):
“The importance of careful and vigilant selection by Soviet publishers in
choosing works for translation from foreign languages has been frequently
stressed by Party and government, and is visible in a number of specialregulations applying to the publication of translations. A publishing-
house considering translation of a foreign work must, unless there is a
special need for speedy publication, obtain at least two recommendations
for the translation from scholarly institutions or specialists, and secure the
agreement of the appropriate chief editorial office in the State Committee
for Publishing before submitting details of the work for ‘coordination’ to
the State Committee or (in the case of scientific and technical works) tothe State Scientific and Technical Library.”
Between 1986 and 1991, control over the publishing industry moved out of
state hands. State-owned publishing houses were joined by a multitude of other
ownership structures, competition entered the industry, and the focus shifted away
from producer-led publishing to consumer-led publishing. The monopoly system of
publishers was scrapped; price controls and many state subsidies were terminated.Through the reforms, firms, organizations, and institutions gained the right to
publish, and Russian authors and publishers gained the right to freely buy or sell
rights, including in transactions with international parties.21
Communist countries post collapse, and take a variation of the following form:
Y it = β 0 + β 1Postt + β 2X it + it (5.1)
where Y it is the (log) number of book translations in country i in year t .22 Postt
is a dummy variable for the years 1991 and onwards,23 and its coefficient measures
the change in translation patterns post collapse. Our control variables X it include
population and real GDP per capita. In some specifications, we include country fixed
effects to account for differences across countries that are constant over time.
We also estimate difference-in-differences models that compare the pre- and post-
collapse translation flows into Communist countries with flows into Western European
countries. The inclusion of Western European countries as a comparison group
accounts for other common factors that may have affected translation patterns over
the sample period 1980-2000. The basic difference-in-differences specification is:
Y it = β 0 + β 1Communisti × Postt + β 2Communisti + β 3Postt + β 4X it + it (5.2)
where Y it and Postt are as before, Communisti is a dummy variable for whether the
translating country was a former Communist country, and Communisti × Postt is
the interaction between these two variables. The coefficient on the latter variable
measures the effect of the collapse of Communism on translations into Communist
countries (relative to into Western European countries). In addition to specifications
that control for population and GDP and include country fixed effects, we also run
specifications with year fixed effects to absorb changes over time that are common to
all regions.
22The trivially few observations with zero values are dropped.23We choose post-1991 because it is midway between the end of Communism in the Satellites
(late in 1989) and the collapse of the Soviet Union (late in 1991). Using alternative Post variables,namely post-1989, post-1990, and post-1992, does not substantially alter the results (not presented).
In both the basic regression and difference-in-differences model, the construction
of the dependent variable is complicated by the lack of a one-to-one mapping betweencountries and languages. We deal with this by only counting translations into the
“main” language for each country, defined as the most widely spoken language in
the country.24 In Section 5.5.6 we show the main results are robust to also including
translations into secondary languages, and to using the number of pages translated
as an alternative dependent variable.
After testing the effect of the collapse of Communism on overall translations
in Section 5.5.1, we investigate heterogeneity in the magnitude of the effect acrossdifferent types of idea to shed light on what sorts of ideas were more restricted during
the Communist era and on what determined the degree of convergence to the West
post-collapse. We begin in Section 5.5.2 by allowing the effect to differ for translations
from Western and Communist languages, expecting mainly translations from Western
languages, which weren’t originally written under a Communist government, to
increase after the collapse of Communism.
To shed light on what determined convergence between East and West, in Section
5.5.3 we test whether the effect of the collapse was bigger for the Satellite countries,
which had a more Western orientation, than for the Soviet countries. Then, in
Section 5.5.4, we test whether the convergence we document reflected catching up
in translating old titles (stocks) or a convergence to Western levels in translating
current titles (flows).
In Section 5.5.5, we show that the changes in translation patterns that occurred
were not simply driven by general changes in the book industry, as total publications
of original books in Communist countries did not increase after Communism’s
collapse.
24“Most widely spoken” is defined in terms of native speakers where these data are available,otherwise in terms of the language spoken at home or spoken on a day-to-day basis.
We test for heterogeneity of the effect across book fields in Section 5.6. To further
shed light on the role of isolation during the Cold War, we test whether the effect of thecollapse was bigger for more “ideological” fields, such as philosophy and economics,
and whether it was bigger for titles that were perceived to be more threatening to the
regime. We note that, given censorship was lifted with the collapse of Communism,
remaining differences between Eastern and Western Europe post collapse reflect either
pre-existing differences in tastes between East and West, or a lack of convergence in
their tastes post collapse. We find that in fields such as history and arts, translations
of Western titles did not converge to Western levels, suggesting a lack of Easterninterest in the Western version of these fields, perhaps because they are relatively
culture specific.
Section 5.7 analyzes the translation patterns of the most influential Western titles
of the 20th century. This analysis reveals that the increase in translations of Western
titles in Communist Europe involved important ideas. Furthermore, it allows us to
see whether specific titles were available in translation in any Communist language
before the collapse of Communism, which could mean countries were accessing the
titles through a secondary language. This might be particularly relevant for the
case of Russian, which could be read by many people in Communist Europe even
outside Russia. In fact, we find most of these titles were not translated anywhere
in Communist Europe before the collapse of Communism. Finally, we collected
additional information on these titles that allowed us to test the extent to which
translations of specific authors and titles considered particularly threatening to
Communism increased more than translations of other titles with the collapse of
Figure 5.3 shows translations per million inhabitants in the Soviet countries, the
Satellites, and the Western European countries. For each set of countries, we show
translations from Communist languages and Western European languages.25,26
This figure shows that before the collapse of Communism, Western European
countries had much higher translation rates into their main language than Communist
countries, and these translations were almost entirely from Western European
languages. The Satellites translated more than the Soviet countries, and both sets
translated primarily from Communist languages.
However, in the few years around 1990, the patterns of translation for Communist
countries changed drastically. The Satellites’ translations of Western European titles
shot up to approach the level of translations of Western European countries, and their
translations of Communist titles fell away.
By the year 2000, the Satellites’ translation patterns had converged to those
of Western European countries to a remarkable degree, though they still showed a
slight bias towards translations from other former Communist countries. The Soviet
countries also experienced a fall in translations from Communist languages, but their
increase in translations from Western European languages was small and short-lived.
These translation patterns stand in contrast to translations from Western European
languages in Western European countries, which increased only gradually and by
25The Communist languages are: Armenian, Azerbaijani, Belarusian, Bulgarian, Czech, Estonian,Georgian, Hungarian, Kazakh, Kirghiz, Latvian, Lithuanian, Moldovan, Polish, Romanian, Russian,Slovakian, Tajik, Turkmen, Ukrainian, and Uzbek. The Western European languages are: Danish,Dutch, English, Finnish, French, Modern Greek, Icelandic, Irish, Italian, Maltese, Norwegian,Portuguese, Spanish, and Swedish. Note the German language is neither classified as a Communistlanguage nor a Western European language.
26Translations from English show very similar changes over time to translations from all WesternEuropean languages.
5.5. THE EFFECT OF THE COLLAPSE ON TOTAL TRANSLATIONS 133
much less over this period. Similarly, translations from Communist languages in
Western Europe, which were few, showed little change over the period. We nextsubject these patterns to regression analysis.
5.5.1 Changes in overall translation patterns
We first estimate a simple OLS regression as in equation (5.1), predicting the number
of book translations in countryi
in yeart . The first three columns of Table 5.1 present
the regression results. The first column is a basic specification with no additional
controls. The second column adds controls for log population and log GDP per capita.
The third column adds country fixed effects. We see that translations in Communist
countries rose when Communism collapsed. We note that the main coefficient in
the specification without controls is positive but statistically insignificant, but we
show next that this simply masks opposite patterns of translations from Western and
Communist languages. When controls for population and GDP per capita are added,
the coefficient on Postt is large and significant, even when country fixed effects are
included. Translations in Communist countries increased by 120% (e0.799 − 1) after
the collapse of Communism (column 3).
In column 2, where country fixed effects are not included, the coefficients on
population and GDP per capita have the expected positive sign and are significant,
indicating richer and more populous countries translate more. When country fixed
effects are included, the coefficient on population becomes large and negative, butthis is based on little variation, and is probably driven by the population decreases
that occurred in many of the Communist countries post collapse.27
27In the specifications with country fixed effects, the coefficients on population and GDP percapita are identified off within-country correlation between population and translations.
5.5. THE EFFECT OF THE COLLAPSE ON TOTAL TRANSLATIONS 135
per capita; we also include specifications that fully interact the dummies for whether
the original language is Communist or Western European with country fixed effects.Under the hypothesis that Communism suppressed information flows from Western
into Communist Europe, we expect β 1a to be positive. The expected sign of β 1b is less
clear, but is expected to be negative if Communist countries substituted Communist
translations for Western ones pre collapse.
We next estimate difference-in-differences regressions that use Western Europe as
the comparison group. To allow translation patterns to differ between translations
from Western languages and those from Communist languages, we in fact estimatethe following regression:
Because translations tended to increase in Western Europe during the 1990s,
the difference-in-difference estimates presented in Table 5.2 are generally smallerthan the OLS estimates, but they are still economically large and statistically
significant. Specifically, the first column of Table 5.2 is a basic difference-in-
differences specification with no additional controls. We see that, as suggested by
the graphs, Communist translations from Western European languages rose by 260%
when Communism collapsed, whereas translations between Communist countries fell
by 71%. These large magnitudes demonstrate just how dramatically the types of
translated titles available in Eastern Europe shifted when Communism collapsed.
The second column of Table 5.2 shows that these effects are robust to controlling
for log population and log GDP per capita.30 The third column adds country
fixed effects interacted with Communist and Western original languages; the main
results hold and remain significant. The fourth column is the most demanding
specification. It allows translations from Communist languages and from Western
European languages to be on different linear time trends in each country, and identifies
the effect of the collapse of Communism off changes in translations over and abovethese time trends. The main results hold up, though the decrease in translations from
Communist languages decreases in significance. Note, however, that this specification
may in fact underestimate the effect of the collapse of Communism on translations
because the changes that constituted the collapse of Communism were many and
occurred over several years around the date we attribute to the collapse, so some of
these changes are likely falsely attributed to the time trends in this specification. The
fifth column includes both country and year fixed effects; the results are unchanged.
Moreover, column 3 of Table 5.2 also shows that Western countries did not
translate more Communist titles post collapse; the coefficient on the interaction of
30We do not have comparable population or GDP data for Iceland, thus this country is excludedin the specifications where these controls are included.
where Satellitet is a dummy variable for whether the translating country is a Satellite
country. The main coefficients of interest are β 1a and β 1b, which capture whether
translations from Western European and Communist languages respectively increased
31As a second alternative, we divide the Communist countries by whether they are Slavic or non-Slavic, and by whether they are primarily Catholic or Orthodox. Translations in the Slavic countriesshow similar patterns to those in the Soviet nations, and translations in the non-Slavic countriesare similar to in the Soviet satellites. However, the Slavic/non-Slavic difference is less pronouncedthan the Soviet/satellite difference. Similarly, the Orthodox countries behave more like the Sovietnations and the Catholic countries more like the satellites, though the distinction here is smalleragain. The Slavic countries are Russia, the Ukraine, Belarus, the Czech Republic, Slovakia, Poland,and Bulgaria. The Catholic countries are Lithuania, Poland, the Czech Republic, Slovakia, andHungary.
5.5. THE EFFECT OF THE COLLAPSE ON TOTAL TRANSLATIONS 139
specification). In contrast, translations of Communist titles decreased by 68% (70%,
i.e. decreased by two thirds) for Satellites and 74% (76%) for Soviet countries. Acomparison of column 6 with column 7 reveals that differences in income can account
for some but not all of the difference between the post-Communism translation
experiences of the Soviet countries and those of the Satellites.
To test how the effect of the collapse of Communism changed over time and how
similar Eastern and Western Europe become, we run a version of column 7 of Table
5.2 that replaces Post and its interactions with year dummies (for each year 1989
and onwards) and their equivalent interactions. The top half of Figure 5.4 showsthat the positive effect of the collapse of Communism on translations from Western
Europe increases until about 1992, and then stabilizes, especially for the Satellite
countries. The lower half of Figure 5.4 shows that the negative effect of the collapse
on translations between Communist countries increases until 1991, at which time it
largely stabilizes.32,33
5.5.4 Convergence in translation flows or catching up on
stocks?
As mentioned earlier, Figure 5.3 suggests that translations of Western titles in
Satellite countries nearly converged to their levels in Western countries. We note
that this figure understates convergence because it doesn’t control for GDP, which
was lower in Communist countries. Indeed, column 7 of Table 5.2 shows that
Satellite translations of Western titles post collapse are actually greater than Western
32Appendix Figure C.1 shows the equivalent graph where we also include country fixed effectsin the regression equation (equivalent to column 3 of Table 5.2). The effects are similar and moreprecisely estimated, but there it is not possible to compare Communist translations with the Westernlevel of translations.
33We present this figure for the difference-in-differences specification, but the equivalent graph forthe OLS specification looks nearly identical.
translations of these titles after controlling for population and GDP.34 Translations of
Western titles by Soviet countries, however, increased to just 8% of such translationsby Western countries.35
Figure 5.4 illustrates the dynamics of how the translation of Western titles in
Satellite countries converged to and even surpassed Western levels, but in Soviet
countries did not. The figure also shows that translations of Communist titles fell
over several years in both Soviet and Satellite countries but remained higher than
their level in the West.
This convergence of Communist to Western countries could reflect a convergencein the rate of translation of new titles (flows), or a catching up on older titles missed
out on during the Communist era (stocks). We now examine this issue.
Our data set does not lend itself easily to infer the years in which the original titles
were published. However, for the years 1985, 1993 and 1996, we sampled over 1,400
translations from Western languages, identified their original dates of publication
from online sources, and used these to estimate the age distribution of translations
of Western titles.We define flows as titles translated within 15 years of their publication, but our
findings hold for other cutoffs (10, 20, 30 years). We find that such titles make up
the majority of translations in most fields.36 Across fields, the median percentage of
translations that were flows in Communist Europe was 58% in the pre period and
71% post; in Western Europe it was 78% in the pre period and 82% post. We adjust
the total number of translations using these percentages corresponding to each field,
and repeat our main analysis for both flows and stocks.Table 5.3 shows our difference-in-differences regressions separately for flows and
340.687 + 1.337 - 3.249 + 1.777<0.35Specifically, the coefficient on Communist countries for translations from Western languages is
-3.249, and its interaction with post is 0.687, so Soviet translations of Western titles remain at 8%(e−2.562) of Western levels.
36Literature is the primary exception, where flows account for roughly half the titles translated.
5.5. THE EFFECT OF THE COLLAPSE ON TOTAL TRANSLATIONS 141
stocks. Both translations of stocks and flows of Western titles show large increases
in Communist Europe upon the collapse of Communism. This suggests Communistcountries both began catching up on older titles, and increased their rate of translation
of current titles. Moreover, Communist countries overtook the West in their
translation of both newer and older titles. This suggests both a convergence in the
flow of new ideas, and a catching up on older ideas.
To illustrate these phenomena graphically, Figure 5.5 replicates Figure 5.3 for
flows and stocks separately. The figure illustrates how the Satellite’s translations of
new titles almost converge to their Western levels even without controlling for GDP,and their translations of old titles overshoot the levels in the West.
5.5.5 The collapse of Communism did not affect original
publications of books
One potential concern is that the increases in Western translations post collapse were
driven by changes in the publishing industry that allowed a larger total number of
books to be published. If this were the case, then the increase in translations could
be mechanical rather than indicating an increased openness to Western ideas.
Table 5.4 presents OLS and difference-in-differences specifications such as in
equations (5.1) and (5.2) with the total number of original books published as the
dependent variable.37 The table shows that the total number of original books
published in Communist countries did not increase with the collapse of Communism,
37Book publication data are from the Unesco Statistical Yearbooks for the years 1985-99 and fromUnesco’s online data on book production available at http://stats.uis.unesco.org/unesco/. Theyare available pre and post collapse for only a subset of our countries, namely the Communistcountries Belarus, Bulgaria, Estonia, Hungary, Latvia, Poland, Romania and the Ukraine, andthe Western European countries Belgium, Denmark, Finland, France, Iceland, Italy, Netherlands,Norway, Portugal, Spain, Sweden, and Switzerland. Note, however, that these data are only availableat an aggregate level and a large number of years are missing, which precludes using them to conductmore complex analysis.
and may have actually declined. Specifically, the coefficient of interest, which is the
coefficient on Post in the OLS specifications and on Post × Communism in thedifference-in-differences specifications, is negative and small in most specifications.
5.5.6 Further robustness checks
Number of pages translated as an alternative dependent variable
For robustness, we use the number of pages translated as an alternative dependent
variable that captures the possibility that longer books contain more ideas. Becausewe are concerned that some of the short publications might not in fact be books, we
limit translations to titles of 49 pages or longer (the minimum length for a “book”
as defined by Unesco). Appendix Table C.1 shows that the results are similar when
using this alternative dependent variable.
The Bertrand et al. critique of difference-in-differences estimators
Bertrand, Duflo and Mullainathan (2004) show that difference-in-differences tech-
niques applied to data with more than two periods generate inconsistent standard
errors because they do not account for serial correlation of the outcomes. To address
this critique, we follow their recommended procedure and collapse our data down to
one pre-collapse and one post-collapse observation. The pre-collapse values of the
variables are the averages for the years 1980 to 1989, and the post-collapse values
are the averages for 1992 to 2000. We discard data from 1990 and 1991, considering
this the transition period. Appendix Table C.2 shows the equivalent difference-in-
differences regressions to Table 5.2, but run with only these two observations for
each country/original language pair. Our main results remain large and statistically
significant. Specifically, the increase in Satellite translations from Western European
languages is significant at the 1% or 5% level in every specification, and the decrease
Accounting for Russian-speaking populations in other Communist coun-
tries
Our main analysis shows Soviet countries lag behind both Satellite and Western
countries in their translations of Western titles post collapse. To create a lower
bound on these differences, we include translations into Russian in each of the Soviet
countries in addition to translations into the country’s main language. The results
(not presented) are very similar to the specifications that include translations into
secondary languages, shown in Appendix Table C.4.38
Accounting for the possibility of Russia translating for other Communist
countries
A potential concern is that many translations into Communist languages might
actually be published in Russia, the largest of the Communist countries and the
political center of Communist Europe, rather than in the home country, in which case
we would under-report the ideas flowing into the other Communist countries. That is,
the concern is that translations from, for instance, English into Czech are published
in Russia. To account for this possibility, we ran specifications including Russia’s
translations into other Communist languages as translations in the appropriate
Communist countries. In fact, the number of such translations was very low and
the results (not presented) are effectively unchanged.
5.6 The effect of the collapse by book field
In this section we investigate how the effect of the collapse of Communism on book
translations varied by field. First we show the change in translations per capita
38We note that the Satellite countries translate very few titles into Russian; including translationsinto Russian as well as into the main language for all the Communist countries instead of just theSoviet countries makes no difference (results not presented).
because it tended to be unthreatening to Communism and was vital for Soviet power
on the world stage. Perhaps surprisingly given the advanced state of Exact Sciencein Communist Europe, Western translations of Communist Exact Science titles were
always very low. When Communism collapsed, Exact Science translations between
Communist countries fell away, but were gradually replaced by translations from
Western European languages.
More generally, Satellite translations of Western titles converged to Western levels
in all fields except for Art and History. Before the collapse of Communism, differences
in translation rates between Eastern and Western Europe reflected both the effect of
censorship in the Communist countries and differences in tastes. However, when
the Communist regime collapsed official censorship was abolished, thus post-collapse
differences are likely indicative of consumer preferences that differ considerably
between the two halves of Europe. Titles in Arts and History seem likely to contain
pervasive culture-specific aspects, which makes differences in preferences probable
and explains the lack of convergence of their translations post collapse.
5.6.2 Regression analysis by book field
We next estimate our second specification from Table 5.2 separately for translations
in each of the eight fields. We run for each field a difference-in-differences regression
predicting the log of translations plus one.40 Figure 5.7 plots the coefficients on
the two interactions of interest against each other. The axes in the figure are the
coefficients of interest multiplied by 100, which can approximately be thought of as
40For each field we also run two separate regressions, a probit regression predicting whether thenumber of translations is positive (extensive margin), and an OLS regression that estimates the lognumber of translations given the number of translations is non-zero (intensive margin). AppendixTable C.5 presents the coefficients on the interactions of interest in both regressions. The resultstell a similar story.
percentage changes in translation when Communism collapsed.41
The figure shows that the change in translations from Western European languagesand the change from Communist languages are positively correlated across fields. This
suggests the types of ideas that were considered helpful or harmful to the Communist
regime tended to be the same whether the original language was Communist or
Western European.
The axes, which show the extent to which translations “rebounded” when
Communism collapsed, can be approximately thought of as the extent to which the
translation of such ideas was suppressed under Communism. Religion translations,in the top right hand corner of the graph, were most highly suppressed under
Communism. Natural Science translations, in the lower left hand corner, were the
most encouraged under Communism from both types of language. However, the
comparatively small increases in translations of Western Arts and History titles
likely reflect a lack of taste for these books in Eastern Europe rather than a lack
of suppression of them under Communism. Another subject of particular interest is
Social Science, which was relatively suppressed from Western European sources underCommunism, but was among the most encouraged from Communist languages. This
seems to suggest that Communist countries had their own version of Social Science,
but they substituted away from it and towards the Western version when Communism
collapsed.
5.6.3 Regression analysis by book subfield
While our translation data divide titles into eight aggregate fields, we disaggregate
further each of these eight fields by searching for the most commonly used keywords
41When we allow the effect of the collapse of Communism to differ for Soviet countries relative toSoviet satellites (figure not presented), the relative positions of the subjects are similar for the twotypes of Communist countries, though the points for the Soviet countries are all shifted to the left.
in the book titles, and grouping these keywords by subfields such as mathematics,
physics and chemistry. We then test the effect of the collapse of Communism oneach subfield. In order to consistently categorize books by keywords in their titles,
we focus on titles translated from English (74% of the titles translated from Western
European languages) for which the original title is non-missing (79% of these titles).42
To select the keywords for which we search in each field, we first identified the words
that appear most frequently in titles translated in that field (e.g. physics, chemistry,
earth, and universe). We then discarded those that select titles that are not primarily
on a consistent topic. To the remaining informative common keywords we addedrelated keywords that also returned consistent topics.43 We then aggregated our
keyword searches into cohesive subfields.44,45 The percentage of titles captured by
this process ranges from roughly 20% to 55% in the various fields.46 Appendix C.4
lists the keywords contributing to each subfield. Appendix C.5 gives examples of the
titles found by each keyword search.
42Our results for the subfields identified by keyword searches are not driven by the restrictions to
titles translated from English or with non-missing original titles. Restricting from titles translatedfrom Western languages to titles translated from English in a difference-in-differences specificationpooling all fields increases the coefficient of interest from 1.34 to 1.78; subsequently restricting totranslations with non-missing original titles decreases it slightly to 1.62. These changes are smallrelative to the standard errors on the coefficient estimates.
43Note our searches also capture variant forms and spellings of the keywords (e.g. British andAmerican spellings), and obvious typographical errors.
44The aggregated subfields for each field are as follows. For Religion and Theology: Christian,Judeo-Christian, Judaism, theology, Islam, Eastern religions; for Education, Social Science and Law:economics, communism, political science, sociology and anthropology, and education; for Natural andExact Science: mathematics, physics, chemistry, biology, geology; for Applied Science: computers,business, medical, engineering, food, gardening. We do not present results from subfield keywordsearches in the fields Arts, Games and Sports, Literature, History, Geography, and Biography, orPhilosophy and Psychology because they are largely uninformative.
45Notice individual titles might be captured by more than one search, in which case they areattributed to both.
46The primary reasons why these percentages were not higher were that many titles areuninformative about the subject of the book (e.g. “Nowhere to Hide” by Susan Francis is anEnglishwoman’s story of her life in Iraq in the time of Saddam Hussein), and many others containonly keywords that appear in multiple contexts (e.g. the keyword “rights” appears in ThomasPaine’s classic on democracy “Rights of Man” and the title “Human Rights Violations In Zaire”.)
To test which subfields jumped the most post collapse, within each field we
run a difference-in-differences regression that compares the effects across constituentsubfields. The coefficients of interest are the interactions of the subfield fixed effects
with the Post × Communist variable.
The coefficients of interest and their confidence intervals are shown in Figure
5.8, which suggests that even within fields, certain subfields increased more post
collapse. We find that within the field of Exact Science, mathematics titles jumped
less than titles in geology, physics, chemistry and especially biology. Within the Social
Science field, books related to economics jumped the most post collapse. Medicaltitles jumped more than any other titles in the Applied Science field; engineering
titles jumped the least. Within the field of Religion, books with Christian-related
words in their titles jumped more post collapse than Eastern Religion books and
books with Jewish-related or Islamic-related words in their titles. Titles in the field
History, Geography and Biography were difficult to categorize by keyword because of
the manner in which such books are titled. However, we were able to isolate early
history titles (approximately the prehistoric period until the renaissance), a periodabout which we expect Western and Eastern Europe to largely agree, and indeed
Communist translations of this category increased very little.
5.7 The effect of the collapse on translations of
influential titles
Since we have a small number of observations in our analysis of influential titles,
we limit ourselves to a simple pre/post, Communist/West comparison. This means
we need to use the same set of countries in every year we include in order to draw
conclusions about relative changes in Eastern compared with Western Europe. Thus
because some countries have missing data for some years, we consider three alternative
sub-samples for which we have consistent data. Our preferred sample, using the wholeperiod 1980-2000, consists of translations in the Communist countries Bulgaria, the
Czech Republic, Poland, Romania, Slovakia, Estonia, and Belarus, and the Western
European countries Spain, France, Denmark, Norway, Austria, and Belgium. The
first alternative sample also includes Russia, but only uses the period 1980-1996. The
second alternative sample differs from the preferred sample in that it also includes
Finland, Lithuania, Latvia, Iceland, and Moldova, but only uses the periods 1980-89
and 1995-2000. We present results for the preferred sample only, but results for thetwo alternative samples are similar.
A glance at the countries that translated the influential titles in the pre and
post periods reveals their translation in the Communist countries greatly increased
after the collapse of Communism. Furthermore, the majority of these titles that
were so influential to Western European thought were not published in translation
anywhere in Communist Europe before the collapse of Communism. Specifically,
only 19% of the titles were translated in the period 1980-88 anywhere in Communist
Europe, compared with 61% in the period 1989-2000. Note this implies the collapse
of Communism didn’t merely cause Communist countries to translate into their own
languages titles they’d previously had access to in another Communist language,
such as Russian; it actually increased the titles available in any Communist language.
In contrast, Western Europe had already translated 72% of the titles in the pre
period. Our sample of the titles most frequently translated in Western Europe was
also strongly affected; 30% were translated in the Communist region in the pre period,
where Y ijt is the log of the number of countries translating title i or alternatively
author i (plus one). The dependent variable is defined over the two periods pre (1980-1988) and post (1989-2000) and the two regions Western Europe and Communist
Europe.48 Postt is a dummy for post Communism’s collapse, and AntiComm Authori
is a dummy for whether the author of title i voiced explicitly anti-Communist
opinions. We also include title (or author) fixed effects to test the effect of the
collapse within a title (or an author). We interact these title fixed effects with the
post dummy to allow each title to be translated differently post. The coefficient of
interest is β 1, which tests the extent to which the translations of anti-Communistauthors increased more than the translations of other authors post collapse.
As an alternative to examining the translation of influential titles, we examine
the translation of titles by influential authors. The authors we use are those with a
47OLS regressions that compare Communist countries before and after the collapse yield similarresults (not shown).
48Note this cutoff date of 1989 for “post” differs to the 1991 used in the analysis of the totalnumber of translations. The reason we prefer the 1989 cutoff for the analysis of individual titlesis that by 1989 Gorbachev’s reforms had greatly reduced the Communist regime’s restrictions oninformation flows, so we don’t want to attribute a translation published in 1989 to the pre-collapseperiod. The results are qualitatively similar when using 1991 as the first “post” year, but they aresometimes less significant because some anti-Communist authors were translated as early as 1989,e.g. von Hayek’s famous “The Road to Serfdom”. When dropping the two transition years 1989 and1990 and using 1991 as the first “post” year, the results are unchanged and highly significant. Wealso note that the results from the analysis of the total number of translations discussed in equations1-6 are robust to defining post as 1989 onwards, but there we choose the 1991 cutoff because we testfor an average effect and because Communism did not collapse in the Satellites until 1991.
book appearing on one of the three lists of influential titles given in Section 5.2.2. As
a second alternative that captures readership rather than critics’ views, we take thetitles most frequently translated in Western Europe in the period 1980-2000 (30 from
each field). Compared with the influential titles, these titles, listed in Appendix C.6,
are more likely to be classics or popular works, and less likely to be academic. We
run alternative specifications that replace the anti-Communist author variable with
dummies for whether the title was published during the Communist era and whether
it was published during the Cold War. The premise is that titles published during the
Communist era, especially during the Cold War, would be more threatening to theCommunist regime and thus more likely to be translated by Communist countries only
post collapse. We also run alternative specifications that test whether authors who
won the Nobel prize, and are thus potentially even more influential, were translated
more by Communist countries post collapse.
Table 5.5 presents the estimation results for our preferred sample of countries and
years. The first six columns are author- and title-level regressions of influential titles,
and the last three columns present results from title-level regressions for the mosttranslated titles. We find that overall Communist translation of titles and authors
considered influential in the West and of the most translated titles increased sharply
and significantly post collapse.
Furthermore, compared with other influential titles, titles written by Nobel
laureates and titles first published during the Communist period were both translated
less pre collapse and increased more post collapse. Similarly, titles whose authors
voiced anti-Communist opinions were translated less in Communist countries than
other influential titles pre collapse (significantly in the author specification), but their
translation increased more post collapse to the point they were actually translated
more than other titles. These patterns suggest such titles were more threatening
to the Communist regime, and later increased in popularity, likely because of their
both an increase in Satellites’ translations of older titles and a jump in translations
of newer titles, which reached Western levels. These findings are consistent with bothcatching up on the stock of ideas that were missed out on under Communism and a
convergence between Satellite countries and Western Europe in the diffusion of new
Western ideas. In contrast, we find that the collapse of Communism had little effect
on Western translations in Soviet countries, suggesting the diffusion of Western ideas
into these countries was limited.
The effects of the collapse of Communism varied substantially by book field.
Specifically, we find evidence consistent with some types of Western ideas flowingmore than others into Communist countries. First, Western ideas that were more
suppressed under Communism jumped more after the collapse. The translation of
religious and philosophy titles was heavily suppressed under Communism and jumped
substantially post collapse, but the translation of scientific titles was affected to
a much smaller degree. When focusing on a subset of titles considered the most
influential, we find titles whose authors voiced anti-Communist opinions, titles written
during the Communist era, and titles written by Nobel Laureates were translated lessthan other titles under Communism, and experienced larger increases in translation
post collapse.
Second, the degree of convergence to Western levels of translations varied
substantially across types of Western ideas. Whereas Satellites’ translations of
Western titles in the more scientific fields, which likely contain knowledge that is
more useful for economic development, reached their levels in Western Europe post
collapse, translations in Art and History, which are more cultural, did not increase
by as much.
A key lesson from our study is that incentives play a major role in shaping the
international flow of knowledge. Distortion of these incentives by institutions can
have long-lasting effects that can only be remedied by institutional change.
Naturally, book translations have a number of limitations as a measure of the flow
of ideas. They only allow us to measure idea flows across language barriers, whichprecludes measuring idea flows between countries that share a language, or between
linguistically similar groups within a country. Furthermore, because of the length of
time it takes to write a book, they tend not to capture very new ideas. In addition,
some people are able to read multiple languages, so have access to ideas before they
are translated.49 Finally, ideas in books must by definition be codifiable as opposed
to tacit. That is, they must be able to be expressed in words and written down.
Despite these limitations, translations are an attractive measure of the interna-tional flow of ideas because they capture flows of non-rival, disembodied ideas, and
their key purpose is to transmit written ideas, information and/or knowledge between
languages. Moreover, they are both quantifiable and classifiable by field and specific
content, and thus lend themselves naturally to empirical work.
49However, it is reasonable to assume that such a person finds it less costly to read in his ownlanguage, thus an increase in translations into his native language implies a reduced cost of accessto information.
Figure 5.3: Translations in Communist and Western Europe
Notes: This figure shows translations from Western European and Communist languages inthe former Soviet countries, the Satellite countries, and Western European countries. Thevalues are averages over the countries in the regions, and include translations into the mainlanguage of the country only.
Figure 5.4: The effects over time of the collapse of Communism ontranslations
Translations from Western European languages
Translations from Communist languages
Notes: The coefficients plotted are from the estimation of a version of equation (5.6) inwhich the post dummy and its interactions have been replaced by year dummies (for 1989-2000) and their equivalent interactions. Controls for population and GDP per capita are also
included. The top two figures show coefficients and 95% confidence intervals on interactionsof the year dummies with Western translations in Soviet countries (left panel) and inSatellite countries (right panel). The Western level line is the negative of the coefficient onSoviet (left panel) or Satellite (right panel). The lower two figures show the equivalent fortranslations from Communist languages.
Notes: This figure plots the coefficients (x100) on Communisti × Postt ×WesternLang j(x axis) and Communisti × Postt × CommunistLang j (y axis) from equation (5.4) (withcontrols for log population and GDP per capita) run separately for each subject. Thedependent variable is the log of translations plus one. These coefficients (approximately)measure the percentage change in Communist translations caused by the collapse of
Figure 5.8: Effects of the collapse on translations from English bysubfield
Notes: The regressions that give rise to these coefficients are difference-in-differencesregressions comparing Communist with Western Europe, run by field as described in Section5.4.
T a b l e 5 . 1 : B e f o r e / a f t e r a n a l y s i s : T h e e f f e c t o f t h e c o l l a p s e o f
C o m m u n i s m o n t r a n s l a t i o n s
D e p e n d e n t v a r i a b l e : l o g n u m
b e r o f t r a n s l a t i o n s
( 1 )
( 2 )
( 3 )
( 4 )
( 5 )
( 6 )
( 7 )
( 8 )
( 9 )
P o s t
0 . 4
3 9
0 . 9
2 6 * * *
0 . 7
9 9 * *
( 0 . 2
9 3 )
( 0 . 2
1 6 )
( 0 . 2
6 7 )
T r a n s l a t i o n s f r o m W e s t e r n
o r i g i n a l l a n g u a g e s i n t e r a c t e d w i t h :
P o s t
1 . 5
8 9 * * *
2 . 0 1 4
* * *
1 . 7
6 1 * * *
0 . 2
7 0
0 . 8
9 3 *
0
. 8 0 6 * *
( 0 . 2
5 9 )
( 0 . 2 2 6 )
( 0 . 1
7 9 )
( 0 . 2
7 4 )
( 0 . 4
8 3 )
( 0 . 3
3 3 )
S a t e l l i t e c o u n t r y * p o s t
1 . 7
4 1 * * *
1 . 2
7 1 * *
1 . 1
6 8 * * *
( 0 . 3
3 0 )
( 0 . 4
5 2 )
( 0 . 3
3 6 )
T r a n s l a t i o n s f r o m C o m m u n i s t o r i g i n a l l a n g u a g e s i n t e r a c t e d w i t h :
P o s t
- 1 . 3
7 0 * * *
- 0 . 9 4 5
* * *
- 1 . 1
6 0 * * *
- 1 . 7
7 6 * * *
- 1 . 1
5 4 * *
- 1
. 4 2 1 * * *
( 0 . 1
7 9 )
( 0 . 1 1 3 )
( 0 . 1
8 6 )
( 0 . 4
1 1 )
( 0 . 4
4 5 )
( 0 . 4
5 3 )
S a t e l l i t e c o u n t r y * p o s t
0 . 5
5 9
0 . 0
9 1
0 . 2
0 6
( 0 . 4
4 5 )
( 0 . 3
7 5 )
( 0 . 4
8 4 )
O t h e r c o n t r o l s :
R e a l G D P p e r c a p i t a ( l n )
1 . 7
1 6 * * *
1 . 2
6 6 *
1 . 4 9 4
* * *
0 . 6
9 1 *
0 . 9
8 9 *
0 . 2
8 8
( 0 . 3
9 7 )
( 0 . 6
1 6 )
( 0 . 2 9 0 )
( 0 . 3
3 1 )
( 0 . 5
5 2 )
( 0 . 3
5 3 )
P o p u l a t i o n ( l n )
0 . 6
2 4 * * *
- 8 . 6
2 1 * *
0 . 5 4 9
* * *
- 4 . 9
5 3 * *
0 . 7
1 7 * * *
- 2 . 9
3 0
( 0 . 0
9 2 )
( 3 . 2
4 2 )
( 0 . 0 9 1 )
( 2 . 0
9 6 )
( 0 . 1
5 6 )
( 1 . 8
1 0 )
W e s t e r n o r i g i n a l l a n g u a g e d u
m m y
Y e s
Y e
s
Y e s
Y e s
Y e s
Y e s
C o m m u n i s t o r i g i n a l l a n g u a g e d u m m y
Y e s
Y e
s
Y e s
Y e s
Y e s
Y e s
S a t e l l i t e c o u n t r y * W e s t e r n o
r i g i n a l l a n g u a g e
Y e s
Y e s
S a t e l l i t e c o u n t r y * C o m m u n i s t o r i g i n a l l a n g u a g e
Y e s
Y e s
C o u n t r y f i x e d e f f e c t s
Y e s
C o u n t r y f i x e d e f f e c t s * W e s t e r n o r i g i n a l l a n g u a g e
Y e s
Y e s
C o u n t r y f i x e d e f f e c t s * C o m m u n i s t o r i g i n a l l a n g u a g e
Y e s
Y e s
R - S q u a r e d
0 . 0
2 8
0 . 3
5 6
0 . 7
4 0
0 . 2
4 5
0 . 4 2 2
0 . 8
6 9
0 . 4
2 5
0 . 6
6 1
0 . 8
8 0
O b s e r v a t i o n s
2 5 6
2 5 6
2 5 6
5 1 1
5 1 1
5 1 1
5 1 1
5 1 1
5 1 1
A n o b s e r v a t i o n i s a :
c o u n t r y , y e a r
c o u n t r y , y e a r , o r i g i n a l l a n g u a g e ( W e s t e r n o r C o m m u n i s t )
N o t e s : A l l c o l u m n s
a r e O L S r e g r e s s i o n s u s i n g a n n u a l d a t a f o r t h e p e r i o d 1 9 8 0 - 2 0 0 0 . C o l u m n s 1 - 3 e s t i m a t e e q u a t i o n ( 5 . 1 )
f r o m t h e p a p e r ; c o l u m n s 4 - 6 e s t i m a t e e q u a t i o n ( 5 . 3 ) ; c o l u m n s 7 - 9 e s t i m a t e e q u a t i o n ( 5 . 5 ) . T h e c o u n t r i e s u s e d
i n t h e
a n a l y s i s a r e R u s s i a ,
B e l a r u s , E s t o n i a , L a t v i a , L i t h u
a n i a , M o l d o v a , t h e U k r a i n e , B u l g a r i a , t h e C z e c h R e p u b l i c , H u
n g a r y ,
P o l a n d , R o m a n i a , a n d S l o v a k i a . W e i n c l u d e t h e t h r e e B a l t i c c o u n t r i e s i n t h e S a t e l l i t
e c o u n t r i e s ( s e e e x p l a n a t i o n i n S
e c t i o n
5 . 3 . 1 ) . T h e C o m m u
n i s t a n d W e s t e r n o r i g i n a l l a n g u
a g e s a r e g i v e n i n f o o t n o t e 2 5 . W e i n c l u d e t r a n s l a t i o n s i n t o t h e
m a i n
l a n g u a g e o f t h e c o u n t r y o n l y .
P o s t i s a d u m m y f o r 1 9 9 1 o n w a r d s . S t a n d a r d e r r o r s , i n p a r e n t h e s e s , a r e c l u s t e r e d
a t t h e
c o u n t r y l e v e l . * p < 0 . 1 0 , * * p < 0 . 0 5 , * * * p < 0 . 0 1 .
i ff er en c e s OL S r e gr e s s i on s u s i n g
ann u al d a t af or t h e p er i o d 1 9 8 0 -2
0 0 0 , wi t h C omm uni s t
E ur o
p e a s t h er e gi on of i n t er e s t an d W
e s t er nE ur o p e a s t h e c om p ar i s on gr o u p. C ol umn s 1 - 5 e s t i m a t e e q u a t i on ( 5 .4 ) f r om t h e
p a p e
r ; c ol umn s 6 -1 0 e s t i m a t e e q u a t i on ( 5 . 6 ) .Th e C omm uni s t c o un t r i e s u s e d i n t h e an al y s i s ar e R u s s i a ,B el ar u s ,E s t oni a ,
L a t v
i a ,L i t h u ani a ,M ol d o v a , t h e Uk r ai n e ,B ul g ar i a , t h e Cz e ch R e p u
b l i c ,H un g ar y ,P ol an d , R om ani a , an d S l o v ak i a.Th e
W e s t er nE ur o p e an c o un t r i e s u s e d ar eA u s t r i a ,B el gi um , S wi t z er l an d ,D enm ar k , S p ai n ,F i nl an d ,F r an
c e ,I c el an d ,I t al y , t h e
N e t h
er l an d s ,N or w a y ,P or t u g al , an d S w e d en. W ei n cl u d e t h e t h r e eB al t i c c o un t r i e s i n t h e S a t el l i t e c o un t r i e s ( s e e ex pl an a t i on
i n S e c t i on 5 . 3 .1 ) .Th e C omm uni s t an d W e s t er n or i gi n al l an g u a g e s a
r e gi v eni nf o o t n o t e2 5 . W ei n c
l u d e t r an s l a t i on s i n t o
t h em ai nl an g u a g e of t h e c o un t r y onl y.P o s t i s a d umm yf or 1 9 9 1
on w ar d s .P o p u l a t i o n a n d GDP
c o n t r o l s ar e t h el o g s
of p o p ul a t i on an d of r e al GDP p er c a pi t a. C o u n t r y- s p e c i fi c t i m e t r e n d s ar el i n e ar . S t an d ar d er r or
T a b l e 5 . 3 : C o n v e r g e n c e a n a l y s i s : T h e e f f
e c t o f t h e c o l l a p s e o f C
o m m u n i s m o n t r a n s l a t i o n s o f
r e c e n t v e r s u s o
l d e r W e s t e r n t i t l e s
D e p e n d e n t v a r i a b l e : l o
g n u m b e r o f t r a n s l a t i o n s f r o m a W e s t e r n o r i g i n a l l a n g u a g e
F l o w s : t i t l e s 1 5 y e
a r s o l d a n d n e w e r
S t o c k s : t i t l e s o l d e r t h a n 1 5 y e a r s
( 1 )
( 2 )
( 3 )
( 4 )
( 5 )
( 6 )
( 7 )
( 8 )
( 9 )
( 1
0 )
P o s t * C o m m u n i s t c o u
n t r y
1 . 4
1 7 * * *
2 . 1 1
4 * * *
1 . 4 8
5 * * *
0 . 7
2 7 *
1 . 4
0 8 * * *
1 . 2
6 3 * * *
1 . 9
6 0 * * *
1 . 3
3 1 * * *
0 . 5
7 3
1 . 2 5
4 * * *
( 0 . 2
8 3 )
( 0 . 3
5 2 )
( 0 . 3
0 5 )
( 0 . 3
8 5 )
( 0 . 3
2 5 )
( 0 . 2
8 3 )
( 0 . 3
5 2 )
( 0 . 3
0 5 )
( 0 . 3
8 5 )
( 0 . 3 2 5 )
C o m m u n i s t c o u n t r y
- 2 . 9
6 6 * * * - 1 . 9
9 7 * * *
- 2 . 0
2 9 * * *
- 1 . 0
6 1 *
( 0 . 4
8 4 )
( 0 . 5
9 7 )
( 0 . 4
8 4 )
( 0 . 5
9 7 )
P o s t
0 . 4
2 8 * * *
0 . 1 1
9
0 . 5 3
0 * * *
0 . 2
3 6
- 0 . 0
2 7
- 0 . 3
3 5 *
0 . 0
7 6
- 0 . 2
1 9
( 0 . 1
2 5 )
( 0 . 1
7 8 )
( 0 . 1
7 3 )
( 0 . 1
3 9 )
( 0 . 1
2 5 )
( 0 . 1
7 8 )
( 0 . 1
7 3 )
( 0 . 1
3 9 )
P o p u l a t i o n a n d G D P c o n t r o l s
Y e s
Y e s
Y e s
Y e s
Y e s
Y e s
Y e s
Y
e s
C o u n t r y f i x e d e f f e c t s
Y e s
Y e s
Y e s
Y e s
Y e s
Y
e s
C o u n t r y - s p e c i f i c t i m e
t r e n d s
Y e s
Y e s
Y e a r f i x e d e f f e c t s
Y e s
Y
e s
R - S q u a r e d
0 . 4
7 8
0 . 6
1 0
0 . 9
3 4
0 . 9
5 7
0 . 9
4 3
0 . 2
6 9
0 . 4
5 3
0 . 9
0 8
0 . 9
4 0
0 . 9 2 0
O b s e r v a t i o n s
5 0 0
4 8 2
4 8 2
4 8 2
4 8 2
5 0 0
4 8 2
4 8 2
4 8 2
4 8 2
A n o b s e r v a t i o n i s a c o
u n t r y , y e a r
N o t e s : A l l c o l u m n s
a r e d i ff e r e n c e - i n - d i ff e r e n c e s O L S r e g r e s s i o n s ( e q u a t i o n ( 5 . 2 ) ) u s i n g a n n u a l d a t a f o r t h e p e r i o d
1 9 8 0 -
2 0 0 0 , w i t h C o m m u n i s t E u r o p e a s t h e r e g i o n o f i n t
e r e s t a n d W e s t e r n E u r o p e a s t h e c o m p a r i s o n g r o u p . T h e d e p e n d e n t
v a r i a b l e f o r c o l u m n s 1 - 5 i s t r a n s l a t i o n s o f r e c e n t t i t
l e s , a n d f o r c o l u m n s 6 - 1 0 i s t r a n s l a t i o n s o f o l d e r t i t l e s . S e e t h e
n o t e s
t o T a b l e 5 . 2 f o r t h e
C o m m u n i s t a n d W e s t e r n c o u n t r
i e s u s e d . T h e W e s t e r n o r i g i n a l l a n g u a g e s a r e g i v e n i n f o o t n o t e 2
5 . W e
i n c l u d e t r a n s l a t i o n s
i n t o t h e m a i n l a n g u a g e o f t h e c o u n t r y o n l y .
P o s t i s a d u m m y f o
r 1 9 9 1 o n w a r d s . P o p u
l a t i o n a n d
G D P
c o n
t r o l s a r e t h e l o g s
o f p o p u l a t i o n a n d o f r e a l G D P p e r c a p i t a .
C o u n
t r y - s p
e c i fi c
t i m e
t r e n
d s a r e l i n e a r . S t a n d a r d e r r o r s , i n
p a r e n t h e s e s , a r e c l u s t e r e d a t t h e c o u n t r y l e v e l . * p < 0 . 1 0 , * * p < 0 . 0 5 , * * * p < 0 . 0 1 .
e n d e n t v a r i a b l e : l o gn um b e r of t r a n s l a t i on s f r om a W e s t e r n or i gi n a l l a n g u a g e
F
l o w s : t i t l e s 1 5 y e ar s ol d an d n e w e r
S t o c k s : t i t l e s ol d e r t h
an1 5 y e ar s
( 1 )
( 2 )
( 3 )
( 4 )
( 5 )
( 6 )
( 7 )
( 8 )
( 9 )
( 1 0 )
P o s t * C omm uni s t c o un t r
1 .4 1 7 * * *
2 .1 1 4 * * *
1 .4 8 5 * * *
0 . 7 2 7 *
1 .4 0 8 * * *
1 .2 6 3 * * *
1 . 9 6 0 * * *
1 . 3 3 1 * * *
0 . 5 7 3
1 .2 5 4 * * *
( 0 .2 8 3 )
( 0 . 3 5 2 )
( 0 . 3 0 5 )
( 0 . 3 8 5 )
( 0 . 3 2 5 )
( 0 .2 8 3 )
( 0 . 3 5 2 )
( 0 . 3 0 5 )
( 0 . 3 8 5 )
( 0 . 3 2 5 )
C om
m uni s t c o un t r
-2 . 9 6 6 * * * -1 . 9 9 7 * * *
-2 . 0 2 9 * * *
-1 . 0 6 1 *
( 0 .4 8 4 )
( 0 . 5 9 7 )
( 0 .4 8 4 )
( 0 . 5 9 7 )
P o s t
0 .4 2 8 * * *
0 .1 1 9
0 . 5 3 0 * * *
0 .2 3 6
- 0 . 0 2 7
- 0 . 3 3 5 *
0 . 0 7 6
- 0 .2 1 9
( 0 .1 2 5 )
( 0 .1 7 8 )
( 0 .1 7 3 )
( 0 .1 3 9 )
( 0 .1 2 5 )
( 0 .1 7 8 )
( 0 .1 7 3 )
( 0 .1 3 9 )
P o p ul a t i on a n d GDP c on t r ol s
Y e s
Y e s
Y e s
Y e s
Y e s
Y e s
Y e s
Y e s
C o u
n t r f i x e d e f f e c t s
Y e s
Y e s
Y e s
Y e s
Y e s
Y e s
C o u
n t r y- s p e c i f i c t i m e t r e n d s
Y e s
Y e s
Y e a r f i x e d e f f e c t s
Y e s
Y e s
R- S q u a r e d
0 .4 7 8
0 . 6 1 0
0 . 9 3 4
0 . 9 5 7
0 . 9 4 3
0 .2 6 9
0 .4 5 3
0 . 9 0 8
0 . 9 4 0
0 . 9 2 0
O b s e r v a t i on s
5 0 0
4 8 2
4 8 2
4 8 2
4 8 2
5 0 0
4 8 2
4 8 2
4 8 2
4 8 2
An o b s e r v a t i oni s a c o un t r , e a r
N o t e
s : Al l c ol umn s u s e ann u al d a t af or t h e p er i o d 1 9 8 0 -2 0 0 0 . C ol um
n s 1 -4 ar e b ef or e / af t er OL S r e gr
e s s i on s u s i n g onl y t h e
C om
m uni s t c o un t r i e s ( e q u a t i on ( 5 .1 ) ) ; c ol umn s 5 - 9 ar e d i ff er en c e-i n- d
i ff er en c e s OL S r e gr e s s i on s wh er e t h er e gi on of i n t er e s t
i s C o
mm uni s t c o un t r i e s an d t h e c om p ar i s on gr o u pi s W e s t er nE ur o p e ( e q u a t i on ( 5 .2 ) ) .Th e C omm un
i s t c o un t r i e s u s e d ar e
B el ar u s ,B ul g ar i a ,E s t oni a ,H un g ar y ,L a t vi a ,P ol an d , R om ani a an d t h e Uk r ai n e , an d t h e W e s t er nE ur o p e an c o un t r i e s u s e d
ar eB
el gi um ,D enm ar k ,F i nl an d ,F r a
n c e ,I c el an d ,I t al y ,N e t h er l an d s ,N or w a y ,P or t u g al , S p ai n , S w e d en , an d S wi t z er l an d .
P o s t
i s a d umm yf or 1 9 9 1 on w ar d s . C o u n t r y- s p e c i fi c t i m e t r e n d s ar e
The existence of multiple languages and the necessity of translating between them
raise the interesting question of what is the cost imposed by language barriers in
terms of access to ideas, relative to the counterfactual of everyone speaking the
same language.1 The one-language world is an interesting counterfactual because it
minimizes barriers to idea diffusion and, once established, involves no more language-
learning than a situation where different populations speak different languages. The
languages spoken by people in different parts of the world are the result of many
historical events and processes, and it may well be the case that such a multi-language
situation is a highly inefficient equilibrium.
To estimate the cost of status quo in terms of access to written ideas, consider
the question of how many books the average person has access to under four different
scenarios. In the first scenario, everyone speaks the same language, thus every written
book is accessible to everyone. In the second scenario, people in different countries
speak different languages, and there is neither multilingualism nor translation of titles.
Here an individual only has access to the titles written in his own country. In the
third scenario, people in different countries speak different languages and there is
no multilingualism, but some titles are translated. Thus an individual has access to
those titles written in his country plus those translated into his language. In the
final scenario, people in different countries speak different languages natively and
there are no translations, but everyone also speaks a second language, namely the
language that gives him access to the most additional titles. Practically, this means
everyone who doesn’t speak English natively learns it as their second language, and
native English speakers learn German. However, original titles still cannot be read
1There are, of course, many more aspects to language and language barriers than access to ideaswritten in books, but quantifying how language barriers affect access to books remains informativeabout one aspect of the cost of multiple languages.
by everyone because they are written in a range of different languages.
I consider access to titles in each scenario under two alternative assumptions
about what constitutes an interesting title. First, I assume that all titles published
are interesting. In both cases, I also assume the original titles that are written in
each country do not vary by scenario. Note this assumption could be violated if
potential market size or competition from other titles affect what books are actually
published. Using average annual data for 1995 to 1999 on original publications and
translations, and population data from 1999 for the 49 countries for which all these
are available, I find average access to titles is 640,326 when everyone speaks thesame language, compared with just 66,766 (10.4% of total titles) in the scenario with
multiple languages, 69,463 (10.8%) with multiple languages and translations, and
163,127 (25.5%) when everyone is bilingual. These (admittedly crude) calculations
demonstrate that access to titles is drastically reduced by the existence of multiple
languages, and translations do relatively little to combat this effect.
However, titles vary in importance, and it may be that most of the titles that are
not translated would be of no interest to people in foreign countries (and perhaps
of relatively little interest to people at home as well). I thus alternatively assume
the extreme case that only titles that are ever translated are of interest to anyone,
and that the number of titles translated out of a language is the maximum of the
number of titles translated from that language into any one target language in a
country. Then average access to titles falls to 11,204 in the one-language scenario,
3,517 (31.4%) in the multiple-language scenario, 6,214 (55.5%) with translations, and
7,366 (65.7%) with bilingualism.
Although under both assumptions universal bilingualism leads to higher access
to titles than translations, the cost is also likely to be much higher. Universal
bilingualism in the countries in the sample would imply 2.6 billion people must learn a
second language; this compares with just under 65,000 titles being translated. Under
Table 6.1: Average access to titles under four counterfactuals:
number and % of titles written
Counterfactual: Onelanguage
Multiplelanguages
Multiplelanguages,translations
Multiplelanguages,bilingualism
Assumption:
All books are 640,326 66,766 69,463 163,127interesting 100% 10.4% 10.8% 25.5%
Only books thatare ever 11,204 3,517 6,214 7,366translated are 100% 31.4% 55.5% 65.7%interesting
Notes: The columns of this table present the average person’s access to titles under fouralternative counterfactual scenarios and two alternative assumptions about what constitutesan interesting title. The counterfactuals are as follows: i) everyone speaks the samelanguage, and all books are written in this language; ii) everyone speaks only their native
language, books are distributed across languages as given by the data, and there are notranslations or multilingualism; iii) as case ii, but translations occur as given by the data;and iv) as case ii, but everyone is bilingual in their native language and the foreign languagethat gives them the greatest additional access to titles. The first row assumes all titlespublished are interesting; the second row assumes only titles that are ever translated areinteresting. Each cell in the table gives the number and percentage of total interesting titlesthe average person can read.
any reasonable assumptions, the cost of the translations is much lower.
Comparisons of either scenario with the case where there is only one language
reveal that the cost of the existence of multiple languages in terms of access to written
works is very large. However, there would be large costs of transitioning to a situation
where everyone speaks the same language natively, even ignoring the cultural heritage
On the other hand, it seems we are increasingly moving towards using English
as the international lingua franca . Historically, lingua francas such as Arabic inthe Islamic Empire, Latin for European scholars until the eighteenth century, and
French within diplomatic circles have emerged organically within specific geographic
or social areas. English is already used widely in many parts of the world where it is
not spoken natively, and it may be that the increasing ease of global communication
through means such as the internet will encourage its spread to all corners of the
globe. Revealingly, translation patterns over the past half century show an increase
in the dominance of English as a source language, especially with the collapse of Communism in Eastern Europe, which caused the former Communist countries to
switch from predominantly translating from Russian to predominantly translating
from English. This suggests an increase in the extent to which titles of international
interest are written in English, though it remains an open question whether this trend
will continue to the point where all communication internationally occurs in English.
One force that may work in the opposite direction is the improvement in machine
translation. Although still far from perfect, the ability of computer programs
to translate between languages (and generate meaningful output) has increased
dramatically over recent years. This reduces the costs of translation, and thus the cost
of the existence of multiple languages. However, it seems unlikely that technology will
be able to fully bridge all language barriers in the foreseeable future, and thus language
differences are likely to remain a barrier to idea transmission, with implications for
economic development and intellectual advancement, for some time to come.
A natural further question to ask is how translation flows affect important economic
outcomes such as GDP and economic growth. This is a challenging question to address
convincingly because of the obvious problems of reverse causality and unobserved
heterogeneity. For example, as countries become richer they inevitably translate
more, and it may be that countries with populations that are more inclined to read
both translate more and grow faster. In addition, the effect of translation flows oneconomic outcomes is expected to be distributed over a number of years after the
flow occurs, making it more difficult to identify empirically. To study the causal
effect of translations on growth will thus require an instrumental variable that affects
translation flows without affecting growth directly. Such a variable is difficult to find.
I thus leave questions of the effects of translations on economic outcomes for future
research.
However, the evidence I present in this dissertation is consistent with a positiveeffect of idea flows on growth. First, as shown in Table 3.7, there is a positive
correlation between inward translation flows and GDP per capita in the country.
Second, as I show in chapter 5, after the collapse of Communism (which drove
the largest change in translation patterns during the period I study), the Satellite
countries both increased their translations of Western titles more than the Soviet
countries, and had better growth outcomes over the following decade.
Language Major countriesJavanese IndonesiaKannada IndiaKazakh KazakhstanKikuyu KenyaKinyarwanda RwandaKonkani* IndiaKorean South Korea, North KoreaKurdish* Iraq, Iran, TurkeyLahnda* PakistanLombard ItalyMaithili India
Malagasy* MadagascarMalay* Malaysia, IndonesiaMalayalam IndiaMarathi IndiaModern Greek Greece, CyprusNeapolitan ItalyNepali NepalNyanja MalawiOriya IndiaOromo* EthiopiaPanjabi India
Persian* Iran, AfghanistanPolish PolandPortuguese Portugal, BrazilPushto* Pakistan, AfghanistanQuechua* Peru, BoliviaRajasthani* IndiaRomanian* Romania, MoldovaRussian Russia, Israel, Kazakhstan, KyrgyzstanSerbo-Croatian* Serbia and Montenegro, Bosnia and Herzegovina, CroatiaShona Zimbabwe
Sindhi Pakistan, IndiaSinhala Sri LankaSomali Somalia, EthiopiaSpanish Spain, Argentina, Bolivia, Chile, Colombia, Costa Rica,
Cuba, Dominican Republic, Ecuador, El Salvador, Guatemala,Honduras, Mexico, Nicaragua, Panama, Peru, Puerto Rico,Uruguay, Venezuela
Swedish SwedenTamil India, Sri LankaTatar RussiaTelugu IndiaThai* ThailandTurkish TurkeyTurkmen Turkmenistan, IranUighur ChinaUkrainian UkraineUrdu Pakistan, IndiaUzbek* Uzbekistan, Tajikistan, AfghanistanVietnamese VietnamXhosa South AfricaYoruba NigeriaZulu South Africa
Notes: This table lists the “top 100” languages (aggregated to the macrolanguage level),and their “major” countries as described in section 3.1.1. Asterisks denote macrolanguages.
Translations from Communist original languages in:
Communist country * post -1.213*** -0.370* -0.562 -1.568*** -0.748* -0.934*
(0.212) (0.194) (0.356) (0.410) (0.429) (0.537)
Satellite country * post 0.512 0.154 0.230
(0.432) (0.305) (0.553)
Communist country 1.783*** 2.857*** 1.813*** 2.043***
(0.330) (0.413) (0.448) (0.490)
Satellite country -0.044 0.619
(0.409) (0.549)
Post -0.193* -0.556*** -0.422* -0.193* -0.450*** -0.389*
(0.110) (0.132) (0.232) (0.112) (0.145) (0.227)
Other controls:
Western original language dummy Yes Yes Yes Yes Yes Yes
Communist original language dummy Yes Yes Yes Yes Yes Yes
Population and GDP controls Yes Yes Yes Yes
Country fixed effects * Western original language Yes YesCountry fixed effects * Communist original language Yes Yes
R-Squared 0.641 0.755 0.982 0.698 0.838 0.986
Observations 104 100 100 104 100 100
An observation is a country, pre/post, original language (Western or Communist)
Notes: All columns are difference-in-differences OLS regressions using using data aggregatedto the pre/post collapse level, with Communist Europe as the region of interest and WesternEurope as the comparison group. Columns 1-3 estimate equation (5.4) from the paper;columns 4-6 estimate equation (5.6).“Pre” values are the average over the years 1980-89;“post” values are the average over the years 1992-2000. See the notes to Table 5.2 forthe Communist and Western countries used. We include the three Baltic countries in the
Satellite countries (see explanation in Section 5.3.1). The Communist and Western originallanguages are given in footnote 25. We include translations into the main language of thecountry only. Population and GDP controls are the logs of population and of real GDP percapita. Standard errors, in parentheses, are clustered at the country level. * p<0.10, **p<0.05, *** p<0.01.
a r e O L S r e g r e s s i o n s u s i n g a n n u a l d a t a . C o l u m n s 1 - 6 a r e f o r
t h e y e a r s 1 9 8 0 - 2 0 0 0 ; c o l u m n s 7
- 9 a r e
f o r 1 9 8 9 - 2 0 0 0 . T h e
c o u n t r i e s u s e d i n t h e a n a l y s i s a r e R u s s i a , B e l a r u s , E s t o n i a , L a t
v i a , L i t h u a n i a , M o l d o v a , t h e U k
r a i n e ,
B u l g a r i a , t h e C z e c h
R e p u b l i c , H u n g a r y , P o l a n d , R o m a n i a , a n d S l o v a k i a . T h e C o m m
u n i s t a n d W e s t e r n o r i g i n a l l a n g u a g e s
a r e g i v e n i n f o o t n o t e
2 5 . W e i n c l u d e t r a n s l a t i o n s i n t o t h e m a i n l a n g u a g e o f t h e c o u n t r y o n l y . T h e v a r i a b l e s I n s t i t u
t i o n a
l i z e d
d e m o c r a c y ,
P o
l i t i c a l
c o m p e
t i t i o n ,
P r i c e
l i b e r a
l i z a t i o n , a n d T r a
d e a n
d f o
r e i g n e x c
h a n g e
s y s t e m
r e f o
r m
a r e m e a s u r e s o f a
s p e c t s
o f t h e d e g r e e o f r e f o
r m f r o m c o m m u n i s t c e n t r a l l y - p l a n n e d e c o n o m y t o d e m o c r a t i c m
a r k e t e c o n o m y . T h e y a r e d e s c r i b e d i n
d e t a i l i n S e c t i o n C . 2 . 1 . P o p u
l a t i o n a n
d G D P c o n t r
o l s a r e t h e l o g s o f p o p u l a t i o n a n d o f r e a l G D P p e r c a p i t a . S t a
n d a r d
e r r o r s , i n p a r e n t h e s e s , a r e c l u s t e r e d a t t h e c o u n t r y l e v e l . * p < 0 . 1 0 , * * p < 0 . 0 5 , * * * p < 0 . 0 1 .
r e d i ff e r e n c e - i n - d i ff e r e n c e s O L S r e g r e s s i o n s u s i n g a n n u a l d a t a f o r
t h e p e r i o d 1 9 8 0 - 2 0 0 0 , w i t h C o m m
u n i s t
E u r o p e a s t h e r e g i o n o f i n t e r e s t a n d W e s t e r n E u r o p
e a s t h e c o m p a r i s o n g r o u p . C o l u m n s 1 - 5 e s t i m a t e e q u a t i o n ( 5 . 4 ) f r o m
t h e p a p e r ; c o l u m n s
6 - 1 0 e s t i m a t e e q u a t i o n ( 5 . 6 ) . S e e t h e n o t e s t o T a b l e 5 . 2 f o r t h e C o m m u n i s t a n d W e s t e r n c o u
n t r i e s
u s e d . W e i n c l u d e t h
e t h r e e B a l t i c c o u n t r i e s i n t h e S a t e l l i t e c o u n t r i e s ( s e e e x p l a n a t i o n i n S e c t i o n 5 . 3 . 1 ) . T h e C o m m
u n i s t
a n d W e s t e r n o r i g i n a
l l a n g u a g e s a r e g i v e n i n f o o t n o t e
2 5 . W e i n c l u d e t r a n s l a t i o n s i n t o t h e m a i n a n d s e c o n d a r y l a n g u a g e s o f
t h e c o u n t r y . P o s t i s
a d u m m y f o r 1 9 9 1 o n w a r d s . P o
p u
l a t i o n a n
d G D P c o n
t r o l s a r e t
h e l o g s o f p o p u l a t i o n a n d o f r e a l G D P
p e r c a p i t a .
C o u n
t r y
- s p e c
i fi c
t i m e
t r e n
d s a r e l i n e a r .
S t a n d a r d e r r o r s , i n p a r e n t h e s e s
, a r e c l u s t e r e d a t t h e c o u n t r y l e
a r e d i ff e r e n c e - i n - d i ff e r e n c e s r e g r e s s i o n s ( e q u a t i o n ( 5 . 4 ) ) u s i n g a n n u a l d a t a f o r t h e p e r i o d 1 9 8 0
- 2 0 0 0 ,
w i t h C o m m u n i s t E u
r o p e a s t h e r e g i o n o f i n t e r e s t a n d W e s t e r n E u r o p e a s t h e c o m p a r i s o n g r o u p . S e e t h e n o t e s t o T a b l e 5 . 2
f o r t h e C o m m u n i s t a n d W e s t e r n c o u n t r i e s u s e d . T h e C o m m u n i s t a n d W e s t e r n o r i g i n a l l a n g u a g e s a r e g i v e n i n f o o t n o t e 2 5 .
W e i n c l u d e t r a n s l a t i o n s i n t o t h e m a i n l a n g u a g e o f t
h e c o u n t r y o n l y .
P o s t i s a d u m m y f o r 1 9 9 1 o n w a r d s . P o p u
l a t i o
n a n
d
G D P
c o n
t r o l s a r e t h e l o g s o f p o p u l a t i o n a n d o f r e a
l G D P p e r c a p i t a .
C o u n
t r y - s p e
c i fi c
t i m e
t r e n
d s a r e l i n e a r . S t a
n d a r d
e r r o r s , i n p a r e n t h e s e s , a r e c l u s t e r e d a t t h e c o u n t r y l e v e l . * p < 0 . 1 0 , * * p < 0 . 0 5 , * * * p < 0 . 0 1 .
is that, because both Western translations and the degree of transition increase over
time in most countries, the effects in this specification may be driven by the presenceof two unrelated time trends. We thus add year fixed effects interacted with original
language in the second column of each group. The concern remains that we are
identifying off levels differences between countries, and countries differ across many
more dimensions than just their degree of transition away from Communism, so we
add country dummies interacted with original language in the third columns. Thus
in the final column of each group, the coefficient of interest is identified solely off
between-country differences in changes over time.The two variables directly related to the political system, institutionalized
democracy and political competition , are both positively and significantly related
to translations from Western European languages. These results suggest that
Communist countries that transitioned more away from Communism experienced
a higher jump in Western European translations. For instance, the regression with
country and year fixed effects shows an increase in institutionalized democracy score
from 7, the 25th percentile in 2000, to 9, the 75th percentile in 2000, corresponds toa 32% increase in translations from the West. The transition away from Communism
consisted of various broad-ranging reforms, and in columns 7 to 9 we test the relative
importance of two relevant reforms, namely price and trade deregulations. The
regressions suggest that while trade and foreign exchange system reform was a more
important driving force of increasing translations from Western European languages,
price liberalization was more important in reducing translations from Communist
languages. These results suggest that, while trade barriers kept translations from theWest artificially low, the Communist price control system kept between-Communist
Hofstede, Geert. 1983. “Dimensions of national cultures in fifty countries and three
regions.” In Explications in Cross-cultural Psychology . , ed. J.B. Deregowski, S.Dziurawiec and R.C. Annis, 335–355. Lisse, Netherlands:Swets and Zeitlinger.