80516 Volume 25 • Number 1 • 2011 ISSN 0258-6770 (PRINT) ISSN 1564-698X (ONLINE) THE WORLD BANK ECONOMIC REVIEW Volume 25 • 2011 • Number 1 THE WORLD BANK ECONOMIC REVIEW SYMPOSIUM ISSUE ON INTERNATIONAL MIGRATION AND DEVELOPMENT Five Questions on International Migration and Development C ¸ ag ¨ zden, Hillel Rapoport, and Maurice Schiff ˘lar O Part I. International Migration Where on Earth is Everybody? The Evolution of Global Bilateral Migration 1960–2000 C ¸ ag ¨ zden, Christopher R. Parsons, Maurice Schiff, ˘lar O and Terrie L. Walmsley Immigration Policies and the Ecuadorian Exodus Simone Bertoli, Jesús Fernández-Huertas Moraga, and Francesc Ortega Do Migrants Improve Governance at Home? Evidence from a Voting Experiment Catia Batista and Pedro C. Vicente Part II. International Remittances What Explains the Price of Remittances? An Examination Across 119 Country Corridors Thorsten Beck and María Soledad Martínez Pería Pages 1–156 Remittances and the Brain Drain Revisited: The Microdata Show That More Educated Migrants Remit More Albert Bollard, David McKenzie, Melanie Morten, and Hillel Rapoport www.wber.oxfordjournals.org 2 THE WORLD BANK ECONOMIC REVIEW editors Alain de Janvry and Elisabeth Sadoulet, University of California at Berkeley Jaime de Melo, Editor until July 1, 2010, has been technically responsible for this issue. assistant to the editor Marja Kuiper editorial board Downloaded from http://wber.oxfordjournals.org/ at International Monetary Fund on August 19, 2013 Harold H. Alderman, World Bank (retired) William F. Maloney, World Bank Pranab K. Bardhan, University of California, David J. McKenzie, World Bank Berkeley Jaime de Melo, University of Geneva Scott Barrett, Columbia University, USA Juan-Pablo Nicolini, Universidad Torcuato di Asli Demirgüç-Kunt, World Bank Tella, Argentina Jean-Jacques Dethier, World Bank Nina Pavcnik, Dartmouth College, USA Quy-Toan Do, World Bank Vijayendra Rao, World Bank Frédéric Docquier, Catholic University of Martin Ravallion, World Bank Louvain, Belgium Jaime Saavedra-Chanduvi, World Bank Eliana La Ferrara, Università Bocconi, Italy Claudia Paz Sepúlveda, World Bank Francisco H. G. Ferreira, World Bank Joseph Stiglitz, Columbia University, USA Augustin Kwasi Fosu, United Nations Jonathan Temple, University of Bristol, UK University, WIDER, Finland Romain Wacziarg, University of California, Paul Glewwe, University of Minnesota, Los Angeles, USA USA Dominique Van De Walle, World Bank Ann E. Harrison, World Bank Christopher M. Woodruff, University of Philip E. Keefer, World Bank California, San Diego Justin Yifu Lin, World Bank Yaohui Zhao, CCER, Peking University, Norman V. Loayza, World Bank China The World Bank Economic Review is a professional journal used for the dissemination of research in development economics broadly relevant to the development profession and to the World Bank in pursuing its development mandate. It is directed to an international readership among economists and social scientists in government, business, international agencies, universities, and development research institutions. The Review seeks to provide the most current and best research in the field of quantita- tive development policy analysis, emphasizing policy relevance and operational aspects of economics, rather than primarily theoretical and methodological issues. Consistency with World Bank policy plays no role in the selection of articles. The Review is managed by one or two independent editors selected for their academic excellence in the field of development economics and policy. The editors are assisted by an editorial board composed in equal parts of scholars internal and external to the World Bank. World Bank staff and outside researchers are equally invited to submit their research papers to the Review. For more information, please visit the Web sites of the Economic Review at Oxford University Press at www.wber.oxfordjournals.org and at the World Bank at www.worldbank.org/research/journals. Instructions for authors wishing to submit articles are available online at www.wber.oxfordjournals.org. Please direct all editorial correspondence to the Editor at wber@worldbank.org. THE WORLD BANK ECONOMIC REVIEW Volume 25 † 2011 † Number 1 SYMPOSIUM ISSUE ON INTERNATIONAL MIGRATION AND DEVELOPMENT Five Questions on International Migration and Development 1 C ¸ ag ¨ zden, Hillel Rapoport, and Maurice Schiff ˘ lar O Part I. International Migration Where on Earth is Everybody? The Evolution of Global Bilateral Migration 1960 –2000 12 C¸ ag ¨ zden, Christopher R. Parsons, Maurice Schiff, ˘ lar O and Terrie L. Walmsley Immigration Policies and the Ecuadorian Exodus 57 Simone Bertoli, Jesu ´ ndez-Huertas Moraga, ´ s Ferna and Francesc Ortega Do Migrants Improve Governance at Home? Evidence from a Voting Experiment 77 Catia Batista and Pedro C. Vicente Part II. International Remittances What Explains the Price of Remittances? An Examination Across 119 Country Corridors 105 Thorsten Beck and Marı ´a Soledad Martı ´a ´nez Perı Remittances and the Brain Drain Revisited: The Microdata Show That More Educated Migrants Remit More 132 Albert Bollard, David McKenzie, Melanie Morten, and Hillel Rapoport SUBSCRIPTIONS:A subscription to The World Bank Economic Review (ISSN 0258-6770) comprises 3 issues. Prices include postage; for subscribers outside the Americas, issues are sent air freight. Annual Subscription Rate (Volume 25, 3 Issues, 2011): Institutions—Print edition and site-wide online access: £168/ $252/E252, Print edition only: £154/$231/E231, Site-wide online access only: £140/$210/E210; Corporate—Print edition and site-wide online access: £251/$376/E376, Print edition only: £230/$345/E345, Site-wide online access only: £209/$314/E314; Personal—Print edition and individual online access: £43/$64/E64. US$ rate applies to US & Canada, EurosE applies to Europe, UK£ applies to UK and Rest of World. There may be other subscription rates available; for a complete listing, please visit www.wber.oxfordjournals.org/subscriptions. Readers with mailing addresses in non-OECD countries and in socialist economies in transition are eligible to receive complimentary sub- scriptions on request by writing to the UK address below. Full prepayment in the correct currency is required for all orders. Orders are regarded as �rm, and payments are not refundable. Subscriptions are accepted and entered on a complete volume basis. Claims cannot be con- sidered more than four months after publication or date of order, whichever is later. All subscriptions in Canada are subject to GST. Subscriptions in the EU may be subject to European VAT. If registered, please supply details to avoid unnecessary charges. For subscriptions that include online versions, a proportion of the subscription price may be subject to UK VAT. Personal rates are applicable only when a subscription is for individual use and are not available if delivery is made to a corporate address. Downloaded from http://wber.oxfordjournals.org/ at International Monetary Fund on August 19, 2013 The current year and two previous years’ issues are available from Oxford University Press. Previous BACK ISSUES: volumes can be obtained from the Periodicals Service Company, 11 Main Street, Germantown, NY 12526, USA. E-mail: psc@periodicals.com. Tel: (518) 537-4700. Fax: (518) 537-5899. CONTACT INFORMATION: Journals Customer Service Department, Oxford University Press, Great Clarendon Street, OxfordOX2 6DP, UK. E-mail: jnls.cust.serv@oup.com. Tel: þ44 (0)1865 353907. Fax: þ 44 (0)1865 353485. In the Americas, please contact: Journals Customer Service Department, Oxford University Press, 2001 Evans Road, Cary, NC 27513, USA. E-mail: jnlorders@oup.com. Tel: (800) 852-7323 (toll-free in USA/Canada) or (919) 677-0977. Fax: (919) 677-1714. In Japan, please contact: Journals Customer Service Department, Oxford University Press, Tokyo, 4- 5-10-8F Shiba, Minato-ku, Tokyo, 108-8386, Japan. E-mail: custserv.jp@oup.com. Tel: þ 81 3 5444 5858. Fax: þ 81 3 3454 2929. POSTAL INFORMATION: The World Bank Economic Review (ISSN 0258-6770) is published three times a year, in February, June, and October, by Oxford University Press for the International Bank for Reconstruction and Development/THE WORLD BANK. Send address changes to The World Bank Economic Review, Journals Customer Service Department, Oxford University Press, 2001 Evans Road, Cary, NC 27513-2009. Periodicals postage paid at Cary, NC and at additional mailing of�ces. Communications regarding original articles and editorial management should be addressed to The Editor, The World Bank Economic Review, The World Bank, 3, Chemin Louis Dunant, CP66 1211 Geneva 20, Switzerland. ENVIRONMENTAL AND ETHICAL POLICIES: Oxford Journals, a division of Oxford University Press, is committed to working with the global community to bring the highest quality research to the widest possible audience. Oxford Journals will protect the environment by implementing environmentally friendly policies and practices wherever possible. Please see http://www.oxfordjournals.org/ethicalpolicies.html for further information on environmental and ethical policies. DIGITAL OBJECT IDENTIFIERS: For information on dois and to resolve them, please visit www.doi.org. PERMISSIONS: For information on how to request permissions to reproduce articles or information from this journal, please visit www.oxfordjournals.org/jnls/permissions. Advertising, inserts, and artwork enquiries should be addressed to Advertising and Special Sales, ADVERTISING: Oxford Journals, Oxford University Press, Great Clarendon Street, Oxford, OX2 6DP, UK. Tel: þ 44 (0)1865 354767; Fax: þ 44(0)1865 353774; E-mail: jnlsadvertising@oup.com. DISCLAIMER: Statements of fact and opinion in the articles in The World Bank Economic Review are those of the respective authors and contributors and not of the International Bank for Reconstruction and Development/THE WORLD BANK or Oxford University Press. Neither Oxford University Press nor the International Bank for Reconstruction and Development/THE WORLD BANK make any representation, express or implied, in respect of the accuracy of the material in this journal and cannot accept any legal responsibility or liability for any errors or omissions that may be made. The reader should make her or his own evaluation as to the appropriateness or otherwise of any experimental technique described. The World Bank Economic Review is printed on acid-free paper that meets the minimum require- PAPER USED: ments of ANSI Standard Z39.48-1984 (Permanence of Paper). INDEXING AND ABSTRACTING: The World Bank Economic Review is indexed and/or abstracted by CAB Abstracts, Current Contents/Social and Behavioral Sciences, Journal of Economic Literature/EconLit, PAIS International, RePEc (Research in Economic Papers), and Social Services Citation Index. COPYRIGHT # 2011 The International Bank for Reconstruction and Development/THE WORLD BANK All rights reserved; no part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise without prior written permission of the publisher or a license permitting restricted copying issued in the UK by the Copyright Licensing Agency Ltd, 90 Tottenham Court Road, London W1P 9HE, or in the USA by the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. Typeset by Techset Composition Limited, Chennai, India; Printed by Edwards Brothers Incorporated, USA. Five Questions on International Migration and Development C ¸ ag ¨ zden, Hillel Rapoport, and Maurice Schiff ˘ lar O JEL codes: f22, f24, j61, 015 The movement of people in search of better economic conditions and a more secure environment is as old as human history. Such movements not only pro- foundly affect the lives of the migrants, but also lead to signi�cant economic Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 and social transformations in migrants’ countries of origin and destination.1 In recent years, a signi�cant increase in the growth of international migration and remittance flows and in awareness of their development impact has led to a resurgence of interest by academics, policymakers, and analysts in what has been referred to as the third leg of globalization (the other two being inter- national trade and international capital flows). The renewed interest in international migration led the World Bank Development Research Group to initiate the Research Program on C¸ ag ¨ zden (corresponding author; cozden@worldbank.org) is a senior economist in the ˘ lar O Development Research Group of the World Bank. Hillel Rapoport (hillel_rapoport@hks.harvard.edu) is visiting research fellow at the Center for International Development at Harvard University and associate professor at Bar-Ilan University and at EQUIPPE, University of Lille. Maurice Schiff (mschiff@ worldbank.org) is a lead economist in the Of�ce of the Chief Economist for Latin America at the World Bank, visiting professor at the University of Chile, and fellow at the Institute for the Study of Labor (IZA-Bonn). 1. Studies have generally shown that international migration has a positive impact on poverty reduction and human capital investments and outcomes—including on children’s short- and long-term physical development, education (especially for girls), and use of birth-related healthcare services. Studies have also shown a positive impact on investments in physical capital (such as land and agricultural implements); entrepreneurship, including the establishment of small and microenterprises; housing; and reduction in child labor (see studies in O ¨ zden and Schiff 2006, 2007 and references therein). Migration has also been found to have a positive impact on trade (Rauch and Trinidade 2002; Iranzo and Peri 2009) and foreign direct investment (Kugler and Rapoport 2007; Javorcik and others 2011); to reduce home-country fertility in the case of migration to low-fertility countries and raise it in the case of migration to high-fertility countries (Beine, Docquier, and Schiff 2008); and the brain drain has been found to promote technology diffusion in some studies (Kerr 2008; Agrawal and others 2011) though not in others (Schiff and Wang 2009). THE WORLD BANK ECONOMIC REVIEW, VOL. 25, NO. 1, pp. 1 – 11 doi:10.1093/wber/lhr021 # The Author 2011. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 1 2 THE WORLD BANK ECONOMIC REVIEW International Migration and Development in 2003.2 More recently, the Research Department of the Agence franc ¸ aise de De´ veloppement (AFD) and the World Bank Development Research Group have collaborated on several research projects and conferences. This symposium issue gathers some of the papers presented at the Second International Migration and Development Conference, held at the World Bank in Washington, DC, on September 10 –11, 2009.3 The success of the conference series and the commitment of the World Bank and AFD to sponsoring the conferences reflect the recognition by inter- national development agencies and the academic community of the importance of international migration to the development agenda. The �ve articles in this symposium issue fall into two groups.4 A �rst group of three articles deal with the measurement, determinants, and political effects of international migration. A new global bilateral migration database for 1960–2000 (O ¨ zden and others 2011) updates and extends the Parsons and others (2007) database back to 1960. The second article takes advantage of Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 existing surveys and matches Ecuadorian migrants in the United States and Spain with migrant households in Ecuador to investigate determinants of the size, selection, and sorting across destinations of the recent migration wave out of that country (Bertoli, Ferna ´ ndez-Huertas Moraga, and Ortega 2011). The third article designs an experiment to examine the impact of Cape Verde’s migrants on the demand for good governance in that country (Batista and Vicente 2011). Two articles on international remittances constitute the second set of contributions. They examine the determinants of remittance costs (Beck and Martı ´a 2011) and the relationship between migrants’ education ´nez Perı and their propensity to remit (Bollard and others 2011). Both articles use orig- inal microdata collected from a large number of countries. 2. The research program was initiated under the guidance of Franc ¸ ois Bourguignon, then Senior Vice President of Development Economics, and Alan Winters, then Director of the Development Economics Research Group, at the World Bank. 3. The �rst conference was held at the University of Lille in June 2008, the third at the Paris School of Economics in September 2010, and the fourth is planned for June 2011 at Harvard. 4. These studies are the latest to come out of the World Bank Research Program on International Migration and Development. Previous studies have been collected in three volumes. (Many papers were also published as World Bank Policy Research Working Papers, and most have appeared in refereed journals.) The �rst volume (O ¨ zden and Schiff 2006) examined the determinants and development impact of migration and remittance on such issues as poverty, health, education, entrepreneurship, and child labor, as well as aspects of brain drain, brain gain, and brain waste. A major contribution was a new database on international migration to countries of the Organisation for Economic Co-operation and Development by Docquier and Marfouk. The second volume (O ¨ zden and Schiff 2007) also examined the impact of migration and remittances on schooling and labor markets, host countries’ immigration policies, and returning migrants’ gains from overseas work experience, as well as a new global bilateral migration database for 2000 by Parsons and others (2007). A third volume (Morrison, Schiff, and Sjoblom 2008) focused on the determinants and impact of the migration of women and the difference between male and female migrants and between no-migrant male and female heads of household. ¨ zden, Rapoport, and Schiff O 3 I . I N T E R N AT I O N A L M I G R AT I O N This section discusses the three articles on international migration dealing with the new database, the determinants of destination choices, and migration’s governance impact. What Are the Regional Speci�cs and Dynamics of International Migration? Evidence on international migration is far sparser than that on trade and capital flows. Bilateral international trade data are classi�ed by a large and detailed set of characteristics and are reported monthly. Some capital flow data are even available daily. And trade and �nancial flow data are generally avail- able from both importing and exporting countries, so the two sources can be compared for accuracy. Bilateral aggregate (country-level) migration data come mostly from censuses conducted every 10 years, and only from destination Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 countries that choose to collect and disseminate these data. In short, migration data are among the scarcest international flow data. Thus, collecting comprehensive and reliable data on international migration patterns and migrant characteristics at aggregate and household levels became an overarching objective of the World Bank international migration research program. A major effort was launched to assemble global migration data- bases.5 Docquier and Marfouk (2004, 2006) constructed a global bilateral database of South–North and North–North migration (from 165 developing countries to 30 Organisation for Economic Co-operation and Development (OECD) countries and between OECD countries for three levels of education for 1990 and 2000. Several extensions followed, including a disaggregation of skilled migrants by age of entry in the host country (Beine, Docquier, and Rapoport 2007)6 and gender (Docquier, Lowell, and Marfouk 2009).7 In a parallel effort, Parsons and others (2007) constructed the most comprehensive global bilateral migration database at the time, consisting of a 226x226 matrix of bilateral migrant stocks for all country pairs in the world for the 2000 census round. The article by O ¨ zden and others (2011) in this issue updates Parsons and others’ (2007) database on bilateral migrant stocks and extends it back from 2000 to 1960 and disaggregates it by gender. The global bilateral matrices 5. The collection effort also included microdata gathered through household surveys, which contained detailed international migration modules in various countries, including Brazil, India, Ghana, Pakistan, and Tonga. 6. The data were disaggregated to identify skilled emigrants who obtained their last degree in their home country and those who obtained it elsewhere. That evidence is not directly available but can be approximated through information on the age at which skilled immigrants entered the host country. 7. The OECD also assembled a database on migration to the OECD (Dumont and Lemaı ˆtre 2004), which was then disaggregated by migrants’ age, gender, educational attainment, and place of birth (OECD 2008). A global bilateral database of the medical brain drain was also put together by Docquier and Bhargava (2006). 4 THE WORLD BANK ECONOMIC REVIEW were assembled by combining more than one thousand censuses and popu- lation register records. The matrices provide for the �rst time a complete set of bilateral migration stocks for the second half of the twentieth century. The article describes in detail the key assumptions made in constructing the bilateral migration matrices and in handling missing observations and the emergence of new countries. The clear and detailed explanations of the meth- odology used to construct these matrices are among the database’s key strengths, enhancing its usefulness and enabling anyone to improve and extend it as new data become available. The new evidence enables the authors to identify migrants’ main source and destination countries, characterize the bilateral structure of migration patterns around the world, and identify the most important migration corridors, as well as the evolution of migration at the bilateral, country, and regional levels for 1960–2000. The authors note several important changes in these patterns. South–North migration grew rapidly over the period, while the shares of Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 North–North, North–South, and South–South migration declined. The United States remained the world’s main migration destination in 2000, home to one in �ve migrants. But the composition of migrant stocks in the United States and across the world underwent major changes. In 1960, most migrants in the United States originated in Europe; by 2000, most came from Latin America and the Caribbean. Worldwide, migrants from Europe and South Asia were important in 1960; by 2000, migrants from Latin America, North Africa and the Middle East had gained prominence. This database constitutes an important extension of the information avail- able on international migration. If the previous global bilateral migration data- bases are any indication, the new data are likely to be a rich source for academics, policy analysts, and others interested in bilateral and overall migration stocks at the country, region, or global level and on their evolution over the second half of the twentieth century. How Do Policy and Incentives Affect the Size, Destination, and Composition of Migration Flows? Ecuador experienced massive emigration following a deep economic crisis in the late 1990s. Bertoli, Ferna ´ ndez-Huertas Moraga, and Ortega (2011) use micro-level data on Ecuador and its main destination countries, Spain and the United States, to examine the impact of wage gaps and immigration policies on the size and composition of migration flows. Detailed data from the two desti- nations enable the authors to focus more precisely on differences across desti- nations. Other studies have combined micro-level data from various countries in their analysis, including Bollard and others’ (2011) article on the relation- ship between migrants’ skill levels and their propensity to remit, and Clemens, Montenegro, and Pritchett’s (2008) paper comparing real wages in migrants’ home and host countries. The determinants of the level and distribution of ¨ zden, Rapoport, and Schiff O 5 skilled and unskilled labor migration for different destination countries have also been examined, for example, by Grogger and Hanson (2011) and Beine, Docquier, and O ¨ zden (2011), though they do so on the basis of aggregate bilat- eral macrodata and do not consider the impact of speci�c immigration policies. Migration policies are typically complex collections of laws, rules, and implementation measures. Their components do not always constitute a cohe- sive whole, possibly because they often originate in different ministries ( justice, interior, foreign affairs, and others) and because they may be influenced by groups with different and even contradictory interests. Thus, identifying changes over time or their impact is likely to be dif�cult. Bertoli, Ferna ´ ndez-Huertas Moraga, and Ortega use data on the large emigration flows from Ecuador in the late 1990s to estimate the impact of a sudden change in Spain’s policy in August 2003 (the mid-point of their sample period) with the introduction of a visa requirement for nonimmigrant admission of Ecuadorians. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Using the microdata to estimate Mincer-type wage equations, the authors �nd that the income gains associated with migration are larger for the United States than for Spain, with the difference increasing with migrants’ level of edu- cation. This �nding is consistent with the higher share of Ecuadorian college graduates residing in the United States but not with the �nding that Spain was the main destination for Ecuadorian emigrants. This seeming anomaly is explained by the fact that Ecuadorians visiting Spain did not need a visa, so they could simply remain in the country to work in the parallel labor market. Entering the United States illegally was substantially more dif�cult, which, together with the higher skill premium in the U.S. labor market, explains why both the number of Ecuadorians and the share of unskilled Ecuadorian migrants were larger in Spain than in the United States. This situation changed in 2003 with the elimination of the visa waiver program, a policy change that is estimated to have led to a two-thirds reduction in the flow of Ecuadorians to Spain. The authors’ �ndings seem to indicate that some changes in immigration policy can have a dramatic impact on immigration. In contrast, McKenzie and Rapoport (2010) and Beine, Docquier, and O ¨ zden (2011) have shown the importance of diaspora networks for immigration, concluding that changes in immigration policy may have a limited impact on future immigration flows because of the strength of the network effects. The �ndings in the Bertoli, Ferna ´ ndez-Huertas Moraga, and Ortega article suggest that the impact of a change in immigration policy may depend on the policy reform itself and on the conditions under which the reform takes place. Do Migrants Improve Governance at Home? Migrants are affected by their experiences in their country of destination and, in turn, they affect their home country in a variety of ways. Batista and Vicente 6 THE WORLD BANK ECONOMIC REVIEW (2011) examine the extent to which migrants from Cape Verde, both current and those that return home, contribute to political change in their home country. This is an important issue because of the centrality of institutions to economic development (Acemoglu, Johnson, and Robinson 2005; Rodrik 2007) and because the literature on the relationship between migration and institutions is limited. Assuming institutions are positively affected by the average level of human capital in migrants’ home country and the size of their diaspora, unskilled migration should improve the quality of institutions. Skilled migration, for its part, would have two opposite effects: a positive impact through the increased size of the diaspora and a negative one associated with a decrease in human capital in the home country. These hypotheses are con�rmed in a paper on migration and democracy by Docquier and others (2011).8 An earlier study based on data for one point in time �nds that emigration has a positive impact on political institutions and a negative one on economic institutions (Li and Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 McHale 2009). Another study (Spilimbergo, 2009) �nds that foreign students returning to their home country have a positive impact on democracy, but only if they studied in a democratic country. These studies are based on country- level data. The article by Batista and Vicente uses microdata to examine whether migration in general and skilled migration in particular contributes to political change in Cape Verde, a nine-island tropical country off the coast of West Africa with half a million inhabitants, good institutional scores by African standards, and a long tradition of migration. Current migrants represent a �fth of the population, and skilled migrants constituted 67 percent of migrants in 2000 (Docquier and Marfouk 2006), a share that remained high (60 percent) even after excluding people who acquired their tertiary education abroad (Beine, Docquier, and Rapoport 2007). Batista and Vicente set up a "voting experiment" along the following lines: after taking a survey on perceived corruption in public services, respondents were asked to mail a prestamped postcard if they wanted the survey results to be published in the national media. Controlling for indi- vidual, household, and locality characteristics, the authors regressed partici- pation in the voting experiment, which they interpret as demand for accountability, on migration prevalence at the locality level. They show that both current and return migrants from the United States, but not from Portugal, the other main destination country, signi�cantly raise participation rates; the effect is stronger for return migrants. They do not �nd evidence of additional effects for skilled migrants. 8. They �nd, based on a panel of cross-country data, that migration has a positive impact on political institutions while skilled migration has an ambiguous impact. In a simple model, Schiff and Docquier (2010) also �nd an ambiguous U-shaped impact of skilled migration on institutions. ¨ zden, Rapoport, and Schiff O 7 Thus, the study provides microeconometric evidence supporting the country- level �ndings on the positive effects of foreign students (Spilimbergo 2009) and emigration (Docquier and others 2011) on democracy at home. I I . I N T E R N AT I O N A L R E M I T TA N C E S Another migration dividend for developing countries, probably the most obvious, is remittances. The rapid rise in South–North migration, as documen- ted in O¨ zden and others (2011), has been accompanied by a dramatic rise in migrants’ remittances. Recorded remittance flows to developing countries rose nearly sixfold from 1995 ($57 billion) to 2008 (more than $328 billion). The recent economic crisis resulted in a 5 percent decline in remittances, though 2010 saw a rebound of about the same amount (World Bank 2010). On average and over the last few years, remittances have approximately equaled Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 the amount of foreign direct investments (a more volatile source of foreign exchange for developing countries) and were about triple the size of of�cial development assistance. There have been two recent structural changes in remittances: in the indus- trial organization of the remittance business, with the entry of new operators (including many banks), and in the skill composition of migration flows, induced largely by increasingly quality-selective immigration policies in rich countries. The common wisdom is that the entry of new operators should lead to more competition, lower remittance costs, and ultimately, higher remittance volumes thanks to income and substitution (from informal to formal channels) effects. The change in skill composition, in contrast, should lead to lower remittance volumes or, at least, to lower remittances per migrant (as educated migrants, presumably, have lower incentives to remit). As the two contributions described here show, however, the reality is more nuanced and complex. What Explains the Price of Remittances? Reducing the cost (or price) of remittances would seem to be the most obvious way to increase the volume of remittances reaching developing countries. International organizations such as the World Bank and various development forums have long called for policy intervention to increase competition in the remittance business. Recently, as Beck and Martinez (2011) recall, world leaders at the L’Aquila 2009 G-8 summit called for cutting the price of remit- tances by half in �ve years (from a current average of 10 percent). The presumption is that more competition (including more transparency) will lead to lower remittance prices. This presumption would seem to be sup- ported by the Mexican experience. In 2008, when the World Bank dataset on remittance costs was launched, Ratha (2008) noted that Mexico’s earlier release of remittance cost data from about a dozen U.S. cities to several Mexican cities had been accompanied by a 56 percent decline in remittance 8 THE WORLD BANK ECONOMIC REVIEW costs between 1999 and 2004. “The hope is,� Ratha wrote, “that the new database will have a similar cost reduction effect.� Publicizing remittance prices would better inform consumers and elicit further competition in remit- tance corridors. That would reduce the drain on poor migrants’ incomes, increase their ability to send more money home, and prevent incumbent �rms from using their market power to extract a large part of the global surplus from international migration in transfer, exchange rate conversion, and other remittance fees. The database also makes remittance data available to research- ers investigating remittance markets. Beck and Martinez Peira (2011) analyze the bilateral costs of sending remit- tances and �nd enormous heterogeneity in the magnitude of these costs and in their determinants across country pairs. Quite surprisingly, they �nd that �nan- cial development and competition in the banking sector are poor predictors of the bilateral costs of remittances. A closer look reveals that these costs are instead driven by competition in the remittance business itself. That segment of Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 the market is dominated by one �rm, Western Union, whose prices seem to be set independently of competitive pressures. The authors suggest that this may be due to Western Union’s better network coverage and to more years in oper- ation than other �rms in a number of corridors. Western Union’s price-setting behavior is consistent with the price leadership model in which a dominant �rm may be the sole operator in a contestable market and yet charge less than the monopoly price to keep potential competi- tors out. Or when its competitive advantage is not too large, it may charge a price equal to its competitors’ marginal cost (minus epsilon) in order to secure market share. In either case, observed competition—as measured by standard concentration indices—is unlikely to accurately predict market prices. How Does Migrants’ Education Level Affect Their Propensity to Remit? As noted, international migration from developing to developed countries is increasingly of the “brain drain� type.9 This has given rise to questions about whether the increasingly high-skilled nature of emigration from developing countries will slow the rise in remittances. The literature on migrant remit- tances shows that the two main motivations to remit are altruism and exchange.10 Altruism is directed primarily toward one’s immediate family and decreases with social distance. The exchange motive posits that remittances simply “buy� various services, such as care of the migrant’s assets (land and cattle, for example) or relatives (children, elderly parents) at home; such trans- fers are typically observed in cases of temporary migration and signal 9. In Organisation for Economic Co-operation and Development (OECD) countries, the number of migrants with a tertiary education and originating from developing countries doubled between 1990 and 2000, while the number for those with a primary school education rose only 20 percent. 10. See Rapoport and Docquier (2006) for a comprehensive survey of the theoretical and empirical literature on migrants’ remittances. ¨ zden, Rapoport, and Schiff O 9 intentions to return. A particular type of exchange takes place when remit- tances are de facto repayments of loans used to �nance the migrant’s invest- ments in education or migration. Thus, it is theoretically unclear whether educated migrants remit more than less educated ones. Educated migrants’ income is a priori higher, providing them with a greater capacity to remit; they may remit more to meet implicit commitments to reimburse the family for funding education investments. On the other hand, educated migrants tend to emigrate with their family and to do so on a permanent or longer term basis and are therefore less likely to remit (or are likely to remit less) than someone moving alone on a temporary basis. The question of whether educated migrants remit more or less than do less educated migrants has been surprisingly understudied, especially at a micro level. Most of the previous literature (Faini 2007; Niimi, O ¨ zden, and Schiff 2010) used aggregate data and found a negative effect of migrants’ education on total remittances received. Bollard and others (2011) question the �ndings Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 of that literature, positing that the many differences across countries could result in a spurious negative relationship between remittances and migrants’ skill levels in cross-country studies. The authors examine this issue by combin- ing household survey data on immigrants in 11 destination countries. They �nd a mixed pattern for the relationship between education and the likelihood of remitting, and a strong positive relationship between education and the amount remitted (intensive margin) conditional on remitting (extensive margin). Combining these intensive and extensive margins gives an overall positive effect of education on the amount remitted, with an expected amount of $1,000 annually for a migrant with a university degree and $750 for someone without one. Data from the surveys containing information on income show, however, that the less educated tend to remit a larger share of their income. The microdata used in this study also allow investigation of why the more educated remit more. Bollard and others (2011) �nd that it is the higher income earned by migrants that explains much of the higher remittances rather than characteristics of their family situations or their intentions to return. Indeed, and in contrast to common wisdom, declared intentions to return do not differ signi�cantly across education groups. And while it is con�rmed that educated migrants do migrate more with their spouse and children, less edu- cated migrants tend to have larger extended families at destination, suggesting compensating effects of family closeness and size on remittance behavior across education categories. ***** The articles in this symposium issue provide original contributions on �ve important questions on the economics of international migration and develop- ment—questions on the measurement, policy determinants and political impact of international migration, and on the determinants of the price of 10 THE WORLD BANK ECONOMIC REVIEW international remittances and their relationship with migrants’ education levels. These articles are expected to elicit wide interest, stimulate additional research and further our knowledge in these areas. The articles are part of an ongoing collaborative research effort between the World Bank Development Research Group and the Research Department of the Agence franc ¸ aise de De´ veloppement, a collaboration of demonstrated value that the two institutions are committed to pursue. REFERENCES Acemoglu, D., S. Johnson, and J. Robinson. 2005. “Institutions and Growth.� In Handbook of Economic Growth, ed. P. Aghion, and S. Durlauff. Amsterdam: North Holland. Agrawal, A.K., D. Kapur, J. McHale, and A. Oettl. 2011. “Brain Drain or Brain Bank? The Impact of Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Skilled Emigration on Poor Country Innovation.� Journal of Urban Economics 69 (1): 43 –55. Batista, C., and P. Vicente. 2011. “Do Migrants Improve Governance at Home? Evidence from a Voting Experiment.� World Bank Economic Review, this issue. Beck, T., and M.S. Martinez Peria. 2011. “What Explains the Price of Remittances? An Examination across 119 Country Corridors.� World Bank Economic Review, this issue. ¨ zden. 2011. “Diasporas.� Journal of Development Economics 95 (1): Beine, M., F. Docquier, and C. O 30 –41. Beine, M., F. Docquier, and H. Rapoport. 2007. “Measuring International Skilled Migration: New Estimates Controlling for Age of Entry.� World Bank Economic Review 21 (2): 249– 54. Beine, M., F. Docquier, and M. Schiff. 2008. “International Migration, Transfer of Norms and Home-country Fertility.� IZA Discussion Paper 3912. Institute for the Study of labor (IZA), Bonn. Bertoli, S., J. Fernandez-Huertas Moraga, and F. Ortega. 2011. “Immigration Policies and the Ecuadorian Exodus.� World Bank Economic Review, this issue. Bollard, A., D. McKenzie, M. Morten, and H. Rapoport. 2011. “Remittances and the Brain Drain Revisited: The Microdata Show That More Educated Migrants Remit More.� World Bank Economic Review, this issue. Clemens, M., C. Montenegro, and L. Pritchett. 2008. “The Place Premium: Wage Differences for Identical Workers across the U.S. Border.� CGD Working Paper 148. Center for Global Development, Washington, DC. Docquier, F., E. Lodigiani, H. Rapoport, and M. Schiff. 2011. “Emigration and Democracy� World Bank Policy Research Paper 5557. World Bank, Washington, DC. Docquier, F., B.L. Lowell, and A. Marfouk. 2009. “A Gendered Assessment of the Brain Drain.� Population and Development Review 35 (2): 297– 321. Docquier, F., and A. Marfouk. 2004. “Measuring the international mobility of skilled workers (1990– 2000): Release I.� Policy Research Working Paper 3381. World Bank, Washington, DC. ———. 2006. “International Migration by Educational Attainment, 1990–2000.� In International Migration, Remittances, and Development, ed. C. O¨ zden, and M. Schiff. New York: Palgrave Macmillan. ˆtre. 2004. “Counting Immigrants and Expatriates in OECD Countries: A Dumont, J.C., and G. Lemaı New Perspective.� Organisation for Economic Co-operation and Development, Paris. Faini, R. 2007. “Remittances and the Brain Drain: Do More Skilled Migrants Remit More?� World Bank Economic Review 21 (2): 177 –91. Grogger, J., and G. Hanson. 2011. “Income Maximization and the Selection and Sorting of International Migrants.� Journal of Development Economics 95 (1): 42– 57. ¨ zden, Rapoport, and Schiff O 11 Iranzo, S., and G. Peri. 2009. Migration and Trade: Theory with an Application to the Eastern– Western European Integration. Journal of International Economics 79 (1): 1–19. ¨ zden, M. Spatareanu, and I.C. Neagu. 2011. “Migrant Networks and Foreign Direct Javorcik, B.S., C. O Investment.� Journal of Development Economics 94 (2): 231–241. Kerr, W.R. 2008. “Ethnic Scienti�c Communities and International Technology Diffusion.� Review of Economics and Statistics 90: 518– 37. Kugler, M., and H. Rapoport. 2007. “International Labor and Capital Flows: Complements or Substitutes?� Economics Letters 94 (2): 155–62. Li, X., and J. McHale. 2009. “Does Brain Drain Lead to Institutional Gain? A Cross-country Empirical Investigation.� Department of Economics, Queen’s University, Kingston, ON. McKenzie, D., and H. Rapoport. 2010. “Self-selection Patterns in U.S.–Mexico Migration: The Role of Migration Networks.� Review of Economics and Statistics 92 (4): 811–21. Morrison, A.R., M. Schiff, and M. Sjoblom. 2008. The International Migration of Women. New York and Washington, DC: Palgrave Macmillan and World Bank. ¨ zden, and M. Schiff. 2010. “Remittances and the Brain Drain: Skilled Migrants Do Niimi, Y., C. O Remit Less.� Annales d’Economie et de Statistique 97– 98. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 OECD (Organisation for Economic Co-operation and Development). 2008. A Pro�le of Immigrant Populations in the 21st Century: Data from OECD Countries. Paris: Organisation for Economic Co-operation and Development. C. O¨ zden, and M. Schiff, ed. 2006. International Migration, Remittances & the Brain Drain. New York and Washington, DC: Palgrave Macmillan and World Bank. C. O¨ zden, and M. Schiff, ed. 2007. International Migration, Economic Development, and Policy. New York and Washington, DC: Palgrave Macmillan and World Bank. ¨ zden, C., C.R. Parsons, M. Schiff, and T.L. Walmsley. 2011. “Where on Earth is Everybody? The O Evolution of Global Bilateral Migration 1960– 2000.� World Bank Economic Review, this issue. Parsons, C.R., R. Skeldon, T. Walmsley, and L.A. Winters. 2007. “Quantifying International Migration: A Database of Bilateral Migrant Stocks.� In International Migration, Economic Development and Policy, ed. C. O¨ zden, and M. Schiff. New York and Washington, DC: Palgrave Macmillan and World Bank. Rapoport, H., and F. Docquier. 2006. “The Economics of Migrants’ Remittances.� In Handbook of the Economics of Giving, Altruism, and Reciprocity, ed. S.-C. Kolm, and J. Mercier Ythier. Amsterdam: North Holland. Ratha, D. 2008. “A New Remittance Prices Database Brings Much-needed Transparency.� People Move: A Blog about Migration, Remittances and Development (blog). World Bank, Washington, DC. September 25. http://blogs.worldbank.org/peoplemove/a-new-remittance-price-database-brings- much-needed-transparency. Rauch, J.E., and V. Trinidade. 2002. “Ethnic Chinese Networks in International Trade.� Review of Economics and Statistics 84 (1): 116– 30. Rodrik, D. 2007. One Economics, Many Recipes: Globalization, Institutions, and Economic Growth. Princeton, NJ: Princeton University Press. Schiff, M., and F. Docquier. 2010. “Brain Drain, Human Capital, and Institutions.� Of�ce of the Chief Economist, Latin America and the Caribbean Region, World Bank. Schiff, M., and Y. Wang. 2009. “North-South Trade-Related Technology Diffusion, Brain Drain, and Productivity Growth: Are Small States Different?� World Bank Policy Research Working Paper 4828. World Bank, Washington, DC. Spilimbergo, A. 2009. “Foreign Students and Democracy.� American Economic Review 99 (1): 528–43. World Bank. 2010. Migration and Remittances Factbook 2011. Washington, DC: World Bank. Where on Earth is Everybody? The Evolution of Global Bilateral Migration 1960–2000 C ¸ ag ¨ zden, Christopher R. Parsons, Maurice Schiff, ˘ lar O and Terrie L. Walmsley Global matrices of bilateral migrant stocks spanning 1960–2000 are presented, disaggregated by gender and based primarily on the foreign-born de�nition of migrants. More than one thousand census and population register records are com- bined to construct decennial matrices corresponding to the �ve census rounds between 1960 and 2000. For the �rst time, a comprehensive picture of bilateral global migration over the second half of the 20th century emerges. The data reveal that the global migrant stock increased from 92 million in 1960 to 165 million in 2000. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Quantitatively, migration between developing countries dominates, constituting half of all international migration in 2000. When the partition of India and the dissolution of the Soviet Union are accounted for, migration between developing countries is remarkably stable over the period. Migration from developing to developed countries is the fastest growing component of international migration in both absolute and rela- tive terms. The United States has remained the most important migrant destination in the world, home to one �fth of the world’s migrants and the top destination for migrants from some 60 sending countries. Migration to Western Europe has come C¸ ag ¨ zden (corresponding author; cozden@worldbank.org) is a senior economist in the ˘ lar O Development Research Group of the World Bank. Christopher Parsons (lexcrp1@nottingham.ac.uk) is a consultant at the World Bank and a doctoral candidate at the University of Nottingham. Maurice Schiff (mschiff@worldbank.org) is a lead economist in the Of�ce of the Chief Economist for Latin America at the World Bank. Terrie Walmsley (twalmsle@purdue.edu) is an associate professor and executive director of the Global Trade Analysis Project, Purdue University, and an associate professor at the University of Melbourne. First and foremost the authors thank the United Nations Population Division for spearheading the creation of the Global Migration Database. In particular, they thank Bela Hovy and Hania Zlotnik for their close support and shared vision that ensured the completion of this project. They are grateful to Richard Black, Ronald Skeldon, and especially L. Alan Winters for having the foresight to initiate this project and for their unwavering support. They also extend thanks to the librarians at the British Library, the Library of Congress, and the London School of Economics and to Lorraine Wright at the United States Census Bureau for assistance beyond the call of duty. The authors thank Steven Vertovec and Norbert Winnige of the Max Planck Institute for the Study of Religious and Ethnic Diversity for helping surmount the data issues concerning Germany and the former Soviet Union. They thank Michel Beine, Fre ´ ric Docquier, and the journal editor, as well as three anonymous ´ de referees, for advice and comments. They gratefully acknowledge �nancial support from the World Bank Knowledge for Change Program and Ivar Cederholm’s help with administration of the funding. The �ndings, interpretations, and conclusions expressed in this article are those of the authors and do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. THE WORLD BANK ECONOMIC REVIEW, VOL. 25, NO. 1, pp. 12– 56 doi:10.1093/wber/lhr024 # The Author 2011. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 12 ¨ zden, Parsons, Schiff, and Walmsley O 13 largely from elsewhere in Europe. The oil-rich Persian Gulf countries emerge as important destinations for migrants from the Middle East and North Africa and South and Southeast Asia. Finally, although the global migrant stock is predominantly male, the proportion of female migrants increased noticeably between 1960 and 2000. The number of women rose in every region except South Asia. JEL codes: F22, O15, J11, J16 International migration—the movement of people across national borders—has important economic, social, and political implications. Despite the recent emer- gence of a dynamic literature, empirical analysis of migration flows and their impact lags behind the policy debate and the theoretical literature. The main reason is the absence of comprehensive and reliable data on international migration patterns and migrant characteristics at either the aggregate or the Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 household level. The objective of this article is to use data from more than one thousand national censuses and population registers to estimate a complete global origin–destination migration matrix for each decade over 1960–2000. These 226*226 matrices, com- prising every country, major territory, and dependency around the world, are divided into periods corresponding to the last �ve completed census rounds. The gender dimension of international migration over this period is also presented. The primary source of the raw data is the United Nations Population Division’s Global Migration Database, created through the collaboration of the United Nations Population Division, the United Nations Statistics Division, the World Bank, and the University of Sussex (United Nations 2008). This unique data repository comprises 3,500 individual census and population register records1 for more than 230 destination countries and territories over the last �ve decades. The database provides information on international bilateral migrant stocks (by citizenship2 or place of birth), sex, and age. There is con- siderable variation, however, in how destination countries collect, record, and disseminate immigration data. Meaningful comparison of destination country records over time is thus often confounded. In constructing global bilateral migration matrices, several challenges arise. First, destination countries typically classify migrants in different ways—by place of birth, citizenship, duration of stay, or type of visa. Using different criteria for a global dataset generates discrepancies in the data. Second, many geopolitical changes occurred between 1960 and 2000, with many international borders 1. Of the 3,500 sources detailed in the overarching UN Global Migration Database, 1,107 were suitable for analysis, once repeated censuses had been removed or combined. The Global Migration Database should not be confused with the Trends in International Migrant Stock Database, which lists aggregate migrant stocks for each destination country in the world at �ve year intervals (United Nations 2006). 2. The article treats the concepts of nationality and citizenship as analogous and uses the terms interchangeably. 14 THE WORLD BANK ECONOMIC REVIEW redrawn as new countries emerged and others disappeared. In addition to creating millions of migrants overnight—as when the Soviet Union collapsed—these events complicate the tracking of migrants over time. Third, even when national censuses of destination countries include data on international migrant stocks, the data are presented along aggregate geographic categories rather than by country of origin. Data therefore need to be disaggregated to the country level. Finally, the greatest hurdle is dealing with omitted or missing census data. Very few destination countries—especially developing countries—have conducted rigorous censuses or population registers during every census round over the second half of the twenti- eth century. Wars, civil strife, lack of funding, and political intransigence are but a few reasons why records may be discontinuous. The main contributions of this article lie in identifying and overcoming these challenges in order to construct a consistent and complete set of origin– destination matrices of international migrant stocks for 1960–2000, disaggre- gated by gender. The starting point is a master set of 226 origin or destination Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 countries and regions. Despite border changes, all migrants are assigned to this master set so that migrations can be meaningfully tracked over time. These assignments, especially in cases where only aggregate data are available, are made using several alternative propensity measures based either on a destina- tion country’s propensity to accept international migrants or on an origin country’s propensity to send migrants abroad. Cases of omitted data occur when destination countries do not collect or publicly disseminate the information on migrants. When data from census rounds are missing altogether, the approach taken depends on the extent of the omission (see appendixes 3 and 4). When suf�cient data are available for other decades, interpolation is used. When not enough data are available, propensity measures are used to generate bilateral data. When a gender breakdown is missing, gender splits are calculated based on supplementary statistics or other data in the matrices (see appendix 5). The resulting migration matrices should be viewed as work in progress, but they are an important step in an ongoing global effort to improve migration data. The matrices can be readily updated as additional or superior information surfaces, and they can easily be extended to include future census rounds. Bilateral datasets of international migration are rare. Attempts to create them have focused almost exclusively on industrialized countries as destinations because these countries have more accurate and more frequently produced data. Harrison and others (2003) calculate bilateral remittances for the countries of the Organisation for Economic Co-operation and Development (OECD) together with the 27 largest nonmembers. These estimates are based on international bilateral migrant stock data that the authors also provide, although many of the data are derived from the Trends in International Migration (OECD 2002). This report, published annually since 1973, was arguably the most comprehensive guide to international migration for many years and has been the basis for many studies (see, for example, Mayda 2007). ¨ zden, Parsons, Schiff, and Walmsley O 15 More recently, the OECD has developed a database that provides a comprehen- sive overview of migration to OECD countries in 2000 (OECD 2008). These data are disaggregated by a number of covariates including age, gender, educational attainment, and place of birth. Another series of papers, again concentrating on the OECD, examines the brain drain in 1990 and 2000 (see, for example, Docquier and Marfouk 2006); migrants’ gender (Docquier, Lowell, and Marfouk 2009); age of entry (Beine, Docquier, and Rapoport 2007); and the medical brain drain (Bhargava and Docquier 2007). Parsons and others (2007) construct a matrix encompassing the entire world for the 2000 census round. Until now, this was the most comprehensive global overview of bilateral migrant movements. Ratha and Shaw (2007) use an earlier version of the dataset in a paper focusing on migration between developing countries (generally referred to as South–South migration in the literature) and bilateral remittance flows. The data in the current article reveal several important patterns. Between 1960 and 2000, the global migrant stock rose from 92 million to 165 million, Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 but fell as a share of world population, from 3.05 percent to 2.71 percent. A large share of the stock in 1960 reflects the partition of India, and in all decades migration within the Soviet Union (and former Soviet Union) accounts for a large proportion of the world migrant stock. A majority of the remaining migrant stocks is due mainly to increasing migration from developing countries to the United States, Western Europe, and the Persian Gulf (referred to as South–North migration). While the growth in South–North migration has been astonishing, North–North, North–South and South–South migrations all represent declining shares of world migration. Even so, South–South migration dominates global trends numerically. The majority of these migrations are intraregional, within the countries of the former Soviet Union, South Asia, and West Africa. Interregional migrations between developing countries are principally to the Persian Gulf countries. The United States continues to be the most important destination, home to around one �fth of the world’s migrant population and the recipient of the largest migrant flows from no less than 60 countries. At the beginning of the period, most migrants in the United States were born in Europe; today the vast majority come from Latin America and the Caribbean. This change in the com- position of migrant stocks mirrors the wider trend. In 1960, except for migration within the Soviet Union, the majority of migrants were born in Europe and South Asia. In 2000, migration from these regions remained impor- tant, but migration from Latin America, East Asia, North Africa, and the Middle East is also prominent. The origin countries most affected by inter- national migration are small, typically island states, mostly in the Paci�c or the Caribbean. The destination countries most affected by migration are the countries of the New World (the United States, Canada, Australia, and New Zealand) and the oil-rich Persian Gulf countries. The data clearly show that international migration is spreading across the globe as migrants widen their destination choices. By 2000, a greater number 16 THE WORLD BANK ECONOMIC REVIEW of migration flows were observed between more country-pairs than at any other time covered in this database. For example, migrants from East Asia and Paci�c who once migrated elsewhere within the region now constitute sizable communities across the world. An increasing number of Africans make their homes in Europe and the United States. This diversi�cation is also reflected in destination countries’ willingness to accept migrants from ever more diverse backgrounds. This is particularly the case for the United States, Australia, New Zealand, and Canada, all of which select migrants based on quali�cations rather than country of origin. The gender composition of international migration flows has also evolved. Although the global migrant stock is still disproportionately male, the percen- tage of women in the global migrant stock rose between 1960 and 2000. This increased feminization of international migration is particularly pronounced in the immigrant stocks of Latin America and the Caribbean, Japan, East Asia and Paci�c, and Sub-Saharan Africa. These four areas have also experienced Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 the greatest increase in the proportion of female emigrants over the period. The article is organized as follows. Section I discusses de�nitions of migrants and how migrants are recorded, describes the raw data, and identi�es gaps in knowledge. Section II considers the comparability of migration data and the major challenges in constructing the matrices. It also discusses the conventions and assumptions adopted in meeting the challenges. Given these assumptions, section III investigates the reliability of the estimates, and section IV analyzes the data, highlighting the key patterns in international migration over 1960–2000. Section V discusses some implications of the study. I. PRELIMINARIES Migration data are complex. They almost always come from destination countries, because it is dif�cult for origin countries to collect demographic data on people who are not living in the country. Unlike trade and �nancial stat- istics, which are recorded by both transacting parties, the quality of migration statistics depends almost entirely on the rigor with which destination countries survey the migrants within their borders. In addition, destination countries’ recording and dissemination methods can differ greatly. Understanding the analysis in this article requires an understanding of the subtle differences in various sources and de�nitions, together with an understanding of the inherent inconsistencies between them.3 Who Are Classi�ed As Migrants? The United Nations (1998, p. 6) de�nes a migrant as “any person that changes his or her country of usual residence.� This broad de�nition implies a 3. This section highlights many of the nuances in the data, but for fuller treatment of the subject, see Bilsborrow and others (1997). ¨ zden, Parsons, Schiff, and Walmsley O 17 movement from one location to another, the most relevant concept for econ- omic analysis. However, of�cial records apply many different de�nitions of what constitutes an international migrant. Most common criteria are based on country of birth, country of citizenship, purpose of visit or visa type, place of last permanent residence, and duration of stay. The two main de�nitions of migration—being born in or being a citizen of a foreign country—are used most consistently over time and across countries. Citizenship is important for determining an individual’s legal rights for employment, voting, and access to public services. The place of birth de�nition is superior for determining physical movement. Destination countries typically publish migration statistics by either category, mainly according to national migration and citizenship laws. Historically, countries in the Americas and Oceania favor the country of birth de�nition whereas countries in Asia, Africa, and Europe traditionally adopt a mix of the two de�nitions. Individuals may be classi�ed as migrants or nonmigrants depending on the Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 de�nition. Many destination countries grant citizenship to foreign-born people who are family members of citizens or who satisfy certain legal and residence requirements. These naturalized citizens continue to be recorded as migrants under the foreign-born de�nition but not under the foreign citizen de�nition. Many destination countries (for example, the United States) grant automatic citizenship to people born within their territory regardless of parents’ citizen- ship. Yet others, such as Japan, require at least one parent to be a citizen for children to acquire citizenship, even if they were born within its borders. Because of these differences in citizenship and naturalization laws, the numbers of migrants will be substantially higher in the United States if the foreign-born criterion is used. In Japan, on the other hand, the number of migrants comes out higher under the foreign citizenship criteria. Where data are available for both de�nitions, priority is given to data by country of birth, for several reasons. First, country of birth is more appropriate in analyzing physical movements and handling the cases of former colonies and dependencies.4 Second, while nationality can change, place of birth cannot.5 Third, naturalization rates vary enormously across destination countries. Differences in laws on citizenship criteria (for both migrants and their children born in the destination country) do not affect data based on place of birth. 4. This discussion of de�nitions highlights the somewhat paradoxical possibility of individuals being classi�ed as migrants without ever having moved across an international border. As mentioned, this is generally possible only in the case of people born in one country but who are citizens only of another country. A similar situation arises with dependencies and former colonies. Residents of Martinique, a French dependency, are automatically granted French citizenship. The statistics for Martinique show all the domestic population as French, possibly leading one to think that Martinique is part of metropolitan France or that most of the population moved to France. In such cases, having data categorized by both foreign born and foreign nationality would enable differentiating between the number of locally born inhabitants of Martinique who are French (referred to as Martiniquais), those born in metropolitan France who moved to Martinique, and people from other countries. 5. Of course the country of birth may be rede�ned, as elaborated in the next section. 18 THE WORLD BANK ECONOMIC REVIEW Fourth, when migrants cannot be assigned to a speci�c origin, they are often recorded under an aggregated umbrella heading. These categories embody ambiguity about a migrant’s origin, and since migrants are assigned to aggre- gated headings more frequently when the citizenship de�nition is used, the foreign born concept is again favored. Last, for migrants living in disputed ter- ritories, such as Kashmir and Western Sahara, an individual’s status or of�cial citizenship may be unclear, while country of birth is usually more certain. How Are Migrants Recorded? Destination countries employ a wide range of tools to enumerate migrants, including population censuses, population registers and registers of foreigners, border statistics, and worker and residence permits.6 This article focuses on census and population register records, which are widely available, have the broadest geographic coverage, and include similar questions, thereby yielding more standardized responses. For these reasons, they are the primary sources Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 for most data in the Global Migration Database. Where both censuses and population registers are available, censuses receive priority. Censuses, generally conducted decennially, are retrospective tools for survey- ing an entire population (or in some cases, a representative sample) at a single point in time. In addition to their universal coverage, their greatest strength is the inclusion of questions on place of birth and nationality. Censuses also typi- cally aim to enumerate the resident population, whether documented or undo- cumented (Bilsborrow and others 1997). So although some migrants have a strong incentive to provide false information to enumerators, many undocu- mented migrants will be captured in these matrices.7 The size and scope of the census questionnaires vary enormously, both over time and in different destina- tion countries. And there is potential variation in the quality of censuses both across countries and over time. Richer countries have many resources at their disposal to design questionnaires, train interviewers, employ statisticians, and disseminate results. Researchers have little choice but to accept the data at face value. However, where the underlying census is clearly substandard (when there are errors that are obviously not coding errors or not easily corrected), these data are omitted from the analysis. Popular in many parts of Europe, population registers are continuous report- ing systems providing up-to-date demographic and socioeconomic information for everyone surveyed. Typically, registers have evolved over time (from parish records, for example). They were never developed speci�cally to record inter- national migration information, and they vary considerably across countries. For example, the laws under which individuals are classi�ed as migrants and 6. This article deals exclusively with migrant stocks. Nothing can be gleaned therefore about when a migration took place, save for inferences that can be made by comparing differences in stocks over time. Nor is anything known about the circumstances (such as visa type) under which an individual entered a particular destination country. 7. The extent to which illegal migration is captured remains unknown. ¨ zden, Parsons, Schiff, and Walmsley O 19 the conditions under which they are inscribed or deregistered differ greatly (Bilsborrow and others 1997). The Raw Data The Global Migration Database is a vast collection of destination country data sources detailing migrant stocks from numerous origin countries and regions (United Nations (2008). Compiling and maintaining the underlying primary sources require herculean efforts to scour the key census collections of the world and enter the data manually. In total, the database comprises records from some 3,500 separate censuses from more than 230 migrant destination countries and territories, by sex and age. Destination countries make numerous revisions between census waves,8 and the database incorporates as many of these revised �gures as possible.9 The starting point is to choose the most relevant source for each destination Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 country from each completed census round.10 Priority is given to data that are superior bilaterally and disaggregated by gender.11 Of the 3,500 sources detailed in the overarching Global Migration Database, 1,107 were suitable for analysis once repeated censuses were removed or combined. Of these, 951 record data disaggregated by gender, as reported in table 1. Despite the large number of primary sources, there are still inevitable gaps (table 2). This might be because a particular destination country did not conduct a census in a given decade or disseminate the relevant bilateral or gender-speci�c information. The majority of the migrants omitted from these censuses are in the Middle East and Africa. The countries of the Middle East are often reticent about releasing data, while many countries in Africa have a long history of conflict. Nonetheless, the 68 countries for which there are complete data account for 68 percent of the world migrant stock in 2000. The 17 countries for which there is only one census account for less than 2 percent of the total stock. The data for earlier decades reflect an identical pattern. II. HARMONIZING THE M AT R I C E S Given the complexities of the underlying data, several major challenges arise in constructing global bilateral migration matrices. The most critical were explained above. In some cases, there is no choice but to recognize that the 8. Census results are also often released in waves, typically beginning with preliminary estimates and following with incremental releases of more detailed data. 9. The raw data are available at http://esa.un.org/unmigration. 10. Bhutan, Colombia, and El Salvador did not conduct censuses during the 2000 round; the relevant censuses for 2005 or 2007 are included instead. Similarly, for seven countries without 1960 censuses, data from the 1950 census round are included. In these cases, each origin countries’ migrant stock as a share of the total is calculated in 1950 and these shares are applied to the 1960 total. 11. There is little standardization in the age brackets that countries use to record migrants’ age. This is the main reason why an analysis of migrants’ age is omitted from the current study. 20 THE WORLD BANK ECONOMIC REVIEW T A B L E 1 . Total Number of Database Sources Census Birthplace Nationality Total national Birthplace by Nationality by round sources sources sources gender gender 1960 124 67 149 103 64 1970 112 52 133 92 49 1980 145 86 164 117 80 1990 151 114 175 129 99 2000 134 122 161 118 100 Total 666 441 782 589 392 Source: Authors’ calculations based on data described in text. T A B L E 2 . Number of Missing Census Rounds Number of missing census Number of destination Share of world migration in 2000 rounds countries (%) Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 0 68 68 1 55 12 2 41 10 3 39 8 4 17 2 5 6 0 Total 226 100 Source: Authors’ calculations based on data described in text. underlying processes that generated the data are less than ideal and to accept the data at face value. In others, every effort has been made to standardize the data. De�ning the Master Country List Over the period covered by the 1960–2000 censuses used to construct the global bilateral matrices of migrant stocks (1955–2004), the global political landscape underwent fundamental changes. Many countries, especially in Africa, Oceania, and the Caribbean, gained their independence. Following the end of the cold war, many countries redrew their political boundaries. Some fragmented into smaller nation states, such as the Soviet Union, Czechoslovakia, and Yugoslavia, and others reuni�ed following an extended period of separation, such as Germany and Yemen.12 A single standard set of countries is speci�ed for the entire timeframe of the database, for both origin and destination locations, so that migration numbers for pairs of countries can be compared over time. Since many new origin and destination countries emerged during the study period, the most current set of countries and regions was chosen. 12. Small border changes and territorial disputes are ignored. ¨ zden, Parsons, Schiff, and Walmsley O 21 A region is de�ned as any geographic entity that conducts its own census and that commonly features as an origin in the others’ censuses. For example, Western Sahara is omitted because it does not conduct a census although it is a commonly designated origin region. In all, 226 countries, territories, and regions are included in this list in each of the �ve migration matrices (see appendix 1). One implication of these inclusion decisions is that migration from Croatia to Germany, for example, is reported in every matrix, even though Croatia did not exist in the early time periods. Researchers interested in migration from Yugoslavia to Germany in 1960 would simply total the individ- ual migration levels from the successor states of Yugoslavia. Performing the analysis according to historical boundaries, though easier, would have masked many recent international movements. Moreover, drawing conclusions about destination countries that no longer exist would offer policymakers less useful information for drawing inferences. Another complication is the 11 additional destinations with census data that Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 do not map perfectly to the master list. Five of these were aggregated into other countries in the master list: Christmas Islands (to Australia), Cocos Islands (to Australia), Kosovo (to Serbia and Montenegro), South Yemen (to Yemen), and West Germany (to Germany). Six additional countries or terri- tories no longer exist, but they map to two or more of the 226 locations on the master list. These are the Gilbert and Ellice Islands, the former Yugoslavia, Czechoslovakia, Ruanda-Urundi, the Trust Territory of the Paci�c Islands, and the Soviet Union. Except for the Soviet Union, the census data for these countries or territories are disaggregated and distributed among the destination countries currently in existence on the basis of more recent migration �gures.13 All of these assignments are made according to the distribution of immigrants of the successor countries in later years. The Soviet Union is a unique challenge. As mentioned, the enforcement of new borders and the creation of new nation states typically create new migrants overnight. According to the foreign-born de�nition, people who cross new borders that are created with the break-up of a country are considered migrants, even if they moved before the break-up while the country was still uni�ed. This is particularly problematic in the case of the Soviet Union because 15 new sovereign nations were created overnight, there have historically been large numbers of internal migrants, and migrants have traditionally been recorded using a de�nition based on ethnicity. Failing to make any adjustment for the Soviet Union, therefore, would result in a large arti�cial jump in the number of migrants at the time of break-up (see appendix 3). 13. For example, the 1988 census data for the Trust Territory of the Paci�c Islands were disaggregated and distributed among the Marshall Islands, the Federated States of Micronesia, the Commonwealth of the Northern Mariana Islands, and the Republic of Palau. However, in years when a country conducted its own census but was also included in the census of a more aggregated region, the country’s own census is prioritized. 22 THE WORLD BANK ECONOMIC REVIEW Last, speci�c adjustments are made in the case of Germany and the Republic of Korea. For Germany, bilateral data are available only by nationality. However, these data fail to take adequate account of the large number of ethnic Germans who arrived from other countries between 1944 and 1950 (mainly expellees) and those who arrived after 1950 (mainly resettlers). Material from the German 2005 micro-census was therefore used to sup- plement the data for Germany (see appendix 3). In the case of Korea, data by nationality are readily available for each census round. However, these data fail to account for the large numbers of migrants from the People’s Democratic Republic of Korea living in the Republic of Korea. Since the United Nations Trends in International Migrant Stock details the total migrant stock in the Republic of Korea by the country of birth de�nition and because citizenship is rarely granted to people from outside, it is simply assumed that the nationality data were comparable to the foreign-born de�nition. The nationality total was then subtracted from the UN total and the remaining migrants were assigned Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 to the People’s Democratic Republic of Korea. Recording and Recoding There is little standardization in the recording and dissemination practices for censuses across destination countries.14 The level of detail with which destina- tion countries record and disseminate migration data depends on the design of the original questionnaire. Some census questionnaires ask for a speci�c country of birth and others simply ask for a general geographic region, such as Africa. Even if the original questionnaire asked detailed questions, some countries disseminate data only on how many residents were born abroad or have foreign citizenship. In general, three types of migrant origin are observed in the disseminated census data: † Speci�c geographic regions: Some of these correspond to exactly one of the 226 countries and territories in the master list. Others pertain to localities that tend to be obscure territories, islands, or regions, such as the Isle of Man or Ceuta. † Aggregate geographic regions: These correspond to two or more countries or territories in the master list. They can be continents (such as Africa), parts of continents (such as South Asia), political alliances (European Union), or other classi�cations (such as Other Ex-French Africa; Algeria, Tunisia, and Morocco; and Melanesia). The data for these aggregate regions need to be allocated to the 226 countries in the master list. The details of the procedures are discussed below. 14. The United Nations (1998) has developed recommendations aimed at promoting standardized recording practices across countries. Until such practices are followed uniformly, harmonization will remain a key issue in understanding and comparing migration statistics. ¨ zden, Parsons, Schiff, and Walmsley O 23 † Miscellaneous categories: These include refugees, stateless, and born at sea. There are generally no geographic correspondences for these. Thousands of geographic regions and categories emerged from the more than one thousand individual destination country sources chosen for the analysis. The vast majority of these are repetitions that refer to identical geographic locations using different expressions. For example, French Upper Volta and the Republic of Upper Volta were relabeled Burkina Faso. In the end, 292 speci�c geographic regions (�rst bullet above) and 236 aggregate geographic regions (second bullet) were identi�ed. The 292 speci�c regions include the 226 countries and territories in the master list and 66 other single locations that can be assigned to one of the 226 in the master list (see appendix 2).15 The 236 aggregate geographic regions pose larger problems. The migrants Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 originating from a given aggregate geographic area need to be allocated to the individual countries that comprise that area. This is one of the greatest dif�cul- ties in this project, and resolving it is one of the main contributions of this work. Several propensity measures were developed depending on the quality of the data. They are based either on a destination country’s propensity to accept migrants from a particular origin or on origin countries’ propensity to send migrants abroad. These propensity shares are then calculated, and the resulting number of migrants are assigned, in order of quality, to speci�c origin countries in the master list. Finally, the miscellaneous categories also needed to be dealt with consist- ently to enable meaningful comparisons between country pairs. There is often a high number of nonresponses to the question about place of birth for foreign- born residents (Bilsborrow and others 1997, p. 60). As a result, some censuses report large numbers of people whose place of birth is unknown. All these indi- viduals are assumed to be natives in the analysis since it is unclear whether they are domestically born or foreign born. These entries are therefore deleted from the matrices. In other cases, calculations were made to check whether these totals contributed to the foreign born in each census. In most circum- stances they did not, and so they were dropped. In cases when these totals did refer to migrants, they were treated as an appropriate aggregate category to be assigned later, as detailed below. Finally, all categories referring to the “state- less�16 were dropped because despite their importance as a minority group in global migrant patterns, there is no way to meaningfully assign them to an origin. 15. For example, the Vatican is assigned to Italy, Wake Island to the United States, and Labuan to Malaysia. 16. Some estimates put the number of stateless people (those lacking any citizenship) as high as 11 million, although many of these people will not be captured in censuses. The stateless represent an important category of migrants; for more information, see www.unhcr.org/pages/49c3646c155.html. 24 THE WORLD BANK ECONOMIC REVIEW Disaggregation of Aggregate Categories The disaggregation of the 236 origin regions identi�ed in the censuses is one of the key steps in creating the bilateral migration matrix. Three propensity equations are used to allocate migrants to one of the 226 countries in the master list. Each measure varies in quality depending on the availability of underlying data. The preferred option is to use migration data from the destina- tion country for the relevant year. If this option is not available, information from the destination country for other years is used. Should that not be poss- ible, subregions17 are created, and countries with insuf�cient data are assumed to have a similar propensity to accept migrants as other countries in the subre- gion. Failing this option, global propensity measures are constructed.18 More than a single method of allocation is chosen so that the data already in the matrices can be used to maximum effect. All these allocations ignore the gender pro�le of migrants. This dimension is accounted for at a later stage, Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 once all the aggregate categories have been assigned. Varying Survey Dates During the 10-year window of each census round, there are no conventions on when a destination country should conduct its census. Although many destina- tion countries conduct their censuses at the turn of the decade, the actual date is up to each country. Attempting to standardize census dates would require changing the numbers reported in the original census documents. Most destination countries conduct their census within two years of the middle year of each census round—between 1998 and 2002 for the 2000 census round, for example (table 3). The census numbers thus are not changed, and the matrices report all censuses as comparable in each round. A full list of census dates is in appendix 1. An alternative version of the database that has been mapped to the United Nations (2006, 2009) Trends in International Migrant Stock database is available from the authors. These data are standar- dized over time in terms of the years to which they refer. Calculating Missing Gender Splits Although common in the underlying data, bilateral migration data disaggre- gated by gender are sparser than aggregate migrant totals (see table 1). An important contribution of the current work is in estimating the gender break- down of all migrants in destination countries in the global migration matrices. Similar to the allocation from aggregated categories in the Global Migration 17. The subregions used for the disaggregations are the 21 UN regions (see http://unstats.un.org/ unsd/methods/m49/m49regin.htm, with the countries of Oceania aggregated into a single subregion. They do not match the large World Bank regions used in the analysis in section IV. 18. While this propensity measure is clearly inappropriate, less than 1 percent of all migrants and observations are assigned on this basis. This method is included so that every migrant in the underlying data is accounted for. ¨ zden, Parsons, Schiff, and Walmsley O 25 T A B L E 3 . Percentage of Censuses Conducted during the Middle of each Census Round Census round Censuses by birthplace Censuses by nationality 1960 78 66 1970 71 71 1980 78 59 1990 80 58 2000 84 57 Source: Authors’ calculations based on data described in text. Database to speci�c origins in the master list, two measures are used for calculating gender splits; they are described in appendix 5. Combining Migrant De�nitions Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Only a single de�nition of a migrant (foreign born or foreign citizen) can be applied to each destination country in the �nal matrices. Switching de�nitions over time for the same destination country would yield inconsistent data. Priority is given to the foreign-born de�nition, and these data are always used if at least three censuses using that de�nition and with detailed bilateral infor- mation are available for a particular country. However, only nationality data are available for many destination countries. For countries such as Japan that rarely offer citizenship to foreigners, this does not pose much of a problem since foreign-born and nationality data will be very similar. For other destina- tion countries, including data based on the nationality concept will lead to dis- parities. When fewer than three foreign born data sources are available and the nationality data are of superior quality, the nationality de�nition is chosen (see appendix 1). Where fewer than three data points by either de�nition are avail- able, several assumptions are made to �ll the missing data. Missing Censuses and Census Data The �nal hurdle in constructing the global migration matrices is dealing with omitted data. No census round is truly complete since no round has ever included every country in existence at the time. Censuses are expensive because of their universal coverage and labor intensity. For those reasons, many countries have started to conduct censuses only recently (Bhutan began in 2005). Censuses can also be abandoned because of civil unrest or military con- flict. They can also be politicized, because they can be used to estimate the size of a particular ethnic group. In other words, data may simply never be released even if they are collected. Nor is there any guarantee that a question on nation- ality or country of birth will even be included in the census questionnaire. Many countries in Central Asia, as well as Fiji, Sri Lanka, and Tonga, have in some years included questions on ethnicity instead, which is useless for 26 THE WORLD BANK ECONOMIC REVIEW identifying migrants. For all these reasons, inevitable gaps in the data emerge (see table 2). Three conventions are adopted for constructing missing data. The one that is ultimately used depends on how many data are missing and for which decades these data are missing relative to the decades for which data are available. MISSING IN-BETWEEN DECADES. Where data are missing for a particular decade but are available for the decade before and after, a linear trend is assumed between the earlier and later bilateral data. In total, 86 country-years of data were interpolated using this method. MISSING BEGINNING OR END DECADES. Where the data are missing at the begin- ning or the end of the time period, the destination country is assumed to have the same bilateral migrant composition as in the decade closest to the missing period. The bilateral shares from the closest decade for which data are avail- able are applied to the destination country’s total number of migrants for the Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 missing decade. The information comes from one of two sources. In some cases, the census provides the total number of migrants without any bilateral information. If these data are not available, the total from the closest decade is taken and adjusted for growth in migration. The growth rates are taken from Trends in International Migrant Stock, which details total migrant stocks for all countries in the world at �ve year intervals (United Nations 2006).19 The missing end decades are calculated for 116 countries for which data are lacking, most of them for the 1960s and 1970s.20 Trends in International Migrant Stock database thus can be used to estimate growth rates by estimating missing totals in years for which censuses are not available, and it provides a consistent set of totals over time for countries for that have data of insuf�cient quality. An important difference between the matrices presented in this article and the Trends in International Migrant Stock database is the treatment of refugees. While refugees are generally enumerated in developed country censuses, this is not always the case for developing countries. Refugees interned in camps are less likely to be surveyed at the time of census. Making allowances for these refugees, the Trends in International Migrant Stock database adds to the number of migrants refugees reported by the United Nations Refugee Agency and the United Nations Relief and Works Agency for developing countries that are not likely to have included the refuges in their census data. Since the majority of developed countries record refugees alongside other migrants on a bilateral basis, there are normally no remedial measures for removing them. 19. The 2008 revision includes data only for 1990– 2010. To ensure consistent �gures over time, the 2005 revision, which covers 1960–2005, was used instead. 20. Taiwan, China, and Norfolk Island pose an additional problem, since the United Nations does not provide data for these locations, so migrant totals in other years cannot be calculated. For these two areas, therefore, the numbers of migrants are set to zero in the earlier decades for which data are lacking. ¨ zden, Parsons, Schiff, and Walmsley O 27 Similarly, for developing countries for which no census data are available, it is impossible to know whether the numbers contained in Trends in International Migrant Stock database include refugees. For the cases that rely on the Trends in International Migrant Stock database, the number of refugees is subtracted from the totals, with the intention of removing refugees in camps from the total, since the focus is on economic migration.21 COUNTRIES WITH VERY POOR DATA. For the 59 destination countries for which there are two or fewer census data points, it is impossible to meaningfully interpolate missing census totals or bilateral numbers. In these cases the census totals detailed in the Trends in International Migrant Stock are used. This has the advantage of ensuring consistent totals for the number of migrants in each of the �ve census periods. The average bilateral shares from the censuses with data are then applied to these totals to derive bilateral data for each census round. Finally, there are six destination countries for which bilateral data are com- Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 pletely lacking.22 In these cases, data for all the other countries in the subregion are used to calculate the propensity of every country in the destination subre- gion to accept migrants from elsewhere in the world. All of the propensities sum to one. These shares are multiplied by the total migrant stock �gures pro- vided in the Trends in International Migrant Stock database to calculate the bilateral numbers. III. RELIABILITY OF THE E S T I M AT E S The previous section described the challenges in constructing the matrices and the range of measures used to generate the missing observations. This section highlights the extent to which the estimates are based on the underlying raw data and their reliability. Categorizing the Methods Used Nine main methods were used to generate the cells: (1) Pure raw: Derived directly from the raw bilateral census data. (2) Raw scaled: Based on the underlying raw bilateral data scaled to the UN numbers. (3) Pure remainder: Assigned directly from the disaggregation of aggregate categories applying one of the propensity measures. 21. In the case of Palestine, for which the UN totals consist entirely of refugees, these totals are not removed. It is possible to calculate migrant totals for Palestine in other decades. 22. The six countries are China, Eritrea, Maldives, Qatar, Somalia, and Democratic People’s Republic of Korea. Of these, Eritrea, and Somalia have been affected by conflict. China has conducted censuses over the period, but their de�nition of migration is not compatible with the de�nitions used throughout the article. 28 THE WORLD BANK ECONOMIC REVIEW (4) Remainder scaled: Based on disaggregations using one of the propensity measures and then scaled to the UN numbers. (5) Raw and remainder combined not scaled: Based primarily on bilateral raw data and to which disaggregations of certain aggregate categories were added. (6) Raw and remainder combined scaled: Similar to R&R not scaled except that the resulting value was scaled to the UN numbers. (7) Pure interpolation: Calculated solely by interpolating missing end and middle censuses, but not scaled to the UN data. (8) Interpolation and scaled: Both interpolated and scaled, for countries with poor data or for cells calculated by interpolating missing and end decades which then had to be scaled. (9) Missing: For countries for which bilateral data were missing for every census round, such as Somalia. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 The data used in the �rst six methods are from the raw census data. The data for the last three methods are missing because of omissions in the underlying data and need to be �lled. Therefore, varying percentages of observations in each decade are assigned by the methods described (table 4). In 1960, 59 percent of observations are directly assigned from the raw bilateral data or from one of the disaggregations of the aggregate raw data (the �rst six cat- egories). By 2000, this proportion rises to 69 percent. However, these obser- vations account for some 84 percent of the total number of international migrants in 1960 (table 5). This proportion rises to 86 percent by 2000 because a small number of corridors (cells) account for a large proportion of global migration stocks. The bulk of the remaining international migrants are assigned on the basis of interpolation. Among the �rst six categories that are based on raw census data, three cat- egories (raw scaled, R&R not scaled, and R&R scaled) are constructed through the summation of bilateral raw numbers and disaggregations of some aggregate categories in the original censuses. Since these categories together constitute around 45 percent of migrants in each census round, the original bilateral portion of each cell was compared with the �nal number assigned to them after the various calculations as a check on accuracy. For each decade, therefore, the overall percentage contribution of the raw bilateral data to the total is calculated (table 6).23 In each census round, at least 92 percent of all those categories are derived from the raw data. Simulating Missing Data Finally, to examine the reliability of the estimated missing census data and test the methodologies, several scenarios are assumed. All bilateral observations for 23. Although only aggregates for each decade are presented here, a full matrix detailing exactly how each cell was generated is available from the authors. T A B L E 4 . Percentage Distribution of Observations by Allocation Method Raw and remainder combined Census round Pure raw Raw scaled Pure remainder Remainder scaled Not scaled Scaled Pure interpolation interpolation and scaled Missing 1960 12.31 0.13 40.65 3.34 2.12 0.02 24.86 12.58 3.98 O 1970 12.07 0.02 34.88 2.03 2.54 0.17 33.64 10.68 3.98 1980 12.00 0.15 45.76 5.90 4.28 0.15 18.88 8.91 3.98 1990 13.82 0.27 47.13 5.88 6.50 0.06 15.23 7.12 3.98 2000 12.85 1.02 39.97 4.29 10.41 0.86 10.66 15.97 3.98 Source: Authors’ calculations based on data described in text. ¨ zden, Parsons, Schiff, and Walmsley 29 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 30 T A B L E 5 . Percentage Distribution of Migrants by Allocation Method Raw and remainder combined Census Pure Raw Pure Remainder Not Pure Interpolation Total THE WORLD BANK ECONOMIC REVIEW round raw scaled remainder scaled scaled Scaled interpolation and scaled Missing (millions) 1960 28.61 8.73 7.10 0.51 39.23 0.20 4.72 10.51 0.40 92.3 1970 42.94 0.00 4.01 1.97 34.25 0.25 12.00 4.11 0.46 102.4 1980 30.12 0.23 4.42 0.07 48.77 0.14 11.31 4.64 0.30 118.6 1990 36.61 0.55 4.79 0.35 43.07 0.15 10.05 3.84 0.58 139.4 2000 35.62 1.18 7.56 0.33 40.22 1.07 7.08 6.23 0.72 165.3 Source: Authors’ calculations based on data described in text. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 ¨ zden, Parsons, Schiff, and Walmsley O 31 T A B L E 6 . Contribution of Raw Bilateral Data to the Total Census round Accounted for by “raw� data (%) 1960 95.9 1970 92.5 1980 92.5 1990 92.1 2000 93.6 Source: Authors’ calculations based on data described in text. a single year for four countries (Australia, United States, Switzerland, and Chile) in different parts of the world are deleted and the missing cells are �lled using one of �ve methods.24 The �rst simulation assumes that all bilateral data for 2000 are missing but that the total number of migrants is available. The missing bilateral numbers then have to be �lled using the propensity measure (equation 1 in appendix 4) based on the data available in other years. The Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 second and third simulations assume that the total is missing as well, and interpolation is used to �ll in all missing data for 1960 and 2000. The fourth and �fth simulations remove all data for all years and then �ll the missing years using data for the remaining portion of the subregion (table 7). The simulations perform well. The four countries are examined one at a time, starting with Australia. The correlation coef�cient between the predicted and actual data in each simulation is at least 0.945. Interpolating the data is the most accurate method of predicting the missing data, and simulation 2 for 1960 is more accurate than simulation 3 for 2000. Simulation 1 does not perform as well: the data from other years fail to adequately account for the fairly signi�- cant shift in the composition of the Australian immigrant stock after 1990. When simple subregional shares are used (simulations 4 and 5), the correlation coef�cients remain high. The actual distribution of immigrants, however, is less accurate, especially in simulation 5. This is because New Zealand, the country in the subregion that has by far the greatest weight for apportioning migrants for Australia’s missing data, did not experience the same influx of migrants from Asia that Australia did. In other words, Australia represents such a large share of immigration in Oceania that when it is removed, the remaining countries (mostly small island countries that are origins, not destinations) are not particu- larly accurate predictors of migration to Australia. The U.S. case is similar. Using interpolation to �ll in the missing years proves effective, while the results from simulation 1 are also reasonable. The results from simulations 4 and 5 are less accurate. The problem with using regional shares for calculating missing coef�cients for the United States is similar to that for Australia. The poor results are due to the differences in the migrant pro�les of the United States and Canada, which provides the weights 24. For all countries, data quality is highest for 2000 and lowest for 1960, except for Chile, for which 1980 has the worst quality data. 32 T A B L E 7 . Five Simulations Testing the Reliability of Generated Cells with Missing Data 1 Propensity (2000 2 Interpolation (1960 3 Interpolation (2000 removed) removed) removed) 4 Missing (1960 removed) 5 Missing (2000 removed) Correlation Log Correlation Log Correlation Log Correlation Log Correlation Log Country coef�cient ratio coef�cient ratio coef�cient ratio coef�cient ratio coef�cient ratio THE WORLD BANK ECONOMIC REVIEW Australia 0.945 – 0.575 0.998 – 0.067 0.990 – 0.126 0.954 –0.242 0.946 – 1.302 United 0.893 – 0.243 0.961 0.169 0.972 – 0.262 0.596 0.284 0.250 0.124 States Switzerland 0.771 0.393 0.971 0.041 0.900 – 0.312 0.899 –0.006 0.818 0.010 Chile 0.688 0.059 0.997 0.183 0.897 0.038 0.498 –0.080 0.376 0.009 Note: A cutoff of 250 migrants is implemented for calculating the log ratios since they can be highly skewed by the predictions of very small corridors. Source: Authors’ calculations based on data described in text. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 ¨ zden, Parsons, Schiff, and Walmsley O 33 for �lling in the missing U.S. values. This methodology signi�cantly underpre- dicts the numbers of migrants from U.S. dependencies, since Canada hosts very few of them, and overpredicts the numbers from former British colonies, popu- lations that are more prominent in Canada. Simulations 4 and 5 perform extremely well for Switzerland: the deviations from the actual data are less than 1 percent. This is due to the fact that several large Western European nations have similar migrant pro�les to Switzerland, unlike the case for Australia and New Zealand and the United States and Canada. The data for 1970–2000 prove better for interpolating the missing data for Switzerland for 1960, while the data for earlier years are somewhat less effective at predicting the missing data for 2000. The results for Chile are also good. Using the data for Chile in other years and the propensity measures yields a margin of error that is under 6 percent (simulation 1). Interpolation proves accurate when data for either 1960 or 200 are removed. With subregional shares, the differences in the log ratios are small, Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 but the correlation coef�cients are not as high as in other cases because Chile’s immigrant pro�le is bimodal. Chile has a small number of large immigrant stocks and a large number of very small stocks. Although the predictions for the size of the stocks are reasonable, the relative rankings are not as accurate. The results indicate that interpolation is the most effective method of allo- cation, although the allocations based on the propensity measures and on the subregional shares fair reasonably well. This is heartening, since around a quarter of the observations and 14 percent of the world migrant stock is allo- cated for 2000 using interpolation. Filling a missing country-year of data using propensities is less effective. Even so, the correlations remain high and the result- ing data are not suf�ciently inaccurate to warrant throwing them away. It is important to remember, however, that simulation 1 represents a worst case. This extreme measure is resorted to only for a few countries for which data are missing. In almost every case, aggregate categories are much narrower in the raw data. Nevertheless, even with this constrained method with extreme assumptions (missing all data for a country in a region with very few comparable countries), the results seem reasonable. And even when the results are skewed, this is gener- ally due to the over- or underpredicting of a handful of key migrant corridors. Finally, the aggregate �gures obtained are compared with those from the Trends in International Migrant Stock database (United Nations 2006, 2009) to highlight key differences. The database provides data by destination only, not for each bilateral corridor, so only aggregate numbers can be compared. For this comparison, mid-year estimates of the world migrant stock for 1990– 2000 are taken from the 2008 edition and estimates for the earlier censuses, 1960–1980, are taken from the 2005 edition (table 8). The analysis subtracts the estimated number of refugees from the total mid-year estimates of the world migrant stock from the Trends in International Migrant Stock database to yield the net number of migrants in each decade. These numbers are then compared with the decadal estimates generated through this project, both the 34 THE WORLD BANK ECONOMIC REVIEW T A B L E 8 . Comparison of Aggregate Numbers with the United Nations Trends in International Migrant Stock Database Unite Nations database Current study Census round Total Refugees Net total Total Within the Soviet Union Germans Net total 1960 75.5 2.2 73.3 92.3 15.8 3.7 72.7 1970 81.3 3.9 77.4 102.4 21.0 3.8 77.6 1980 99.3 9.1 90.2 118.6 23.6 3.8 91.3 1990 155.5 18.5 137.0 139.4 – 4.7 134.7 2000 178.5 15.6 162.9 165.3 – 3.8 161.5 Source: Authors’ calculations based on data described in text and United Nations (2006, 2009). total and the net, after subtracting estimates of migrants within the Soviet Union for 1960–1980 (data for 1990 and 2000 should be directly comparable) Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 and the number of ethnic German migrants added to the German censuses. The aggregate estimates are remarkably close (the two net totals), differing at most by around 1 million migrants, except in 1990. There are several poss- ible explanations for these differences. First, the census totals from the current work may not match because censuses do not always make allowances for tem- porary workers. For example, Singapore’s of�cial 2000 census records 563,430 foreign-born migrants. The United Nations, however, reports 1,351,806 foreign-born migrants for 2000. Second, there are cases where the current study reports data by nationality, but the corresponding �gure in the Trends in International Migrant Stock refers to the foreign born. This situation generally arises when a census does not report the number of foreign-born migrants on a bilateral basis. Examples include Austria and Co ˆ te d’Ivoire. Third, differences in the years to which the data refer can generate large disparities. For example, this study uses the 1966 data for Australia, whereas Trends in International Migrant Stock reports data for 1970. Overall, however, the fact that the totals are remarkably close in every decade adds credence to the estimates here. I V. T H E E V O L U T I O N OF G LO B A L B I L AT E R A L M I G R AT I O N The greatest strengths of the global migration matrices are their bilateral cover- age, the number of decades covered, and the disaggregation by gender. These data are too rich for a full analysis of all movements between all pairs of countries. Instead, this section summarizes the major trends in the evolution of bilateral migrant stocks, based primarily on World Bank regions.25 25. Appendix 1 details the World Bank regions: South Asia, East Asia and Paci�c, Sub-Saharan Africa, Latin America and the Caribbean, Europe and Central Asia, and Middle East and North Africa. High-income Middle East and North Africa refers to the predominantly oil producing countries in the Persian Gulf (Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, and the United Arab Emirates) and to Israel. ¨ zden, Parsons, Schiff, and Walmsley O 35 Global Trends The migration matrix for the 1960 census round reflects a realigning world in the postcolonial era. Over the 1960–2000 period, the composition of world migration fundamentally changed, driven by world events and increasingly selective immigration policies in developed countries, which led to greatly diversi�ed migrant stocks. Mirroring this pattern, most countries now send migrants to an increasing number of destinations. Migration to developing countries has been driven largely by the partitioning of India26 and the breakup of the Soviet Union, both events that need be reconciled when inter- preting the data. However, while the United States and Western Europe remained throughout the most important destinations, there have been signi�- cant migration movements to the other countries of the ‘New World’ (Australia, New Zealand, and Canada) as well as to the oil-rich Persian Gulf countries ( primarily from South and East Asia), reflecting a huge increase in Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 demand for labor following the oil shocks of the 1970s. Between 1960 and 2000, the total global migrant stock increased from 92 million to 165 million.27 At the beginning of the period, one �fth of the world’s migrant population was born in Europe, and one sixth was attributable to the partition of India and migration within the Soviet Union. Two-thirds of the growth up to 2000 was due to migrant flows to Western Europe and the United States, and the rest was due mostly to increased mobility between the countries of the former Soviet Union, the emergence of the Gulf States as key migrant destinations, greater intra-Africa migration flows, and migration to Australia, New Zealand, and Canada. The number of migrants in South Asia fell over the period, reflecting a falloff after the migrations that followed par- tition (see �gure 2 later in this article). Despite the sustained increase in the global migrant stock over the period, migrants declined as a share of the world population between 1960 and 1990 (from 3.05 percent to 2.63 percent), then rose again slightly to 2.71 percent in 2000. The importance of migration for destination and origin countries depends on the size of the migrant stock relative to the population. As might be expected, many countries with the highest concentrations of immigrants are small countries with comparatively few people. The countries or territories with a population or more than 1 million people and immigrant ratios over 20 percent in 2000 include the United Arab Emirates (41 percent), Kuwait (38 percent), the Occupied Palestinian Territories (31 percent), Israel (25 percent), and Oman (20 percent). Countries with immigrant ratio less than 1 percent include Indonesia, Madagascar, and Cuba. By destination subregion, migration has become more concentrated in all developed country regions and less 26. It is not possible to differentiate among migrants who moved before, during, or immediately after the partition of India because these migrations occurred before the beginning period of the matrices. 27. This increase would be starker had it not been for the special treatment of the Soviet Union. 36 THE WORLD BANK ECONOMIC REVIEW concentrated in many developing country regions, especially South and Southeast Asia, South America, and Southern, Eastern, and Central Africa. Emigration ratios (ratio of emigrants to the sum of the emigrant and dom- estic populations) were calculated for origin countries. Unsurprisingly, small island states and those experiencing political upheaval or environmental cata- strophe have the highest emigration concentrations. In 2000 these included Niue (80 percent), Tokelau (64 percent), Montserrat (56 percent), Cook Islands (53 percent), and Palau (47 percent). Countries or territories with more than 1 million residents and the highest emigration concentrations include Jamaica (26 percent), the Occupied Palestinian Territories (24 percent), Albania (23 percent), Bosnia and Herzegovina (23 percent), Republic of Ireland (23 percent), and Armenia (22 percent). Those at the other end of the spectrum include Mongolia (2 percent), Madagascar (4 percent), Ethiopia (4 percent), and Brazil (5 percent). By subregion of origin, emigrant concen- trations have remained far more stable over the period than immigrant ratios Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 across most of the world. Notable changes have occurred, however, in emigra- tion ratios in the Paci�c and the Caribbean and Central America (higher) and South Asia (lower). Global Migration between the “North� and the “South� Dividing the world into two regions, the North (developed countries) and the South (developing countries),28 highlights some of important patterns under- pinning international migration over the second half of the twentieth century. The number of migrants from the North remained fairly stable, while the number from the South increased (�gures 1 and 2). Much of the growth in the number of migrants is driven by migrations from the South to the North, which rose from 14 million to 60 million between 1960 and 2000. Numerically, South–South migration dominates global trends, although this migration is declining as a proportion of total world migration. In 1960, South–South migration accounted for 61 percent of the total migrant stock; by 2000, it had fallen to 48 percent. When the migrant-creating effects of South Asia and the Soviet Union are factored in, however, South–South migration remains stable over the period, at about a quarter of the total (see �gure 2). As a proportion of total migrant stock, only South–North migration rose between 1960 and 2000. Increasingly liberal immigration policies in developed countries have been paralleled by large movements from developing countries. The data show that the proportion of world migration attributable to South– North migration rose from 16 percent to 37 percent. This dramatic increase is 28. The developed countries are Australia, Canada, Japan, New Zealand, the United States, and the EU-15 and the European Free Trade Association, which have all been relatively affluent over the entire period of interest. The EU-15, rather than some other European Union grouping, is included because the latest year to which the data refer is 2004. All other countries are classi�ed as developing. ¨ zden, Parsons, Schiff, and Walmsley O 37 F I G U R E 1. Changes in the Number of Migrants in Developed to Developing Country Migration Corridors, 1960–2000 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Source: Authors’ calculations based on data described in text. F I G U R E 2. Changes in the Share of Migrants by Migration Corridors, 1960-2000 (percentage contribution) Source: Authors’ calculations based on data described in text. unquestionably one of the de�ning trends of the period, surpassing migration between developed countries from 1970 to 1980, both in numbers and as a proportion of the total migrant stock. 38 THE WORLD BANK ECONOMIC REVIEW Global Migration to Developed Countries The growth in the South–North migration has been driven largely by move- ments to the United States and Western Europe. Between 1960 and 2000, migrant stocks grew by 24.3 million in the United States and 22 million in Western Europe, accounting for some 42 percent of the world total in 2000. However, there are notable differences in the migrant compositions of these two regions. Whereas the U.S. immigrant pro�le has changed dramatically, Europe’s has remained more stable, reflecting in part its continuing ties with former colonies. Immigration to the United States in 1960 was dominated by Europeans, who accounted for around 60 percent of the total and 6 of the top-10 migrant corridors. Of the 10.4 million migrants in the United States at that time, 1.26 million were born in Italy, 990,000 in Germany, 835,000 in Great Britain, 750,000 in Poland, 360,000 in Ukraine, 340,000 in Ireland, and 305,000 in Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Austria. By 2000, the share of these origin countries declined, to around 15 percent. Balancing this trend, the number of migrants from Latin America and the Caribbean and East Asia and Paci�c rose sharply. In 2000, 52 percent of the immigrant stock in the United States were born in Latin America and the Caribbean and 17 percent in East Asia and Paci�c. The United States is an important destination for migrants from all regions except Southern and Central Africa. In 2000, the United States received the largest number of migrants29 from 60 countries, including Germany, Vietnam, Cuba, and the Republic of Korea. Moreover, 13 of the 50 largest migration corridors in the world and 6 of the 10 largest South–North corridors in 2000 were to the United States. The two largest corridors to the United States were from Mexico and the Philippines, the largest and 12th largest developing to developed country migration corridors in the world. They accounted for 10.8 million migrants, equivalent to 31 percent of the migrant stock in the United States, or nearly 7 percent of the world migrant stock. Western Europe has been instrumental in many of the largest migrations in history, as both a major sending and receiving region. Between 1960 and 2000, many Western European countries transformed from net migration senders to net migration receivers. Today, Western Europe remains a key destination region for migrants from every other part of the world except the high-income Middle East and North Africa region. Increasingly over the period, Western Europeans began migrating to other countries in the region. In 2000, two-�fths of Western European migrants lived elsewhere in Western Europe, driven largely by the expansion and economic and political integration of the European Union. This is a signi�cant increase from 1960, when far greater numbers of Europeans chose to migrate to the United States and to Latin 29. Migration corridors are discussed to highlight the most important global migrant stocks; at no point does the discussion relate to migration flows. The focus is on stock data, and the term “migration corridor� simply refers to the bilateral migrant stock for a particular pair of countries. ¨ zden, Parsons, Schiff, and Walmsley O 39 America and the Caribbean. Despite these increases, however, intra-Western European migrants are increasingly becoming a minority proportion of the migrant stock, especially after 1970 as migration from developing countries increased. Migrants from Turkey and Poland in Germany constitute the two largest diasporas in Western Europe and the second and third largest develop- ing to developed countries migration corridor globally. Elsewhere in Europe, the most signi�cant migrant corridor from developing countries is from Algeria to France. In all decades except 2000, this corridor is among the top four most important developing to developed country migrations in the world. Other notable corridors from the South to Western Europe include South Asia to Great Britain, the former Yugoslavia to Germany, and North Africa (countries in addition to Algeria) to France. Modern day Australia, New Zealand, and Canada were all founded through immigration; in 1960, 71 percent of migrants to Australia, New Zealand, and Canada were born in Western Europe—39 percent of them in the United Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Kingdom. By 2000, however, that share had fallen to 36 percent of the total, as migrants from the East Asia and Paci�c region ( particularly China and Vietnam) gained prominence; they now account for more than a �fth of migrants. Germans in the United States and British in Australia are the two largest migration corridors between developed countries. Facing a chronic skills short- age, Australia implemented the Ten Pound Pom scheme in the postwar period as part of its Populate or Perish policy. Opening the country to all British citi- zens, including those from Cyprus and Malta, the Australian government managed to persuade over one million people to migrate before 197330 for the price of just 10 British pounds. Given the cultural similarities between Australia and the United Kingdom and the relaxed reciprocal visa restrictions, bilateral migration flows remain strong to this day. Japan has historically been more reticent than other OECD members to admit migrants. Immigration to Japan is mainly from Korea and elsewhere in East Asia, although from 1960 onwards, Japan did admit larger proportions of migrants from both Southeast Asia and South America, speci�cally Brazil, the Nikkei burajiru-jin. Global Migration to Developing Countries Statistically, the most important events affecting migrant movements to the South over the study period are the partition of India and the disintegration of the Soviet Union. There have been other important changes as well since 1960, particularly the large shift in global migration toward the Persian Gulf countries. In 2000, 15 percent of the migrant stock in developing countries (including both India partition and intra-Soviet Union migrants) was in the high-income Middle East and North Africa region, up from under 3 percent in 1960. These 30. From 1973 onward, the price of assisted migrant’s passage rose. 40 THE WORLD BANK ECONOMIC REVIEW migrants reflect movements predominantly from South and Southeast Asia (45 percent in 2000) and the low-income Middle East and North Africa region (33 percent) to the Gulf and from the countries of the former Soviet Union to Israel.31 Of total migration to developing countries, the low-income Middle East and North Africa and the Latin America and Caribbean regions continue to attract steady shares. Compared with 1960, however, both regions attract proportionally far fewer Western Europeans and more migrants from other developing countries. Although the number of migrants across Africa increased by some 4 million over the period, Sub-Saharan Africa accounted for only 14 percent of total migrants in developing countries in 2000, down from 11 percent in 1960. The numbers of migrants in Southeast Asia, Europe other than European Free Trade Association and the EU 15, and Eastern Africa fell over the period, reflecting a sharp drop in migrants from East Asia in Southeast Asia, fewer migrants from the former Soviet Union in Eastern Europe, and fewer migrants from South Asia and East Africa to other developing countries Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 in the subregions. Intra-Soviet Union and intra-South Asia migration constituted 42 percent of South–South migration globally in 2000 (�gure 3). The largest migrant corri- dors were between countries of the former Soviet Union, between Russia and Ukraine (in both directions), and between Kazakhstan and Russia. Migrant corridors between Bangladesh, India, and Pakistan are very large in both direc- tions, with Bangladeshi migrants in India the largest migrant population in South Asia. In the Persian Gulf, the largest migrant groups are Indian and the Egyptian migrants in Saudi Arabia, Indian migrants in the United Arab Emirates, and Pakistani migrants in Saudi Arabia. Migration from the North to the South, although still large, is declining (see �gure 2). In 1960, developed country migrants constituted the majority of migrants to the Paci�c Islands, Central and South America, and Central Africa; today, that is no longer the case. Migrants from developed to developing countries have declined in both absolute and relative importance. Today, the most important developed to developing country movements are from Western Europe to South America and to other European countries and from the United States to Central America and the Caribbean. Migrants from the United States to Mexico constitute the largest developed to developing country migration corridor in the world today, at more than 340,000 people. Before 2000, migration between Italy and Argentina was the largest developed to developing country migration corridor in every decade. Other notable devel- oped to developing country corridors are from Spain to Argentina and from Great Britain to South Africa. 31. In 1960, over half of all migrants in Israel were born in the Eastern Europe and Central Asia. ¨ zden, Parsons, Schiff, and Walmsley O 41 F I G U R E 3. Inter- and Intra- regional Migration between Developing Countries, 2000 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Source: Authors’ calculations based on data described in text. Gender Assessment of International Migrant Stocks In 1960, men made up a larger share of all regional immigrant stocks except in the United States and Eastern Europe and Central Asia (�gure 4a). Between 1960 and 2000, the gender composition of immigrant stocks changed consider- ably. The United States, Eastern Europe and Central Asia, and South Asia all experienced slight declines in the share of women in total migrants. The largest percentage increases over the period in the share of women in the total migrant stocks were Latin America and the Caribbean (14.8 percent); Japan (14.3 percent); East Asia and Paci�c (13.3 percent); Sub-Saharan Africa (11.2 percent); Australia, New Zealand, and Canada (8.3 percent); and Western Europe (4.9 percent). The proportion of women in the migrant stock fell sharply in both the high-income Middle East and North Africa region (23.8 percent) and the low-income Middle East and North Africa region (9.1 percent drop) . In absolute terms, however, the number of female migrants in all regions but South Asia rose. Despite the high-income Middle East and North Africa region hosting fewer women than men, the region experienced the largest rise in the number of female migrants (up 3.5 million or 540 percent) over the period. Other regions that experienced large increases in the number of female migrants include the United States (up 12.1 million or 228 percent); Western Europe (11.2 million, 190 percent); and Australia, New Zealand, and Canada (3 million, 130 percent). The biggest absolute decline in the numbers of female migrants between 1960 and 2000 was in South Asia (down 3 million or 40 percent). In 2000, the countries with the highest proportion of female migrants were Nepal (70 percent), Mauritius (63 percent), and Moldova (60 percent). 42 THE WORLD BANK ECONOMIC REVIEW Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 F I G U R E 4. The Percentage of Women in Immigrant Stock by Region, 1960 and 2000 In terms of emigrant stocks in 1960, only two regions sent higher numbers of women abroad relative to men, Australia, New Zealand, and Canada and Eastern Europe and Central Asia (�gure 4b). They did so again in 2000, along with Western Europe, East Asia and Paci�c, and Japan. In percentage terms, the ratio of female to male emigrants declined slightly in the United States; Australia, New Zealand, and Canada; and Eastern Europe and Central Asia and more substantially in South Asia (9.6 percent) and in both Middle East and North Africa regions (high income, 6.2 percent; low-income, 7.8 percent). The four regions that experienced the greatest increases also experienced the ¨ zden, Parsons, Schiff, and Walmsley O 43 largest increase in women as a share of their total immigrant stocks: East Asia and Paci�c (17.9 percent), Japan (15.5 percent), Sub-Saharan Africa (15.4 percent), and Latin America and the Caribbean (6.9 percent). In absolute terms, all regions of the world sent more women abroad in 2000 than in 1960. The largest proportional increase was from Latin America and the Caribbean (up 10.9 million or 630 percent), followed by the high-income Middle East and North Africa region (500,000, 290 percent), the low-income Middle East and North Africa region (3.3 million, 250 percent), Japan (330,000, 210 percent), East Asia and the Paci�c (6.3 million, 180 percent), and Sub-Saharan Africa (4.4 million, 180 percent). In 2000, the countries with the highest pro- portion of women in their emigration stocks were Ukraine (61 percent), the Philippines (60 percent), and Singapore (60 percent). Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 V. C O N C L U S I O N This article draws on the largest collection of censuses and population registers providing information on international bilateral migration and constructs con- sistent square matrices for the last �ve completed census rounds (1960 to 2000). Problems in the underlying data that confound meaningful comparisons include differences in recording and recoding practices among destination countries and missing and omitted data. The main contribution of this article is in recognizing and overcoming these obstacles by making a series of simplifying assumptions. Tradeoffs between pragmatism and accuracy are inevitable, and one of the largest hurdles is estab- lishing a set of rules for achieving a �xed set of countries. Researchers face daunting challenges when working with migration data, and any attempt to resolve them will inevitably fall short of the ideal, especially when compared to international statistics on trade and �nancial flows. Nevertheless, given the paucity of comparable data on international migration, especially outside of the OECD, the completed database represents an important step in an ongoing effort to understand trends in international migration. The matrices provide a reasonably accurate portrait of global migration over the second half of the twentieth century and should provide a useful starting point for researchers and policymakers working on a broad range of issues. 44 THE WORLD BANK ECONOMIC REVIEW APPENDIX 1. LIST OF SOURCES Ta b l e A 1 . List of Database Sources by Census Round Country or territory De�nitiona 1960 1970 1980 1990 2000 Australia and New Zealand Australia FB 1961 1966 1981 1991 2001 New Zealand FB 1961 1971 1981 1986 2001 Japan Japan NT 1960 1970 1980 1990 2000 Canada Canada FB 1961 1981 1986 2001 United States United States FB 1960 1970 1980 1990 2000 Western Europe Andorra NT 1969 1984 1994 2004 Austria NT 1961 1971 1981 1991 2001 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Belgium NT 1961 1970 1981 1991 2001 Cyprus FB 1960 1992 2001 Denmark FB 1960 1965 1981 1991 2001 Faeroe Islands NT 1994 2004 Finland FB 1960 1970 1980 1990 2000 France FB 1962 1968 1982 1990 1999 Germany NT(FB) 1960 1970 1980* 1990* 2000 Gibraltar FB 1961 1970 1981 1991 2001 Greece NT 1961 1971 1981 1991 2001 Iceland FB 1960 1965 1980 1990 2000 Ireland FB 1961 1970 1981 1986 2002 Italy FB 1961 1971 1981 1991 2001 Liechtenstein NT 1960 1970 1980 1990 1998 Luxembourg FB 1960 1970 1981 1991 2001 Malta NT 1957 1967 1995 Monaco FB 1961 1968 1982 1990 2000 Netherlands FB 1960 1992 2002 Norway FB 1960 1970 1980 1990 2000 Portugal FB 1960 1981 1991 2001 San Marino NT 1972 1980 Spain FB 1960 1981 1991 2001 Sweden FB 1960 1970 1980 1990 2000 Switzerland NT 1960 1970 1980 1990 2000 United Kingdom FB 1961 1971 1981 1991 2001 Eastern Europe and Central Asia Albania NT 1989 Armenia ETH(FB) 1959 1970 1979 1989 2001 Azerbaijan ETH(FB) 1959 1970 1979 1989 Belarus ETH(FB) 1959 1970 1979 1989 1999 Bosnia & Herzegovina FB 1981* Bulgaria FB 2001 Croatia FB 1981* 1991 2001 Czech Republic FB 1991* 2001 Estonia ETH(FB) 1959 1970 1979 1989 2000 Georgia ETH(FB) 1959 1970 1979 1989 (Continued ) ¨ zden, Parsons, Schiff, and Walmsley O 45 TABLE A1. Continued Country or territory De�nitiona 1960 1970 1980 1990 2000 Hungary NT 1960 2003 Kazakhstan ETH(FB) 1959 1970 1979 1989 Kyrgyzstan ETH(FB) 1959 1970 1979 1989 1999 Latvia ETH(FB) 1959 1970 1979 1989 2000 Lithuania ETH(FB) 1959 1970 1979 1989 2001 Macedonia FB 1981* 1994 Moldova ETH(FB) 1959 1970 1979 1989 Poland FB 1970 2002 Romania FB 1966 1992 2002 Russian Federation ETH(FB) 1959 1970 1979 1989 2002 Serbia & Montenegro FB 1981* 1991 2002 Slovakia FB 1991* 2001 Slovenia FB 1981* 1991 2002 Tajikistan ETH(FB) 1959 1970 1979 1989 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Turkey FB 1960 1965 1980 1990 2000 Turkmenistan ETH(FB) 1959 1970 1979 1989 Ukraine ETH(FB) 1959 1970 1979 1989 2001 Uzbekistan ETH(FB) 1959 1970 1979 1989 High income Middle East and North Africa Bahrain NT 1959 1971 1981 1991 2001 Israel FB 1961 1972 1983 2001 Kuwait NT 1957 1970 1975 1985 2001 Oman NT 1993 2004 Qatar FB Saudi Arabia NT 1992 1995 United Arab Emirates NT 1980 1993 2003 Rest of Middle East and North Africa Algeria NT 1966 Egypt NT 1960 1976 1986 1996 Iran (Islamic Republic of) NT 1986 1996 Iraq FB 1957 1997 Jordan NT 1961 1979 1994 2004 Lebanon FB 1996 Libyan Arab Jamahiriya NT 1964 1973 Morocco NT 1960 1971 2004 Occupied Palestinian Territory FB 1997 Syrian Arab Republic NT 1960 1970 1981 1994 Tunisia NT 1956 1966 1984 1994 2004 Yemen NT 1986 2004 Africa Angola FB 1960 1983 1993 Benin NT 1979 2002 Botswana NT 1971 1981 1991 2001 Burkina Faso FB 1975 1985 1996 Burundi FB 1979 1990 Cameroon FB 1976 1987 Cape Verde NT 1980 1990 Central African Republic NT 1975 1988 Chad FB 1993 (Continued ) 46 THE WORLD BANK ECONOMIC REVIEW TABLE A1. Continued Country or territory De�nitiona 1960 1970 1980 1990 2000 Comoros FB 1958 1980 1991 Congo NT 1974 1984 Coˆ te d’Ivoire NT 1975 1988 1998 Democratic Republic of the Congo NT 1958* 1984 Djibouti FB 1991 Equatorial Guinea NT 1950 1983 Eritrea FB Ethiopia NT 1961 1994 Gabon NT 1960 1993 Gambia NT 1963 1973 1983 1993 Ghana FB 1960 1970 1984 2000 Guinea NT 1983 1996 Guinea-Bissau FB 1950 1979 1991 Kenya FB 1962 1969 1979 1989 1999 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Lesotho NT 1956 1976 1986 1996 Liberia FB 1962 1974 1984 Madagascar NT 1965 1975 1993 Malawi FB 1966 1977 Mali FB 1976 1987 1998 Mauritania NT 1977 1988 Mauritius NT 1972 1983 1990 2000 Mayotte FB 1991 1997 Mozambique NT 1955 1980 1997 Namibia NT 1991 2001 Niger NT 1977 1993 2001 Nigeria NT 1963 1991 Rwanda NT 1958* 1978 1991 2002 Re´ union FB 1961 1974 1982 1990 1999 Saint Helena FB 1966 1976 1987 1998 Sao Tome and Principe NT 1981 1991 Senegal FB 1960 1976 1988 2002 Seychelles NT 1960 1982 1987 1997 Sierra Leone FB 1985 2004 Somalia FB South Africa FB 1961 1970 1980 1985 2001 Sudan FB 1956 1983 1993 Swaziland FB 1956 1966 1976 1986 1997 Togo NT 1981 Uganda NT 1969 1991 2002 United Republic of Tanzania FB 1967 1978 1988 2002 Zambia FB 1963 1969 1980 1990 Zimbabwe FB 1956 1969 1992 South Asia Afghanistan FB 1975 Bangladesh FB 1961 1974 Bhutan FB 2005 India FB 1961 1971 1981 1991 2001 Maldives FB Nepal FB 1961 1971 1981 1991 2001 (Continued ) ¨ zden, Parsons, Schiff, and Walmsley O 47 TABLE A1. Continued Country or territory De�nitiona 1960 1970 1980 1990 2000 Pakistan FB 1961 1973 1998 Sri Lanka NT 1963 1971 1981 East Asia and the Paci�c American Samoa FB 1960 1970 1980 1990 2000 Brunei Darussalam FB 1960 1971 1981 1991 Cambodia FB 1998 China FB China, Hong Kong Special Administrative FB 1961 1971 1981 1991 2001 Region China, Macao Special Administrative FB 1981 1991 2001 Region Cook Islands FB 1956 1966 1976 1996 Democratic People’s Republic of Korea FB Democratic Republic of Timor-Leste FB 2004 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Fiji FB 1956 1966 1976 1986 French Polynesia FB 1962 1977 1988 1996 Guam FB 1960 1970 1980 1990 2000 Indonesia NT 1971 1990 2000 Kiribati FB 1963 1973 1978 1990 2000 Lao People’s Democratic Republic NT 1995 Malaysia FB 1957 1970 1980 1991 2000 Marshall Islands NT 1988 1999 Micronesia (Federated States of) FB 1973 1994 2000 Mongolia NT 2000 Myanmar NT 1973 1994 2002 Nauru FB 1961 1966 1977 2002 New Caledonia FB 1963 1969 1983 1989 1996 Niue FB 1956 1966 1976 1986 Norfolk Island FB 1981 1991 2001 Northern Mariana Islands FB 1980 1990 2000 Palau FB 1980 1990 2000 Papua New Guinea FB 1966 1980 Philippines NT 1960 1970 1980 1990 2000 Republic of Korea NT(FB) 1960 1970 1980 1990 2000 Samoa FB 1956 1971 1986 2001 Singapore FB 1957 1970 1980 1990 2000 Solomon Islands FB 1970 1976 1986 1999 Taiwan NT 1990 2000 Thailand NT 1960 1970 2000 Tokelau FB 1961 1972 1976 1986 2001 Tonga FB 1956 1966 1976 1986 1996 Tuvalu FB 1963* 1973* Vanuatu FB 1967 1979 1989 1999 Viet Nam FB 1989 Wallis and Futuna Islands FB 1969 1976 1990 2003 Latin America and the Caribbean Anguilla FB 1984 1992 2001 Antigua and Barbuda FB 1960 1970 1991 2001 Argentina FB 1960 1970 1980 1991 2001 (Continued ) 48 THE WORLD BANK ECONOMIC REVIEW TABLE A1. Continued Country or territory De�nitiona 1960 1970 1980 1990 2000 Aruba FB 1960 1981 1991 2000 Bahamas FB 1960 1970 1980 1990 Barbados FB 1960 1980 1990 Belize FB 1960 1980 1991 2000 Bermuda FB 1960 1970 1980 1991 2000 Bolivia FB 1950 1976 1992 2001 Brazil FB 1960 1970 1980 1991 2000 British Virgin Islands FB 1960 1970 1980 1991 Cayman Islands FB 1960 1979 1989 2000 Chile FB 1960 1970 1982 1992 2002 Colombia FB 1964 1970 1993 2005 Costa Rica FB 1963 1973 1984 2000 Cuba FB 1953 1970 2000 Dominica FB 1960 1981 1991 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Dominican Republic FB 1960 1970 2002 Ecuador FB 1962 1974 1982 1990 2001 El Salvador FB 1961 1971 1992 2007 Falkland Islands (Malvinas) FB 1962 1972 1986 2001 French Guiana FB 1961 1974 1982 1990 1999 Greenland FB 1960 1970 1976 Grenada FB 1960 1981 1991 Guadeloupe FB 1961 1974 1982 1990 1999 Guatemala FB 1963 1973 1981 1994 2002 Guyana FB 1960 1980 1991 2002 Haiti FB 1950 1971 1982 Honduras FB 1961 1988 2001 Jamaica FB 1960 1970 1982 1991 2001 Martinique FB 1961 1974 1982 1990 1999 Mexico FB 1960 1970 1980 1990 2000 Montserrat FB 1960 1970 1980 1991 Netherlands Antilles FB 1971 1981 1992 2001 Nicaragua FB 1963 1971 1995 Panama FB 1960 1970 1980 1990 2000 Paraguay FB 1950 1972 1982 1992 2002 Peru FB 1960 1972 1981 1993 Puerto Rico FB 1970 1980 1990 2000 Saint Kitts and Nevis FB 1960 1970 1980 1991 2001 Saint Lucia FB 1960 1980 1991 2001 Saint Pierre et Miquelon FB 1962 1974 1982 1999 Saint Vincent and the Grenadines FB 1960 1980 1991 Suriname NT 1964 2004 Trinidad and Tobago FB 1960 1970 1980 1990 2000 Turks and Caicos Islands FB 1960 1980 1990 United States Virgin Islands FB 1960 1970 1980 1990 2000 Uruguay FB 1963 1975 1985 1996 Venezuela FB 1961 1971 1981 1990 2001 *The census year was derived from splitting an aggregated census. a. FB is foreign born, NT is nationality, and ETH is ethnic group. FB(NT) means that the original data by nationality were amended and the resulting numbers are closer to foreign-born de�nition. Source: Authors’ calculations based on data described in text. ¨ zden, Parsons, Schiff, and Walmsley O 49 APPENDIX 2. LIST OF AGGREGATIONS T A B L E A 2 . List of Aggregations Aggregated region Master region Aggregated region Master region Aden Yemen Northern Ireland United Kingdom Alaska United States of Palmyra United States of America America Alboran and Perejil Spain Panama Canal Zone Panama Ascension Island Saint Helena Penang Malaysia Azores Portugal Pitcairn Island United Kingdom Bonaire Netherlands Antilles Providencia Island Colombia Born abroad of U.S. United States of Saint Croix United States Virgin parent(s) America Islands British Indian Ocean United Kingdom Saint Martin Netherlands Antilles Territory Canary Islands Spain Saint Thomas United States Virgin Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Islands Canton and Enderbury Kiribati San Andres Island Saint Pierre and Islands Miquelon Ceuta and/or Melilla Spain Sarawak Malaysia Channel Islands United Kingdom Scotland United Kingdom Channel Islands and the United Kingdom South Senegal Senegal Isle of Man Christmas Island Australia South Vietnam Vietnam Cocos (Keeling) Islands Australia South Yemen Yemen Curacao Netherlands Antilles Spanish Sahara Morocco Dubai United Arab Emirates Svalbard and Norway J. Mayen Islands East Germany Germany Terre Nova Canada Easter Island Chile Tristan de Cunha Saint Helena England United Kingdom Vatican Italy England and Wales United Kingdom Wake Island United States of America French India India Wales United Kingdom Galapagos Ecuador West Germany Germany Gaza Strip Occupied Palestinian Western New Guinea Indonesia Territory Germany (East Berlin) Germany Western Sahara Morocco Germany (unspeci�ed) Germany Zanzibar Tanzania Great Britain United Kingdom Hawaii United States of America Howland Island United States of America Isle of Man United Kingdom Jammu India Johnston Islands United States of America Kashmir India Kosovo Serbia and Montenegro (Continued ) 50 THE WORLD BANK ECONOMIC REVIEW TABLE A2. Continued Aggregated region Master region Aggregated region Master region Labuan Malaysia Madeira Portugal North Borneo Malaysia North Senegal Senegal North Vietnam Vietnam North Yemen Yemen Source: Authors’ calculations based on data described in text. APPENDIX 3. ADJUSTMENTS TO THE DATA This appendix describes the adjustments made to the data for the former Soviet Union and Germany. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Former Soviet Union Censuses for the Soviet Union for 1959, 1970, 1979, and 1989 were collected to address the data issues created by the dissolution of the Soviet Union. These censuses all use ethnicity to identify migrants. Crucially, for 1989, comparable country of birth data exist for all 15 republics. The censuses based on ethnicity document intra-Soviet migrants (Uzbeks in Turkmenistan, for example) and external nationalities (such as Afghans). In addition, there are miscellaneous Soviet nationalities (such as the Chuvash, Tatars, and Uyghurs), many of whose homelands span several Soviet republics/countries and who should therefore not be counted as international migrants since they were born on one side of the border or the other as opposed to moving across it. First, people of these miscellaneous nationalities were broadly aggregated to one or more of the 15 former Soviet republics on the basis of country by country research and a close inspection of the numbers over time. Similarly, external nationalities were assigned, with particular attention to determining whether these people were actually migrants. For example, people recorded as Germans will likely be ethnic Germans who migrated long before the census period examined in this study. Those recorded as Poles, however, are more likely to have been forcibly deported. Once the aggregations were completed, the ratios of foreign-born migrants to migrants de�ned by ethnicity in 1989 were calculated for people who were both born in one of the 15 former Soviet republics and resided there. These ratios were then applied to these republics/ countries in every census period before adding the “external� migrants. These corrections captured a large proportion of the most important migrants to and between the Soviet republics. This process adds many millions of migrants to the totals in the early decades and avoids the problem of a very large arti�cial jump in international migration between 1980 and 1990, after the dissolution of the Soviet Union. ¨ zden, Parsons, Schiff, and Walmsley O 51 Germany The 2005 German micro-census includes data on emigrants of German origin from Eastern Europe who arrived between 1944 and 1950 (referred to as expel- lees, Vertriebene) or between 19502005 (referred to as resettlers, Aussiedler). These data are recorded by year of birth and year of migration; country of birth is not recorded. As of 1950, there were 11.96 million expellees and 4.48 million resettlers residing in Germany. According to the data provided by the Max Planck Institute for the Study of Religious and Ethnic Diversity, 3.61 million were still in Germany as of 2005. Mortality data from the United Nations Population Division (United Nations 2010) on Germany for each decade and age group were used to calculate the number of migrants who would have been residing in Germany at the beginning of each decade from 1960 to 2000, taking into account migrants’ age and year of entry. After calcu- lating the total number residing in Germany in each decade, shares were esti- Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 mated by country of origin using the nationality shares from the 1950 data on expellees and post-1950 data on resettlers. The numbers of expellees and reset- tlers were then added to the existing totals. APPENDIX 4. PROPENSITY MEASURES This appendix presents the propensity measures used to disaggregate the 236 aggregate origin regions/countries identi�ed in the censuses. Let Mo,d,t denote the number of migrants from origin country o in destination country d in year t. These are the entries in the bilateral matrices that need to be completed. Now, instead of Mo,d,t, suppose a census in country d gives the number of migrants originating from region R (which includes country o), denoted as MR,d,t. The problem is to �nd an allocation rule (so,d,t) for estimating the bilat- eral stock from this aggregate amount. The allocation rule can be written as Mo,d,t ¼ so,d,t MR,d,t . One type of aggregation problem occurs in the case of migrants from Czechoslovakia, the Soviet Union, and Yugoslavia and their successor states. For example, in many cases, migrants are recorded from Czech Republic, Slovakia, and Czechoslovakia in the same year. Belgium’s 2001 reports 308 migrants from Czechoslovakia, 554 from the Czech Republic, and 412 from Slovakia. Presumably, migrants who left before the partition reported Czechoslovakia as their origin country, whereas most postpartition migrants reported the successor countries. In such cases, it is assumed that the distri- bution of migrants from these two countries was the same before and after the break-up of Czechoslovakia. Of the 308 migrants recorded as originating from Czechoslovakia, 177 migrants (308*[554/966]) were assigned to the Czech Republic and 131 (308*[412/966]) to Slovakia. In other cases of aggregated migrant stock data, migrant data from other decades were used as the basis for disaggregation. Migrants were allocated 52 THE WORLD BANK ECONOMIC REVIEW according to a relative propensity, which is averaged over time. This can be for- mally written as:  X  X 1 1 À � so;d;t ¼ so;d;k ¼ Mo;d;k =MR;d;k ð1Þ n k[K n k[K where K denotes the set of census years other than t for which bilateral data exist, and n is the number of such observations in set K. This propensity is simply the likelihood that a particular destination country will accept migrants from a speci�c origin country, relative to all the other countries comprising that aggregate origin region. For example, Australia records 29,311 migrants from the Soviet Union in 1966. This total needs to be disaggregated among the 15 successor countries in the master list. While the data for Australia cover Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 census material for each of the �ve census rounds, only the 2001 census pro- vides details for all 15 successor countries. According to the �rst method for allocating aggregate categories, the 2001 census is used to calculate the contri- bution of each of these countries towards the total. Those shares are then used to allocate the 29,311 migrants from the Soviet Union in 1966 among the con- stituent republics to yield the bilateral numbers for Australia (table A3). T A B L E A 3 . Allocation of Aggregate Origin Region by Migrant Shares over Time for Australia Origin country listed Total immigrants Share of Soviet Union Number of migrants in 2001 Australian to Australia in migration to Australia allocated in 1966 across census 2001 in 2001 (%) constituent countries Azerbaijan 145 0.3 93 Armenia 899 2.0 576 Belarus 1,041 2.3 667 Estonia 2,386 5.2 1,529 Georgia 310 0.7 199 Kazakhstan 438 1.0 281 Kyrgyzstan 101 0.2 65 Latvia 6,690 14.6 4,287 Lithuania 3,689 8.1 2,364 Moldova 483 1.1 309 Russian Federation 15,022 32.8 9,625 Tajikistan 41 0.1 26 Turkmenistan 26 0.1 17 Ukraine 14,062 30.7 9,010 Uzbekistan 412 0.9 264 Total Soviet Union 45,745 100 29,311 Source: Authors’ calculations based on data described in text. ¨ zden, Parsons, Schiff, and Walmsley O 53 In this simple example, only the data for 2001 are available. Where data are available for more than one census, the shares across all decades are averaged before estimating the bilateral numbers. In the absence of such data (disaggregated data for the same destination country in other census years), the world is disaggregated into destination sub- regions. Origin countries in the same subregion are then assumed to have a similar propensity over time to send migrants to a particular destination country in a subregion for which data are lacking as they do to other countries in that subregion. For example, assume that the census data for Morocco in a particular year include the origin category All West Africa but no individual data on migrants from Ghana and that there are no bilateral data on Ghanaian migrants in other Moroccan censuses. In this instance, migrants from Ghana are assumed to have a similar propensity to migrate to Morocco as they have to other countries in North Africa. Data from other countries in North Africa (Algeria, Egypt, Libya, and Tunisia) are then used to calculate the propensity Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 of Ghanaians—relative to migrants from other West African countries—to migrate to each country in North Africa. These propensity shares, which sum to one, can be applied to the All West Africa aggregate category from the Moroccan census to disaggregate it into the constituent West African countries. Equation 2 expresses this propensity measure:  XX 1 À � so;d;t ¼ Mo;g;k =MR;g;k ð2Þ nf k[K g[G In equation (2), G denotes the set of comparable destination countries (Algeria, Egypt, Libya, and Tunisia in the example above); R is the set of origin countries (All West Africa); n is the number of census years for which data exist; and f is the number of countries in region G. In short, this is the relative propensity of origin country o to send migrants to subregion G relative to other countries in its own region (R). Where appropriate data for the subregion cannot be found, the set of all countries in the world is used. APPENDIX 5. CALCULATING GENDER SPLITS When gender splits are missing, the preferred option is to divide the world into subregions. Then it is assumed that the gender ratio of an origin country’s emi- grant stock in a speci�c decade is the same for each destination country in that subregion. The missing gender ratio in an origin country’s emigrant stock can then be calculated using data disaggregated by gender from all destinations in the same subregion as the destination country for which data are lacking. Using the same notation as in the previous section, assume that Mo,d,t is the aggregate migrant stock from origin country o to destination country d in year t and that Wo,d,t is the female migrant stock for the same origin-destination 54 THE WORLD BANK ECONOMIC REVIEW T A B L E A 4 . Calculation of Sex Ratios Based on Concurrent Subregional Shares. Number of male Number of female Destination country migrants in 1990 from migrants in 1990 from Males Females in Scandinavia Uruguay Uruguay (%) (%) Denmark 92 90 51 49 Finland 11 21 39 66 Norway 67 78 46 54 Average across subregion 47 53 Source: Authors’ calculations based on data described in text. pair in the same year t. The ratio of female migrants to male migrants is denoted as go,d,t, which is given by go,d,t ¼ Wo,d,t/Mo,d,t. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 For example, imagine that in a given decade, the gender splits of emigrants from Uruguay in Scandinavian countries are known, except for Sweden. In this situation, it is assumed that the ratio of female migrants to male migrants from Uruguay to Sweden is the ratio of female migrants to male for all of the other Scandinavian countries (Denmark, Finland, Norway) in that decade. Formally, this can be stated as go;d;t ¼ Wo;G;t =Mo;G;t ð3Þ where G is the destination region (the Scandinavian countries except Sweden), o is the origin country (Uruguay), and d is the destination country (Sweden). Once this proportion go, d, t is calculated, it can be multiplied by the total number of migrants Mo, d, t to Sweden to calculate the number of female migrants. There is considerable variation in the balance between male and female migration from Uruguay to Scandinavian countries other than Sweden (Denmark, Finland, Norway) during the 1990 census round (table A4). On average, however, 47 percent of Uruguayan migrants are men and 53 percent are women. In the 1990 census, Sweden records 2,640 migrants as originating from Uruguay. Then 1,390 (0.53*2,640) of these migrants are women and 1,250 (0.47*2,640) are men. These calculations based on concurrent shares can be calculated only if data disaggregated by gender exist for all other countries in the destination subre- gion. If not, the world is divided into destination subregions, and gender splits are calculated based on regional shares over time. Continuing from the pre- vious example, assume the data for Denmark, Finland, and Norway are una- vailable in 1990, so that the gender split for Uruguayan migrants in Sweden cannot be calculated based on Scandinavian data for 1990. In this case, the data for Scandinavia across all other decades are used to calculate the average ¨ zden, Parsons, Schiff, and Walmsley O 55 ratios of female migrants to total migrants over time. This can be written for- mally as:  X  X 1 1 À � go;d;t ¼ go;G;t ¼ Wo;G;k =Mo;G;k ð4Þ n k[K n k[K The expression in brackets (Wo,G,k/Mo,G,k) is the ratio of female migrants to male migrants from origin o to all destination countries in the destination subregion G, across all decades k, for which data exist. Of course, complete data are not available for the current decade t since, were that the case, equation (4) would be preferred. Again, once calculated, this share is multi- plied by the total number of migrants to determine the number of female migrants. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 REFERENCES Beine, M., F. Docquier, and H. Rapoport. 2007. “Measuring International Skilled Migration: A New Database Controlling for Age of Entry.� The World Bank Economic Review 21 (2): 249– 54. Bhargava, A., and F. Docquier. 2008. “HIV Pandemic, Medical Brain Drain, and Economic Development in Sub-Saharan Africa.� The World Bank Economic Review 22 (2): 345– 66. Bilsborrow, R.E., G. Hugo, A.S. Oberai, and H. Zlotnik. 1997. International Migration Statistics, Guidelines for Improving Data Collection Systems. Geneva: International Labour Of�ce. Docquier, F., and A. Marfouk. 2006. “International Migration by Educational Attainment (1990-2000): Release 1.1.� In International Migration, Remittances and Development, ed. C. O¨ zdenM. Schiff. New York: Palgrave Macmillan. Docquier, F., B.L. Lowell, and A. Marfouk. 2009. “A Gendered Assessment of Highly Skilled Emigration.� Population and Development Review 35 (2): 297–321. Harrison, A., T. Britton, and A. Swanson. 2003. Working Abroad: the Bene�ts from Nationals Working in Other Economies. Paris: Organisation for Economic Cooperation and Development. Integrated Public Use Microdata Series. 2008. International: Version 6.0 [Machine-readable database]. Minnesota Population Center, University of Minnesota, Minneapolis, MN. Mayda, A. M. 2007. “International Migration: A Panel Data Analysis Of The Determinants Of Bilateral Flows.� CReAM Discussion Paper Series 0707, Centre for Research and Analysis of Migration, Department of Economics, University College London. OECD (Organisation for Economic Co-operation and Development). 2002. Trends in International Migration database. SOPEMI 2002 Edition. Paris: OECD. OECD (Organisation for Economic Co-operation and Development). 2008. A Pro�le of Immigrant Populations in the 21st Century: Data from OECD Countries. Paris: OECD. Parsons, C.R., R. Skeldon, T.L. Walmsley, and L.A. Winters. 2007. “Quantifying the International Bilateral Movements of Migrants.� In International Migration, Economic Development and Policy, ed. C. O¨ zden, and M. Schiff. New York: Palgrave Macmillan. Ratha, D., and W. Shaw. 2007. “South-South Migration and Remittances.� World Bank Working Paper 102, World Bank, Washington, DC. United Nations Statistics Division. 1998. Recommendations on Statistics of International Migration Revision 1. New York: United Nations. United Nations, Department of Economic and Social Affairs, Population Division. 2008. United Nations Global Migration Database. New York: United Nations. http://esa.un.org/unmigration 56 THE WORLD BANK ECONOMIC REVIEW ———. 2006. Trends in Total Migrant Stock 1960– 2000, 2005 Revision. Database. POP/DB/MIG/ Rev.2005/Doc. New York: United Nations. ———. 2009. Trends in International Migrant Stock: The 2008 Revision. Database. POP/DB/MIG/ Stock/Rev.2008. New York: United Nations. http://www.un.org/esa/population/. ———. 2010. “World Population Prospects: The 2009 Revision, Highlights.� Working Paper. ESA/P/ WP.210. New York: United Nations. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Immigration Policies and the Ecuadorian Exodus Simone Bertoli, Jesu ´ ndez-Huertas Moraga, ´ s Ferna and Francesc Ortega Ecuador recently experienced an unprecedented wave of emigration following the severe economic crisis of the late 1990s. Individual-level data for Ecuador and its two main migration destinations, Spain and the United States, are used to examine the size and skill composition of these migration flows and the role of wage differences in accounting for these features. Estimations of earnings regressions for Ecuadorians in all three countries show substantially larger income gains following migration to the United States than to Spain, with the wage differential increasing with migrants’ edu- cation level. While this �nding can account for the pattern of positive sorting in edu- cation toward the United States, it fails to explain why most Ecuadorians opted for Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Spain. The explanation for this preference appears to lie in Spain’s visa waiver program for Ecuadorians. When the program was abruptly terminated, monthly inflows of Ecuadorians to Spain declined immediately. JEL codes: O15, J61, D31 Following the seminal contributions of Roy (1951), Sjaastad (1962), and Borjas (1987), most studies of international migration have focused on how wage differentials shape the decision on whether and where to migrate. There is also consensus that many nonwage factors are important: demographic changes in Simone Bertoli (simone.bertoli@eui.eu) is a Jean Monnet Fellow at the Robert Schuman Centre for Advanced Studies, European University Institute. Jesu ´ s Ferna´ ndez-Huertas Moraga (corresponding author; jesus.fernandez@iae.csic.es) is a researcher at the Institute for Economic Analysis of the Spanish Council for Scienti�c Research, Barcelona. Francesc Ortega (fortega@qc.cuny.edu) is a professor at Queens College of the City University of New York. The authors are grateful to the journal editor and to three anonymous referees for careful comments and suggestions and to Gordon Hanson, Hillel Rapoport, participants at the World Bank–French Development Agency Second International Conference on Migration and Development and at the Third Insights on Immigration and Development Economics (INSIDE) Workshop. They also thank Lı ´dia Brun and Feray Koc ¸ for helpful research assistance. The article is part of the INSIDE research projects. Simone Bertoli received �nancial support from the RBNE03YT7Z project, funded by the Italian Ministry for Education, University and Research. Jesu ´ ndez-Huertas Moraga received �nancial support from the ´ s Ferna ECO2008-04785 project, funded by the Spanish Ministry for Science and Innovation. Jesu ´ s Ferna´ndez- Huertas Moraga works under a JAE–Doc contract from the Junta de Ampliacio ´ n de Estudios Program, co�nanced by the European Social Fund; he also acknowledges the support of the Barcelona Graduate School of Economics Research Network and of the government of Catalonia. The usual disclaimers apply. A supplemental appendix to this article is available at http://wber.oxfordjournals.org/. THE WORLD BANK ECONOMIC REVIEW, VOL. 25, NO. 1, pp. 57 – 76 doi:10.1093/wber/lhr004 Advance Access Publication March 18, 2011 # The Author 2011. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 57 58 THE WORLD BANK ECONOMIC REVIEW the origin countries (Hanson and McIntosh 2010a), cultural and linguistic proximity (Grogger and Hanson forthcoming), ethnic networks (McKenzie and Rapoport 2010; Beine, Docquier, and O ¨ zden forthcoming), and immigration policies in the main host countries (Mayda 2010; Ortega and Peri 2009). Identifying the impact of immigration policies on migration decisions is often problematic because immigration policies are multifaceted; tracking their differences over time and across countries is a challenge. This article focuses on the massive Ecuadorian migration of the late 1990s and early 2000s1 by isolat- ing the effects of a change in one policy dimension: the introduction of a visa requirement for visitors from Ecuador to Spain in August 2003. Individual-level data from comparable sources in Ecuador, the United States and Spain, the two main destination countries, were assembled to identify the composition and distribution of the recent Ecuadorian migration. The information was used to assess to what extent these features can be explained by wage and nonwage factors. The destination of Ecuadorian migrants was examined by education level Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 and gender. More women than men and more people without a college degree2 emigrated to Spain, while more college graduates opted for the United States. Individual observations on labor earnings for Ecuadorians were used to run country-speci�c Mincer regressions and to estimate the income gain associated with migration to the two main destinations. The estimated differences in labor earnings across countries and levels of schooling are consistent with the higher average level of education of migrants to the United States. Still, wage factors are starkly at odds with the relative scale of migration to the two destination countries. The much larger income gains associated with migration to the United States do not help explain why most Ecuadorians who left in the aftermath of the crisis opted for Spain. This choice is all the more puzzling considering that precrisis Ecuadorian migration networks were denser in the United States than in Spain,3 yet the postcrisis migration was character- ized by a large shift “from New York to Madrid� (Jokisch 2001). The litera- ture on networks and migration suggests that the denser U.S. networks should have contributed to an increase in the scale of Ecuadorian migration to the United States relative to Spain (Beine, Docquier, and O ¨ zden forthcoming; 4 McKenzie and Rapoport 2010). This puzzle can be explained by a key difference in the immigration policies of the United States and Spain. Spain had introduced a visa waiver program for 1. See Beckerman and Solimano (2002), Ja ´ come (2004), Larrea (2004), and Laeven and Valencia (2008) for an analysis of the causes and economic consequences of the late 1990s crisis. 2. College degree or college graduate is de�ned as a person with at least four years of college education. 3. Before 1999, there were 272,000 Ecuadorian-born individuals in the United States (U.S. Census Bureau 2000) but just 76,000 in Spain (INE 2001). 4. Figure S.1 and table S.1 and the related discussion in the supplemental appendix to this article (available at http://wber.oxfordjournals.org/) provide some suggestive evidence that this was the case; observe that networks could have also contributed to reduce the level of education of Ecuadorian migrants to the United States, as the empirical results in Bertoli (forthcoming) show. ´ ndez-Huertas Moraga, and Ortega Bertoli, Ferna 59 Ecuadorians since 1963. Its influence on the distribution of Ecuadorian migrants across the two main destinations can be gauged by what happened when the program was terminated in August 2003, at a time when other rel- evant facets of immigration policies in the two destination countries remained unchanged: monthly inflows of Ecuadorians into Spain fell sharply. The article is structured as follows. Section I describes the timing of the Ecuadorian exodus. Section II presents descriptive statistics. Section III analyzes the skill composition of migration flows. Section IV reports Mincerian regressions and attempts to reconcile the implied wages with the data on migration flows. Section V discusses the most relevant differences in immigra- tion policies between the United States and Spain. Finally, section VI discusses some implications of the �ndings. I . D ATA S O U R C E S AND TIMING OF MI G R AT I O N Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Ecuador experienced a severe economic and �nancial crisis in the second half of the 1990s, prompting a large wave of international migration. Information on this migration episode comes from three sources: the December 2005 round of the National Survey of Employment and Unemployment in Urban and Rural Areas (ENEMDU; INEC 2005) for Ecuador, the 2007 American Community Survey (ACS; U.S. Census Bureau 2007) for the United States, and the 2007 National Immigrant Survey (ENI; INE 2007) for Spain.5 These data sources provide comparable individual-level information on Ecuadorians residing in the three countries on age, year of migration, gender, education, marital status, employment status, sector of occupation, and pretax labor earnings.6 The three datasets contain information on 73,758 individuals residing in Ecuador and 2,030 who migrated to Spain or to the United States between 1999 and 2005. Figure 1 plots the distribution of the Ecuadorian migrants in the ACS 2007 and in the ENI 2007 by year of arrival. The time pro�le of migration flows from Ecuador to the two destinations is very similar, with a surge in flows to the United States and Spain around 2000, in the aftermath of the economic crisis. Though the timing is similar, the scale differs substantially. Some 137,148 Ecuadorians emigrated to the United States during 1999–2005,7 and some 318,243 emigrated to Spain—more than twice as many. Ecuadorian data 5. The ENEMDU 2005 is a nationally representative labor market survey covering a sample of 73,758 people (INEC 2005). The ACS 2007 sample covers approximately 2.5 percent of the resident population in the United States (U.S. Census Bureau 2007; Ruggles and others 2008). The ENI 2007 is a nationally representative survey of the foreign-born population in Spain, with a sample size of 15,500 (INE 2007). 6. Other relevant variables, such as province of residence in Ecuador or English language pro�ciency, are not available on a comparable basis in the three datasets; some of these variables were used to perform robustness checks, described in the supplemental appendix. 7. The period covers migration episodes that occurred at least two years before the ENI 2007 and the ACS 2007, as surveys in destination countries might be unable to adequately enumerate recently arrived migrants (Hanson 2006). 60 THE WORLD BANK ECONOMIC REVIEW F I G U R E 1. Arrivals of Ecuadorians to the United States and Spain, 1991–2006 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Note: The �gure plots the distribution of migrants by their year of �rst arrival at destination; the two vertical lines delimit the reference period of the analysis. Source: Authors’ calculations using data from U.S. Census Bureau (2007) and INE (2007). sources show a similar picture of the scale and timing of migration to the two main destinations (see �gure S.2 in the supplemental appendix). Two issues arise concerning the representativeness of the sample. First, the sample does not account for temporary migrants who had returned home by the time of the survey. However, the size of the return flow is very small. Based on ENEMDU 2005 data, 9,890 people returned to Ecuador from Spain or the United States between 1999 and 2005, a very small number compared with the roughly half a million Ecuadorian migrants to the United States and Spain. Thus, any bias due to return migration is likely to be very small. Second, the sample enumerates most Ecuadorians who moved to the United States or Spain over the 1999–2005 period, irrespective of their legal status at destination. The number of Ecuadorians who entered the United States between 1999 and 2005 according to the 2007 ACS is very close to the sum of the number of Ecuadorians who became legal permanent residents and the best available esti- mate of the size of illegal flows of Ecuadorians over the same period.8 Spain 8. Hoefer, Rytina, and Baker (2008) estimate that 10,000 Ecuadorians entered the United States illegally every year over 2000–06. The 1999– 2005 issues of the Yearbook of Immigration Statistics (U.S. Department of Homeland Security various years) reveal that 64,034 Ecuadorians became permanent residents over �scal years 1999–2005. Adding this �gure—which also includes adjustment of status—to the estimated undocumented inflow yields approximately 135,000, which is reassuringly close to the 137,148 Ecuadorian migrants recorded by the ACS 2007 over the period. ´ ndez-Huertas Moraga, and Ortega Bertoli, Ferna 61 extended three amnesties to illegal immigrants (in 2000, 2001 and 2005), so most Ecuadorians had a legal residence permit at the time of the ENI 2007. II. SAMPLE SELECTION AND D E S C R I P T I V E S TAT I S T I C S Since the aim of the analysis was to understand the determinants of migration decisions by prime working age Ecuadorians, the sample was restricted to people born during 1949–82 who were 16–49 years old and living in Ecuador in 1998, at the onset of the economic crisis, and who then left Ecuador between 1999 and 2005 or stayed in the country. Some 205 individuals who reported past international migration experience were excluded, so that the subsample of stayers includes only those who had never migrated. These sample selection criteria deliver a sample of 509 migrants to the United States, 915 migrants to Spain, and 27,917 stayers. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 The distribution of migrants in the selected sample between the two destina- tion countries is similar to that depicted in �gure 1: the migration flow to Spain was almost three times as large as the flow to the United States, with some differences by education and gender. The ratio of migrants to Spain to migrants to the United States was 3.2 for non-college graduates compared with 1.9 for college graduates, and 3.2 for women compared with 2.8 for men (see table S.2 in the supplemental appendix). These �gures suggest that the incentives and the ability to migrate to the United States differed by education and gender. Migrants to the two destinations were similar in age and younger than stayers (table 1). They had been residing there an average of 6 years at the time of the surveys. Male migrants to Spain were on average less educated (8 percent had a college degree) than were stayers and migrants to the United States (14 percent each). Female migrants to the United States were more highly educated (22 percent had a college degree) than were stayers (13 percent).9 Thus, for both genders, Ecuadorians who migrated to the United States were more educated than those who migrated to Spain. They had com- pleted 1.3 more years of schooling, and the share of college graduates was 6 percentage points higher. The employment rate for Ecuadorian men—for both those with a college degree and those without—is the same in the United States and Spain, suggesting that this played a limited role in influencing prospective male migrants’ destination choice. For women, the employment rate is substantially higher in Spain than in the United States, which probably reflects the fact that tied movers (individuals who follow a migrating household member) were a greater share of female Ecuadorian migrants in the United States. Ecuadorian migration to the United States had traditionally been male dominated, so in 9. The same picture emerges when the sample is restricted to individuals born in 1949– 73, who had already completed their education by the onset of the late-1990s crisis. 62 T A B L E 1 . Descriptive Statistics on Ecuadorians in Ecuador, the United States, and Spain Ecuador United States Spain Statistic Mean Standard deviation Mean Standard deviation Mean Standard deviation Men Age at migration 37.97 9.73 28.78 8.42 28.66 7.91 Years since migration 0.00 0.00 5.73 1.99 6.25 1.38 College graduatea 0.14 0.35 0.14 0.35 0.08 0.27 Years of education 9.23 4.86 10.86 4.25 9.58 4.02 Employment rate 0.95 0.22 0.90 0.30 0.90 0.30 College graduate 0.93 0.26 0.92 0.27 0.92 0.28 Non-college graduate 0.95 0.22 0.90 0.30 0.90 0.30 Labor income (U.S. dollars) 3,829 5,447 26,896 20,344 15,979 4,214 College graduate 8,793 9,817 43,219 35,581 16,308 5,743 THE WORLD BANK ECONOMIC REVIEW Non-college graduate 3,023 3,766 23,991 14,368 15,951 4,054 Women Share of total 0.52 0.50 0.46 0.50 0.49 0.50 Age at migration 37.97 9.49 30.22 8.54 28.91 7.54 Years since migration 0.00 0.00 5.68 1.84 6.07 1.43 College graduatea 0.13 0.34 0.22 0.41 0.15 0.36 Years of education 8.91 4.90 11.55 4.29 10.29 4.09 Employment rate 0.60 0.49 0.63 0.48 0.81 0.39 College graduate 0.82 0.39 0.63 0.49 0.84 0.37 Non-college graduate 0.57 0.50 0.63 0.48 0.81 0.39 Labor income (U.S. dollars) 2,883 3,775 18,189 12,718 10,767 3,317 College graduate 5,631 5,692 21,314 10,492 10,891 3,081 Non-college graduate 2,199 2,718 17,445 13,084 10,744 3,358 Number of observations 27,917 509 915 a. De�ned as at least four years of college. Source: Authors’ analysis based on data from INEC (2005), U.S. Census Bureau (2007), and INE (2007). Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 ´ ndez-Huertas Moraga, and Ortega Bertoli, Ferna 63 this postcrisis migration wave, women were more likely than men to be able to take advantage of the family reuni�cation provisions of U.S. immigration law ´ nchez 2004). Conversely, women made up most of the early migrants to (Sa Spain, where they were often employed as domestics and in elderly care (Jokisch and Pribilsky 2002). Pretax labor earnings for Ecuadorians in the United States are well above those in Spain and Ecuador for both men and women and for all levels of edu- cation (see table 1). Average annual labor earnings for an Ecuadorian male college graduate in the United States are $27,000 more than in Spain; for non- college graduates the differences is $8,000.10 These data were collected in 2007 for migrants, and the U.S. dollar depreciated substantially over the seven-year reference period, implying that the data underestimate the difference in earn- ings at the time when most migrants decided to leave Ecuador.11 The three countries also differ in the variability of labor earnings. Earnings dispersion is greatest for Ecuador, while earnings appear to be compressed Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 around the mean for Ecuadorians in Spain (see table 1).12 III. SELECTION AND SORTING IN E D U CAT I O N The descriptive statistics reported above suggest that the average Ecuadorian in the United States was substantially more educated than the average Ecuadorian in Spain. This section provides a more rigorous comparison, controlling for individual differences in observable characteristics, such as age and gender. Migrants are said to be positively selected in education if their average edu- cational attainment is higher than that of stayers and negatively selected if it is lower (Borjas 1999). And migrants to one destination can be said to be posi- tively sorted if their average education is higher than that of migrants to other destinations and negatively sorted if it is lower (Grogger and Hanson forthcoming). To assess the degree of selection and sorting in education, two probit models are estimated for the probability of being a college graduate for a sample that includes stayers and migrants (selection) or migrants to both 10. The labor earnings �gures in table 1 are adjusted for inflation but not for differences in purchasing power parity, because of the large size of remittances, both in absolute terms and relative to migrants’ earnings. As a result, the appropriate price index is some unknown combination of the price level in Ecuador and in the destination country. At any rate, the difference in the price levels in the United States and Spain is very small. Taking the United States as the base (100 in 2007), Spain’s cost of living was 95.5 (World Bank 2008). Ecuador’s cost of living was 42.2 in the same year. 11. The exchange rate stood at $0.92 per euro in 2000, when postcrisis migration reached its peak, rising to $1.37 per euro in 2007 (World Bank 2008), when the labor earning �gures were collected (see also �gure S.3 in the supplemental appendix). 12. The supplemental appendix contains additional descriptive statistics that are helpful in understanding the likely labor market effects of Ecuadorian immigration in the United States and Spain (see table S.3). 64 THE WORLD BANK ECONOMIC REVIEW destinations (sorting). Several alternative speci�cations are considered, varying in the control variables included (table 2). The top two panels in table 2 present estimates for selection. For men, there is clear evidence of negative selection in education for migrants to Spain (com- pared with stayers), as evidenced by the negative and signi�cant coef�cient on the dummy variable for migration to Spain across all speci�cations. The esti- mated coef�cient on the U.S. dummy variable is positive but not signi�cant. For women, there is signi�cant positive selection for migrants to the United States and a much smaller and not signi�cant coef�cient for migrants to Spain. This �nding is robust to controlling for year of birth, marital status, and Ecuadorian province of origin (not available for the U.S. data). The two bottom panels of table 2 present estimates for sorting of migration by education. The main explanatory variable takes a value of one if the migrant opted for Spain and zero if for the United States. The estimated coef�- cient for the dummy variable for Spain is negative and highly signi�cant across Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 all speci�cations for both genders, meaning that both male and female Ecuadorian migrants to Spain were negatively sorted in education relative to migrants to the United States. Ideally, the estimation would control for some measure of networks, but the ACS 2007 data do not enable linking Ecuadorian immigrants to their commu- nities of origin. Still, it is highly unlikely that networks can account for the observed pattern of negative sorting in education to Spain, as their greater density in the United States should have contributed to the opposite pattern to that found in the data.13 I V. E A R N I N G S AND THE DECISION TO MI G R AT E With individual-level data, Mincer regressions can be run using observed earnings for Ecuador, Spain, and the United States to estimate the returns to education for Ecuadorians in each location, without having to rely on extrapolations from income �gures for the general population, as in most empirical studies (Belot and Hatton 2008; Grogger and Hanson forthcoming; Ortega and Peri 2009). The dependent variable in the Mincer equations—which are gender- and country- speci�c—is the log of pretax annual earnings in 2005 dollars, and the regressions are estimated on the subsample of employed individuals.14 The regressions 13. Bertoli (forthcoming) �nds that the greater the density of migration networks (measured as the share of households in each Ecuadorian county that had a member in the United States before the late 1990s crisis), the lower the average level of schooling of migrants that opted for the United States in the aftermath of the crisis. 14. Following Heckman (1979), the robustness of the estimates was tested controlling for selection into employment and adding household size among the regressors in the �rst stage. This had little influence on estimated returns to education for men, given the high rates of employment in the three countries (see table 1); the impact is larger for women, but it does not alter the differences across countries that emerge in table 3 (see table S.4 in the supplemental appendix). T A B L E 2 . Selection and Sorting Regressions, Probit Model Speci�cation Variable (1) (2) (3) (4) (5) (6) Birth cohorts No Yes No Yes Yes Yes Year of birth dummy variable No No Yes No No No Marital status No No No Yes Yes Yes Province dummy variable No No No No No Yes Selection in education Men born in 1949– 82 U.S. migrant dummy variable –0.004 0.057 0.055 0.009 (0.102) (0.103) (0.103) (0.103) Spain migrant dummy variable –0.355 – 0.301 – 0.313 – 0.312 – 0.279 – 0.282 (0.101)*** (0.101)*** (0.101)*** (0.103)*** (0.105)*** (0.107)*** Number of observations 13,991 13,991 13,991 13,991 13,781 13,781 Women born in 1949– 82 U.S. migrant dummy variable 0.340 0.332 0.341 0.322 (0.102)*** (0.103)*** (0.104)*** (0.103)*** Bertoli, Ferna Spain migrant dummy variable 0.095 0.069 0.067 0.067 0.071 0.063 (0.090) (0.091) (0.091) (0.091) (0.086) (0.086) Number of observations 15,350 15,350 15,350 15,350 15,157 15,157 Sorting in education Male migrants born in 1949– 82 Dummy Spain migrant dummy variable –0.35 – 0.383 – 0.454 – 0.404 (0.141)** (0.145)** (0.149)*** (0.143)*** Number of observations 688 688 688 688 Female migrants born in 1949– 82 Dummy Spain migrant dummy variable –0.245 – 0.292 – 0.292 – 0.254 (0.134)* (0.140)** (0.140)** (0.140)* Number of observations 736 736 736 736 ´ ndez-Huertas Moraga, and Ortega ***Signi�cant at the 1 percent con�dence level; ** signi�cant at the 5 percent con�dence level; * signi�cant at the 10 percent con�dence level. Note: Numbers in parentheses are standard errors. The dependent variable is having a college degree (completing at least four years of college). 65 Source: Authors’ calculations using data from INEC (2005), U.S. Census Bureau (2007), and INE (2007). Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 66 THE WORLD BANK ECONOMIC REVIEW include as explanatory variables a proxy for potential labor market experience and its square, marital status, and a measure of educational attainment (either a dummy variable for having a college degree or number of years of schooling). Potential labor market experience is de�ned as age minus the age at which edu- cation was completed, following Mincer (1974).15 For Spain and the United States, years since migration are included as a measure of labor market experience at destination. The �ndings are in line with the descriptive statistics in table 1. First, there is a high college wage premium in Ecuador (98 percent for men and a 107 percent for women) and in the United States (45 percent for men and 37 percent for women).16 In contrast, college-educated Ecuadorians in Spain earned virtually the same as non-college-educated ones. That is, the earnings pro�le for Ecuadorians in Spain appears to be flat across education levels for both men and women.17 The differences in the estimated college premia across the two destination Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 countries are due neither to differences in time elapsed since migration, con- trolled for in speci�cation 2 and 3, nor to differences in the legal status of the Ecuadorian migrants in the two countries. In 2007, the share of Ecuadorian legal residents was 60 percent in the United States18 and 91 percent in Spain.19 Therefore, accounting for legal status—for which individual-level data are not available for the United States—would likely result in an even larger gap in college wage premia between the two destinations. Speci�cation 3 includes years of schooling as a measure of education: for men, the estimated return to an additional year of schooling is 9.8 percent in 15. This is de�ned as the number of years of schooling plus 6. Since it is reasonable to assume that child labor experience does not increase adult wages, potential experience before the age of 16 is not counted. 16. The ACS 2007 provides information on self-reported fluency in English for immigrants; differential English fluency across education groups is likely to influence the observed college wage premium for Ecuadorians. In the sample, 20 percent of non-college graduates do not speak any English compared with less than 1 percent of college graduates. Once controls are included for English pro�ciency in the Mincer equation for the United States, the estimated college premium for men falls from 45 percent to 34 percent, though the difference is not statistically signi�cant (see table S.4 in the supplemental appendix). 17. The low R2 in the Mincer regressions for Spain can be related to the limited dispersion in earnings among Ecuadorian migrants to Spain documented in table 1. This reflects the extreme wage compression in Spain’s labor market for recent migrants, which is due mainly to the highly centralized wage bargaining. In addition, Ecuadorians working in Spain were heavily concentrated in a few occupations and sectors (mainly construction and household services; see table S.3 in the supplemental appendix). 18. The ACS 2007 reports that 403,643 Ecuadorian-born people who were residing in the United States as of January 1, 2007; for the same date, Hoefer, Rytina, and Baker (2008) report that an estimated 160,000 Ecuadorians were residing illegally in the country, putting the share of legal migrants at 60.4 percent. The share would be lower if only postcrisis migrants were considered. 19. Spain’s Local Population Registry recorded 434,673 Ecuadorians as of January 1, 2007. Of these, 376,233 had legal residence permits and 19,345 were naturalized, putting the share of Ecuadorian-born individuals residing legally in the Spain at 91 percent. ´ ndez-Huertas Moraga, and Ortega Bertoli, Ferna 67 Ecuador, 3.7 percent in the United States, and 0.7 percent in Spain, with the last �gure not being signi�cant. For both destination countries, the estimated rate of return is signi�cantly lower than the corresponding rate of return for the general population, reinforcing the argument that relying on countrywide �gures to gauge income gains from migration can be misleading. Mincer regressions estimated on the ACS 2007 for the United States and the 2006 Wage Structure Survey for Spain (INE 2006) show an 11.6 percent return for men in the United States and 5.4 percent in Spain.20 The Mincer regressions provide a basis for gauging the income gains from migration provided that the non-random selection in unobservables across the three countries does not signi�cantly bias the returns to observable character- ´ ndez-Huertas Moraga, and Ortega (2010) adopt the semi- istics. Bertoli, Ferna parametric approach proposed by Dahl (2002) to correct for selection in unobservables when predicting counterfactual earnings for Ecuadorians in the three countries focused on here; their results suggest negligible selection bias.21 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Table 4 displays the predicted average annual earnings based on speci�cation 1 in table 3, by gender and level of education, for Ecuadorians in each of the three countries. College graduates enjoyed a larger earnings gain from migrating to the United States (around $35,000 annually) than did non-college graduates (around $21,000; table 4). Migration to Spain entailed larger expected gains in earnings for non graduates (around $11,000 annually) than for graduates (about $7,000). What are the implications of these estimates for expected earnings for the scale, selection, and sorting of immigrants across destinations? First, wage differences by themselves are unable to account for the differences in the scale of migration to the United States and Spain, since most Ecuadorians migrated to Spain, the lower earnings destination. This implies that other factors must have played a key role. The education composition of migration across destinations is considered next. The �ndings in the previous section on selection in education were incon- clusive, with two signi�cant coef�cients (male migration to Spain and female migration to the United States) and two non-signi�cant ones.22 However, the comparison of the average educational attainment of migrants to the United 20. These two regressions were estimated for the same set of controls as the results reported in table 3 (except for marital status in Spain, which is not available in the 2006 Wage Structure Survey) and for individuals born in 1949–82. The rate of return for women is 13.4 percent in the United States and 5.6 percent in Spain (see table S.5 in the supplemental appendix). 21. Internal migrants in Ecuador were compared with Ecuadorian migrants abroad to further address the concern of nonrandom selection in unobservables. Descriptive statistics show that internal and international migration flows are similar in gender and education composition (see table S.6 in the supplemental appendix); Mincer regressions estimated separately for stayers and internal migrants in Ecuador show no signi�cant differences in the returns to schooling (see table S.7 in the supplemental appendix), which is reassuring about the limited influence on wages exerted by a possible nonrandom selection in unobservables. 22. The signi�cant results on selection are consistent with a linear utility speci�cation of the Roy model, as in Rosenzweig (2007) and Grogger and Hanson (forthcoming). T A B L E 3 . Determinants of Labor Earnings in Ecuador, the United States, and Spain Ecuador United States Spain 68 Variable (1) (3) (1) (2) (3) (1) (2) (3) Men born in 1949 –82 Years of schooling 0.098 0.037 0.007 (0.002)*** (0.012)*** (0.004) College graduatea 0.976 0.449 0.446 0.006 –0.000 (0.033)*** (0.152)*** (0.151)*** (0.065) (0.065) Experience 0.022 0.028 – 0.008 – 0.005 0.001 0.004 0.003 0.004 (0.004)*** (0.004)*** (0.023) (0.023) (0.023) (0.010) (0.010) (0.010) Experience, squared – 0.001 – 0.001 0.000 0.000 – 0.000 – 0.000 –0.000 –0.000 (0.000)*** (0.000)*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Years since migration 0.064 0.059 0.020 0.018 (0.022)*** (0.023)** (0.011)* (0.011)* Marital status 0.292 0.217 – 0.002 – 0.019 0.027 – 0.013 –0.012 –0.008 (0.020)*** (0.020)*** (0.105 (0.104) (0.105) (0.033) (0.033) (0.033) R2 0.18 0.28 0.07 0.13 0.11 0.004 0.02 0.03 THE WORLD BANK ECONOMIC REVIEW Number of observations 11,985 11,985 196 196 196 386 386 386 Women born in 1949 –82 Years of schooling 0.122 0.034 0.005 (0.003)*** (0.023) (0.005) College graduatea 1.067 0.371 0.340 0.029 0.028 (0.035)*** (0.142)*** (0.145)** (0.051) (0.051) Experience 0.003 0.009 0.008 0.005 0.011 0.002 0.000 0.001 (0.006) (0.006) (0.030) (0.029) (0.028) (0.014) (0.015) (0.015) Experience, squared – 0.0001 0.000 – 0.0003 – 0.0003 – 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.001) (0.001) (0.001) (0.000) (0.000) (0.000) Years since migration 0.044 0.042 0.007 0.005 (0.037) (0.039) (0.015) (0.015) Marital status 0.105 0.074 – 0.176 – 0.166 – 0.149 – 0.094 –0.096 –0.097 (0.030)*** (0.028)*** (0.142) (0.143) (0.142) (0.043)** (0.041)** (0.041)** R2 0.18 0.31 0.05 0.06 0.06 0.04 0.04 0.04 Number of observations 7,055 7,055 139 139 139 371 371 371 ***Signi�cant at the 1 percent con�dence level; ** signi�cant at the 5 percent con�dence level; * signi�cant at the 10 percent con�dence level. Note: Numbers in parentheses are standard errors. The dependent variable is the log of yearly pretax labor earnings. a. De�ned as having at least four years of college. Source: Authors’ analysis based on data from INEC (2005), U.S. Census Bureau (2007), and INE (2007). Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 ´ ndez-Huertas Moraga, and Ortega Bertoli, Ferna 69 T A B L E 4 . Predicted Earnings Ecuador United States Spain Variable Mean Standard error Mean Standard error Mean Standard error Men born in 1949– 82 College graduatea 6,066 210 40,976 3,569 13,403 911 Non-college graduate 2,164 41 23,868 1,313 13,181 475 Women born in 1949– 82 College graduatea 4,175 161 28,593 2,771 9,074 540 Non-college graduate 1,400 41 15,847 1,155 9,036 437 Note: Predictions are based on speci�cation (1) in table 3. a. De�ned as having at least four years of college. Source: Authors’ analysis based on data from INEC (2005), U.S. Census Bureau (2007), and INE (2007). Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 States and to Spain (sorting) turned out to be much more revealing, showing positive sorting toward the United States, for both male and female migrants. The pattern of sorting is consistent with the estimated earnings for Ecuadorians in each location reported earlier. The United States offered a sub- stantially higher college wage premium than Spain, which can account for the positive sorting toward the United States. In conclusion, wages differences across the three locations can account for the differences in skill composition of the migration episode analyzed here. However, other factors must be incorporated to account for the differences in the size of the flows. V. I M M I G R A T I O N P O L I C I E S AND THE CHOICE OF D E S T I N AT I O N A country’s attitude toward immigration is manifested in a host of policies, including amnesties for illegal aliens, pension rights portability, quotas on legal immigrants, enforcement of border controls, and visa requirements for nonim- migrant admissions. While the literature acknowledges that these factors affect immigration, much less is known about their individual effects. Why did most Ecuadorians go to Spain despite the substantially larger income gains from migrating to the United States? Several factors might have had a role, but identifying their individual influence is dif�cult. Such factors include the cultural and linguistic ties between Ecuador and Spain, Spain’s more generous welfare services, characteristics of Ecuadorian networks in both countries, and the greater ease of legally entering and of becoming a resident, among others.23 23. Additional time-invariant factors that are not accounted for by the wage differential are represented by the lower costs of living in Spain, lower income taxes in the United States (Bertoli, Ferna´ ndez-Huertas Moraga, and Ortega 2010), and lower cost of sending remittances from the United States, because dollarization in Ecuador enabled Ecuadorians to avoid the unfavorable exchange rates that usually apply to these transfers (see http://remittanceprices.worldbank.org). 70 THE WORLD BANK ECONOMIC REVIEW This section looks at the role played by just one factor: the visa waiver program that eased Ecuadorian travel to Spain. Its termination in 2003 permits isolating the effect of this one dimension of immigration policy on Ecuadorians’ migration choices. But �rst consider some of the differences for Ecuadorians in legally entering and becoming a resident in the United States and in Spain. Legal migration to the United States between 1999 and 2005 occurred mostly through family reuni�cation provisions (17,396 cases of family-based preferences and 36,412 cases of close relatives of naturalized immigrants); few Ecuadorians (7,705) obtained a legal residence permit through employment-based preferences (U.S. Department of Homeland Security various years). More than half of the immi- grants over the reference period were undocumented residents, with few options to regularize their status, as the United States has not approved a general amnesty since the Immigration Reform and Control Act of 1986. Legal migration to Spain depended mainly on obtaining a work visa, but undocu- Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 mented Ecuadorians also had several opportunities to legalize their status through one of Spain’s frequent amnesties in the early 2000s. Spain also had faster access to citizenship. Ecuadorians become eligible for naturalization after two years of legal residence (a shorter period is required for Ecuadorians of proven Spanish descent). In the United States, Ecuadorians can apply for citi- zenship only �ve years after obtaining legal residency documentation (green card).24 An apparently small but important difference in immigration policies was the need for Ecuadorians to obtain a visa to enter the United States, while they could visit Spain for up to three months without a visa provided that they had approximately $2,000, a credit card, a travel plan, hotel reservations, con- �rmed return flight, and justi�cation for visiting (Jokisch and Pribilsky 2002). Most Ecuadorians who wished to immigrate to Spain simply overstayed the three-month period, became undocumented workers, and waited for a general amnesty. Conditions for undocumented workers were much easier in Spain than in the United States. Government raids on workplaces were rare, and everyone residing in Spain had access to free healthcare regardless of immigra- tion status. Illegal immigrants to the United States, by contrast, often experi- enced expensive and risky travel, a hostile social environment, fear of apprehension and deportation, and exclusion from most government services. When Spain’s visa waiver program was terminated in the summer of 2003, there were no other relevant changes in U.S. or Spanish immigration policy toward Ecuadorians, including in immigrants’ access to public services, in Ecuadorian networks, or in cultural or economic conditions. Thus the change in Ecuadorian inflows into Spain in the months following termination of the 24. Access to Spanish citizenship is regulated by the Constitution and by the Ley Orga´ nica 4/2000, while criteria for access to U.S. citizenship are set by the Constitution and the 1952 Immigration and Naturalization Act, partially revised in the early 2000s. ´ ndez-Huertas Moraga, and Ortega Bertoli, Ferna 71 F I G U R E 2. Monthly Inflows of Ecuadorians to Spain, 1999–2007 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Source: Authors’ calculations using data from INE (various years). visa waiver helps to isolate its role in Ecuadorians’ destination decision. In March 2003, the European Union included Ecuador among the countries whose nationals had to have a visa to enter any EU member state (Council of the European Union 2003). Spain complied with this regulation on June 3, 2003, notifying Ecuador that the visa waiver would be suspended as of August 3, 2003 (Boletı´n O�cial del Estado 2003). The inflow of Ecuadorians to Spain dropped sharply immediately after the visa requirement went into effect (�gure 2).25 Average monthly inflows fell from 7,862 in the 12 months before the change to 1,566 in the following 12 months.26 The United States became the main destination for Ecuadorians in 2004 and 2005 (see �gure 1). Such a dramatic effect from termination of the visa waiver might seem sur- prising. Visa waivers do not receive as much attention in the literature as some 25. These data are from the Local Population Registry. Its accuracy is very high, particularly since January 2000, when the Ley Orga ´ nica 4/2000 increased the incentives for illegal migrants to register by allowing them to document their residence in Spain for future amnesties (see Ferna ´ ndez-Huertas Moraga, Ferrer, and Saiz, 2009). 26. A regression of the monthly inflows of Ecuadorians into Spain between January 1999 and December 2005 was run for a set of monthly and yearly dummy variables to control for seasonality in the data and for the confounding effect of macroeconomic conditions and a dummy variable for introduction of the visa requirement. The estimated coef�cient on the visa requirement variable was – 4,790 and highly statistically signi�cant. Similar results were obtained when GDP per capita in the three countries was included among the regressors; the estimated coef�cient for the change in visa policy was –5,026, con�rming that the policy change introduced a structural break in the series. The results are available from the authors on request. 72 THE WORLD BANK ECONOMIC REVIEW T A B L E 5 . Apprehensions and Deportations of Ecuadorian Migrants to the United States and Spain Migrants to the United States Migrants to Spain Year Mexicoa At seab INSc Total Expulsionsd Devolutionse Returnsf Total 1999 — 298 822 1,120 170 10 1,686 1,866 2000 — 1,244 913 2,157 52 120 1,106 1,278 2001 1,055 1,020 960 3,035 70 91 1,021 1,182 2002 1,427 1,608 729 3,764 314 92 4,675 5,081 2003 808 703 722 2,233 614 178 4,950 5,742 2004 1,076 1,189 1,116 3,381 — — — — 2005 3,276 1,149 1,490 5,915 — — — — Total 7,642 7,211 6,752 21,605 1,220 491 13,438 15,149 — is not available. a. Apprehensions and deportations by Mexican authorities. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 b. Alien migrants interdiction by the U.S. Coast Guard, �scal year. c. Aliens removed by the Immigration and Naturalization Service (now Immigration and Customs Enforcement), �scal year. d. Repatriation of illegal aliens resident in the country. e. Individuals who attempted to enter Spain illegally through nonborder areas. f. Individuals rejected at Spanish borders. Source: Authors’ analysis based on data from the INAMI (various years), U.S. Coast Guard (2010), U.S. Department of Homeland Security (various years), and Ministerio de Trabajo e Inmigracio´ n (various years). other dimensions of immigration policy, such as quota size and skill require- ments.27 However, the great distance between Ecuador and Spain means that air travel is virtually the only channel of entry, which simpli�ed enforcement of the new visa requirement. A related question is why the visa waiver had such a large effect on the desti- nation choice of Ecuadorian migrants. A reasonable hypothesis is that for Ecuadorians for whom illegal migration was the only feasible alternative, Spain was a much cheaper, and considerably safer, destination than the United States. Anecdotal evidence suggests that illegal migration to the United States costed $7,000–$9,000 in the late 1990s (Jokisch and Pribilsky 2002), com- pared with $1,800 per migrant to Spain (based on self-reported data from the ENI 2007). That difference was surely important for Ecuadorians, who faced tight liquidity constraints in the years following the crisis. Additionally, attempts to enter the United States illegally entailed a much higher risk of deportation (table 5). Between 1999 and 2005 some 21,605 Ecuadorian 27. Grogger and Hanson (forthcoming) control for visa waivers, which they �nd “are associated with higher migration rates, although the effect is marginally signi�cant.� Ortega (2005, 2010) studies the political-economy determinants of immigration policy but focuses exclusively on quotas and skill requirements. ´ ndez-Huertas Moraga, and Ortega Bertoli, Ferna 73 migrants were caught at sea by the Coast Guard, in Mexico, or by U.S. border patrols.28 Over 1999–2003, some 15,149 Ecuadorian migrants were deported from Spain, nearly all of them rejected at the border. Combining the data on deportations from table 5, the data on total migration flows from �gure 1, and information on legal migration from second- ary sources gives an approximate measure of the probability of apprehension when attempting to migrate illegally to Spain or the United States (ratio of number of deportations to the estimated number of illegal migrants plus depor- tations).29 The estimated probability was 23.6 percent for migrants to the United States and 5.7 percent for Spain (before the end of the vise waiver program).30 Illegal migrants to the United States also faced a high risk of death in transit, whereas the voyage from Ecuador to Spain was safe and comfortable. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 VI. CONCLUSIONS While the analysis in the article found that the skill composition of Ecuadorian migration flows was consistent with the wages received by Ecuadorians at each destination, the larger size of the Ecuadorian migration flows to Spain was puz- zling considering the large college wage premium in the United States. The puzzle is resolved by taking into account that the options for migrating legally to either country were severely limited and that migrating illegally to the United States was much more costly than migrating illegally to Spain, largely for policy-induced reasons. The evidence presented here shows that changes in some dimensions of immigration policy can have very large effects on immigration flows. Most likely, the U.S. tightening of controls over illegal immigration since the mid-1990s, combined with factors that made Spain an attractive destination, was effective in diverting the Ecuadorian exodus toward Spain. When the visa 28. A concern with deportation �gures is that the same would-be migrant can be apprehended and deported more than once and may eventually succeed in migrating; Pribilsky (2007, p. 166) observes that “it is a common practice for Border Patrol agents to ‘throw back’ alien Mexicans caught crossing illegally� and most Ecuadorians can successfully pretend to be Mexicans when apprehended, so that they can make another attempt to cross the border. Still, the �gures in table 5 include only those who were identi�ed as Ecuadorians by U.S. authorities and hence were deported to Ecuador. 29. The number of Ecuadorians who entered the United States illegally over 1999– 2005 (70,000) is from Hoefer, Rytina, and Baker (2008). The number for Spain takes the 303,555 Ecuadorians who entered Spain between 1999 and 2003 from �gure 1 and subtracts the 52,828 who were granted visas over the same period, leaving approximately 250,727 Ecuadorians who entered Spain through nonimmigrant admission provisions. 30. As with the monetary costs of migration, the income gain was still larger for migrating to the United States rather than to Spain even after discounting the differences in the probability of failing to reach the two countries. Still, the crisis of the 1990s probably increased the risk aversion of Ecuadorian households, who would be more unwilling (and unable) to bear the costs of a migration attempt that entailed a high risk of failure. 74 THE WORLD BANK ECONOMIC REVIEW waiver granted to Ecuadorians was repealed in August 2003, the inflow of Ecuadorians to Spain halted almost immediately. The inflows of Ecuadorians increased the relative supply of unskilled labor in both the United States and Spain. The effects on the U.S. labor market were probably very limited, as Ecuadorians represented just 1.3 percent of immigra- tion inflows to the United States in 1999–2005. Their share of immigration inflows to Spain was substantially larger, at 12 percent.31 Still, there is wide- spread agreement among researchers that the largest effects of migration are on migrants’ themselves, rather than on natives, in the form of income gains, part of which can be remitted back to the country of origin. Additionally, as Hanson and McIntosh (2010b) argue for the case of Mexico, the large emigra- tion of Ecuadorians may have kept wages in Ecuador from falling as much as they would otherwise have in the aftermath of the late 1990s crisis. Globalization can provide relief in times of severe economic distress. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 REFERENCES Beckerman, P., and A. Solimano, ed. 2002. Crisis and Dollarization in Ecuador. Washington, DC: World Bank. Beine, M., F. Docquier, and C ¨ zden. Forthcoming. “Diasporas.� Journal of Development Economics. ¸. O Bertoli, S. Forthcoming. “Networks, Sorting and Self-selection of Ecuadorian Migrants.� Annales d’Economie et de Statistique. ´ ndez-Huertas Moraga, and F. Ortega. 2010. “Crossing the Border: Self-selection, Bertoli, S., J. Ferna Earnings and Individual Migration Decisions.� IZA Discussion Paper 4957. Institute for the Study of Labor, Bonn, Germany. Borjas, G.J. 1987. “Self-selection and the Earnings of Immigrants.� American Economic Review 77: 531 –53. ———. 1999. “The Economic Analysis of Immigration.� In Handbook of Labor Economics Vol. 3.1, ed. O. Ashenfelter, and D. Card. Amsterdam: Elsevier. ´n O�cial del Estado. 2003. No. 159, July 4th. www.boe.es. Madrid. Boletı Council of the European Union. 2003. “Council Regulation (EC) 453/2003.� Of�cial Journal of the European Union L 69 : 10– 11, Dahl, G.B. 2002. “Mobility and the Return to Education: Testing a Roy Model with Multiple Markets.� Econometrica 70 (6): 2367– 420. ´ ndez-Huertas Moraga, J., A. Ferrer, and A. Saiz. 2009. “Localizacio Ferna ´ n de los Inmigrantes y Preferencias Residenciales de la Poblacio ´ ctona: ¿Nuevos guetos?�. Institute for Economic ´ n Auto Analysis (CSIC), Bellaterra (Barcelona), Spain. Grogger, J., and G. Hanson. Forthcoming. “Income Maximization and the Selection and Sorting of International Migrants.� Journal of Development Economics. Hanson, G. 2006. “Illegal Migration from Mexico to the United States.� Journal of Economic Literature 44 (4): 869–924. Hanson, G., and C. McIntosh. 2010a. Birth Rates and Border Crossings: Latin American Migration to the US, Canada, Spain, and the UK. NBER Working Paper 16471. Cambridge, MA: National Bureau of Economic Research. 31. See also the discussion for table S.3 in the supplemental appendix. ´ ndez-Huertas Moraga, and Ortega Bertoli, Ferna 75 ———. 2010b. “The Great Mexican Emigration.� Review of Economics and Statistics 92 (4): 798– 810. Heckman, J.J. 1979. “Sample Selection Bias as a Speci�cation Error.� Econometrica 47: 153–61. Hoefer, M., N. Rytina, and B.C. Baker. 2008. Estimates of the Unauthorized Immigrant Population Residing in the United States: January 2007. Of�ce of Immigration Statistics, Policy Directorate, United States Department of Homeland Security, Washington, DC. INAMI (Instituto Nacional de Migracio ´n Estadı ´ n). Various issues. Boletı ´stico Anual. Mexico Distrito Federal: Instituto Nacional de Migracio´ n. ´stica). 2001. Censo de Poblacio INE (Instituto Nacional de Estadı ´ n y Viviendas. Madrid: Instituto ´stica. Nacional de Estadı ´stica. ———. 2006. Encuesta de Structura Salarial. Madrid: Instituto Nacional de Estadı ´stica. ———. 2007. Encuesta Nacional de Inmigrantes. Madrid: Instituto Nacional de Estadı ´stica de Variaciones Residenciales. Madrid: Instituto Nacional de ———. Various years. Estadı ´stica. Estadı INEC (Instituto Nacional de Estadı ´ n y Vivienda. Quito: Instituto ´stica y Censo). 2001. Censo Poblacio ´stica y Censo. Nacional de Estadı Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 ———. 2005. Encuesta Nacional sobre el Empleo y Desempleo en el A ´ rea Urbana y Rural. Quito: ´stica y Censo. Instituto Nacional de Estadı ´ come, L.I. 2004. “The Late 1990s Financial Crisis in Ecuador: Institutional Weaknesses, Fiscal Ja Rigidities, and Financial Dollarization at Work.� IMF Working Paper 04/12. International Monetary Fund, Washington, DC. ´ n Ecuatoriana.� Ecuador Jokisch, B. 2001. “Desde Nueva York a Madrid: Tendencias en la Migracio Dedate 54: 59–84. Jokisch, B., and J. Pribilsky. 2002. “The Panic to Leave: Economic Crisis and the ‘New Emigration’ from Ecuador.� International Migration 40 (4): 76– 101. Laeven, L., and F.V. Valencia. 2008. “Systemic Banking Crisis: A New Database.� IMF Working Paper 08/224. International Monetary Fund, Washington, DC. ´ n y Crisis en el Ecuador. Quito: Ediciones Abya-Yala. Larrea, C. 2004. Pobreza, Dolarizacio Mayda, A.M. 2010. “International Migration: A Panel Data Analysis of the Determinants of Bilateral Flows.� Journal of Population Economics 23: 1249–74. McKenzie, D., and H. Rapoport. 2010. “Self-selection Patterns in Mexico– U.S. Migration: The Role of Migration Networks.� Review of Economics and Statistics 92 (4): 811– 21. Mincer, J. 1974. Schooling, Experience and Earnings. New York: Columbia University Press. Ministerio de Trabajo y Inmigracio ´stico de Inmigracio ´ n. Various years. Anuario Estadı ´ n, Madrid: Ministerio de Trabajo y Inmigracio´ n. Ortega, F. 2005. “Immigration Policy and Skill Upgrading.� Journal of Public Economics 89: 1841–63. ———. 2010. “Immigration, Citizenship, and the Size of Government.� The B.E. Journal of Economic Analysis & Policy: Contributions 10 (1): Article 26. Ortega, F., and G. Peri. 2009. The Causes and Effects of International Migrations: Evidence from OECD Countries 1980– 2005. NBER Working Paper 14883. Cambridge, MA: National Bureau of Economic Research. Pribilsky, J. 2007. La Chulla Vida: Gender, Migration and the Family in Andean Ecuador and New York City. Syracuse, NY: Syracuse University Press. Rosenzweig, M. 2007. “Education and Migration: A Global Perspective.� Yale University, New Haven, CT. Roy, A.D. 1951. “Some Thoughts on the Distribution of Earnings.� Oxford Economic Papers 3 (2): 135–46. 76 THE WORLD BANK ECONOMIC REVIEW Ruggles, S., M. Sobek, A. Trent, C.A. Fitch, R. Goeken, P. Kelly Hall, M. King, and C. Ronnander. 2008. Integrated Public Use Microdata Series: Version 4.0 [Machine-readable database]. Minneapolis, MN: Minnesota Population Center. ´ nchez, J. 2004. “Ensayo sobre la Economı Sa ´ n en Ecuador.� Ecuador Debate 63: ´a de la Emigracio 47 –62. Sjaastad, L.A. 1962. “The Costs and Returns of Human Migration.� Journal of Political Economy 70 (5): 80– 93. U.S. Census Bureau. 2000. United States Census. Washington, DC: Census Bureau. ———. 2007. American Community Survey. Washington, DC: Census Bureau. U.S. Coast Guard. 2010. “Alien Migrant Interdiction.� Department of Homeland Security, United States Coast Guard. www.uscg.mil/hq/cg5/cg531/amio.asp. U.S. Department of Homeland Security. Various years. Yearbook of Immigration Statistics. Washington, DC: Department of Homeland Security, Of�ce of Immigration Statistics. World Bank. 2008. World Development Indicators. Washington, DC: World Bank. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Do Migrants Improve Governance at Home? Evidence from a Voting Experiment Catia Batista and Pedro C. Vicente Can international migration promote better institutions at home by raising the demand for political accountability? A behavioral measure of the population’s desire for better governance was designed to examine this question. A postcard was distributed to house- holds promising that if enough postcards were mailed back, results from a survey module on perceived corruption would be published in the national media. Data from a tailored household survey were used to examine the determinants of this behavioral measure of demand for political accountability (undertaking the costly action of mailing the postcard) and to isolate the positive effect of international emigration using locality- Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 level variation. The estimated effects are robust to the use of instrumental variables, including past migration and macro shocks in the destination countries. The estimated effects can be attributed mainly to migrants who emigrated to countries with better gov- ernance, especially migrants who return home. JEL codes: F22, O12, O15, O43, P16 Keywords: international migration, governance, political accountability, institutions, effects of emigration in origin countries, household survey, Cape Verde, sub-Saharan Africa Recent research has examined the importance of international migration to development in countries of origin. The positive effects on economic growth are well documented for international remittances, return migrants, diaspora Catia Batista (catia.batista@tcd.ie) is assistant professor at Trinity College Dublin and research af�liate at the Institute for the Study of Labor (IZA). Pedro C. Vicente (vicentep@tcd.ie) is assistant professor at Trinity College Dublin, research associate at the Centre for the Study of African Economies (CSAE), University of Oxford, and research af�liate at the Bureau for Research and Economic Analysis of Development (BREAD). The authors gratefully acknowledge useful comments from the journal editor and three anonymous referees. Additional valuable suggestions were provided by Alan Barrett, Michel Beine, Ron Davies, Claudia Martinez, Franco Mariuzzo, John McHale, Kevin O’Rourke, Pia Orrenius, Hillel Rapoport, Frances Ruane, Maurice Schiff, Antonio Spilimbergo, Dean Yang, and participants in a number of seminars and conferences. The authors are indebted to Paul Collier for his initial encouragement of this research project. They thank the dedicated team of local enumerators with whom they worked, and Deolinda Reis and Francisco Rodrigues at the National Statistics Of�ce of Cape Verde. Research assistance was provided by Mauro Caselli. Batista gratefully acknowledges �nancial support from the George Webb Medley Fund at the University of Oxford. Vicente gratefully acknowledges �nancial support from the Economic and Social Research Council (ESRC)–funded Global Poverty Research Programme for the household survey conducted in Cape Verde on which this article is based. THE WORLD BANK ECONOMIC REVIEW, VOL. 25, NO. 1, pp. 77 – 104 doi:10.1093/wber/lhr009 Advance Access Publication May 12, 2011 # The Author 2011. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 77 78 THE WORLD BANK ECONOMIC REVIEW effects promoting foreign investment and international trade, and emigration of the most educated.1 Less attention has gone to the influence of international migration on the quality of institutions, which can be crucial to economic development (see Acemoglu, Johnson, and Robinson 2005). The traditional perspective views emigration as a safety valve that allows individuals unhappy with their political institutions to leave their home country.2 Emigration could therefore be detrimental to the domestic political system (a form of “brain drain�) by undermining demand for political account- ability and, if those who leave are especially quali�ed to improve political insti- tutions, by weakening the capacity to supply better quality institutions. Emigration may also promote improved political institutions in several ways: emigrants may create strong diaspora effects influencing political change (for example, by influencing local authorities on the supply side or by exposing the domestic population to better institutions abroad on the demand side). If return emigrants bene�ted from an enriching experience abroad, that could Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 also translate into improvements in the quality of domestic political institutions (on the supply side by increasing direct participation in the political system and on the demand side by raising awareness and demand for political accountability). Because emigration could affect political institutions differently depending on the context, what actually happens is an empirical question that remains unanswered in the literature. This article tests the hypothesis that international migration experiences promote better institutions at home by boosting demand for political accountability. Examining this question requires understanding popular demand for political accountability. A simple voting experiment was used to capture a behavioral measure of demand for better governance at home. Following a survey of perceived corruption in public services, respondents were asked to mail a prestamped postcard if they wanted the (anonymous) results of this survey to be made publicly available in the media. They were told that at least 50 percent of respondents would have to return postcards for the information to be released publically. 1. Evidence of the positive effects of remittances is provided, among others, by Edwards and Ureta (2003) for El Salvador and Yang (2008) for the Philippines. Dustmann and Kirchkamp (2003), Mesnard and Ravallion (2006) and Batista, McIndoe-Calder, and Vicente (2010) examine the role of return migration. Gould (1994), Rauch and Trindade (2002), Kugler and Rapoport (2007), Iranzo and Peri (2009), and Javorcik and others (forthcoming) evaluate the relationship between migrant networks, and trade and foreign investment. The possibility of a “brain gain� as opposed to traditional “brain drain� is empirically supported by Beine, Docquier, and Rapoport (2008) and Batista, Lacuesta, and Vicente (forthcoming). 2. Hirschman (1970) proposed the “exit� vs. “voice� dichotomy by which citizens unhappy with the domestic situation choose either to emigrate (exit) or to protest and contribute to political change (voice). In this setting, emigration may be understood as a “safety valve,� which releases protest intensity in the home political system and therefore reduces demand for political improvements. Batista and Vicente 79 This voting experiment was not a randomized controlled trial but simply a way to obtain a behavioral measure of demand for political accountability. This measure is likely superior to standard self-reported measures from survey data, which may suffer from “conformity bias� (respondents may want to conform to the perceived anticorruption message of the survey). This behavioral measure of demand for better institutions is therefore a methodological contri- bution of this article. Tailored data from a purposely designed and conducted household survey in Cape Verde is used to examine the determinants of voting behavior and to isolate the positive effect of international emigration on the demand for politi- cal accountability. A simple political economy framework takes voting behav- ior as the outcome of an expected cost-bene�t analysis. A detailed survey was customized to control for potentially varying voting costs (such as the distance to post mail and the ease and frequency of doing so) and for characteristics affecting perceived voting bene�ts (such as con�dence in surveyors, income, Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 and family structure). Overall, the results show that international emigration positively affects demand for improved political accountability, with stronger effects for migrants to countries with better governance and for return migrants than for current migrants. Empirical evidence on the impact of emigration on the quality of political institutions in origin countries is scarce, but there are a few recent contri- butions. Docquier and others (2010) present cross-country evidence that unskilled emigration from a large sample of developing countries to Organisation for Economic Co-operation and Development countries over 1975–2000 positively affected institutional quality in origin countries (measures of democracy and economic freedom). Though skilled emigration had an ambiguous effect in the short run, simulations found signi�cant insti- tutional gains from “brain drain� over the long run, after considering incentive effects of the brain drain on human capital formation. Li and McHale (2009) describe possible mechanisms through which skilled emigration could affect political and economic institutions at home, presenting cross-country evidence for 1990–2006 consistent with the hypothesis of a positive effect on political institutions ( particularly on political accountability) but not on economic insti- tutions. Spilimbergo (2009) uses evidence from 1960 to show that foreign edu- cation acquired in democratic countries seems to promote democracy in home countries. These empirical contributions are consistent with the results reported here, but they cannot distinguish between supply and demand forces nor capture the mechanisms underlying the identi�ed effects because they use aggregate data and explore cross-country variation. This article uses tailored household survey data for a single country, which allows focusing more speci�cally on the impact of emigration on the demand for improved political accountability, while discriminating between the impact of return and current migrants. This approach relies on within country variation, rather than the traditional 80 THE WORLD BANK ECONOMIC REVIEW cross-country source of variation. Reliance on data for a single country may, however, raise external validity concerns, so that contributions by these differ- ent lines of work are both important and complementary. Section I presents an overview of Cape Verde, to provide context for the study. Section II describes the experimental design and theoretical framework supporting the empirical strategy. Section III details the tailored household survey used in the empirical work, including the main descriptive statistics. These data are then used in the empirical analysis reported in section IV. Section V presents some concluding remarks. I. AN INTRODUCTION TO CAPE VERDE Cape Verde is an island country off the coast of West Africa whose 441,000 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 inhabitants live on nine islands. Much of the population is concentrated on Santiago, the largest island and home to the country’s capital, Praia (INE 2002). The country is religiously and ethnically homogeneous: the index of religious fractionalization is 7.66 percent 3 (96 percent of the population is Roman Catholic). The ethnolinguistic fractionalization index is 41.74 percent, comparable to that in Spain and New Zealand and in contrast with high frac- tionalization indexes of more than 80 percent in 20 Sub-Saharan countries. Cape Verde won its independence from Portugal in 1975, and a socialist regime took power. The �rst free elections took place in 1991, and a stable democracy has been in place since then. Governance has been good, particu- larly for a Sub-Saharan African country: Cape Verde ranked 46 of 180 countries according to Transparency International (2009), slightly behind Botswana and Mauritius. The World Bank’s (2011) Worldwide Governance Indicators heralded Cape Verde as having the Best Control of Corruption in Sub-Saharan Africa in 2005, after Botswana. In economic performance, Cape Verde is ranked as a lower middle-income economy by the World Bank (2006), with a 2003 GDP per capita of $5,900 (in purchasing power parity terms; Heston, Summers, and Aten 2006). With an average annual per capita economic growth rate of 4.4 percent over 1981– 2004 (and 5.8 percent over 1991–2000), it has greatly outperformed the Sub-Saharan African average of 0.6 percent, with only Equatorial Guinea (11 percent) and Botswana (5 percent) growing faster (Heston, Summers, and Aten 2006). While these countries have exports accounting for 47 percent and 55 percent of GDP and are rich in natural resources, Cape Verde grew despite a much smaller export share of 20 percent and a dearth of natural resources—in fact, Cape Verde has been plagued by droughts and famines. 3. This index is computed as one minus the Her�ndahl index of group shares and expresses the probability that two randomly selected individuals from a population belong to different groups. See Alesina and others (2003) for details. Batista and Vicente 81 Those droughts and famines have been closely related to the country’s massive emigration. Based on the stock of immigrants in most destination countries, Batista, Lacuesta, and Vicente (forthcoming) estimate that there are around 100,000 Cape Verdean current emigrants, or about 23 percent of the population. Also striking is the magnitude of brain drain emigration: an aston- ishing 68 percent of the educated labor force of Cape Verde lives abroad (Docquier and Marfouk, 2006). While these results depend on how educational attainment is de�ned, this is arguably the highest rate in Africa. Finally, inter- national remittances are high, accounting for 16 percent of GDP over 1987– 2003 (World Bank 2006).4 Remittances have always surpassed foreign direct investment and have nearly duplicated the amount of foreign aid, particularly since 2000. Freedom House (2011) classi�es Cape Verde as “among the freest media environments in Africa.� According to the Press Freedom Index, Cape Verde ranked 44 of 175 countries in freedom of the press, close to France, Spain, and Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Argentina (Reporters without Borders 2009). II. EXP ER IM ENTAL D ES IGN AND EMPIRICAL STRATEGY To empirically evaluate the hypothesis that international emigration may promote demand for better governance at home, this study offered respondents to a survey on perceived corruption in public services the opportunity to (anon- ymously5) make the results available in the national media by participating in a “special referendum.� Following their completion of the corruption question- naire, respondents were offered the opportunity to vote for political account- ability by taking the incentive-compatible voting action of mailing a prepaid postcard that read: “I wish that the conclusions of the survey on the quality of national public services (health, education, justice, . . .), conducted by the University of Oxford (UK) in the �rst months of 2006 to 1,000 households in ˜ o Vicente, Santo Anta the islands of Santiago, Sa ˜ o, and Fogo, be made public in the Cape Verdean media.� Interviewers told each respondent that “it is very important that you put the postcard in the mail if you want Cape Verdeans to be able to demand higher quality public services.�6 4. This share is likely an underestimate as it is based on of�cial statistics, which exclude informal channels, both legal and illegal. 5. Postcards were anonymous in the sense that respondents did not have to write their names on the postcard. This is the message that interviewers were instructed to convey. However, each postcard had a six-digit number that linked each postcard to each interviewed household, so that the household and respondent characteristics were known for each returned postcard. 6. The interview, which averaged 60 minutes, asked explicit questions about the need to bribe public of�cials or to otherwise influence them in order to receive public services. The postcard referred to “the quality of public services� instead of “corruption� in order to minimize behavior correlated with public opinion about corruption and thus to elicit a more accurate behavioral measure of the demand for political accountability. 82 THE WORLD BANK ECONOMIC REVIEW The results on perceived corruption in public services would be made public if 50 percent or more of the postcards were returned. To add credibility to the survey, a “media contract� between survey �eldworkers and respondents detailed the promise that the survey results would be publicized in the national media provided at least half of respondents returned the postcards. News reports and interviews on national television, radio, and newspapers helped to publicize the media contract and to confer legitimacy on the effort. This voting experiment was not a randomized controlled trial but rather a simple means to elicit a behavioral measure of demand for political account- ability. Using a behavioral measure is likely superior to standard self-report measures, which may be tainted by “conformity bias�—respondents would be more likely to conform to what they believe are the interviewers’ expectations about anticorruption attitudes. This hypothesis cannot be rejected from the empirical evidence in this article, as discussed in section IV. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Theoretical Framework Before testing whether international emigration increases the desire for political accountability at home, a theoretical framework was developed to elucidate the determinants of voting in this postcard experiment. There are many potentially relevant political economy theories of turnout and voting, as surveyed by Merlo (2006). Following the traditional literature on electoral participation, voter turnout was modeled as the outcome of an expected cost-bene�t analysis.7 The postcard distributed to the survey respondents was prestamped, so the cost of voting was largely the opportunity cost of mailing it. This cost could depend on how familiar respondents were with posting mail and with how practical it was to do so. The cost will be higher for people who are not used to posting mail, those for whom it is more dif�cult to do so, and those with higher labor income. The literature emphasizes the importance of considering an individual’s cal- culation of expected bene�ts. The expected bene�t of mailing the postcard arises from the desire for political accountability, which is the focus of this article. Crucially, survey respondents who are more con�dent about the trust- worthiness and independence of the foreign institution sponsoring the survey (and about the reliability of the Cape Verdean postal system) will attribute a higher probability to the public dissemination of the results on perceived cor- ruption. The expected bene�t is �nally a function of other variables directly affecting the desire for political accountability. Of greatest interest is the effect of international emigration, but factors such as gender, age, education, wealth, 7. Downs (1957) �rst provided a “calculus of voting� framework, which was later formalized by Tullock (1967) and Riker and Ordeshook (1968). Because of the simple nature of the voting experiment (a simple decision to vote or not), we can ignore strategic voting considerations and assume sincere voting behavior. Batista and Vicente 83 and family ties must also be considered (see, for instance, Alesina and Giuliano 2009). Empirical Strategy An individual respondent i’s voting decision on the survey (and therefore the demand for better political accountability) can be summarized by the following latent variable model: Vi ¼ 1 ðVià !ފŠ > Vià ¼ a0 þ a1 Ml þ a02 Xi þ 1i : This decision will be made whenever the (unobserved) expected net bene�t from voting, V* i , is positive. The expected net bene�t from voting depends on the local proportion of migrants, Ml, with impact a1 on voting behavior, which is the primary estimate of interest.8 The main explanatory variable is Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 computed as: Proportion of international migrants within the household’s spatial area of residence Number of migrants in the locality ¼ Number of residents in the locality where migrants includes both current and return migrants. The effect of the local proportion of migrants on an individual’s demand for good governance includes direct and indirect effects—effects arising directly from the presence of return migrants and indirect effects due to the influence return migrants exert on their peers (think, for instance, of neighbor families with no migrants who become more sensitive to governance issues after talking to a return migrant neighbor who lived in the United States for some years). An additional source of indirect migrant impact on the demand for accountability by local residents is the influence of current migrants who keep in touch with family and friends. Thus the proportion of migrants within a household’s spatial area of residence can be understood as a proxy for how frequently a resident can be expected to meet migrants (or their relatives and friends) in this locality. Recall that even though the results intuitively point to the importance of return migrants, the framework is suf�ciently wide to encompass the impact on the locality of origin of current migrants—through their contacts with family and friends, for instance. Second, the empirical speci�cation includes a vector of individual, house- hold, and locality characteristics, Xi, determining the costs and bene�ts of mailing the voting postcard. This vector includes individual demographics (such as age as a determinant of the ease of mailing the postcard and of the 8. Locality here is a census area in Cape Verde, which corresponds roughly to a small neighborhood, where social interaction would be expected to occur. 84 THE WORLD BANK ECONOMIC REVIEW demand for accountability) and individual controls for how familiar someone is with posting mail and how practical it is. In addition, there is an individual indicator of con�dence in the foreign institution sponsoring the survey and experiment. At the household level, vector Xi includes variables such as family structure and asset ownership, which are likely determinants of an individual’s subjective valuation of the bene�t of improved governance. At the locality level, the analysis controls for average expenditure per capita and for the share of local residents working in agriculture, construction, and retail trade, which may also influence the perceived bene�t of better governance. All regressions also include island �xed effects. Probit regressions are used to estimate this empirical model. Variation of migration behavior across localities, after controlling for individual, household, and local characteristics, is the source of variation that enables identi�cation of the main coef�cient of interest,a1. Unlike with family-level variation, using locality-level variation mitigates Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 self-selection concerns based on unobservable characteristics: unobserved ability (which may increase both migration and demand for good governance) may be correlated across family members but that is not likely at the locality level. Indeed, using locality-level variation should permit averaging out unob- served heterogeneity to some extent, thus avoiding the most apparent endo- geneity problems. Moreover, Cape Verde is a small, homogeneous country, which rules out the most obvious (potentially omitted) factors that could promote migration and accountability demand simultaneously at the locality level. I I I . D ATA D E S C R I P T I O N : TA I LO R E D H O U S E H O L D S U RV E Y The empirical work is based on a household survey on migration and the quality of public services designed to answer the research questions. The survey was conducted in Cape Verde from December 2005 to March 2006 by the authors, who were af�liated with the University of Oxford. (Additional details on the �eldwork and survey are at www.csae.ox.ac.uk/resprogs/corruption/cv/ cv.htm.) The survey was submitted to a representative sample of 1,066 resident households (997 complete interviews) in 5 percent of the 561 census areas of Cape Verde. This sample provided information on resident nonmigrants and return migrants and on a large sample of current emigrants. The questionnaire included a module on the perceived quality/corruption of public services and one on migration characteristics of the household (including full migration his- tories). The interviewed household member, who had to be at least 30 years old, was asked to provide the socio-demographic characteristics of all house- hold members, including children living elsewhere. The respondent was also asked to characterize all migration spells of household members, including who emigrated, where, and when. Finally, there were questions about the Batista and Vicente 85 household’s economic situation, such as living standards, income, and whether any member had received remittances the previous year. (An English trans- lation of the questionnaire is available at www.csae.ox.ac.uk/resprogs/ corruption/cv/questcveng.pdf.) Face to face survey interviews were conducted by teams of local interviewers and the authors, who recruited and trained the local teams. Interviewers received at least 18 hours of training in groups of two or three on understand- ing the content/objectives of the survey, answering the questionnaire, and piloting. Census areas for the sample were chosen randomly, with weighting by number of households, and households within a census area were chosen ran- domly using standard techniques (nth house, with second visits attempted the same day). To be eligible, members of the household had to be resident in the country any time during 1985–2006. The random sampling of households had two weaknesses: differences in the Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 number of attempts to interview a selected household in the different census areas and differences in the number of nonresponses. Weighted data were used to account for these problems, although differences from unweighted data are negligible. Data collected on nonrespondents on their gender, approximate age, approximate schooling, and approximate income were used for this purpose. About half the respondents did not provide information on income, so regressions were run with 452 observations, at most.9 The survey data on nonmigrants, return migrants, and current migrants show that relative to nonmigrants, current emigrants are slightly more likely to be male and in their prime working years (ages 21–50; table 1). They are also more likely to have a postsecondary education. Return migrants are strongly more likely to be male (compared with both residents and current migrants) and most are over 50 years old. They tend to be less educated than current migrants, but are still more likely than residents to have a postsecondary education. The survey results on annual migration flows over 2000–05 are close to those for the last national census period 1995–2000 (INE 2002), for both migrant outflows (around 4 percent) and returns (about 20 percent). Portugal (55 percent) and the United States (20 percent) are the main destination 9. An attrition analysis was conducted to evaluate the impact of the missing observations on the baseline econometric results (with and without controls) using multiple imputation methods. It showed that comparing the effect of local migration on voting behavior when observations without income information are excluded has a large impact on the magnitude and signi�cance of the estimated results. When multiple imputation methods were used to recover the missing information, the magnitude of estimated coef�cients falls but the statistical signi�cance remains. This suggests that the missing income observations influence the magnitude of estimated effects, which would likely be smaller were income data available for all respondents, but that the positive sign and statistical signi�cance of the estimates remain in all possible speci�cations. The results are fairly stable, however, regardless of the number of imputations performed (if anything, results improve as the number of imputations rises). 86 THE WORLD BANK ECONOMIC REVIEW T A B L E 1 . Characteristics of Cape Verdean Migrants and Nonmigrants Characteristic Nonmigrants Current migrants Return migrants Sample size 4997 907 241 Gender (%) Male 48 52 64 Female 52 48 36 Age (%) 0 –10 years 21.4 0.4 2.4 11– 20 years 28.6 11.2 4.9 21– 30 years 12.9 33.9 5. 5 31– 40 years 13.1 25.0 17.6 41– 50 years 10.1 20.5 15.8 51– 60 years 4.4 8.0 11.5 61– 70 years 4.2 0.9 18.8 71– 80 years 3.8 0.2 20.6 81– 90 years 1.2 0.0 3.0 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 . 91 years 0.02 0.0 0.00 Education (males ages 15 – 64; %) No education 3.7 3.6 5.2 Preschool 1.5 0.7 0.0 Basic reading and writing 11.4 8.2 14.3 Primary 59.7 62.4 50.7 Intermediate secondary 18.8 9.9 19.5 Secondary 1.1 0.4 3.9 Postsecondary 3.8 14.9 6.5 Note: Numbers may not sum to 100 because of rounding. Source: Authors’ survey. countries for migrants; again, these results are similar to of�cial census stat- istics (INE 2002). The next most popular destinations are European countries (France with 12%, Netherlands and Luxemburg with 2% each) and Brazil (with 3%). Because only 43 percent of the postcards were returned, the results of the survey were not published in the national media. I V. E M P I R I C A L R E S U L T S This section on the main empirical results focuses on the robustness of the esti- mates of a gain in the demand for political accountability arising from inter- national migration. Baseline Results The baseline estimation of the probability of a given survey respondent return- ing the postcard is shown in table 2 (column 1). Even before controlling for other covariates (except for urban locality and island �xed effects), there is T A B L E 2 . Probability of Mailing Voting Postcard (Columns 1–6, Marginal Effects of Probit Estimates) and Probability of Self-reported Demand for Accountability (Column 7, Ordered Probit Estimates): Baseline Results Variable (1) (2) (3) (4) (5) (6) (7) Proportion of international migrants 0.9419 0.9446 1.0103 1.0724 1.1034 1.0886 2.0218 (relative to residents) in locality (0.3465)*** (0.3512)*** (0.3623)*** (0.3510)*** (0.3859)*** (0.3677)*** (1.3204) Trust in Oxford University 0.0077 0.0211 0.0228 0.0334 0.0348 0.0365 (0.0232) (0.0231) (0.0226) (0.0238) (0.0237) (0.0665) Habit of posting 0.0045 0.0083 0.0100 0.0089 0.0092 – 0.0152 (0.0132) (0.0127) (0.0127) (0.0137) (0.0138) (0.0267) Male – 0.0863 – 0.0928 – 0.0751 – 0.0751 0.1954 (0.0485)* (0.0467)** (0.0485) (0.0485) (0.1179)* Age 0.0207 0.0161 0.0131 0.0140 – 0.0602 (0.0134) (0.0143) (0.0143) (0.0142) (0.0304)** Age squared – 0.0002 – 0.0002 – 0.0001 – 0.0001 0.0004 (0.0001) (0.0001) (0.0001) (0.0001) (0.0003) Individual labor income – 0.0002 – 0.0002 – 0.0003 – 0.0003 0.0001 (0.0001)** (0.0001)** (0.0001)** (0.0001)** (0.0003) Number of children 0.0205 0.0212 0.0215 0.0189 (0.0120)* (0.0120)* (0.0121)* (0.0296) Household asset ownership – 0.1401 – 0.1242 – 0.1244 – 0.0016 (0.0626)** (0.0651)* (0.0647)* (0.1569) (Continued ) Batista and Vicente 87 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 88 TABLE 2. Continued Variable (1) (2) (3) (4) (5) (6) (7) Average private consumption expenditure 0.9789 0.6916 – 0.9512 per capita in locality (0.6637) (0.7866) (1.8530) Share of residents working in agriculture in – 0.8688 – 1.2906 – 0.1134 locality (0.5572) (0.9234) (2.1112) Share of residents working in construction – 0.5909 – 0.7994 – 6.8096 THE WORLD BANK ECONOMIC REVIEW in locality (1.1277) (1.1626) (2.5307)*** Share of residents working in retail trade 1.2200 0.9100 4.2588 in locality (1.6264) (1.7060) (3.3857) Share of households receiving international 0.9963 remittances in locality (1.3944) Number of observations 452 452 452 452 452 452 451 * Signi�cant at the 10 percent level; ** signi�cant at the 5percent level; *** signi�cant at the 1percent level. Note: Numbers in parentheses are robust standard errors clustered at the locality level. All regressions include urban locality dummy variable and island �xed effects. Self-reported demand for accountability is expressed as a 1 – 7 scale. Source: Authors’ analysis based on authors’ survey. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Batista and Vicente 89 a striking statistically signi�cant difference between the postcard voting prob- ability of localities depending on the ratio of migrants to residents (each per- centage point increase in the ratio of migrants to residents, including both current and return migrants, leads to a 0.94 percentage point increase in the probability of voting). After controlling for several individual- and household-level covariates, the observed voting differences remain (see table 2, columns 2–4). The signs of all signi�cant coef�cients are as expected and do not vary as additional controls are included. Because of the potential for omitted-variable bias, several locality-level controls are added, such as average private consump- tion expenditure per capita and the occupational structure in the locality. The addition of these controls does not alter the magnitude and signi�cance of the estimated effect (table 2, column 5). Another concern is that international migration may be proxying for important local �nancial characteristics, so that international remittances may also matter as determinants of the desire for Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 better governance. That does not seem to be the case: including the proportion of local households receiving international remittances has an insigni�cant economic and statistical impact and almost no affect on the estimated coef�- cients and signi�cances of the other determinants included in the regression (table 2, column 6). The baseline estimates are therefore those presented in column 5 of table 2. There is a strong negative income/wealth effect on the demand for more accountability. Having annual labor income with a negative estimated coef�- cient would be dif�cult to interpret directly as a negative income effect as this could simply be proxying the opportunity cost (time value) of mailing the post- card. However, this effect is also strong for asset ownership: wealthier people seem to place less value on the bene�ts of political accountability, which is consistent with Minier’s (2001) �nding that democracy is not a normal good. At the local level, though, the results consistently point to average expenditure per capita as positively influencing postcard mailing behavior. Baseline Robustness Checks Several robustness checks were conducted. Drawing from the evidence on “brain gain� (Batista, Lacuesta, and Vicente, forthcoming), the �rst robustness check addresses whether local education affects the way international migration for a locality generates a desire for political accountability. When controlling for local educational attainment, intermediate secondary and secondary school- ing do not change the sign, magnitude, and statistical signi�cance of the impact of local migration on the demand for political accountability (table 3, columns 1 –3). A postsecondary education, however, increases the migration effects, even though the positive coef�cient on postsecondary education is not signi�- cant at conventional levels. A potential concern with these estimated effects is that the probability of mailing a postcard may depend on respondents’ experience with and 90 T A B L E 3 . Probability of Mailing Postcard (Marginal Effects of Probit Regressions): Robustness Checks Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Proportion of international 1.1091 1.0818 1.3132 0.9140 1.1349 0.9889 1.0639 0.9920 0.9012 1.0104 migrants (relative to (0.3863)*** (0.4054)*** (0.3708)*** (0.4334)** (0.4159)*** (0.4066)** (0.3813)*** (0.3680)*** (0.3780)** (0.4157)** residents) in locality Ratio of residents – 0.0471 0.5362 completing (0.2704) (0.3314) relative to residents THE WORLD BANK ECONOMIC REVIEW not completing 9 years of schooling in locality Ratio of residents – 0.3184 – 2.0897 completing (0.4014) (0.7537)*** relative to residents not completing 12 years of schooling in locality Ratio of residents 1.8460 6.1358 completing (1.1037)* (1.9907)*** relative to residents not completing 15 years of schooling in locality Perceived corruption in 0.0381 0.0150 health sector (0.0147)*** (0.0182) Perceived corruption in 0.0382 0.0364 education sector (0.0154)** (0.0191)* Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Con�dence in postal system – 0.0144 –0.0061 (0.0263) (0.0288) Wait to mail until passes a –0.0366 –0.0195 postbox (0.1955) (0.2044) Gives to (taxi) driver to post 0.1933 0.0552 (0.1696) (0.2303) Gives to family member to 0.0884 –0.0237 post (0.1390) (0.1515) Gives to letter carrier 0.3767 (0.2607) Makes intentional trip to 0.0945 0.0180 postbox (0.1167) (0.1287) Time to postbox –0.0078 –0.0128 (0.0149) (0.0219) Comfort in posting mail 0.0119 0.0358 (0.0130) (0.0251) Number of observations 452 452 452 426 400 435 451 443 445 363 * Signi�cant at the 10 percent level; ** signi�cant at the 5percent level; *** signi�cant at the 1percent level. Note: Numbers in parentheses are robust standard errors clustered at the locality level. All regressions include the same controls as the baseline regression in column 5 of table 2. Source: Authors’ analysis based on authors’ survey. Batista and Vicente 91 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 92 THE WORLD BANK ECONOMIC REVIEW perceptions of corruption. That is indeed the case: respondents who perceive more corruption in the health and education sectors (the sectors most respon- dents had contact with) are signi�cantly more likely to mail the postcard (table 3, columns 4–5). The impact of perceived corruption also affects the magnitude and signi�cance of the impact of international emigration, but the impact is not systematically in one direction. Overall, the sign, magnitude, and broad statistical signi�cance of the effect of international migration remain stable throughout the different speci�cations. This result points to an intuitive, crucial role of perceived corruption in creating incentives for greater demand for accountability. Another important issue is to control properly for the cost of mailing the postcard and the con�dence when doing so. The sign, magnitude, and signi�- cance of the estimated coef�cients on local international emigration are not strongly affected by the choice of these controls (table 3, columns 6–9). None of these controls is ever statistically signi�cant. This is consistent with the idea Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 that, although incentive compatible, the costs of mailing the postcard are of slight importance to the results. When all the alternative controls are used simultaneously in a single regression, the main coef�cient of interest has the same magnitude as that esti- mated using other important controls and is again signi�cant at the 5 percent level despite the loss of observations implied by using all controls simul- taneously (table 3, column 10). Mechanics 1: Migrant Destination Having established the relevance of local migration in determining voting be- havior in the experimental setting, it is reasonable to wonder about the mech- anisms underlying this result. How does local migration affect behavior? One approach is to examine how the destination of local migrants affects the results. A comparison of the effect of the two main migrant destinations, Portugal and the United States, is striking: only migration to the United States has a sizable and signi�cant impact on the desire for better governance (table 4, columns 1 and 2). The effects of local migration to Portugal are not statistically signi�cant. Mechanics 2: Current and Return Migrants Continuing along this line of investigation, it is possible to distinguish between the effects of current and return migrants by country of destination (table 5, columns 1 and 2). The magnitude and signi�cance of effects are much higher for return migrants than for current migrants, regardless of country of destina- tion. This is an intuitive result, as migrants’ experience is more likely to affect the community of residence once migrants return and interact with residents than while they are away. Note also that the effects of both return and current migrants to the United States are positive (although insigni�cant for current migrants), whereas the effect of migrants returning from Portugal is negative. T A B L E 4 . Probability of Mailing Voting Postcard (Columns 1–7, Marginal Effects of Probit Estimates) and Probability of Self-reported Demand for Accountability (Columns 8 and 9, Ordered Probit Estimates) Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) Proportion of 1.1210 0.5435 0.6185 0.7998 0.6865 0.1818 0.2358 – 3.0466 – 0.3748 international (1.0271) (1.1214) (1.3669) (1.1458) (1.0633) (1.2426) (1.2329) (2.3588) (2.4081) migrants to Portugal (relative to residents) in locality Proportion of 2.7384 2.6239 2.6069 2.5595 2.6833 2.3141 3.1322 11.0254 12.6180 international (0.8777)*** (1.0761)** (1.0271)** (1.1184)** (0.9900)*** (1.1519)** (1.1275)*** (2.1924)*** (1.6359)*** migrants to United States (relative to residents) in locality Ratio of residents – 0.0343 completing (0.3374) relative to residents not completing 9 years of schooling in locality Ratio of residents – 0.3994 completing (0.3812) relative to residents not completing Batista and Vicente 12 years of schooling in locality 93 (Continued ) Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 94 TABLE 4. Continued Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) Ratio of residents 1.1284 completing (1.0349) relative to residents not completing 15 years of schooling in locality Perceived 0.0372 THE WORLD BANK ECONOMIC REVIEW corruption in health sector (0.0148)** Perceived 0.0390 corruption in education sector (0.0157)** Controls included No Yes Yes Yes Yes Yes Yes No Yes Number of 452 452 452 452 452 426 400 451 451 observations * Signi�cant at the 10 percent level; ** signi�cant at the 5percent level; *** signi�cant at the 1percent level. Note: Numbers in parentheses are robust standard errors clustered at the locality level. All regressions include the same controls as the baseline regression in column 5 of table 2. Self-reported demand for accountability is expressed as a 1 – 7 scale. Source: Authors’ analysis based on authors’ survey. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 T A B L E 5 . Probability of Mailing Voting Postcard (Columns 1–7, Marginal Effects of Probit Estimates) and Probability of Self-reported Demand for Accountability (Columns 8 and 9, Ordered Probit Estimates) Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) Proportion of current 2.0057 0.9979 0.8934 0.9949 1.8130 0.7794 0.9772 – 4.2651 – 2.0208 international (1.1713)* (1.3858) (1.4444) (1.3717) (1.4379) (1.5497) (1.5515) (3.2703) (2.9193) migrants to Portugal (relative to residents) in locality Proportion of current 0.9286 0.0938 0.4789 0.0001 0.8021 – 1.2650 0.9654 4.5386 6.9932 international (2.1526) (3.0738) (3.0037) (3.1214) (2.8120) (3.8085) (3.3864) (7.2504) (7.0402) migrants to United States (relative to residents) in locality Proportion of – 4.8152 – 4.9271 – 6.0974 – 4.0434 – 6.6494 – 7.0375 – 5.8158 – 4.4922 0.9119 international return (2.4159)** (2.8707)* (3.6964)* (3.3599) (2.4733)*** (3.5251)** (3.4314)* (12.3565) (12.8589) migrants to Portugal (relative to residents) in locality Proportion of 4.5445 5.0953 4.7343 5.1888 4.2942 5.7620 5.0397 19.7322 19.8166 international return (2.5979)* (2.3956)** (2.4130)** (2.4798)** (2.3635)* (2.7759)** (2.5711)** (6.9132)*** (7.1887)*** migrants to United States (relative to residents) in locality Ratio of residents 0.1610 completing relative (0.3662) to residents not Batista and Vicente completing 9 years of schooling in locality 95 (Continued ) Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 96 TABLE 5. Continued Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) Ratio of residents – 0.2466 completing relative (0.3932) to residents not completing 12 years of schooling in locality Ratio of residents 2.0197 completing relative (0.8948)** to residents not completing 15 years of THE WORLD BANK ECONOMIC REVIEW schooling in locality Perceived corruption 0.0375 in health sector (0.0153)** Perceived corruption 0.0370 in education sector (0.0162)** Controls included No Yes Yes Yes Yes Yes Yes No Yes Number of 452 452 452 452 452 426 400 451 451 observations * Signi�cant at the 10 percent level; ** signi�cant at the 5percent level; *** signi�cant at the 1percent level. Note: Numbers in parentheses are robust standard errors clustered at the locality level. All regressions include the same controls as the baseline regression in column 5 of table 2. Self-reported demand for accountability is expressed as a 1 – 7 scale. Source: Authors’ analysis based on authors’ survey. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Batista and Vicente 97 Robustness Checks: Self-selection How can we ensure that the estimated local migration effects are really causing the demand for accountability? One might conjecture that selection (for instance, on observable characteristics such as education) is driving the �nd- ings. To examine this possibility, the differences in means were estimated between localities with strong migration to Portugal (migrants to Portugal con- stitute at least 5 percent of the resident population) and those without; the same was done for migration to the United States. Households in areas prone to migration to Portugal are usually less well off than those in areas prone to migration to the United States, although those prone to migrate to the United States seem to possess above mean assets that could allow them to overcome the �nancial costs of an international move (table 6). Migrants to Portugal tend to originate in areas where agriculture and construction dominate over services, such as retail trade; the reverse is true for Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 migrants to the United States. Education pro�les differ as well: the most edu- cated migrants move to the United States, an expected outcome considering the higher costs involved (�nancial, language, and distance). Finally, there is a slightly higher perception of corruption in the health sector in areas with strong migration to Portugal. With such a pro�le, it is desirable to control for local educational attainment in regressions evaluating the impact of migration by destination country. The effects of education are not visible at the aggregate level, when the impact of all migrants to different destinations is considered (see table 4, column 3-5). Only when the analysis is decomposed into current and return migrants does the impact become apparent. The most striking dimension of selectivity in migration, college education, also has the greatest impact on the results. After controlling for tertiary education, the impact of return migration from Portugal becomes signi�cantly negative (see table 5, columns 3 –5)—this may be related to the fact that Cape Verde had no universities until 1995 and that Portugal was the usual destination for Cape Verdeans seeking a college education. Apart from this strong impact on the coef�cient on return migration from Portugal, the estimated results are not very sensitive to this or other dimensions along which migrants seem to self-select when choosing a migration destination. This result indicates that self-selection is not likely to underlie the impact of migration on the demand for political accountability. Indeed, migrant assimila- tion of the accountability norms in the destination country is a better expla- nation for this impact.10 In the latest Transparency International (2009) 10. This is consistent with the �ndings of Fidrmuc and Doyle (2004) and Spilimbergo (2009), which also provide evidence supporting migrant assimilation effects in the destination country. Fidrmuc and Doyle (2004) focus on Czech and Polish migrants and also �nd that self-selection (by political attitudes and economic characteristics) is not likely to explain migrants’ political attitudes. Spilimbergo (2009) describes how the political attitudes of migrants differ depending on the political characteristics of the destination countries. 98 THE WORLD BANK ECONOMIC REVIEW T A B L E 6 : Descriptive Statistics for Survey Respondents in Areas with Strong Migration to Portugal and Areas with Strong Migration to the United States Strong migration to Strong migration to Variable Portugal United States Male – 0.0001 0.0726 (0.0500) (0.0732) Age 1.18987 0.8926 (1.4803) (2.0500) Individual labor income – 82.8924 19.6443 (26.9850)*** (41.7703) Number of children 0.1787 – 0.3183 (0.2490) (0.2903) Household asset ownership 0.1252 – 0.0280 (0.0317)*** (0.0564) Trust in Oxford University 0.2551 – 0.1278 (0.1089)** (0.1679) Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Habit of posting – 0.3219 – 0.2356 (0.1967) (0.2681) Average private consumption expenditure per 0.0077 0.0316 capita in locality (0.0058) (0.0112)*** Fraction of residents working in agriculture in 0.0322 0.0017 locality (0.0043)*** (0.0047) Fraction of residents working in construction in 0.0227 – 0.0181 locality (0.0029)*** (0.0026)*** Fraction of residents working in retail trade in – 0.0057 – 0.0119 locality (0.0024)** (0.0021)*** Fraction of households receiving international 0.0028 0.0279 remittances in locality (0.0020) (0.0044)*** Ratio of residents completing relative to residents – 0.0239 0.0886 not completing 9 years of schooling in locality (0.0229) (0.0479)* Ratio of residents completing relative to residents – 0.0265 0.0448 not completing 12 years of schooling in locality (0.0110)** (0.0200)** Ratio of residents completing relative to residents – 0.0097 0.0172 not completing 15 years of schooling in locality (0.0031)*** (0.0058)*** Perceived corruption in health sector 0.4074 – 0.1839 (0.2149)* (0.2659) Perceived corruption in education sector – 0.0358 – 0.3119 (0.1845) (0.2377) * Signi�cant at the 10 percent level; ** signi�cant at the 5percent level; *** signi�cant at the 1percent level. Note: Numbers in parentheses are robust standard errors clustered at the locality level. Strong migration to a certain destination is de�ned as migrants to that destination representing at least 5 percent of the resident population. Table shows mean difference relative to areas where migration to the same destination is not strong. Source: Authors’ analysis based on authors’ survey. cross-country governance ranking, the United States places 19th, Portugal ranks 35th and Cape Verde 46th. This evidence can be interpreted to show that the experience of emigrants to the United States is more conducive to promoting demand for better governance than that of emigrants to Portugal. Also, the Batista and Vicente 99 negative impact of return migrants from Portugal should be viewed in the context of the baseline destinations against which migrants to the United States and Portugal are being compared; those are mostly European countries, such as France and the Netherlands, which rank closer to the United States in govern- ance than to Portugal or Cape Verde. Robustness Check: Potential Endogeneity and Instrumental Variable Estimation Despite the supportive evidence that observable self-selection does not seem to explain the estimated results, there may still be endogeneity concerns related to potential unobserved heterogeneity and locality-level omitted variables. To examine these concerns, the baseline regressions are reestimated using two sets of instrumental variables: �ve-year lagged local migrant stocks based on the full migration history available for all household members in the survey, and Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 external sources in destination countries (unemployment rates, nominal GDP per capita, and GDP growth rates in the United States and Portugal) in the 10 years before the survey. These variables are aggregated using a weighted sum in which the weight is the �ve-year lagged local migrant stock to each destination relative to the �ve- year lagged overall stock of migrants to that destination in each 10-year period. This weight can be understood as a �ve-year lagged proxy for migration net- works in the destination country, which combined with macro information from the destination country, should constitute an exogenous source of vari- ation for migration, enabling identi�cation of the coef�cients of interest. Note that the weighting procedure guarantees enough variation to identify the effects of interest at the locality level. The second set of instruments also enables testing for overidenti�cation in all three estimated speci�cations. This second set of instruments is also stronger—lagged instrument strength could be a problem for certain regressions, as displayed in table 7, column 7. After �nding that the instruments used seem strong and exogenous in all possible speci�cations (see table 7), it is also reassuring to observe that the esti- mates are not substantially different from those obtained using probit methods. This �nding points to the small importance of any endogeneity concerns at the local level, after controlling for all relevant covariates. Robustness Check: Alternative Measures of the Demand for Accountability One additional potential concern with the analysis is that the postcard exper- iment might not be exactly measuring a desire for political accountability. To strengthen the contention that that is the case, a survey variable is used that asks respondent directly whether they agree or disagree (on a 1 –7 scale) with the statement: “As a common citizen of Cape Verde, I believe I should require competence in the public services (health centers, schools, courts, police) that are aimed at my needs.� 100 T A B L E 7 . Probability of Mailing Voting Postcard (Columns 1, 2, 4, 5, 7, and 8) and Probability of Self-reported Demand for Accountability (Columns 3, 6, and 9): Instrumental Variable Estimates Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) Proportion of international 1.0298 1.4380 1.8467 THE WORLD BANK ECONOMIC REVIEW migrants (relative to (0.3895)*** (0.3472)*** (1.0169)* residents) in locality Proportion of international – 1.0262 – 0.4735 – 0.3474 migrants to Portugal (relative (1.8587) (1.5247) (2.8333) to residents) in locality Proportion of international 2.7575 3.3261 7.7153 migrants to United States (0.9659)*** (0.9731)*** (1.5838)*** (relative to residents) in locality Proportion of current 1.2976 1.6291 0.7949 international migrants to (9.0557) (1.4478) (2.4863) Portugal (relative to residents) in locality Proportion of current 3.3171 – 0.6921 4.4871 international migrants to (5.6915) (5.8877) (6.7337) United States (relative to residents) in locality Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Proportion of international – 2.7566 – 4.8785 – 0.8931 return migrants to Portugal (11.2768) (3.6941) (8.3697) (relative to residents) in locality Proportion of international 1.9758 4.9559 9.6918 return migrants to United (3.6264) (3.2493) (4.8695)** States (relative to residents) in locality Instrument seta A B B A B B A B B F-statistics on excluded 394.1 28.1 28.2 9.5; 124.2 9.2; 52.3 9.2; 52.4 2.1; 11.2; 13.9; 11.5; 13.9; 11.6; instruments in �rst stage 129.5; 180.7; 178.2; regressions 378.1 1156.5 1169.1 Overidenti�cation test ( p-value) NA 0.18 0.42 NA 0.33 0.76 NA 0.15 0.38 Controls included Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of observations 452 452 451 452 452 451 452 452 451 * Signi�cant at the 10 percent level; ** signi�cant at the 5percent level; *** signi�cant at the 1percent level. Note: Numbers in parentheses are robust standard errors clustered at the locality level. All regressions include the same controls as the baseline regression in column 5 of table 2. a. Instrument set A includes �ve-year lagged regressors of interest. Instrument set B uses macroeconomic variables at destination weighted by �ve-year lagged local migration stock size indicators, as described in the text. Batista and Vicente 101 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 102 THE WORLD BANK ECONOMIC REVIEW This self-reported measure of demand for better governance is used to verify whether the determinants of postcard voting behavior are similar. The results are reassuring. The sign and signi�cance of the main estimated coef�cients remain stable except for the effect of the proportion of international migrants in a locality on the probability of mailing a postcard, which has a p-value of only 12.6 percent (see table 2, column 7). The impact of migration to the United States is also still strongly positive and signi�cant and that of migration to Portugal is statistically insigni�cant (see table 4, columns 8 and 9). The same results hold when disaggregated by current and return migration status: return migration from the United States is a powerful positive determinant of the demand for accountability, whereas current migration and return migration to Portugal are not statistically signi�cant (see table 5, columns 8 and 9). Overall, the most salient outcome when of using self-reported survey data instead of the postcard behavioral measure is that the magnitude of the esti- mated effects is much larger, an outcome that could be related to survey Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 respondents’ desire to conform to the perceived anticorruption message of the survey (conformity bias). Mechanics 3: Direct and Social Effects of Local Migration In summary, the evidence points to international emigration to countries with good governance (in particular, to the presence of return migrants) as promot- ing demand for political accountability in the origin country. It is important to emphasize that the focus here is on the impact of locality- level migration. The variable used, the proportion of international migrants within the household’s spatial area of residence, can be understood as a proxy for the frequency of potential interactions between migrants and residents (who are not necessarily migrants and who do not necessarily have a return or current migrant in the household). The larger this proportion, the more likely such interactions will be and the more likely that people in the locality are more open to demanding accountability. The effects of local migration are both direct and indirect. Return migrants, for instance, should have both a direct and an indirect impact on households in the locality. Current migrants can also have indirect effects through communi- cations with their network of friends and family back home. The empirical question left unanswered is that of the different magnitude of the direct and indirect effects identi�ed at the local level. If additional data on migrant networks became available, this could be an important way forward in the literature. V. C O N C L U D I N G R E M A R K S This article contributes to the understanding of a largely unmeasured but important potential effect of international emigration: its impact on insti- tutional quality, a determinant of economic growth. Batista and Vicente 103 The �ndings point to an overall positive impact of international emigration on the demand for improved political accountability in the country of origin. In particular, the results emphasize the importance of the migration destination country: the impacts are stronger for migration to countries with better govern- ance. The impacts are also stronger for return migrants than for current migrants, who can only indirectly influence their relationship networks in the home country. International emigration likely affects the supply side of domestic political institutions as well as the demand side, a part of the lively ongoing “brain drain� vs. “brain gain� debate. Total effects could presumably be negative if there were positive selection in current emigration flows and positive if skilled migrants return. This is an empirical question to be answered by future research. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 REFERENCES Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2005. “Institutions as a Fundamental Cause of Long-Run Growth.� In Handbook of Economic Growth, ed. Philippe Aghion, and Steven Durlauff. Amsterdam: North-Holland. Alesina, Alberto, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and Romain Wacziarg. 2003. “Fractionalization�. Journal of Economic Growth 8 (2): 155– 94. Alesina, Alberto, and Paola Giuliano. 2009. “Family Ties and Political Participation.� IZA Discussion Paper 4150. Institute for the Study of Labor, Bonn. Beine, Michel, Frederic Docquier, and Hillel Rapoport. 2008. “Brain Drain and LDCs’ Growth: Winners and Losers�. Economic Journal 118: 631–52. Batista, Catia, Tara McIndoe-Calder, and Pedro C. Vicente. 2010. “Return Migration and Entrepreneurship in Mozambique�. Trinity College, Dublin. Batista, Catia, Aitor Lacuesta, and Pedro C. Vicente. Forthcoming. “Testing the ‘Brain Gain’ Hypothesis: Micro Evidence from Cape Verde.� Journal of Development Economics. ´ de Docquier, Fre ´ ric, Elisabetta Lodigiani, Hillel Rapoport, and Maurice Schiff. 2010. “Emigration and the Quality of Home Country Institutions�. Catholic University of Louvain, University of Luxembourg, Harvard University, and World Bank. ´ de Docquier, Fre ´ ric, and Abdeslam Marfouk. 2006. “International Migration by Education Attainment, 1990–2000.� In International Migration, Brain Drain and Remittances, ed. Caglar Ozden, and Maurice Schiff. New York, NY: Palgrave Macmillan. Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper Collins. Dustmann, Christian, and Oliver Kirchkamp. 2003. “Migration Duration and Activity Choice after Re-Migration�. Journal of Development Economics 67: 351–72. Edwards, A., and M. Ureta. 2003. “International Migration, Remittances, and Schooling: Evidence from El Salvador.� Journal of Development Economics 72: 429– 61. Fidrmuc, Jan, and Orla Doyle. 2004. “Voice of the Diaspora: An Analysis of Migrant Voting Behaviour�. CEPR Discussion Paper 4619. Center for Economic and Policy Research, Washington, DC. Gould, D. M. 1994. “Immigrant Links to the Home Country: Empirical Implications for U.S. Bilateral Trade Flows,� Review of Economics and Statistics 76 (2): 302–16. Freedom House. 2010. Freedom of the Press 2010. www.freedomhouse.org/template.cfm?page=16 104 THE WORLD BANK ECONOMIC REVIEW Heston, Alan, Robert Summers, and Bettina Aten. 2006. Penn World Table Version 6.2. Center for International Comparisons of Production, Income, and Prices at the University of Pennsylvania, Philadelphia. Hirschman, Albert. 1970. Exit, Voice, and Loyalty: Responses to Decline in Firms, Organizations, and States. Cambridge, MA: Harvard University Press. ¸a INE (Instituto Nacional de Estatistica). 2002. Recenseamento Geral da Populac ¸a ˜ o e da Habitac ˜ o. Cidade da Praia, Cabo Verde: INE. ´ rito a ———. 2004. Inque ` s Despesas e Receitas Familiares. Cidade da Praia, Cabo Verde: INE. Iranzo, S., and G. Peri. 2009. “Migration and Trade: Theory with an Application to the Eastern-Western European Integration.� Journal of International Economics 79 (1): 1–19. Javorcik, Beata, C ¨ zden, Mariana Spatareanu, and Cristina Neagu. Forthcoming.“Migrant ¸ aglar O Networks and Foreign Direct Investment�. Journal of Development Economics. Kugler, Maurice, and Hillel Rapoport. 2007. “International Labor and Capital Flows: Complements or Substitutes?�. Economics Letters 92 (2): 155 –62. Li, Xiaoyang, and John McHale. 2009. “Emigrants and Institutions�. University of Michigan, Ann Arbor, and National University of Ireland, Galway. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Merlo, Antonio. 2006. “Whither Political Economy? Theories, Facts, and Issues.� In Advances in Economics and Econometrics, Theory, and Applications: Ninth World Congress of the Econometric Society, vol. I, ed. R. Blundell, W. Newey, and T. Persson. Cambridge, UK: Cambridge University Press. Mesnard, Alice, and Martin Ravallion. 2006. “The Wealth Effect on New Business Startups in a Developing Economy.� Economica 73: 367–92. Minier, Jenny. 2001. “Is Democracy a Normal Good?� Southern Economic Journal 67 (4): 996– 1009. Rauch, J., and Vitor Trindade. 2002. “Ethnic Chinese Networks in International Trade.� Review of Economics and Statistics 84: 116– 30. Reporters without Borders. 2009. Press Freedom Index 2009. http://en.rsf.org/ Riker, William H., and Peter C. Ordeshook. 1968. “A Theory of the Calculus of Voting.� American Political Science Review 62: 25– 42. Spilimbergo, Antonio. 2009. “Foreign Students and Democracy.� American Economic Review 99 (1): 528 –43. Transparency International. 2009. Transparency International Annual Report 2009. www.transparency .org . Tullock, Gordon. 1967. Toward a Mathematics of Politics. Ann Arbor: University of Michigan Press. World Bank. 2006. World Development Indicators. Washington, DC: World Bank. ———. 2011. Worldwide Governance Indicators. http://info.worldbank.org/governance/wgi/ Yang, Dean. 2008. “International Migration, Remittances, and Household Investment: Evidence from Philippine Migrants’ Exchange Rate Shocks.� The Economic Journal 118: 591–630. What Explains the Price of Remittances? An Examination Across 119 Country Corridors ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı Remittances are a substantial source of external �nancing for developing countries that influence many aspects of their development. Though research has shown that remittances are both expensive and price sensitive, little is known about what explains their price. Newly gathered data across 119 country pairs or corridors are used to explore the factors associated with the price of remittances. Corridors with larger numbers of migrants and more competition among providers are found to exhibit lower prices for remittances, when average prices across all types of remittance service providers are considered. Corridors with lower barriers to access banking services and Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 broader regulation of remittance service providers also have lower prices. Remittance prices are higher in richer corridors and in corridors with greater bank participation in the remittance market. Few signi�cant differences emerge when results are compared across banks and, separately, across money transfer operators. However, estimations for Western Union, a leading player in the remittances business, suggest that its prices are less sensitive to competition. JEL classi�cation: F22, F24. In 2008, remittances to developing countries reached $328 billion, more than twice the amount of of�cial aid and over half of foreign direct investment flows (World Bank 2009a). Numerous studies have shown that remittances have a positive and signi�cant impact on economic development along multiple dimensions, including poverty alleviation, education, entrepreneurship, infant Thorsten Beck (corresponding author; T.Beck@uvt.nl) is a professor of economics and CentER fellow and chair of the European Banking Center at Tilburg University. Marı ´nez Perı ´a Soledad Martı ´a (mmartinezperia@worldbank.org) is a senior economist in the Finance and Private Sector Development Research Group of the World Bank. The authors thank Diego Anzoategui and Subika Farazi for excellent research assistance. They received helpful comments from participants at the Second International Conference on Migration and Development and the International Conference on Diaspora for Development, as well as from World Bank colleagues in the Finance and Private Sector Development Research Group and the Payment Systems Unit. The authors are particularly grateful to the journal editor and to three anonymous referees for constructive comments and suggestions. This article’s �ndings, interpretations, and conclusions are entirely those of the authors and do not necessarily represent the views of the organizations with which they are af�liated. THE WORLD BANK ECONOMIC REVIEW, VOL. 25, NO. 1, pp. 105 –131 doi:10.1093/wber/lhr017 Advance Access Publication May 23, 2011 # The Author 2011. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 105 106 THE WORLD BANK ECONOMIC REVIEW mortality, and �nancial development.1 Hence, understanding the market for remittance transactions can be critical for promoting the development process in many low-income countries. Remittance transactions are known to be expensive, with estimates averaging 10 percent of the amount sent (World Bank 2009b). At the same time, these costs are widely dispersed across corridors and range from 2.5 percent to 26 percent. Furthermore, case studies have shown that remittance flows are very sensitive to prices and are likely to increase substantially as prices fall. For example, Gibson, McKenzie, and Rohorua (2006) estimate a 22 percent price elasticity in the New Zealand–Tonga corridor and calculate that lowering the fees to the levels found in the most competitive corridors would raise remit- tances by 28 percent. Using a randomized experiment, Aycinena, Martinez, and Yang (2009) estimate that a $1 lower fee in the United States –El Salvador corridor would boost remittances $25 a month from an average of $290. Because remittances are important for economic development and appear to Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 be sensitive to price, lowering the price has become a priority for policymakers. At the L’Aquila 2009 G-8 Summit, leaders pledged to reduce the price of remittances by half (from 10 to 5 percent) in �ve years.2 Yet, little is known empirically about what explains the price of remittances.3 Are high prices due mainly to factors in the sending or the receiving country? Are high prices related to socioeconomic factors, industry market structure, or government pol- icies and regulations that affect remittance service providers and the mark-ups they are able to charge? Are there signi�cant differences between banks and money transfer operators (MTOs)? Explaining the variation in prices is thus of interest for academics and policymakers alike. Using a new dataset assembled by the World Bank Payment Systems Group on the price of remittances across 119 country pairs or corridors (Remittance Prices Worldwide database (World Bank 2009b), this article explores the 1. For example, see Adams and Page (2003), Adams (2005), IMF (2005), Lopez-Co ´ rdova (2005), Maimbo and Ratha (2005), and Taylor, Mora, Adams and Lopez-Feldman (2005) for studies on the impact of remittances on poverty. Studies such as Cox-Edwards and Ureta (2003), Hanson and Woodruff (2003), Lo ´ rdova (2005), and Yang (2008) �nd that by helping to relax household ´ pez-Co constraints, remittances are associated with improved schooling outcomes for children. Remittances have also been shown to promote entrepreneurship (see Massey and Parrado 1998; Maimbo and Ratha 2005; Yang 2008; Woodruff and Zenteno 2007). Furthermore, a number of studies on infant mortality and birth weight have documented that, at least in the Mexican case, migration and remittances help lower infant mortality and are associated with higher birth weight among children in households that receive remittances (see Kanaiaupuni and Donato 1999; Hildebrandt and McKenzie 2005; Duryea et al. 2005; and Lo ´ rdova 2005). Aggarwal, Demirgu ´ pez-Co ¨c¸ -Kunt, and Martinez Peria (2010) show that remittances can have a positive impact on �nancial development. 2. Paragraph 134, page 49 of the L’Aquila 2009 G8 Summit. http://www.g8italia2009.it/static/ G8_Allegato/G8_Declaration_08_07_09_�nal,0.pdf. 3. Orozco (2006) and Freund and Spatafora (2008) are the exception, but their data is limited to a few countries or a few providers. While Orozco’s work focuses exclusively on Latin America, the Freund and Spatafora study analyzes only the costs of remittances sent from the United States and the United Kingdom exclusively via MoneyGram or Western Union to 66 countries. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 107 factors associated with remittance prices in 2009.4 It studies corridors that include 13 major remittance sending countries and 60 receiving countries representing approximately 60 percent of total remittances to developing countries. Using data at the corridor level permits bilateral analysis of prices rather than analysis of prices aggregated at the receiving or sending country level only. Furthermore, unlike previous studies focusing on a certain type of remittance service provider (usually the largest international MTOs), the analy- sis here considers the largest providers, whatever the type, in each corridor.5 And by averaging across all types of providers and across each type of provider (banks and MTOs) separately, the factors associated with the price of remit- tances can be compared across different types of institutions. Finally, analyzing the prices charged by Western Union across 98 corridors (80 percent of the sample) alleviates concerns about bias due to differences across �rms and thus sheds light on the factors correlated with the prices charged by a leading remit- tance service provider with worldwide operations. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 The analysis distinguishes three groups of variables that can be associated with cross-corridor variation in the price of remittances. One is the impact of government policies—including exchange rate policies, capital controls, and regulation of remittance service providers—that can influence the price of remittances through their impact on the cost structure of remittance service providers. The second is the role of factors that might affect the ability of remittance service providers to increase their mark-up, such as extent of competition, market structure, and level of education of the migrant population. The third is the role of socioeconomic characteristics in sending and receiving countries that might influence fees through their impact on the cost structure of remittance service providers and on provi- ders’ ability to raise the mark-up. Estimations of the price of remittances across all types of remittance service providers show that prices are consistently lower in corridors with a larger number of migrants and more competition and in corridors with lower access barriers to the banking system and broader regulation of remittance service providers. Remittance prices are higher in richer corridors and in corridors with a higher share of banks among providers. The prices of sending remit- tance using banks or MTOs are associated with similar factors. Western Union prices appear to be less sensitive to competition, perhaps a reflection of the �rm’s market power. This article is related to the literature on the price of banking services. Beck, Demirgu ¸ -Kunt, and Martı ¨c ´nez Perı ´a (2008) document large cross-country 4. The original World Bank database for the period analyzed here contains information on 134 corridors. From that total, 13 corridors (where Russia is the sending country) are missing exchange rate data and 2 other corridors are missing information for some explanatory variables. 5. On average, 8– 10 providers are included for each corridor. In some corridors, primarily those including the United States and Spain as sending countries, the number of providers surveyed exceeds 10. 108 THE WORLD BANK ECONOMIC REVIEW variation in the costs to customers of opening and maintaining bank accounts and in the fees for using automated teller machines and for transferring funds, �nding that �rms report lower �nancing constraints in countries with lower costs of �nancial services. Freund and Spatafora (2008) and Orozco (2006) also present data on remittance prices, but for few countries and providers and not at the corridor level. In a broader sense, this article is also related to the literature on bank inter- est rate spreads (the differences between deposit and lending rates), with higher spreads indicating more expensive banking services. Both institution-speci�c characteristics, such as size and ownership, and market and country character- istics, such as competition and the legal and institutional framework, have been shown to be signi�cant predictors of interest rate spreads (see Demirgu ¸ -Kunt, Laeven and Levine 2004; Laeven and Majnoni 2005; and ¨c Beck 2007 for a general discussion). This article is a �rst exploration of corridor variation in the price of remit- Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 tances and is subject to two important caveats. First, as a pure cross-sectional analysis, it is potentially subject to reverse causation and omitted-variable biases. Hence, only limited, if any, inference can be made on causality.6 Second, the analysis is also limited in scope since it includes data only from formal providers of remittance services. By some estimates, at least a third of remittances are sent through informal channels (Freund and Spatafora 2008). These limitations notwithstanding, the article offers interesting evidence that should stimulate further data collection and analysis. The article is organized as follows. Section I describes the data on the price of remittances. The empirical methodology is in section II and the results are in section III. Section IV summarizes the �ndings and recommends further research. I . D ATA ON THE PRICE OF R E M I T TA N C E S The data on the price of remittances are from the Remittance Prices Worldwide database, a recent survey of remittance service providers conducted by the Payment System Unit of the World Bank in the �rst quarter of 2009 (World Bank 2009b).7 The price of remittances consists of a fee component and an exchange rate spread component. The World Bank dataset covers 14 sending and 72 receiving countries. However, because spread information is missing for remittances from the Russian Federation and data are missing for some explanatory variables, the focus is on 119 corridors, at most, including 6. Most of the variables, however, are likely to be exogenous to remittance prices, including migration flows, distance, and even banking market structure, given the small share of bank pro�ts stemming from remittances. 7. Since then, the data have been updated, and prices are now available through the �rst quarter of 2010. However, because data for most of the correlates used in this analysis are not updated with the same frequency, the panel dimension of the data could not be exploited. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 109 13 sending countries and 60 receiving countries.8 In most cases, data cover the prices from the main sending location for the corridor in question to the capital city or most populous city in the receiving market. Data were collected by interviewers posing as customers and by contacting individual �rms. Within each corridor, the data were gathered on the same day to control for exchange rate fluctuations and other changes in fee structures. In general, price data were collected for 8–10 major service providers in each cor- ridor, including the main MTOs and banks active in the market.9 Companies surveyed in each segment were selected to cover the maximum remittance market share possible.10 Since the dataset does not provide information on the market shares of each provider, it is not possible to compute weighted averages. Hence, the regression analysis uses both the simple average and the median prices calculated across all providers in a corridor as dependent vari- ables.11 Results are reported using only the simple averages, however, because average and median prices are highly correlated (0.96). Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Prices based on two amounts are available per corridor: the local equiv- alent of $200 and the local equivalent of $500. Since previous studies have found that a typical remittance transaction involves sending close to $200, the analyses are based on the prices associated with this amount.12 Furthermore, the prices of sending $200 and $500 (expressed as a percen- tage of the amount sent) are highly correlated (0.91), so the results do not vary signi�cantly.13 Figure 1 illustrates the variation in average prices across the 119 corridors, calculated across all surveyed remittance service providers in each corridor. The average remittance prices are lowest in the Saudi Arabia–Pakistan corridor (2.5 percent of $200) and highest in the Germany–Croatia corridor (25.8 percent). Averaged across all corridors and providers, the price is 10.2 percent. There is considerable heterogeneity in prices even when the same sending or remittance receiving country is considered. Prices of remittances sent from the United States vary from 3.7 percent to Ecuador to 14.1 percent to Thailand (�gure 2). Remittance prices to India vary from 3.1 percent from Saudi Arabia to 13.3 percent from Germany (�gure 3). This variation underlines the 8. The full data are available at www.remittanceprices.org. Data on exchange rate spreads are also missing for some institutions in Germany, France, and Japan. These institutions are excluded from the calculations of the average remittances costs from those countries. 9. The number of respondents by corridors varies depending on the number of �rms active in the corridor. Some corridors (like the Spain– China corridor) include only two �rms, while others (like the United States– Mexico corridor) go as high as 18. 10. No more information is provided on how �rms were selected. For a discussion of the methodology, see http://remittanceprices.worldbank.org/Methodology/. 11. A priori, it is not clear how having weighted averages instead of simple averages would change the estimations. This problem is interpreted as a potential case of measurement error in the dependent variable, which should not bias the estimates but would affect their ef�ciency. 12. Freund and Spatafora (2008) use the same amount in their study. 13. These results are available on request. 110 THE WORLD BANK ECONOMIC REVIEW F I G U R E 1. Average Price of Remittances Sent Across 119 Migration Corridors Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Source: Remittance Prices Worldwide database. F I G U R E 2. Average Price of Remittances from the United States to 22 Receiving Countries Source: Authors’ analysis based on data from World Bank (2009b). ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 111 F I G U R E 3. Average Price of Remittances to India from Eight Sending Countries Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Source: Authors’ analysis based on data from World Bank (2009b). importance of analyzing the price of remittances at the corridor rather than at the sending or receiving country level. Remittance prices also vary across provider types (table 1). On average, banks charge substantially higher fees (12.4 percent) than do MTOs (8.8 percent). This differential does not control for the fact that banks and MTOs are not active in all corridors and that different banks and different MTOs are active in different corridors. When the analysis focuses on corri- dors where both types of institutions are active, average prices for banks exceed those for MTOs in 43 of the 63 corridors. Furthermore, when prices at the provider level are regressed on a set of corridor dummy variables and a bank dummy variable, bank prices average 3 percentage points higher than MTO fees. Western Union’s prices (10.8 percent) are slightly higher than the average for all MTOs (8.8 percent). Western Union prices also exhibit high cross-corridor variation, ranging from 2.7 percent in the Saudi Arabia –Yemen corridor to 29.9 percent in the United Kingdom–Albania corridor (�gure 4). II. E M PI R I CA L ME T H O D O LO GY To examine the determinants of remittance prices, the average price of sending remittances, Pij, is regressed on a set of sending and receiving country T A B L E 1 . Summary Statistics and Data Sources 112 Number of Description Abbreviation observations Mean Median Date Source All providers 119 10.24 9.47 2009 Remittance Prices Worldwide (World Bank 2009b) Banks’ average prices (% of $200) Banks avg price 70 12.40 11.78 2009 Remittance Prices Worldwide (World Bank 2009b) Money transfer operators’ average prices MTOs avg price 112 8.78 8.07 2009 Remittance Prices Worldwide (World (% of $200) Bank 2009b) Western Union’s average prices (% of $200) WU avg price 98 10.84 10.33 2009 Remittance Prices Worldwide (World Bank 2009b) Log of number of migrants in the corridor Log bil mig 119 11.61 11.88 2006 Ratha and Shaw (2007) Log of GDP per capita in receiving country Log GDP rec 119 7.15 7.40 Average for World Development Indicators (World 2006-07 Bank 2009c) Log of GDP per capita in sending country Log GDP send 119 10.02 10.17 Average for World Development Indicators (World THE WORLD BANK ECONOMIC REVIEW 2006-07 Bank 2009c) Dummy for pegged exchange rate or Peg rec 119 0.33 0.00 2008 Annual Report on Exchange dollarization Arrangements and Restrictions (IMF) Share rural population (%) in receiving Rural pop rec 119 49.48 50.22 Average for World Development Indicators (World country 2006-07 Bank2009c) Share rural population (%) in sending Rural pop send 119 20.56 18.99 Average for World Development Indicators (World country 2006-07 Bank 2009c) Log of distance between sending and Log dist 119 8.22 8.39 - Distances database (CEPII 2010) receiving countries Common language Com Language 119 0.44 0.00 - Distances database (CEPII 2010) Controls on remittances in receiving country Ctrl remit rec 105 0.21 0.00 2007 Annual Report on Exchange Arrangements and Restrictions (IMF 2007) Share of educated (secondary or tertiary Mig educ 88 54.14 53.47 2000 Database on Immigrants and education) migrants Expatriates (OECD 2010) Branches per capita (100,000 people) in Brchs pc rec 89 6.62 6.30 2008 Database on Access to Financial receiving country Services (World Bank 2007) Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Branches per capita (100,000 people) in Brchs pc send 119 33.64 30.86 2008 Database on Access to Financial sending country Services (World Bank 2007) Index of regulations for remittance providers Reg rec 91 2.20 2.00 2008 World Bank Global Payment Systems in receiving country Survey (World Bank 2008) Index of regulations for remittance providers Reg send 119 2.25 2.00 2008 World Bank Global Payment Systems in sending country Survey (World Bank 2008) H-statistic for banking sector in receiving H-Stat rec 111 0.54 0.52 1994-2006 Bankscope database (Bureau van Dijk country 2009) and author calculations H-statistic for banking sector in sending H-Stat send 119 0.52 0.50 1994-2006 Bankscope database (Bureau van Dijk country 2009) and author calculations Number of respondents per corridor Resp per corr 119 7.97 8.00 2009 Remittance Prices Worldwide (World Bank 2009b) Share of banks per corridor (%) % of banks 119 31.35 20.00 2009 Remittance Prices Worldwide (World Bank 2009b) Index importance of banks receiving country Bank imp rec 90 5.32 6.00 2007/8 World Bank Global Payment Systems Survey (World Bank 2008) Index importance of banks sending country Bank imp send 108 5.11 5.50 2007/8 World Bank Global Payment Systems Survey (World Bank 2008) Log of bilateral trade Log Bil Trade 111 21.90 21.87 Average for Direction of Trade Statistics (IMF 2006-07 2009) Savings account annual fee (% of GDP pc) Annual fee rec 85 0.55 0.07 2003 Beck, Demirgu ¸ -Kunt, Martı ¨c ´a ´nez Perı receiving country (2008) Thorsten Beck and Marı Savings account annual fee (% of GDP pc) Annual fee send 72 0.12 0.00 2003 Beck, Demirgu ¸ -Kunt, Martı ¨c ´a ´nez Perı sending country (2008) Minimum amount to open a savings account Min amnt open rec 85 7.36 1.59 2003 Beck, Demirgu ¸ -Kunt, Martı ¨c ´a ´nez Perı (% of GDP pc) receiving country (2008) Minimum amount to open a savings account Min amnt open send 72 0.11 0.00 2003 Beck, Demirgu ¸ -Kunt, Martı ¨c ´a ´nez Perı (% of GDP pc) sending country (2008) ´a Soledad Martı Source: Authors’ analysis based on data referenced in the table. 113 ´nez Perı ´a Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 114 THE WORLD BANK ECONOMIC REVIEW F I G U R E 4. Average Price of Remittances sent through Western Union Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Source: Authors’ analysis based on data from World Bank (2009b). characteristics and on some corridor-speci�c variables captured by the matrix X in equation (1): Pij ¼ a0 þ a1 Sending country factorsi þ a2 Receiving country factors j þ a3 Xij þ uij ð1Þ where Pij is the price of sending $200 from country i to country j expressed as a percentage of $200. All explanatory variables are lagged relative to the price variable. Since this does not completely rule out reverse causation or endogene- ity bias, the results are interpreted as associations rather than as causal impacts. Table 1 provides the summary statistics and data sources for each vari- able, and table 2 reports correlations across the main variables. Government Policies Equation (1) captures an array of factors that might be correlated with remit- tance prices through their association with the costs faced by remittance service providers and the mark-up providers can charge over their marginal costs. First, it controls for different government policies relating to the exchange rate, T A B L E 2 . Correlation Matrix: Pairwise Correlations among Main Variables Min Log Log Log Rural Rural Brchs Brchs Resp Share Log Annual Annual amnt Avg bil GDP GDP Peg pop pop Com pc pc Reg. Reg. H-Stat H-Stat per of bil fee fee open Price mig rec send rec rec send language rec send rec send rec send corr banks trade rec send rec Avg price 1 Log bil mig 2 0.38** 1 Log GDP rec 0.09 0.26** 1 Log GDP send 2 0.14 0.32** 0.18** 1 Peg rec 2 0.08 2 0.10 2 0.12 2 0.14 1 Rural pop rec 0.03 2 0.12 2 0.75** 2 0.19** 0.07 1 Rural pop send 0.36** 2 0.10 0.09 2 0.51** 0.06 2 0.09 1 Com language 2 0.20** 0.15 2 0.11 0.02 2 0.04 0.05 2 0.29** 1 Brchs pc rec 0.05 0.16 0.58** 0.04 0.20 2 0.59** 0.12 2 0.19 1 Brchs pc send 2 0.11 0.12 0.21** 0.09 0.01 2 0.26** 0.22** 2 0.18** 0.19 1 Reg. rec 0.04 0.08 0.18 0.10 0.01 2 0.25** 0.11 2 0.06 2 0.17 0.14 1 Reg. send 2 0.51** 0.03 2 0.14 0.12 0.03 0.06 2 0.66** 0.19** 2 0.15 0.07 2 0.04 1 H-Stat rec 2 0.21** 0.02 0.20** 0.13 2 0.02 2 0.03 2 0.12 2 0.03 0.05 2 0.02 0.00 0.06 1 H-Stat send 2 0.27** 0.35** 0.28** 0.56** 2 0.14 2 0.33** 0.05 2 0.07 0.16 0.55** 0.22** 2 0.11 0.16 1 Resp per corr 2 0.33** 0.35** 0.26** 0.18 0.00 2 0.22** 2 0.20** 0.30** 0.20 0.30** 0.13 0.22** 0.12 0.19** 1 Share of banks 0.55** 2 0.08 0.05 2 0.46** 0.07 0.09 0.60** 2 0.22** 0.17 2 0.26** 0.01 2 0.63** 2 0.16 2 0.46** 2 0.17 1 Log bil Trade 0.02 0.36** 0.35** 0.25** 2 0.25** 2 0.08 0.03 0.00 0.04 2 0.12 0.21 2 0.17 0.12 0.25** 0.18 0.11 1 Annual fee rec 0.29** 2 0.51** 2 0.33** 2 0.25** 0.07 0.28** 0.06 0.26** 2 0.18 2 0.17 2 0.02 2 0.07 2 0.10 2 0.26** 2 0.14 0.10 2 0.27** 1 Thorsten Beck and Marı Annual 0.29** 2 0.42** 2 0.21 2 0.94** 0.11 0.25** 0.65** 0.10 2 0.17 2 0.30** 2 0.12 2 0.50** 2 0.10 2 0.55** 2 0.28** 0.53** 2 0.11 0.40** 1 fee send Min amnt 2 0.01 2 0.23** 2 0.55** 0.01 0.17 0.31** 2 0.12 0.22** 2 0.24** 2 0.16 0.11 0.08 2 0.10 2 0.12 2 0.11 2 0.18 2 0.41** 0.40** 0.07 1 open rec Min amnt 0.39** 2 0.44** 2 0.22 2 0.93** 0.15 0.27** 0.56** 0.17 2 0.16 2 0.38** 2 0.14 2 0.47** 2 0.13 2 0.72** 2 0.20 0.61** 2 0.10 0.46** 0.96** 0.09 open send **Signi�cant at least at the 5 percent level. ´a Soledad Martı Note: For de�nitions, see table 1. Only the main variables are included in this table. Full results are available on request. Source: Authors’ analysis based on data described in the text. 115 ´nez Perı ´a Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 116 THE WORLD BANK ECONOMIC REVIEW the capital account, and regulation of the remittance market that might influ- ence the costs faced by remittance service providers. It includes a dummy vari- able for receiving countries with pegged exchanged rates (including cases of currency boards, de facto pegged regimes, and no separate legal tender). Lower exchange rate volatility should be associated with lower prices, by lowering the exchange rate costs and uncertainty faced by providers and, thus, the spreads they charge to customers. At the same time, the price of sending remittances is expected to be higher in countries that impose controls on remittance trans- actions, since controls operate like a tax that is likely to be passed onto recipi- ents. Both the dummy variable for pegged exchange rate regimes and the capital controls dummy variable are from the International Monetary Fund (IMF 2007). In 39 corridors (almost 33 percent of the sample), the exchange rate is pegged or the economy is fully dollarized, so there is no exchange rate variability, and in 22 corridors (18 percent of the sample) there are controls on gifts from abroad or remittances. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 The analysis controls for the breadth of regulation of remittance service pro- viders in sending and receiving countries by creating an index of regulation that takes a value of 0 –5 depending on whether providers must be registered, must be licensed, are subject to speci�c safety and ef�ciency requirements, need to comply with anti–money laundering regulations, or need to comply with laws and regulations of general applicability. Data for creating the indexes are from the 2008 Global Payment Systems Survey conducted by the World Bank (2008). While a broader regulatory framework might make the remittance market more transparent and more competitive, greater exposure to regulations can also increase the costs for regulated institutions, so the impact is ambigu- ous a priori.14 Similarly, greater breadth of regulation might reduce the number of service providers, with negative repercussions for competitiveness. The index averages 2.2 among remittance receiving countries and 2.3 among remittance sending countries. Remittance Mark-ups The regressions also include proxies for factors that might be associated with remittance prices because of their effect on the mark-up remittance service pro- viders can charge their customers. The analysis posits that providers can more readily charge a mark-up if there is little competition in the remittance market and if customers are not well informed. Two indirect measures of competition among remittance service providers are used (direct measures are lacking). One is the number of remittance service providers in the database for each corridor. Since the Remittances Prices Worldwide survey tries to cover the most impor- tant providers in a corridor, corridors with more providers are assumed to have 14. Note that the index does not measure the severity of regulations, but only the scope of the regulatory framework. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 117 more active �rms and, other things equal, to be more competitive.15 The average number of respondents across all corridors is 8, and the number varies from 2 in the Spain–China corridor to 18 in the United States–Mexico corridor. The second measure is of competition among banks in receiving and sending countries. The rationale is that more competitive banking sectors will offer cheaper services, including for remittance transactions. This will create pressure for other providers to lower prices as well. Of course, this implicitly assumes that banks are important players in the remittance business. Following Panzar and Rosse (1982, 1987), the H-statistic is used to measure the degree of com- petition in the banking sector by calculating the sum of the elasticities of banks’ interest revenues to different input prices (see the appendix for a discus- sion of the methodology used to calculate the H-statistic).16 Under perfect competition, an increase in input prices raises both marginal costs and revenues by the same amount and, thus, the H-statistic will equal 1. In a monopoly, an Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 increase in input prices results in a rise in marginal costs, a fall in output, and a decline in revenues, leading to an H-statistic of less than or equal to 0. When H is between 0 and 1, the system operates under monopolistic competition. A negative relationship is expected between the H-statistic in sending and receiv- ing countries and the price of sending remittances. Data for 1994–2006 from Bankscope database (Bureau van Dijk 2009) are used to compute the H-statistic. Among both remittance receiving and sending countries, the H-statistic averages close to 0.53. But the standard deviation is larger for remit- tance sending countries. The signi�cance of the relative importance of banks in the remittance market in explaining cross-corridor variation in remittance prices is also explored using the share of bank respondents among all remittance service pro- viders in the database. To the extent that, as some have argued, banks view remittances as a marginal product and are less likely to offer competitive prices (Ratha and Riedberg 2005), a positive correlation is expected between the share of bank respondents and the average price of remittances. Across the 119 corridors, the share of bank respondents varies from 0 percent in the Italy–Sri Lanka corridor to 100 percent in the South Africa –Swaziland corridor. On average, the share across corridors is 31 percent. Because data were lacking on the share of the remittance market captured by each provider, the percentage of bank respondents described above may not reflect the actual importance of commercial banks. Hence, an alternative measure, obtained from the Global Payment Systems Survey (World Bank 15. Because in most cases, mystery shoppers were used to gather data on the price of remittances, the number of respondents should not be affected by the willingness of certain providers to cooperate. However, it is still possible that in some corridors the number of respondents is small simply because interviewers had dif�culty reaching or locating some providers. 16. Other studies that use this methodology to estimate competition include Bikker and Haaf (2002), Gelos and Roldos (2004), Claessens and Laeven, (2004), and Levy-Yeyati and Micco (2007). 118 THE WORLD BANK ECONOMIC REVIEW 2008), is used to test the sensitivity of the �ndings. The measure indicates the degree to which central banks consider commercial banks to be signi�cant remittance service providers, on a scale from 1 (least relevant) to 6 (most rel- evant).17 The correlation between this variable and the percentage of bank respondents is 0.37 and is signi�cant at the 5 percent level. The �nancial literacy of remittance senders can also affect mark-ups. Since �nancial literacy cannot be captured directly, a measure of the education level of migrants in each corridor is used (migrants with a secondary or tertiary edu- cation as a share of total migrants from the remittance receiving country resid- ing in the remittance sending country). This variable comes from the OECD Database on Immigrants and Expatriates (OECD 2010). This variable is expected to be correlated with �nancial literacy, and to the degree that �nan- cial literacy enables consumers to make better informed choices, prices should be lower. The ratio of secondary and tertiary educated migrants varies from 21 percent for Chinese migrants in Italy to 91 percent for Nigerian migrants in the Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 United States. Because this variable is available for only 88 of the 119 corri- dors, it is not included in the baseline estimations but is shown as an additional variable. Socioeconomic Variables Several socioeconomic variables are included that can influence remittance prices by affecting both costs and mark-ups. One, a proxy for the volume of remittance transactions within corridors, is the number (bilateral stock) of migrants residing in the remittance sending country who are originally from the remittance receiving country. These data are from the World Bank (Ratha and Shaw 2007). Unlike the flow of actual remittances, migrant stock is less likely to be endogenous to the price variable. A negative relationship is conjec- tured between the stock of migrants and the price of remittances; a higher volume of migrants might imply scale economies and, hence, lower costs for providers and more competition among them, resulting in smaller mark-ups.18 The number of migrants is negligible in the South Africa–Zambia corridor and exceeds 10 million people in the United States–Mexico corridor. The average is 379,200 migrants. GDP per capita is also included, as a proxy for economic development and standard of living. This variable comes from the World Bank’s World Development Indicators database (World Bank 2009c). The cost of goods and services will be higher in countries with higher standards of living, so remit- tance prices would also be expected to be higher. Countering that tendency, economic development may be associated with greater ef�ciencies and lower 17. The Global Payment Systems Survey scale uses 1 to indicate most relevant and 6 the least relevant. The scale is inverted here so that higher values indicate that banks are more important. 18. The presence of more migrants might encourage entry of a larger number of remittance service providers, leading to more contestability and lower mark-ups. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 119 costs for �nancial intermediation (Harrison, Sussman, and Zeira 1999) and thus lower remittance prices. In the sample, GDP per capita for receiving countries varies from $148 in Malawi to close to $14,000 in the Republic of Korea. Among remittance sending countries, GDP per capita varies from $3,640 in South Africa to $40,200 in Japan (all prices in U.S. dollars).19 The geographic distribution of the population in sending and receiving countries might also be an important driver of the price of remittances, as a more sparsely distributed population might be harder to reach, thus raising transaction costs for providers. A more sparsely distributed population might also increase the pricing power of providers, as geographic access is more dif�- cult for senders and recipients of remittances. The share of rural population in both sending and receiving countries is used to proxy for the disparity in geo- graphic distribution.20 These data come from the World Bank’s World Development Indicators (World Bank 2009c). Among receiving countries, the rural population varies from 13 percent of the total in Lebanon to 87 percent Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 in Uganda and averages 48 percent. By contrast, among sending countries, the rural population varies from 0 for Singapore to 40 percent for South Africa and averages 21 percent. Bank Variables To measure access to �nancial services more directly, some estimations also control for the number of bank branches per capita in sending and receiving countries.21 This variable is expected to have a negative association with the price of sending remittances, as higher branch penetration will reduce trans- action costs and increase scale. The ratio of branches per capita averages about 6 per 100,000 inhabitants in receiving countries and close to 34 per 100,000 in sending countries. Measures of the costs of accessing formal banking services in both sending and receiving countries (the minimum amount to open a savings account and the annual fee to maintain an account) are also included (Beck, Demirgu ¸ -Kunt, and Martı ¨c ´a 2008). Easier and cheaper access is conjec- ´nez Perı tured to increase the options for both senders and recipients of remittances and thus to boost competition. The minimum balance to open a savings account averages 7.36 percent of GDP per capita in receiving countries and 0.11 19. Regressions were also run that controlled separately for the level of �nancial development using the ratio of liquid liabilities to GDP. The results are very similar to those including GDP per capita. Since these variables are highly correlated—(0.2) among receiving countries and (0.4) among sending countries—these estimations are not reported, and GDP per capita is included instead as a broader measure of development. 20. The share of rural population is a better proxy for the effect of service delivery than population density, which is an average within a country and does not take into account which share of the population actually lives in more remote areas. The population density variable yielded similar results. 21. These data are from Beck, Demirgu ¸ -Kunt, and Martı ¨c ´nez Perı´a (2007) and can be accessed at http://go.worldbank.org/EZDOBVQT20. Because these data are not available for all corridors, this variable is not included in all estimations. 120 THE WORLD BANK ECONOMIC REVIEW percent in sending countries; fees average 0.55 percent of GDP per capita in receiving countries and 0.12 percent in sending countries. Corridor-speci�c Variables Finally, several corridor-speci�c variables are included that might influence the extent and ease of remittance transactions and, therefore, their costs. These are the distance between sending and receiving countries (from capital city to capital city) and a dummy variable for a common language (takes a value of one if both countries have at least one common language spoken by more than 9 percent of the population). These data come from the French Research Center in International Economics (CEPII) Distances database (CEPII 2010). Smaller geographic and linguistic distances might also foster the emergence of informal remittance service providers, adding competitive pressure to the formal remittance market. Some estimations also include the log of bilateral Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 exports and imports, a measure of bilateral trade. These data come from the IMF Direction of Trade Statistics (IMF 2009). Correlations Between Variables in Our Dataset The average prices of remittances are signi�cantly lower in corridors with a higher number of migrants, smaller share of rural population, and a common language (see table 2). Prices are also lower in corridors where competition is higher and bank participation in the remittance industry is lower. Finally, prices are lower in corridors where sending countries have a broader regulatory framework for remittance service operators and where minimum balances to open a savings account and annual fees to maintain them are lower. Some explanatory variables are highly correlated with others. For instance, GDP per capita in receiving and sending countries is signi�cantly correlated with competition among providers, rural population share, branch penetration, cost of using banking services, and extent of bilateral trade. II I. EM P I R I CA L R ES ULT S Table 3, column 3.1 reports the baseline estimation considering average remit- tance prices charged across all providers with variables for which data are available for all 119 corridors. Though information on the number of respon- dents and the percentage of banks among respondents is also available across all corridors, these variables are not included in the baseline estimations since, as discussed earlier, they might not adequately capture the degree of compe- tition and the importance of banks in the remittance market. The baseline regression shows that, across all providers in 119 corridors, remittance prices are signi�cantly associated with the number of migrants in the corridor, the level of income, and the share of rural population in receiving and sending countries. Corridors with a higher number of migrants exhibit T A B L E 3 . Regressions for the Average Prices of Sending $200 in Remittances for all Remittance Service Providers Variable (3.1) (3.2) (3.3) (3.4) (3.5) (3.6) (3.7) (3.8) (3.9) (3.10) (3.11) (3.12) Log number of migrants 2 1.09 2 0.98 2 1.12 2 1.10 2 1.05 2 1.02 2 1.47 2 0.44 2 0.35 2 1.07 2 0.87 2 1.06 ( 2 4.96)*** ( 2 4.40)*** ( 2 6.20)*** ( 2 4.37)*** ( 2 4.44)*** ( 2 4.40)*** ( 2 10.54)*** ( 2 1.25) ( 2 2.26)** ( 2 4.41)*** ( 2 2.50)** ( 2 4.33)*** Log GDP per capita 1.91 2.11 2.76 1.31 3.57 1.96 3.20 3.18 3.23 2.25 2.64 2.13 receiving (2.40)** (2.66)*** (3.74)*** (1.88)* (3.98)*** (2.28)** (3.60)*** (2.30)** (2.50)** (2.69)*** (3.17)*** (2.51)** Log GDP per capita 2.04 1.95 3.90 2.73 2.38 1.03 2.33 6.43 10.64 1.94 2.36 2.19 sending (2.21)** (2.11)** (3.96)*** (3.32)*** (2.82)*** (0.91) (2.92)*** (2.56)** (7.19)*** (2.02)** (1.12) (2.03)** Pegged or dollarized 2 1.16 2 1.02 2 0.77 2 1.20 2 0.22 2 0.53 2 1.40 1.06 0.98 2 1.37 2 0.07 2 1.16 ( 2 1.38) ( 2 1.23) ( 2 1.03) ( 2 1.64) ( 2 0.20) ( 2 0.49) ( 2 1.46) (0.57) (0.60) ( 2 1.45) ( 2 0.07) ( 2 1.21) Share rural population 0.09 0.09 0.10 0.05 0.13 0.09 0.12 0.14 0.14 0.12 0.11 0.10 receiving (2.91)*** (2.75)*** (3.23)*** (1.83)* (4.01)*** (2.64)*** (2.59)** (2.16)** (2.63)** (3.34)*** (2.83)*** (2.85)*** Share rural population 0.22 0.21 0.33 0.08 0.23 2 0.01 0.25 0.16 0.16 0.22 0.08 0.22 sending (4.43)*** (4.27)*** (7.19)*** (1.71)* (4.67)*** ( 2 0.11) (4.81)*** (1.47) (2.31)** (4.21)*** (1.00) (3.83)*** Log distance 2 0.36 2 0.29 0.23 2 0.02 0.47 0.19 0.58 2 0.10 0.75 2 0.20 2 0.12 2 0.36 ( 2 0.70) ( 2 0.57) (0.55) ( 2 0.05) (0.87) (0.34) (1.00) ( 2 0.08) (0.90) ( 2 0.39) ( 2 0.18) ( 2 0.65) Common language 0.06 0.49 0.23 0.46 1.31 2 0.41 0.15 1.35 0.78 2 0.36 2 1.32 2 0.29 (0.08) (0.66) (0.34) (0.69) (1.43) ( 2 0.44) (0.19) (0.75) (0.61) ( 2 0.44) ( 2 1.46) ( 2 0.34) Number respondents 2 0.30 per corridor ( 2 2.82)*** H 2 statistic receiving 2 5.15 ( 2 2.65)*** Thorsten Beck and Marı H 2 statistic sending 2 16.12 ( 2 5.08)*** Share of banks per 0.09 corridor (6.70)*** Index banks importance 0.70 receiving ´a Soledad Martı (1.65) Index banks importance 1.88 sending (4.40)*** ´nez Perı ´a Index of regulation 0.24 receiving (Continued ) 121 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 122 TABLE 3. Continued Variable (3.1) (3.2) (3.3) (3.4) (3.5) (3.6) (3.7) (3.8) (3.9) (3.10) (3.11) (3.12) (0.74) Index of regulation 2 2.78 sending ( 2 2.93)*** Bank branches per 0.12 capita receiving (0.62) Bank branches per 2 0.05 capita sending ( 2 3.20)*** Savings accounts fee 0.49 receiving (1.33) Savings accounts fee 16.28 sending (1.76)* THE WORLD BANK ECONOMIC REVIEW Min. amount to open 2 0.02 account receiving ( 2 0.42) Min. amount to open 27.97 account sending (7.24)*** Controls on remittances 0.06 (0.06) Share of educated 0.02 migrants (0.73) Log bilateral trade 2 0.06 ( 2 0.25) Observations 119 119 111 119 84 91 89 53 53 105 88 111 R-squared 0.36 0.38 0.56 0.53 0.54 0.45 0.52 0.38 0.62 0.39 0.25 0.36 Diff. max-min prices 0.26 0.25 0.27 0.35 0.41 0.33 0.28 0.36 0.25 0.23 0.23 0.28 predicted over actual *Signi�cant at the 10 percent level; **signi�cant at the 5 percent level; ***signi�cant at the 1 percent level. Note: Number in parentheses are robust t- statistics. Constant is included but not shown. Source: Authors’ analysis based on data described in the text. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 123 lower prices, while those with higher incomes per capita and a larger percen- tage of rural population face higher prices. These results are consistent across all estimations reported in table 3. As expected, greater competition among providers (measured by number of respondents or the H-statistic for the banking sector) is associated with lower remittance prices (table 3, columns 3.2 and 3.3). Corridors where banks play a larger role in the remittance market exhibit higher prices (columns 3.4 and 3.5). Corridors with broader regulation of remittance service providers in the sending country have lower prices, while the regulatory breadth in the receiving country does not seem to matter (column 3.6). Greater access to and lower costs of banking services are associated with lower prices of remittances (columns 3.7–3.9). In particular, corridors with more bank branches per capita in the sending country face lower prices, while corridors with higher minimum amounts to open accounts and higher annual fees have higher remittance prices. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 The results discussed so far are economically as well as statistically signi�- cant. For example, an increase in the number of migrants from the corridor at the 25th percentile (United Kingdom–China with 56,774) to the corridor at the 75th percentile (Spain–Colombia with 384,621) is associated with a 2 per- centage point drop in average prices. An increase in competition (as measured by the H-statistic) in the sending country from the 25th percentile to the 75th implies a 4.4 percentage point reduction in remittance prices, while an increase in the receiving country is associated with a 1.4 percentage point reduction. A similar change in the number of remittance service respondents (from 6, the 25th percentile, to 10, the 75th percentile) is associated with a 1.2 percentage point drop in prices, while an increase in the scope of remittance regulation in the sending country implies a reduction of 2.8 percentage points. A comparable increase in the number of branches per capita in the sending country is associ- ated with a 1.6 percentage point decline in prices. Even stronger, an increase in the percentage of banks among survey respondents from the 25th (0 percent) to the 75th percentile (50 percent) implies an increase in prices of more than 4 percentage points. Note that the average price across corridors associated with these changes is close to 10 percent, so the effects are considerable. In contrast, no robust association is found between remittance prices and measures of exchange rate stability or the presence of capital controls on remit- tances (columns 3.1 and 3.10). Similarly, the distance between sending and receiving countries, the extent of bilateral trade, and whether countries share a common language are not correlated with remittance prices (columns 3.1 and 3.12).22 Finally, the share of educated migrants does not have a signi�cant effect (column 3.11). 22. If common language is replaced with a dummy variable for whether the receiving and sending countries have colonial ties, the main results do not change and the dummy variable for colonial ties tends to be positive and signi�cant. These results are available on request. 124 THE WORLD BANK ECONOMIC REVIEW Using alternative indicators for several variables, such as the Parson and others (2007) data on bilateral migration and a Barro and Lee (2001) measure of educational attainment, yields similar �ndings.23 Also, running the regressions for median instead of average prices does not change the results sig- ni�cantly; neither does using prices based on sending $500 instead of $200. The results are not reported here but are available on request. Overall, the estimations have good predictive power. The R-squared for the baseline regression (table 3, column 3.1) is 0.36 and varies from 0.25 (column 3.11) to 0.56 (column 3.3), depending on the additional controls included. Similarly, the estimations are reasonably good at predicting the difference between extreme observations (the difference between the corridors with the maximum and minimum prices). Depending on the estimation, the share of the actual difference between the maximum and minimum prices that is predicted by the estimations varies from 0.23 (column 3.11) to 0.41 (column 3.5). Finally, partial plots of remittance prices against the variables found to be Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 consistently signi�cant (log of migrants, log of GDP per capita in sending country, share of rural population in sending country, number of respondents in the corridor, H-statistic for bank competition in the sending country, share of bank respondents, index of importance of banks among remittance service providers, and index of regulation of remittance service providers) show that these variables do a good job of predicting prices and that the correlations are not driven by outliers (�gure 5). The log of migrants appears to be an excep- tion, with large outliers for the South Africa–Zambia and South Africa – Angola corridors (top left corner of �gure 5). However, when these two out- liers are removed, the log of migrants remains signi�cant at the 1 percent level and the other results in the baseline estimations do not change signi�cantly. Next are the factors that influence remittance prices across service provider types. Table 4 shows separate estimations for average prices among banks (columns 4.1–4.4), MTOs (columns 4.5–4.8), and Western Union (4.9–4.12). To save space, only some of the speci�cations shown to be signi�cant in the regressions for all providers (see table 3) are reported here; others are available on request. In columns 4.1–4.4, the dependent variable is the average price across all bank respondents in a corridor. Since there are corridors where banks do not play a signi�cant role in the remittance market (and so were not included in the database), the sample size is smaller than that in table 3. Most of the results discussed so far hold when the sample is restricted to banks. In particular, a larger number of migrants, lower levels of per capita income in the receiving country, and a smaller share of rural population are still 23. The correlation between the World Bank bilateral migration data and the Parson and others (2007) data is 0.66, and results do not change when the Parson and others data are used. These results are available on request. Barro and Lee’s (2001) average years of schooling of the population over 25 for the receiving country was used. Results remain unchanged. The results using the data on the education of migrants are presented here, since those data more directly relate to the population that conducts remittance transactions. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 125 F I G U R E 5. Partial Plots of Selected Regressors against Remittance Prices Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Source: Authors’ analysis based on data described in the text. 126 T A B L E 4 . Regressions for the Average Prices Charged by Banks, Money Transfer Operators, and Western Union on $200 in Remittances Banks Money transfer operators Western Union Variable (4.1) (4.2) (4.3) (4.4) (4.5) (4.6) (4.7) (4.8) (4.9) (4.10) (4.11) (4.12) Log number of 2 1.07 2 1.57 2 1.07 2 1.04 2 1.34 2 1.20 2 1.24 2 0.71 2 2.00 2 1.86 2 2.07 2 2.03 migrants THE WORLD BANK ECONOMIC REVIEW ( 2 2.64)** ( 2 3.97)*** ( 2 2.15)** ( 2 2.50)** ( 2 7.85)*** ( 2 6.42)*** ( 2 5.40)*** ( 2 2.23)** ( 2 7.44)*** ( 2 6.97)*** ( 2 7.32)*** ( 2 4.10)*** Log GDP per capita 3.06 6.28 5.92 15.79 1.02 1.59 0.69 1.63 1.62 1.95 1.33 2.06 receiving (1.79)* (3.42)*** (3.63)*** (2.14)** (1.82)* (2.95)*** (0.87) (2.15)** (2.19)** (2.56)** (1.42) (1.97)* Log GDP per capita 4.81 3.28 0.26 14.63 1.41 1.67 1.01 3.63 2.31 3.37 2.23 3.84 sending (2.88)*** (1.46) (0.14) (2.44)** (2.42)** (2.58)** (0.93) (2.70)** (2.93)*** (2.59)** (1.68)* (0.49) Pegged or dollarized 2 1.22 0.51 1.15 3.97 2 0.91 2 0.84 2 0.70 0.18 2 2.10 2 2.08 2 2.00 2 1.16 ( 2 0.72) (0.23) (0.45) (0.85) ( 2 1.58) ( 2 1.64) ( 2 0.88) (0.19) ( 2 2.70)*** ( 2 2.56)** ( 2 1.87)* ( 2 0.72) Share rural population 0.07 0.15 0.18 0.49 0.04 0.07 0.04 0.09 0.04 0.05 0.05 0.07 receiving (1.18) (2.15)** (3.02)*** (2.04)* (1.67)* (2.86)*** (1.47) (2.62)** (1.13) (1.45) (1.23) (1.53) Share rural population 0.02 0.33 2 0.23 2 0.14 0.06 0.12 0.04 0.04 0.06 0.15 0.05 0.05 sending (0.23) (3.77)*** ( 2 1.33) ( 2 0.87) (1.44) (2.56)** (0.56) (0.54) (0.90) (1.55) (0.43) (0.31) Log distance 0.14 1.14 0.99 2.69 2 0.26 2 0.19 2 0.27 0.14 2 0.28 2 0.28 2 0.48 2 0.34 (0.18) (1.13) (1.20) (1.14) ( 2 0.77) ( 2 0.61) ( 2 0.60) (0.25) ( 2 0.54) ( 2 0.52) ( 2 0.74) ( 2 0.32) Common language 0.73 2 0.87 2 1.71 2 6.93 0.39 0.13 2 0.23 1.01 2 0.20 2 0.35 2 1.14 2.62 (0.49) ( 2 0.57) ( 2 0.93) ( 2 2.41)** (0.58) (0.20) ( 2 0.27) (0.74) ( 2 0.22) ( 2 0.38) ( 2 1.07) (1.02) Share of banks per 0.19 0.03 0.02 corridor (5.71)*** (2.48)** (0.89) Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 H 2 statistic receiving 2 7.27 2 4.55 2 2.61 ( 2 1.47) ( 2 3.46)*** ( 2 1.20) H 2 statistic sending 2 14.34 2 5.13 2 7.99 ( 2 1.80)* ( 2 2.02)** ( 2 1.60) Index of regulation 0.54 2 0.03 0.48 receiving (0.82) ( 2 0.10) (1.18) Index of regulation 2 6.39 2 0.63 2 0.25 sending ( 2 3.70)*** ( 2 0.72) ( 2 0.21) Min. amount to open 0.54 2 0.02 2 0.04 account receiving (0.99) ( 2 0.48) ( 2 0.70) Min. amount to open 50.75 52.99 2 19.66 account sending (3.85)*** (1.02) ( 2 0.27) Observations 70 66 58 26 112 106 86 50 98 92 75 38 R-squared 0.54 0.47 0.56 0.63 0.36 0.42 0.33 0.35 0.44 0.46 0.50 0.44 Diff. max-min prices 0.45 0.38 0.61 0.72 0.28 0.18 0.37 0.19 0.66 0.68 0.70 0.69 predicted over actual *Signi�cant at the 10 percent level; **signi�cant at the 5 percent level; ***signi�cant at the 1 percent level. Note: Number in parentheses are robust t-statistics. Constant is included but not shown. Source: Authors’ analysis based on data described in the text. Thorsten Beck and Marı 127 ´a Soledad Martı ´nez Perı ´a Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 128 THE WORLD BANK ECONOMIC REVIEW associated with lower prices, as is broader regulation of remittance service pro- viders in the sending country. As before, a higher share of banks among respondents and higher minimum balances to open accounts are positively cor- related with prices. The measures of competition are no longer signi�cant at the 5 percent level, a result likely due to the lower number of observations.24 Most of the earlier �ndings are con�rmed when the sample is restricted to MTOs (columns 4.5–4.8 of table 4). A larger number of migrants and greater competition in the banking system are associated with lower prices, while higher levels of income and bank participation are associated with higher prices. A larger share of rural population is associated with higher remittance prices among MTOs, but regulation of remittance service providers and costs of opening bank accounts are not signi�cantly associated with remittance prices among MTOs. Columns 4.9–4.12 of table 3 show results for the prices charged by Western Union, one of the world’s largest MTOs, active in 98 corridors of the sample. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Focusing on one �nancial institution permits controlling for any bias arising from differences in institutions across corridors (composition bias), even within the group of banks and MTOs. The price data for Western Union con�rm that a larger number of migrants and lower GDP per capita in the receiving and sending countries are associated with lower prices. In addition, exchange rate stability (as a result of pegged rates or dollarization) is also correlated with lower prices. Contrary to previous estimations, however, none of the competition-related indicators enter signi�cantly, which could be due to Western Union’s dominant position in the remittance business across most cor- ridors.25 Similarly, the share of rural population is generally not signi�cantly associated with remittance prices across corridors for Western Union. I V. C O N C L U S I O N S This article on 119 migration corridors �nds that remittance prices are associ- ated with a number of factors. First, the number of migrants is negatively and signi�cantly associated with the price of remittances across different samples and providers. This seems to suggest an important volume effect that works through scale economies and lower costs for providers or through higher com- petition in a larger market leading to a lower mark-up. Second, remittance prices are higher in corridors with higher income per capita, which could reflect higher prices of nontradable goods, such as services, in general. Third, competition and market structure matter, except in the case of Western Union. Corridors with a larger number of providers and countries with more 24. This is established by rerunning the regression for the average fee across all providers for the same sample as used in table 4. 25. This could be due to the fact that Western Union might have been operating longer in some corridors than other �rms. Also, Western Union might have better network coverage than other providers in some countries. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 129 competitive banking sectors exhibit lower prices, although prices are higher in corridors with a higher share of banks among providers. Fourth, banking sector outreach, as measured by branch penetration and cost barriers, is associ- ated with lower remittance prices. Finally, a broader regulatory framework for remittance service providers in the sending country is associated with lower remittance prices, especially among banks. Several factors were not found to be consistently correlated with remittance prices, In particular, exchange rate stability, capital controls, and �nancial lit- eracy. However, this might be due to the use of imperfect variables to capture these policies. While this article offers some interesting �ndings on an important topic, it is only a �rst exploration into what drives remittance prices. Future research should be able to exploit panel variation to get deeper into the issues, while also addressing some of the limitations of the data and analysis. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 REFERENCES Adams, Richard H. 2005. “Remittances and Poverty and Investment in Guatemala.� In International Migration, Remittances and the Brain Drain, ed. M. Schiff, and C. Ozden. Washington DC: World Bank. Adams, Richard H., and John Page. 2003. “International Migration, Remittances and Poverty in Developing Countries.� Policy Research Working Paper 3179. World Bank, Washington DC. ¨c Aggarwal, Reena, Asli Demirgu ´a Soledad Martı ¸ -Kunt, and Marı ´a. Forthcoming. “Do Workers’ ´nez Perı Remittances Promote Financial Development?� Journal of Development Economics. Aycinena, Diego, Claudia Martinez, and Dean Yang. 2009. “The Impact of Remittance Fees on Remittance Flows: Evidence from a Field Experiment among Salvadoran Migrants.� Department of Economics, University of Michigan, Ann Arbor. Barro, Robert J., and Jong-Wha Lee. 2001. “International Data on Educational Attainment: Updates and Implications.� Oxford Economic Papers 53 (3): 541–63. Beck, Thorsten. 2007. “Ef�ciency in Financial Intermediation: Theory and Empirical Measurement.� In Bernd Balkenhol, ed., Micro�nance and Public Policy: Outreach, Performance and Ef�ciency, New York, Palgrave, MacMillan. ¨c Beck, Thorsten, Asli Demirgu ´a Soledad Martı ¸ -Kunt, and Marı ´a. 2007. “Reaching Out: Access ´nez Perı to and Use of Banking Services across Countries.� Journal of Financial Economics 85: 234 –66. ———. 2008. “Banking Services for Everyone? Barriers to Bank Access and Use around the World�. World Bank Economic Review 22 (3): 397– 430. Bikker, Jacob A., and Katharina Haaf. 2002. “Competition, Concentration, and Their Relationship: An Empirical Analysis of the Banking Industry.� Journal of Banking and Finance 26 (11): 2191–214. Bureau van Dijk. 2009. Bankscope Database. www.bvdinfo.com/ProductsContainer/Bankscope.aspx. CEPII (French Research Center in International Economics). 2010. Distances database. CEPII, Paris. http://www.cepii.fr/anglaisgraph/bdd/bdd.htm. Claessens, Stijn, and Luc Laeven. 2004. “What Drives Bank Competition? Some International Evidence.� Journal of Money, Credit, and Banking 36 (3): 562– 83. Cox-Edwards, Alejandra, and Manuelita Ureta. 2003. “International Migration, Remittances, and Schooling: Evidence from El Salvador.� Journal of Development Economics 72 (2): 429– 61. ¨c Demirgu ¸ -Kunt, Asli, Luc Laeven, and Ross Levine. 2004. “Regulations, Market Structure, Institutions, and the Cost of Financial Intermediation.� Journal of Money, Credit, and Banking 36 (3): 593–622. 130 THE WORLD BANK ECONOMIC REVIEW Duryea, Suzanne, Ernesto Lo ´ rdova, and Alexandra Olmedo. 2005. “Migrant Remittances and ´ pez Co Infant Mortality: Evidence from Mexico.� Inter-American Development Bank, Washington, DC. Freund, Caroline L., and Nikola Spatafora. 2008. “Remittances, Transaction Costs, Determinants, and Informality.� Journal of Development Economics 86 (2): 356–66. ´ n, and Jorge Roldo Gelos, R. Gasto ´ s. 2004. “Consolidation and Market Structure in Emerging Banking Systems.� Emerging Markets Review 5 (1): 39–59. Gibson, John, David J. McKenzie, and Halahingano Rohorua. 2006. “How Cost Elastic Are Remittances? Evidence from Tongan Migrants in New Zealand.� Paci�c Economic Bulletin 21 (1): 112 –28. Hanson, Gordon H., and Christopher Woodruff. 2003. “Emigration and Educational Attainment in Mexico.� School of International Relations and Paci�c Studies, University of California, San Diego. Mimeo. Harrison, Paul, Oren Sussman, and Joseph Zeira. 1999. “Finance and Growth: Theory and New Evidence.� Finance and Economics Discussion Series, FEDS Working Paper 99-35. Federal Reserve Board, Washington, DC. Hildebrandt, Nicole, and David McKenzie. 2005. “The Effects of Migration on Child Health in Mexico.� Economia 6 (1): 257– 89. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 IMF (International Monetary Fund). 2005. World Economic Outlook 2005. Washington, DC: IMF. ———. 2007. Annual Report on Exchange Arrangements and Restrictions. Washington, DC: IMF. ———. 2009. Direction of Trade database. DOT Online. IMF, Washington, DC. http://www2. imfstatistics.org/DOT/help/DOThelp.htm. Kanaiaupuni, Shawn Malia, and Katharine M. Donato. 1999. “Migradollars and Mortality: The Effects of Migration on Infant Survival in Mexico.� Demography 36 (3): 339 –53. Laeven, Luc, and Giovanni Majnoni. 2005. “Does Judicial Ef�ciency Lower the Cost of Credit?� Journal of Banking and Finance 29 (7): 1791– 812. Levy-Yeyati, Eduardo, and Alejandro Micco. 2007. Concentration and Foreign Penetration in Latin American Banking Sectors: Impact on Competition and Risk. Journal of Banking and Finance 31 (6): 1633–47. ´ pez Co Lo ´ rdova, Ernesto. 2005. “Globalization, Migration, and Development: The Role of Mexican Migrant Remittances.� Economia 6: 217–49. Maimbo, Samuel Munzele, and Dilip Ratha. 2005. Remittances: Development Impact and Future Prospects. Washington, D.C.: World Bank. Massey, D., and E. Parrado. 1998. “International Migration and Business Formation in Mexico.� Social Science Quarterly 79: 1 –20. OECD (Organisation for Economic Co-operation and Development). 2010. Database on Immigrants and Expatriates. OECD, Paris. http://stats.oecd.org/Index.aspx?lang=en#. Orozco, Manuel. 2006. “International Flows of Remittances: Cost, Competition and Financial Access in Latin America and the Caribbean—Toward an Industry Scorecard.� Inter-American Dialogue, Washington, DC. Panzar, John C., and James N. Rosse. 1982. “Structure, Conduct, and Comparative Statistics.� Economic Discussion Paper 248. Bell Laboratories, Murray Hill, NJ. ———. 1987. “Testing for ‘Monopoly’ Equilibrium.� The Journal of Industrial Economics 35 (4): 443 –56. Parsons, Christopher, Ronald Skeldon, Terrie Lousie Walmsley, and L. Alan Winters. 2007. “Quantifying International Migration: A Database of Bilateral Migrant Stocks.� Policy Research Working Paper 4165. World Bank, Washington, DC. Ratha, Dilip, and Jan Riedberg. 2005. “On Reducing Remittance Costs.� Development Prospects Group. World Bank, Washington, DC. Ratha, Dilip, and William Shaw. 2007. South– South Migration and Remittances. World Bank Working Paper 102. Washington, DC: World Bank. http://go.worldbank.org/U4RXL56V20. ´a Soledad Martı Thorsten Beck and Marı ´a ´nez Perı 131 Taylor, J. Edward, Jorge Mora, Richard Adams, and Alejandro Lopez-Feldman. 2005. “Remittances, Inequality, and Poverty: Evidence from Rural Mexico.� Working Paper 05-003. Department of Agricultural and Resource Economics, University of California, Davis. Woodruff, Christopher, and Rene Zenteno. 2007. “Migration Networks and Microenterprises in Mexico.� Journal of Development Economics 82 (2): 509– 28. World Bank. 2007. Data on Access to Financial Services. World Bank, Washington, DC. http://econ. worldbank.org/WBSITE/EXTERNAL/EXTDEC/EXTRESEARCH/0,,contentMDK:21546633 page PK:64214825 piPK:64214943 theSitePK:469382,00.html ———. 2008. Payment Systems Worldwide, a Snapshot: Outcomes of the Global Payment Systems Survey 2008. Washington, DC: World Bank. http://siteresources.worldbank.org/ INTPAYMENTREMMITTANCE/Resources/Global_Survey_Book.pdf ———. 2009a. Migration and Remittances Website. http://go.worldbank.org/SSW3DDNLQ0. ———. 2009b. Remittance Prices Worldwide database. World Bank, Washington, DC. http:// remittanceprices.worldbank.org. ———. 2009c. World Development Indicators database. World Bank, Washington, DC. http://data. worldbank.org/data-catalog/world-development-indicators. Yang, Dean. 2008. “International Migration, Remittances, and Household Investment: Evidence from Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Philippine Migrants’ Exchange Rate Shocks.� Economic Journal 118 (528): 591– 630. A P P E N D I X : O B TA I N I N G T H E PA N Z A R AND ROSSE (1987) H- STATIS TIC Based on the Panzar and Rosse (1987) methodology and following the empiri- cal strategy pursued by Classes and Laeven (2004), the H-statistic is obtained by estimating equation (A1): LnðPit Þ ¼ ai þ b1 lnðW1;it Þ þ b2 lnðW2;it Þ þ b3 lnðW3;it Þ þ g lnðZ;it Þ ðA1Þ þ dD þ eit where P is the ratio of gross interest revenues to total assets ( proxy for banks’ output price); W1 is the ratio of interest expenses to total deposits and money market funding ( proxy for input price of deposits); W2 is the ratio of personnel expenses to total assets ( proxy for input price of labor); W3 is the ratio of other operating and administrative expenses to total assets ( proxy for input price of equipment/�xed capital); Z is a matrix of controls including the ratio of equity to total assets, the ratio of net loans to total assets, and the logarithm of assets; D is a matrix of year dummies; ai denotes bank-level �xed effects; i denotes banks; and t denotes years. Annual balance sheet and income state- ments from Bureau van Dijk’s Bankscope database (Bureau van Dijk 2009) were used to calculate the H-statistic for each sending and receiving country banking sector during 1994–2006. The H-statistic equals b1 þ b2 þ b3, the sum of the input price elasticities of total revenues. Conceptually, the statistic measures the responsiveness of bank revenues to input prices. An H-statistic less than or equal to 0 is a sign of a monopoly, H equal to 1 indicates perfect competition, and H between 0 and 1 indicates monopolistic competition. Remittances and the Brain Drain Revisited: The Microdata Show That More Educated Migrants Remit More Albert Bollard, David McKenzie, Melanie Morten, and Hillel Rapoport Two of the most salient trends in migration and development over the last two decades are the large rise in remittances and in the flow of skilled migrants. However, recent lit- erature based on cross-country regressions has claimed that more educated migrants remit less, leading to concerns that further increases in skilled migration will impede remittance growth. Microdata from surveys of immigrants in 11 major destination countries are used to revisit the relationship between education and remitting behavior. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 The data show a mixed pattern between education and the likelihood of remitting, and a strong positive relationship between education and amount remitted (intensive margin), conditional on remitting at all (extensive margin). Combining these intensive and exten- sive margins yields an overall positive effect of education on the amount remitted for the pooled sample, with heterogeneous results across destinations. The microdata allow investigation of why the more educated remit more, showing that the higher income earned by migrants, rather than family characteristics, explains much of the higher remit- tances. remittances, migration, brain drain, education JEL codes: O15, F22, J61 Two of the most salient trends in migration and development over the last two decades are the large rise in remittances and in the flow of skilled migrants. Of�cially recorded remittances to developing countries have more than tripled over the last decade, rising from $85 billion in 2000 to $305 billion in 2008 Albert Bollard (abollard@stanford.edu) is a PhD student at Stanford University. David McKenzie (dmckenzie@worldbank.org; corresponding author) is a senior economist in the Finance and Private Sector Research Unit of the Development Research Group at the World Bank. Melanie Morten (morten@yale.edu) is a PhD student at Yale University. Hillel Rapoport (hillel@mail.biu.ac.il) is professor of economics at Bar Ilan University and at EQUIPPE, University of Lille and is currently a visiting research fellow at the Center for International Development at Harvard University. The authors are grateful for funding for this project from the Agence Franc ´ veloppement (AFD). They ¸ aise de De thank all the individuals and organizations that graciously shared their surveys of immigrants, and they are grateful to Michael Clemens, participants at the 2nd Migration and Development Conference held at the World Bank in September 2009, three anonymous referees, and the journal editor for helpful comments. All opinions are those of the authors and do not necessarily represent those of AFD or the World Bank. A supplemental appendix to this article is available at http://wber.oxfordjournals.org. THE WORLD BANK ECONOMIC REVIEW, VOL. 25, NO. 1, pp. 132– 156 doi:10.1093/wber/lhr013 Advance Access Publication May 12, 2011 # The Author 2011. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 132 Bollard, McKenzie, Morten, and Rapoport 133 (World Bank 2008, 2009). The number of highly educated migrants from developing countries residing in Organisation for Economic Co-operation and Development (OECD) countries doubled over 1990–2000 (Docquier and Marfouk 2005) and likely has grown since then as developed countries have increasingly pursued skill-selective immigration policies.1 However, despite this positive association at the global level between rising remittances and rising high-skill migration, there are concerns—stemming from the belief that more educated individuals may remit less—that increasingly skill-selective immigration policies may slow or even end the rise in remit- tances. This belief is taken as fact by many; for example, an OECD (2007, p. 11) report says that “low skilled migrants tend to send more money home.� The main empirical evidence to support this assertion across a range of countries comes from two recent studies (Faini 2007; Niimi, O ¨ zden, and Schiff forthcoming) whose cross-country macroeconomic analyses �nd that the highly skilled (de�ned as those with tertiary education) remit less. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Yet there are many reasons to question the results of these cross-country esti- mations. Both studies relate the amount of remittances received at a country level to the share of migrants with tertiary education, at best telling us whether countries that send a larger share of highly skilled migrants receive less or more in remittances than countries that send fewer skilled migrants. The studies do not answer the factual question of whether more educated migrants remit more or less. There are a host of differences across countries that could cause a spur- ious relationship to appear between remittances and skill level across countries. For example, if poverty is a constraint to both migration and education, richer developing countries might be able to send more migrants (yielding more remittances) and those migrants might also have more schooling. Faini (2007) treats the share of migrants who are skilled as exogenous. Niimi, O ¨ zden, and Schiff (forthcoming) try to instrument for the education mix of migrants, but their instruments seem unlikely to satisfy the exclusion restrictions. For example, public spending on education is likely a function of a country’s overall institutional and economic development, which should independently affect the incentive to remit; migrants might send money to overcome poor public spending or for investment when complementary infrastructure and institutions are in place. This article revisits the relationship between remittances and education level using microdata that permit computing the association between a migrant’s education level and remitting behavior. The authors assembled the most com- prehensive micro-level database on remitting behavior currently available, com- prising data on 33,000 immigrants from developing countries from 14 surveys 1. In contrast, the number of low-skill migrants ( primary education or less) increased only 15 percent over the period. Immigration to OECD countries (as de�ned by the number of foreign born) was estimated at 90 million in 2000, about half of total world migration. Of the 90 million immigrants, 60 million were ages 25 or older and were split equally across education categories (primary, secondary, and tertiary; Docquier and Marfouk 2005). 134 THE WORLD BANK ECONOMIC REVIEW in 11 OECD destination countries. The analysis begins by establishing the factual relationship between the propensity to remit and education. No attempt is made to estimate the causal impact of education on remittances.2 From a policy perspective, the concern is whether migration policies that shift the edu- cation composition of migrants affect remittances, not whether education pol- icies that change how much education individuals have affects remittances. Microdata enables asking whether more educated individuals are more or less likely to remit (the extensive margin) and whether they send more or less remit- tances if they do remit (the intensive margin). A mixed association is found between education and remittances at the extensive margin, and a strong posi- tive relationship at the intensive margin. Combining both the extensive and intensive margins reveals that, at least in this large sample, more educated migrants do remit signi�cantly more—migrants with a university degree remit $300 more yearly than migrants without a university degree, where the mean annual remittance over the entire sample is $730. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 The article is organized as follows. Section I summarizes several theories of remitting behavior and the predictions they give for the relationship between education and remittances. Section II then describes the dataset of immigrant surveys with remittances. Section III provides results, and section IV draws some implications. I . T H E O R E T I CA L B AC K G RO U N D Theoretically, there are several reasons to believe that there will be differences in the remitting patterns of highly skilled and less-skilled emigrants. However, a priori, it is not clear which direction will dominate and thus whether the highly skilled will remit more or less on average. On the one hand, several factors tend to lead highly skilled migrants to be more likely to remit and to send a larger amount of remittances. First, highly skilled individuals are likely to earn more as migrants, potentially increasing the amount they can remit. Second, their education may have been funded by family members in the home country, with remittances serving as repayment. Third, skilled migrants are less likely to be illegal migrants and more likely to have bank accounts, lowering the �nancial transaction costs of remitting. On the other hand, several other factors might lead highly skilled migrants to be less likely to remit and to remit less. First, highly skilled migrants may be more likely to migrate with their entire household, so they would not have to send remittances in order to share their earnings with their household. Second, they might come from richer households, which have less need for remittances to alleviate liquidity con- straints. Third, they might have less intention of returning to their home country, reducing the role of remittances as a way of maintaining prestige and ties to the home community. 2. Convincing instruments are lacking to identify this impact. Bollard, McKenzie, Morten, and Rapoport 135 Before turning to the empirical analysis, it is useful to clarify the theoretical relationship between education and remittances and the implied testable predic- tions about education. This will allow identifying the role of several variables that, once interacted with education and various possible motivations to remit, have the potential to explain differences in remitting behavior by education level. The discussion is limited to three possible motives for remittances: altru- ism, exchange, and investment. These were selected for general empirical rel- evance and as the motives through which education is most likely to affect remittances.3 Altruism Altruistic preferences are generally captured by weighting one’s own (the migrant’s) and others’ (relatives) consumption in an individual utility function, Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 with weights reflecting the individual’s degree of altruism, which can itself depend on the closeness among the relatives considered (both family and geo- graphic proximity). For given weights and initial distribution of income, altruistic individuals maximize their utility by transferring (remitting) income so as to reach the desired distribution between themselves and the bene�ciaries of their altruism. Altruistic transfers take place if pretransfer income differences are suf�ciently large or altruism is strong enough and increases with the donor’s income (the extensive margin) and decreases with the recipients’ income (the intensive margin) What does this basic theoretical framework imply for the comparative remit- ting behavior of highly educated and less well educated migrants? First, edu- cated migrants tend to earn more, which all else equal should induce more remittances (at both margins). Second, the conventional wisdom is that edu- cated migrants tend to have more family members with them because of a higher propensity to move with their immediate family, which all else equal should lower remittances.4 Methodologically, this suggests that the location and composition of the family (which fraction of the family accompanies the migrant and which fraction stays in the country of origin) is jointly determined with remittances. This makes it dif�cult to estimate the causal impact of family composition on remittances. Instead, the analysis simply looks at whether differences in remitting patterns by education level disappear when they are conditioned on family composition. Empirically, the analysis will show that while less educated migrants have more relatives in the home country, they also have larger households and more relatives with them in the destination country. 3. See Rapoport and Docquier (2006) for a comprehensive survey of the economic literature on migrants’ remittances. 4. In this basic framework, education has no impact beyond its effect on the migrants’ income and family size, composition, and location, and altruistic preferences are independent of education. 136 THE WORLD BANK ECONOMIC REVIEW Exchange and Investment Motives There are many situations of Pareto-improving exchanges in which remittances “buy� various types of services, such as taking care of the migrant’s assets (land and cattle, for example) or relatives (children, elderly parents) at home. Such motivations are generally a sign of temporary migration and signal a migrant’s intention to return. In such exchanges, there is a participation constraint deter- mined by each partner’s external options, with the exact division of the pie (or surplus) to be shared depending on each partner’s bargaining power. How does education interact with such exchange motives? Two directions emerge from the short discussion above: one through the effect of education on intentions to return, and another through education’s effect on threat points and bargaining powers. The conventional wisdom is that migrants with higher education have less intention (and propensity) to return than do migrants with lower education (see Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Faini 2007), because they are better integrated or can obtain permanent resident status more easily. If that is the case, more educated migrants should transfer less for an exchange motive, reflecting their lower propensity to return.5 What about bargaining powers? Exchange models allow for different possible contractual arrangements reflecting the parties’ outside options and bargaining powers (see, for example, Cox 1987; Cox, Eser, and Jimenez 1998). This has two complemen- tary implications for education as a determinant of remittances in an exchange model. First, to the extent that education is associated with higher income, this relationship is likely to increase a migrant’s willingness to pay, leading to higher remittances; and second, to the extent that educated migrants come from more affluent families, this relationship is likely to increase the receiving household’s bargaining power, also leading to higher remittances. On the whole, an exchange motive therefore predicts that education will have an ambiguous effect on remit- tances, with the sign of the effect depending on whether return intentions or bargaining issues matter more to remittance behavior. The investment motive can be seen as a particular exchange of services in a context of imperfect credit markets. In such a context, remittances can be seen as part of an implicit migration contract between migrant and family, allowing the family access to higher income (investment motive) or less volatile income (insurance motive; Stark 1991). Since the insurance motive does not in theory give rise to clear differences in transfer behavior between highly educated and less educated migrants, the focus here is on the investment motive. The amount of investment �nanced by the family may include the physical costs (such as transportation) and informational costs of migration, as well as education expenditures, and repayment of this implicit loan through remittances is obviously expected to depend on the magnitude of the loan. Thus, the 5. Again, as shown later in the article, this conventional wisdom is not supported by the data; exchange motives are equally relevant for highly educated and less educated migrants as far as return intentions are concerned. Bollard, McKenzie, Morten, and Rapoport 137 investment motive clearly predicts that, all else equal, more educated migrants should remit more to compensate the family for the additional education expenditures incurred. Summary of Predictions Both the altruistic and the exchange motives for remittances yield unclear theoretical predictions as to whether more educated migrants remit more or less than do less education migrants. Once migrants’ incomes are controlled for, their education level should not play a role under the altruistic hypoth- esis (assuming preferences are exogenous to education) except for its effect on the spatial distribution of the family. As already noted, the conventional wisdom here is that the highly educated tend to move with their immediate family, which would lower remittances. Similarly, education is expected to lower remittances under the exchange hypothesis if educated migrants have lower propensities to return; bargaining mechanisms work in the other Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 direction and should translate into higher remittances, with the sign of the total expected effect being theoretically uncertain. Finally, education is likely to have a clear positive impact on remittances under the investment hypothesis. Given these expected mechanisms and the fact that the descriptive statistics for the sample do not support the conjecture that more educated migrants have a substantially higher propensity to move with their family or a substantially lower propensity to return, the other forces at work should be expected to dominate, so that migrants with more education would remit more, which is indeed what the analysis shows. I I . D ATA The micro-level database on remitting behavior created for this study is the most comprehensive available, comprising data on 33,000 immigrants from develop- ing countries derived from 14 surveys in 11 OECD destination countries that were the destination for 79 percent of global migrants to OECD countries in 2000 (Docquier and Marfouk 2005). The focus on destination country data sources enables looking directly at the relationship between education and remit- tance sending behavior by analyzing the migrants’ decision to remit. It also permits capturing the remittance behavior of individuals who emigrate with their entire household; using household surveys from the remittance receiving countries would typically miss such individuals. Since more educated individuals are believed to be more likely to emigrate with their entire household than less educated individuals (Faini 2007), using surveys from migrant sending countries would not be appropriate for examining the relationship between remittances and education. Most of the empirical literature on immigrants uses data from censuses or labor force surveys, but neither contains information on remittances. That 138 THE WORLD BANK ECONOMIC REVIEW requires special purpose surveys of immigrants. The authors pulled together all publicly available datasets they were aware of6 and six additional surveys that are not publicly available but that other researchers generously shared. Table 1 provides an overview of the database of migrants, summarizing the datasets, sample population, and survey methodology. Full details of the source of each dataset are in the supplemental appendix, available online at http://wber. oxfordjournals.org/. The database covers a wide range of populations. It includes both nationally representative surveys, such as the New Immigrant Survey (NIS) in the United States (drawn from green card recipients) and the Spanish National Survey of Immigrants (ENI), which draws on a neighborhood sampling frame, as well as surveys focusing on speci�c migrant communities within the recipient country, such as the Black/Minority Ethnic Survey (BME) in the United Kingdom and the Belgium International Remittance Senders Household Survey (IRSHS) of immigrants from the Democratic Republic of Congo, Nigeria, and Senegal. In all cases, the database includes only migrants Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 who were born in developing countries.7 For each country dataset, comparable covariates were constructed to measure household income, remittance behavior, family composition, and demographic characteristics. Remittances are typically measured at the house- hold rather than individual level. The level of analysis is therefore the house- hold, and variables are de�ned at this level whenever possible—for example, by taking the highest level of schooling achieved by any adult migrant in the household. All �nancial values are reported in 2003 U.S. dollars. In addition, any reported annual remittances that are more than twice annual household income are dropped. While remittance data in surveys can be subject to measurement error, the use of survey �xed effects will capture any common survey-level effects, and there is no strong reason to believe such measurement error would be correlated with education status. Mean and median reported remittances also seem to be of the right order of magnitude when compared with other surveys and migrant incomes. The sample weights provided with the data are always used. Data are pooled by poststratifying by country of birth and by education, so that the combined weighted observations match the distribution of developing country migrants to all OECD countries in 2000 (Docquier and Marfouk2005). The supplemen- tal appendix provides further details. Table 2 presents summary statistics for each country survey and the pooled samples of all destination countries. Overall, 37 percent of migrants in the database have completed a university degree, ranging from 4 percent in the Spanish Netherlands Interdisciplinary Demographic Institute (NIDI) survey to 6. Exceptions include longitudinal surveys of immigrants from Canada and New Zealand, which can only be accessed through datalabs in these countries, and so are not included here. 7. High income countries are de�ned based on the World Bank Country Classi�cation Code, April 2009. T A B L E 1 . Migrant Datasets Number of Dataset Name Year observationsa Population Methodology Australia LSIA Longitudinal Survey of Immigrants 1997 2,537 Primary applicant migrant arrivals Sample of of�cial records of those to Australia September 1993– August 1995 living in cities Belgium IRSHS International Remittance Senders 2005 377 Immigrants from DR Congo, Referrals through Embassies. Household Survey Nigeria, and Senegal France 2MO Survey of Households’ Transfer of 2007 713 Remitters to Algeria, Morocco, Interviews of remitters at post Funds to their Countries of Tunisia, Turkey and the of�ces in high-migrant regions Origin countries of Sub-Saharan Africa France DREES Pro�le and Tracking of Migrants 2006 4,278 New and regularized migrants with Sample of of�cial records Survey 1 þ year residence permits Germany SOEP German Socio-Economic Panel 2000 854 Resident population of the Federal Sample of of�cial records Study Republic of Germany in 1984. Italy NIDI Netherlands Interdisciplinary 1997 1,072 Egyptians and Ghanaians who Interviews at migrant meeting Demographic Institute immigrated within past 10 years places International Migration Survey Japan IDB Survey of Brazilians and Peruvians 2005 846 Latin American immigrant adults Interviews in 15 cities in Japan living in Japan Netherlands CSR Consumentenbond Survey of 2005 648 Major immigrant populations: Face-to-face interviews Remittances Moroccans, Turks, Surinamese, Antilleans, Somalis, and Ghanaians Norway LKI Living Conditions of Immigrants 1996 2,466 Immigrants from 10 countries: Representative survey of immigrant Survey Bosnia and Herzegovina, Chile, population from these countries Iraq, Iran, Pakistan, Serbia, Somalia, Sri Lanka, Turkey, and Vietnam Spain ENI National Survey of Immigrants 2006 9,234 Foreign-born who (intend to) live Sample of of�cial neighborhood Bollard, McKenzie, Morten, and Rapoport in Spain for 1 þ years rosters (Continued ) 139 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 TABLE 1. Continued 140 Number of Dataset Name Year observationsa Population Methodology Spain NIDI Netherlands Interdisciplinary 1997 1,020 Moroccans and Senegalese who Geographical sampling, & Demographic Institute immigrated within past 10 years references from sampled International Migration Survey UK BME Black/Minority Ethnic Remittance 2006 993 Migrant minorities who have Sampling of geographical blocks Survey remitted in past 12 months U.S. NIS New Immigrant Survey 2003 7,046 Migrants receiving green cards Sample of of�cial records May – November 1993 Pew National Survey of Latinos 2006 1084 Nationally representative sample of Sampled phone numbers in Latino respondents ages 18 and high-Latino areas older U.S. Pew Pew National Survey of Latinos 2006 1,084 Nationally representative sample of Sampled phone numbers in Latino respondents ages 18 and high-Latino areas THE WORLD BANK ECONOMIC REVIEW older Note: Number of observations used to calculate the �rst result in each column of table 2. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 T A B L E 2 . Survey Means, by Education Australia Belgium France France Germany Italy Japan Netherlands Norway Spain Spain UK U.S. U.S. Pooled Pooled Pooled Variable and education level LSIA IRSHS 2MO DREES SOEP NIDI IDB CSR LBI ENI NIDI BME NIS Pew Extensive Intensive Total Number of observations 2,656 451 717 4,280 900 1,153 1,065 836 2,466 10,282 1,113 1,152 7,352 1,304 33,022 28,981 26,276 Fraction with university education 0.32 0.59 0.07 0.18 0.20 0.21 0.14 0.11 0.12 0.23 0.04 0.20 0.34 0.06 0.36 0.37 0.37 Total remittances ($ per year) No university 286 1,681 1,380 368 2,724 2,662 993 988 3,099 2,691 375 1,530 699 793 650 University 379 2,475* 1,652 511 2,227 2,920 1,405* 743** 2,835 2,629 1,145** 671** 868 897 874 Fraction who remit No university 0.41 0.91 0.23 0.18 0.60 0.80 0.34 0.49 0.78 0.15 0.54 0.31 0.32 0.32 University 0.37 0.86 0.23 0.20 0.45** 0.90** 0.29 0.37** 0.48** 0.17 0.43 0.27** 0.27** 0.27** Log remittances No university 5.78 6.92 6.62 6.97 7.89 7.76 6.49 7.15 7.99 6.77 7.01 7.34 6.96 6.82 6.91 University 6.23** 7.29** 6.92 7.01 8.11 7.70 6.81** 7.22 8.49* 6.92 7.40** 6.97 7.02 6.97* 7.00 Household income ($ per year) No university 14,457 16,918 23,173 18,612 19,526 10,903 34,014 32,467 14,066 9,074 44,631 33,297 22,417 22,624 23,583 21,964 University 13,556 25,534** 31,301* 28,674** 21,984 13,302* 43,624** 41,995** 19,914** 10,168 50,565 61,084 34,729** 38,948** 38,669** 39,087** Log income No university 9.5 9.5 9.8 9.6 9.8 9.3 10.2 10.1 9.4 9.0 10.3 9.2 9.7 9.6 9.5 9.5 University 9.8** 9.8** 10.0 9.9** 9.8 9.4 10.4 10.3** 9.7** 9.2 10.4 10.0** 10.2** 9.9** 9.9** 9.9** Working No university 0.48 0.70 0.87 0.80 0.63 0.82 0.93 0.48 0.68 0.81 0.82 0.66 0.66 0.65 0.66 0.64 University 0.67** 0.74 0.86 0.86** 0.67 0.87 0.93 0.70** 0.73** 0.66 0.90** 0.78** 0.77* 0.75** 0.74** 0.73** Household size No university 3.81 1.88 2.51 2.90 1.80 1.53 3.82 1.84 3.33 4.10 3.44 3.73 3.76 University 3.44** 2.55** 1.90** 2.58 2.16** 1.76** 3.19** 1.95 3.04* 3.49** 3.17** 3.35** 3.36** Married No university 0.73 0.72 0.65 0.67 0.61 0.56 0.47 0.64 0.66 0.54 0.63 0.63 0.63 University 0.80** 0.51** 0.71* 0.59 0.60 0.48* 0.56** 0.51 0.86** 0.56 0.73** 0.74** 0.74** Spouse outside country No university 0.03 0.25 0.05 0.06 0.42 0.05 0.05 0.05 0.06 University 0.01* 0.19 0.01** 0.05 0.10** 0.03** 0.03** 0.03** 0.03** Number of children No university 1.29 1.16 1.78 1.06 2.50 2.06 1.58 2.25 2.37 1.99 2.05 2.03 University 1.22 0.89** 1.27** 1.00 2.15** 1.85** 0.62** 1.35** 1.81** 1.37** 1.37** 1.37** Children outside country No university 0.21 0.10 0.25 0.71 0.20 0.16 0.38 1.10 0.73 0.49 0.45 0.48 0.50 University 0.07** 0.06 0.17** 0.49* 0.15 0.09 0.26** 0.21** 0.31** 0.37 0.24** 0.25** 0.25** Number of parents Bollard, McKenzie, Morten, and Rapoport No university 1.97 1.13 0.95 1.35 1.42 1.27 2.18 1.81 1.84 1.83 (Continued ) 141 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 TABLE 2. Continued 142 Australia Belgium France France Germany Italy Japan Netherlands Norway Spain Spain UK U.S. U.S. Pooled Pooled Pooled Variable and education level LSIA IRSHS 2MO DREES SOEP NIDI IDB CSR LBI ENI NIDI BME NIS Pew Extensive Intensive Total University 2.32** 1.03 0.70** 1.32 1.35** 1.37 2.74** 2.18** 2.21** 2.23** Parents outside country No university 1.48 0.81 0.42 0.94 1.03 1.01 1.23 0.88 0.98 0.98 1.00 University 2.00** 0.88 0.54 0.67** 1.17* 1.04 1.33 1.26** 1.30** 1.31** 1.31** Years spent abroad No university 3.70 9.32 17.90 4.00 19.20 6.69 8.35 18.46 10.06 7.27 14.89 7.35 16.43 9.20 11.17 10.29 University 3.91** 12.28** 12.70** 4.21 13.51** 7.02 9.18 19.36 12.41** 6.74 14.66 7.05 18.34 8.06** 8.75** 8.40** Legal immigrant No university 1.00 1.00 0.84 0.51 0.66 1.00 0.87 0.84 0.85 University 1.00 1.00 0.85 0.39** 0.82* 1.00 0.85** 0.84 0.84 Will return home No university 0.02 0.45 0.06 0.23 0.39 0.01 0.08 0.35 0.63 0.09 0.19 0.09 0.16 0.11 University 0.04 0.65** 0.10* 0.17 0.53** 0.02 0.08 0.51 0.70 0.13** 0.14 0.09 0.12** 0.09* *Signi�cant at the 5 percent level ** signi�cant at the 1 percent level Note: Signi�cant results indicate that the mean of the variable is statistically different between university-educated and non-university educated house- THE WORLD BANK ECONOMIC REVIEW holds. See table 1 for full survey names. Source: Authors’ analysis based on data described in the text. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Bollard, McKenzie, Morten, and Rapoport 143 59 percent in the Belgium IRSHS. The table also summarizes the covariates by the maximum educational attainment of all adult migrants in the household. Altogether, including both the extensive and intensive margins, more highly educated migrants send home an average of $874 annually, compared with $650 for less educated migrants. There are two opposing effects of education: negative on the extensive margin, and positive on the intensive margin. At the extensive margin, migrants with a university degree are less likely to remit any- thing than migrants without a degree: 32 percent of low-skilled migrants send some money home, compared with 27 percent of university-educated migrants. However, conditional on remitting (the intensive margin), highly educated migrants send about 9 percent more than do less educated migrants. Characteristics that can affect remittance behavior differ between less and more educated migrants. First, more skilled migrants are both more likely to live in a household with working adults and to have a higher household income than are low skilled migrants. But contrary to conventional wisdom, Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 household composition does not differ much for migrants by education level: on average, only 6 percent of low skilled migrants have a spouse outside the country, compared with 3 percent of high skilled migrants. Low skilled migrants are signi�cantly less likely to be married (63 percent) than are high skilled migrants (74 percent). Low skilled migrants have more children (an average of 2.03 compared with 1.37 for high skilled migrants), as well as more children living outside the destination country (0.50) than do high skilled migrants (0.25). However, low skilled migrants also have more family inside the destination country than do high skilled migrants: the average household size for low skilled migrants is 3.76 people, statistically different from the mean household size of 3.36 people for high skilled migrants.8 Another piece of conventional wisdom, that more educated people are less likely to return home, is also not supported by the microdata. Indeed, more educated migrants have spent less time abroad (mean of 8.4 years) than have less educated migrants (10.3 years). Reported plans to return home are similar between the two groups: 9 percent of high skilled migrants report planning to return home, compared with 11 percent of low skilled migrants. While one should be cautious with treating both measures as truly reflecting return prob- abilities; at the very least, they do not indicate a strong tendency for the low skilled to be more likely to return. The simple comparison of means in table 2 shows differences in remittance behavior by education status. However, these comparisons show only that more educated developing country emigrants remit more than less educated developing country emigrants. This risks confounding differences in remittance 8. In some cases this might reflect households in which poorer, less skilled migrants live with other immigrants who are not family members. The database can identify the presence of a spouse, child, or parent in the home country household but cannot identify who migrants live with abroad or the extent to which they share resources within the household abroad. 144 THE WORLD BANK ECONOMIC REVIEW behavior among migrants from different countries with differences in remittance behavior by education level: the next section aims to separate these two differences. I I I . RE S U LT S : T H E RE L AT I O N SH I P B E T W E E N E D U CAT I O N AND REMITTANCES Results are reported for regressions of three remittance measures on education: total remittances (both extensive and intensive margins), an indicator for having remitted in the previous year (extensive margin), and log total remit- tances conditional on remitting (intensive margin; table 3). All regressions include country of birth �xed effects and dataset �xed effects. The key result is that more educated migrants remit more. In the pooled sample, migrants with a university degree remit $298 more per year than non- university educated migrants (row last, last column), with a mean annual remit- Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 tance for all migrants of $734. This overall effect is composed of a negative (statistically insigni�cant) effect at the extensive margin and a highly signi�cant positive effect at the intensive margin. The results are consistent when the second measure of education, years of schooling, is considered. Results for individual countries are mixed at the extensive margin, with edu- cation signi�cantly positively associated with the likelihood of remitting in two surveys (the U.S. NIS and the Survey of Brazilians and Peruvians in Japan), sig- ni�cantly negatively associated with this likelihood in three surveys (the U.S. Pew survey and both Spanish surveys), and no signi�cant relationship in the other six surveys, with three positive and three negative point estimates. One general observation is that a more negative relationship appears in surveys that focus on sampling migrants through community-sampling methods, such as the NIDI surveys, which take their sample from places where migrants cluster, and the Pew Hispanic surveys, which randomly dial phone numbers in areas with dense Hispanic populations. One might expect that educated migrants who live in such areas (and who take the time to respond to phone or on-the-street surveys) would be less successful than educated migrants who live in more inte- grated neighborhoods and thus who would not be picked up in these surveys. In contrast, at the intensive margin, 10 of 12 individual surveys show a posi- tive relationship between remittances and education, 5 of them statistically sig- ni�cant, and 2 show a negative and insigni�cant relationship. Thus it is not surprising that when the data are pooled there is a strong positive association at the intensive margin and that it outweighs the small negative and insigni�- cant relationship at the extensive margin in the total effect. This point is made graphically on a log scale in �gure 1, which plots the nonparametric relationship between total remittances and years of schooling, after linearly controlling for dataset �xed effects using a partial linear model (Robinson 1988), together with a 95 percent con�dence interval. The vertical lines demarcate the quartiles of years of schooling. Average remittances steadily T A B L E 3 : Coef�cients from Regressions of Remittance Measures on Education Australia Belgium France France Germany Italy Japan Netherlands Norway Spain Spain UK U.S. U.S. Pooled Pooled Pooled Variable LSIA IRSHS 2MO DREES SOEP NIDI IDB CSR LBI ENI NIDI BME NIS Pew Extensive Intensive Total Education measured by university degree Total remittances ($ per year) 58.4 922.8** 291.0 -526.6 237.5 -92.6 -168.8 769.5** -554.0* 298.0* Number of observations 2,537 377 854 1,072 846 9,234 1,020 7,046 1,084 24,033 Extensive: Remits indicator 2 0.019 -0.055 0.014 0.042 -0.065 0.091** 0.012 -0.049** -0.232** 0.038** -0.140* -0.018 -0.010 Number of observations 2,654 451 4,278 854 1,153 1,030 2,466 10,282 1,112 7,113 1,296 32,651 25,907 Intensive: Log remittances 0.341* 0.433** 0.363 0.492 0.073 -0.057 0.333** 0.093 0.430* 0.168 0.397* -0.199 0.249** 0.226** Number of observations 958 317 713 184 545 690 648 3,966 761 993 1,118 514 11,392 9,038 Education measured in years Total remittances ($ per year) 19.08* 86.50 26.39 -7.56 -3.03 2.40 -13.65 86.53 64.89 57.81 Number of observations 2,531 377 854 1,072 846 9,164 1,020 7,033 1,084 23,944 Extensive: Remits indicator 0.0080 -0.0042 0.0018 0.0145 0.0010 0.0024** 0.0008 -0.0023 -0.0072** 0.0034** 0.0010 0.0006 0.0014 Number of observations 2,648 451 5,529 854 1,153 1,030 2,450 10,201 1,112 7,100 1,296 32,535 25,807 Intensive: Log remittances 0.0441* 0.0341 0.0224* -0.0085 -0.0032 -0.0040 0.0247* 0.0199** 0.0091 0.0548* 0.0329* 0.0369 0.0256** 0.0229** Number of observations 956 317 713 184 545 690 648 3,942 761 993 1,116 514 11,364 9,010 Means Total remittances ($ per year) 316 2,159 1,399 396 2,621 2,692 1,040 932 3,089 2,679 633 1,479 764 2,466 734 Fraction who remit 0.40 0.85 0.23 0.19 0.53 0.77 0.34 0.41 0.75 0.15 0.46 0.30 1.00 0.27 Fraction with university 0.32 0.60 0.07 0.18 0.20 0.21 0.12 0.11 0.12 0.23 0.04 0.20 0.33 0.06 0.36 0.31 0.38 Years of education 13.4 14.2 7.7 12.0 11.5 14.1 13.3 10.7 12.2 11.4 7.5 13.4 13.4 9.4 12.9 12.3 13.0 *Signi�cant at the 5 percent level ** signi�cant at the 1 percent level Note: See table 1 for full survey names. Source: Authors’ analysis based on data described in the text. Bollard, McKenzie, Morten, and Rapoport 145 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 146 THE WORLD BANK ECONOMIC REVIEW Figure 1. Total Remittances by Years of Schooling Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Note: Figure depicts a semiparametric regression line from a partial linear model with dataset dummy variable evaluated at means; 95 percent pointwise con�dence intervals shown from 500 bootstrap repetitions. Vertical lines separate quartiles. Source: Authors’ analysis based on data described in the text. increase from around $500 in the lowest education quartile to close to $1,000 for those with university degrees. Moreover, the positive association increases most strongly for migrants with postgraduate education, which shows that not only do migrants with some university education remit more than those without, but also that migrants with postgraduate degrees remit more than those with only a couple of years of university. Robustness Although this database on remittances is the most comprehensive available, there are clear limitations, which make it important to see how sensitive the results are to alternative ways of using these surveys. First note that the results pertain only to migration in a sample of OECD countries. The surveys cover a large share of OECD destinations, but they omit other important destinations for developing country migrants such as the Gulf countries and South Africa. This is a limitation shared by the macro studies (Faini 2007; Niimi and others 2008), which also have data only for migrants in OECD countries. Nevertheless, the same forces acting on migrants in the OECD countries are likely to apply in these other destinations: more educated migrants will earn higher incomes and therefore remit more. Although data are rare, there is some evidence to support this is in a study of Pakistani migrants in the Gulf countries, which found that conditional on age and duration of Bollard, McKenzie, Morten, and Rapoport 147 stay, more educated Pakistani migrants remitted more (Abbasi and Hashmi 2000). Moreover, it is still the case that there are a large number of low skilled migrants in the OECD. A large majority of migrants in the pooled sample (63 percent) do not have a university education. A reasonable concern is whether surveys like the U.S. NIS, which capture only legal immigrants, are missing most of the low-skilled migrants. Comparing the skill distribution of immigrants included in the NIS with that of immigrants included in the U.S. Census (which is generally believed to do a good job surveying both legal and illegal migrants), does show a higher skill level in the NIS (12.26 mean years of education) than in the Census (10.84 years). However, once Mexican immi- grants are excluded (the group with the largest number of illegal immigrants), the skill distributions of the NIS (12.96 mean years of education) and the Census (12.21) are much closer, and16 percent of immigrants in both the NIS and the Census have 8 years of education or less. The �rst two columns of Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 table 4 then show that the results for the association between remittances and education continue to hold in the NIS (and, if anything, are more strongly posi- tive) when Mexican immigrants are excluded (table 4, columns 1 and 2). Columns 3 and 4 show that this is also true for the pooled sample of all surveys, which suggests that failure to capture illegal migrants in the survey is not driving the main result. A second potential concern is whether it is valid to pool so many different surveys with different sampling methods and differing degrees of representa- tiveness. Note that survey �xed effects are included in the regression analysis, so that only within-survey variation is used to identify the effect of education; the pooled estimate is thus a consistent estimate for the average association among the surveys. Nonetheless, as an alternative, the regressions are run only for the �ve surveys based on representative sampling from a list of migrants: (Longitudinal Survey of Immigrants to Australia (LSIA), the French Pro�le and Tracking of Migrants Survey (DREES), the German Socioeconomic Panel Study (SOEP), and the Spanish ENI and the NIS). The results show point esti- mates and levels of statistical signi�cance that are very close to those for the full pooled sample (see table 4, column 5). This demonstrates that the results are not being driven by the specialized surveys of particular migrant groups, such as the Japanese and Belgian surveys. Finally, one might query whether the results are being driven by students. That could influence the results based on university education if there were many students studying for undergraduate degrees who do not send remittances and do not yet have a college degree. There are several reasons to believe that this is not the main factor driving results. First, the LSIA and NIS surveys do not include students, which eliminates from the sample students in the countries that are among the most popular destinations for international study. Second, many international students come for postgraduate education, so they would be classi�ed as having a college education and remitting little, which 148 T A B L E 4 . Robustness Checks U.S. New Immigrant Survey(NIS) sample Pooled sample Excluding Nationally Excluding NIS representative Only migrants Only working Variable Full sample Mexicans Full sample Mexicans samples ages 25 þ migrants Education measured by university degree Total remittances ($ per year) 769.5** 839.7** 298.0* 306.8* 318.5* 267.7 308.7 Number of observations 7,046 5,922 24,033 22,909 19,643 21,343 16,693 Extensive: Remits indicator 0.038** 0.046** -0.010 0.000 -0.011 -0.023* 0.048** Number of observations 7,113 5,984 25,907 24,778 20,875 23,043 18,147 Intensive: Log remittances 0.397* 0.458** 0.226** 0.236** 0.220** 0.192** 0.230** Number of observations 1,118 982 9,038 8,902 6,220 8,303 7,360 THE WORLD BANK ECONOMIC REVIEW Education measured in years Total remittances ($ per year) 86.53 108.09 57.81 68.17 62.52 58.78 78.49 Number of observations 7,033 5,909 23,944 22,820 19,554 21,263 16,639 Extensive: Remits indicator 0.0034** 0.0021 0.0014 0.0016 0.0017 0.0007 0.0031* Number of observations 7,100 5,971 25,807 24,678 20,775 22,954 18,083 Intensive: Log remittances 0.0329* 0.0391* 0.0229** 0.0260** 0.0307** 0.0235** 0.0242** Number of observations 1,116 980 9,010 8,874 6,192 8,279 7,335 Means Total remittances ($ per year) 633 719 734 813 614 772 935 Fraction who remit 0.15 0.16 0.27 0.29 0.24 0.28 0.33 Fraction with university 0.33 0.40 0.38 0.38 0.38 0.38 0.38 Years of education 13.4 14.0 13.0 13.1 13.1 13.0 13.4 *Signi�cant at the 5 percent level ** signi�cant at the 1 percent level. Note: Nationally representative surveys are Australia LSIA, France DREES, Germany SOEP, Spain ENI and U.S. NIS. See table 1 for full survey names. Source: Authors’ analysis based on data described in the text. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Bollard, McKenzie, Morten, and Rapoport 149 would offset any effect of undergraduates.9 As a �nal check, the analysis is restricted to individuals who are working (table 4, last column). Since more educated individuals are more likely to be working, this eliminates one channel through which the more educated can earn more and thereby remit more. Nevertheless, even with this restriction, there is a signi�cant positive coef�cient at both the extensive and intensive margins, and the point estimate for total remittances is similar in magnitude, although it is not statistically signi�cant. Taken together, these results indicate that the basic �nding of a positive relationship between total remittances and education appears reasonably robust to alternative ways of combining the surveys. Channels This section uses these microdata to explore some of the channels through which education might influence remittances. Proxies are added to the model Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 to control for differences in household income and work status, in household demographics and the presence of family abroad, in time spent abroad, in legal status, and in intentions to return home. Table 5 shows the results of adding this full set of variables to the pooled model, using years of education as the measure of educational attainment. These channels operate as theory would predict. Households with more income and with adults who work more are more likely to remit: households where a migrant member is working send $345 more annually, with an extra $38 remitted annually for each 10 percent increase in income. As expected, family composition variables are also strongly signi�cant both overall and for the extensive and inten- sive margins: a spouse outside the country is associated with a colossal additional $1,120 remitted each year, approximately one and a half times the mean annual remittance for all migrants. Each child living outside the destination country is associated with an additional $340 remitted annually and each parent for an additional $180. Residing in the destination country legally is associated with an additional $400 annually, providing no evidence that legal migrants lose their desire to remain in contact with their home country. Migrants who plan to move back home also remit signi�cantly more, but this effect is primarily through the extensive margin rather than the intensive margin. Which channels account for the association between education and remit- tance behavior? Tables 6, 7, and 8 report how the coef�cient on education in an ordinary least squares regression changes as controls are added for total remittances, the extensive margin, and the intensive margin. Each panel of each table �rst shows the baseline education coef�cient from regressing remittances only on education and country of birth and dataset �xed effects (from table 3). Each succeeding row then shows changes in this coef�cient when controls are 9. In the United States, 47 percent of international students are studying for postgraduate degrees, compared with 12 percent for associate degrees and 32 percent for bachelor degrees (http://opendoors .iienetwork.org/?p=150827). 150 THE WORLD BANK ECONOMIC REVIEW T A B L E 5 . Remittances on Years of Education for Pooled Sample with All Controls Total Extensive Intensive Variable remittances Remits Log remittances Years of education 37.81 2 0.002* 0.017** (29.64) (0.001) (0.005) Log income 384.59** 0.023** 0.364** (105.37) (0.003) (0.034) Working 345.06** 0.113** 0.514** (90.80) (0.010) (0.065) Household size 2 8.14 2 0.002 0.015 (17.67) (0.002) (0.016) Married 2 89.77 0.004 2 0.097 (68.78) (0.010) (0.061) Spouse outside country 1,120.95** 0.145** 0.568** (236.04) (0.020) (0.097) Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Number of children 2 121.56** 2 0.006 2 0.099** (36.44) (0.003) (0.027) Children outside country 337.78** 0.048** 0.228** (75.14) (0.006) (0.039) Number of parents 2 47.07 2 0.020** 2 0.125** (53.56) (0.005) (0.045) Parents outside country 182.58** 0.063** 0.243** (38.02) (0.006) (0.045) Years spent abroad / 100 2,539.77 0.251** 1.744** (2,533.08) (0.095) (0.656) Years spent abroad squared / 100 2 31.43 2 0.010** 2 0.033* (27.14) (0.002) (0.015) Legal immigrant 398.79** 0.096** 0.167** (121.36) (0.018) (0.061) Will return home 692.30** 0.095** 0.085 (201.83) (0.021) (0.072) Number of observations 23,944 32,535 11,364 *Signi�cant at the 5 percent level ** signi�cant at the 1 percent level. Note: Includes dummy variables for missing covariates and �xed effects for country of birth and survey. Trimmed remittances greater than twice income. Pooled samples poststrati�ed by country and education. Source: Authors’ analysis based on data described in the text. added for income and work status, family composition, and all controls from table 5 (income and family controls, as well as legal status, time spent abroad, and intent to return home). Remittance behavior is accounted for primarily by income and not by differ- ences in family composition. The baseline result for total remittances from table 3, controlling only for country of birth and dataset �xed effects, is that migrants with a university degree remit $300 more than migrants without one. Controlling for the full set of covariates (the all row) reduces the coef�cient on university degree by two-thirds, and it becomes statistically insigni�cant. The third row adds just the family composition variables to the baseline T A B L E 6 . Education Coef�cient as Controls Are Added: Total Annual Remittances (U.S. dollars) Australia Belgium Germany Italy Japan Spain Spain U.S. U.S. Pooled Variable LSIA IRSHS SOEP NIDI IDB ENI NIDI NIS Pew total University education Baseline 58.4 922.8** 291.0 2 526.6 237.5 2 92.6 2 168.8 769.5** 2 554.0* 298.0* (61.1) (351.4) (275.6) (411.6) (374.1) (62.8) (749.4) (254.4) (227.2) (137.6) Income 2 10.1 557.0* 238.5 2 623.9 166.5 2 189.3** 24.7 396.6* 2 741.5** 102.3 (62.4) (281.4) (262.2) (407.2) (359.8) (63.5) (729.0) (174.4) (263.8) (92.8) Familya 29.8 534.7 237.8 2 306.7 317.5 2 112.8 2 6.9 623.6** 2 698.6** 228.2* (61.4) (310.5) (243.5) (394.7) (380.3) (57.6) (725.9) (204.7) (241.9) (103.1) All 2 16.5 475.8 144.6 2 539.6 328.7 2 181.7** 266.2 402.2** 2 835.7** 99.9 (62.1) (272.7) (179.8) (383.3) (365.3) (58.6) (698.6) (154.3) (269.9) (71.6) Number of observations 2,537 377 854 1,072 846 9,234 1,020 7,046 1,084 24,033 Years of education Baseline 19.08* 86.50 26.39 2 7.56 2 3.03 2.40 2 13.65 86.53 64.89 57.81 (9.01) (45.11) (29.37) (34.05) (7.92) (7.36) (19.95) (46.50) (44.97) (37.08) Income 7.99 47.80 3.51 2 32.44 2 2.59 2 13.39 2 26.95 44.98 49.18 32.12 (8.69) (38.28) (27.33) (33.39) (11.50) (7.41) (19.68) (40.00) (45.09) (31.98) Familya 17.03 29.28 25.56 47.31 2 1.86 3.93 10.32 80.78 47.95 55.43 (8.98) (38.45) (27.79) (34.93) (8.62) (6.84) (19.98) (44.75) (46.37) (34.24) All 8.86 33.77 9.66 22.64 1.99 2 7.57 4.50 54.81 27.01 37.81 (8.91) (36.94) (22.82) (32.79) (10.63) (6.84) (19.32) (37.32) (46.38) (29.64) Number of observations 2,531 377 854 1,072 846 9,164 1,020 7,033 1,084 23,944 *Signi�cant at the 5 percent level ** signi�cant at the 1 percent level. Note: Baseline row includes only country of birth and dataset �xed effects. Income row adds working dummy and log income to baseline. Family row adds seven family member controls to baseline. All row is full speci�cation from table 3. Trimmed remittances greater than twice income. Pooled samples poststrati�ed by country and education. See table 1 for full survey names. a. Includes household size, dummy variable if married, dummy variable if spouse is outside the country, number of children, number of children Bollard, McKenzie, Morten, and Rapoport outside the country, number of parents, and number of parents outside the country Source: Authors’ analysis based on data described in the text. 151 Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 152 T A B L E 7 . Education Coef�cient as Controls are Added: Remits Indicator Pooled Australia Belgium France Germany Italy Japan Norway Spain Spain U.S. U.S. Variable LSIA IRSHS DREES SOEP NIDI IDB LBI ENI NIDI NIS Pew Extensive Total University education Baseline 2 0.019 2 0.055 0.014 0.042 2 0.065 0.091** 0.012 2 0.049** 2 0.232** 0.038** 2 0.140* 2 0.018 2 0.010 (0.032) (0.029) (0.027) (0.048) (0.043) (0.034) (0.030) (0.015) (0.081) (0.011) (0.060) (0.010) (0.010) Income 2 0.052 2 0.112** 2 0.027 0.023 2 0.074 0.082* 2 0.020 2 0.062** 2 0.185* 2 0.000 2 0.165** 2 0.043** 2 0.033** (0.032) (0.029) (0.027) (0.047) (0.042) (0.035) (0.030) (0.015) (0.078) (0.011) (0.058) (0.010) (0.011) Familya 2 0.062 2 0.069* 0.015 0.039 2 0.046 0.088* 2 0.004 2 0.067** 2 0.234** 0.022 2 0.148* 2 0.031** 2 0.026* (0.032) (0.032) (0.027) (0.048) (0.041) (0.035) (0.030) (0.013) (0.081) (0.012) (0.060) (0.010) (0.011) All 2 0.080* 2 0.113** 2 0.027 0.028 2 0.065 0.083* 2 0.031 2 0.073** 2 0.177* 0.006 2 0.161** 2 0.043** 2 0.033** (0.031) (0.030) (0.027) (0.048) (0.038) (0.035) (0.030) (0.014) (0.075) (0.012) (0.059) (0.010) (0.011) Number of 2,654 451 4,278 854 1,153 1,030 2,466 10,282 1,112 7,113 1,296 32,651 25,907 observations Years of education THE WORLD BANK ECONOMIC REVIEW Baseline 0.0080 2 0.0042 0.0018 0.0145 0.0010 0.0024** 0.0008 2 0.0023 2 0.0072** 0.0034** 0.0010 0.0006 0.0014 (0.0043) (0.0040) (0.0025) (0.0084) (0.0040) (0.0005) (0.0025) (0.0018) (0.0021) (0.0012) (0.0060) (0.0009) (0.0010) Income 0.0014 2 0.0117** 2 0.0016 0.0071 2 0.0027 0.0035** 2 0.0027 2 0.0049** 2 0.0074** 2 0.0015 2 0.0035 2 0.0027** 2 0.0018 (0.0042) (0.0040) (0.0025) (0.0084) (0.0037) (0.0011) (0.0026) (0.0018) (0.0021) (0.0012) (0.0057) (0.0010) (0.0010) Familya 0.0018 2 0.0060 0.0050* 0.0152 0.0062 0.0019** 2 0.0006 2 0.0040* 2 0.0054* 0.0029* 0.0006 0.0000 0.0006 (0.0044) (0.0042) (0.0025) (0.0087) (0.0041) (0.0006) (0.0026) (0.0017) (0.0021) (0.0012) (0.0060) (0.0009) (0.0010) All 2 0.0025 2 0.0115** 0.0012 0.0130 0.0031 0.0034** 2 0.0037 2 0.0061** 2 0.0046* 0.0002 2 0.0059 2 0.0019* 2 0.0011 (0.0041) (0.0041) (0.0024) (0.0087) (0.0037) (0.0010) (0.0026) (0.0017) (0.0020) (0.0012) (0.0055) (0.0010) (0.0010) Number of 2,648 451 5,529 854 1,153 1,030 2,450 10,201 1,112 7,100 1,296 32,535 25,807 observations *Signi�cant at the 5 percent level ** signi�cant at the 1 percent level. Note: Baseline row includes only country of birth and dataset �xed effects. Income row adds working dummy and log income to baseline. Family row adds seven family member controls to baseline. All row is full speci�cation from table 3. Trimmed remittances greater than twice income. Pooled samples poststrati�ed by country and education. a. Includes household size, dummy variable if married, dummy variable if spouse is outside the country, number of children, number of children outside the country, number of parents, and number of parents outside the country Source: Authors’ analysis based on data described in the text. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Bollard, McKenzie, Morten, and Rapoport 153 speci�cation. The main hypothesis for why less skilled migrants remit more is that they are more likely to have family members outside the country. Therefore, controlling only for this variable would be expected to increase the coef�cient on education, but the opposite occurs: the coef�cient on education drops from $300 to $230 and remains statistically signi�cant. This casts doubt on the idea that low skilled migrants remit more because of their family com- position. One explanation is the earlier observation that low skilled migrants are not only likely to have more family abroad, but they are also likely to live in larger households in the host country. The second row of the table adds just the income variables (a dummy variable for working and log income) to the baseline speci�cation. The coef�cient on university degree falls by more than half and is no longer statistically signi�cant. This suggests that the income effect is a key channel through which education affects remittances—more edu- cated people send more money because they have higher incomes. Although education becomes insigni�cant after controlling for income in the Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 pooled sample, this result masks the heterogeneity in the individual surveys. For example, the education coef�cient remains statistically signi�cant even after con- trolling for all available covariates for three datasets: the Spanish ENI survey, the U.S. Pew dataset, and the U.S. NIS survey. There are several reasons why the education coef�cient might remain signi�cant in some datasets and not in others that cannot be examined with the dataset. One key variable that cannot be con- trolled for is the socioeconomic status of the family in the home country. More educated individuals might come from better-off families and therefore not need to send back as much money. This could explain the negative coef�cient in the ENI and the Pew datasets.10Or more educated individuals might have fewer ties to their home country. Time spent away from the home country and desire to return home are used to control for this, but they may not fully capture the strength of the ties. Also lacking are data on whether migrants are using remit- tance to repay family loans—for example, for education. One additional key issue is that the use of cross-section data does not yield any information about economic shocks that affect the migrant or the migrant’s family. Table 7 examines the extensive margin. In the baseline speci�cation, more educated migrants are less likely to remit anything, but this result is not statisti- cally signi�cant. The negative effect of education on the decision to remit any- thing is strengthened by the inclusion of different sets of covariates. The coef�cient on education (measured by university degree) is negative and signi�- cant once any covariates are included. The alternative measure of education, years of schooling, is not statistically signi�cant. Table 8 examines the intensive margin result, which again appears to be driven by the income effect. Adding 10. An alternative explanation may be that the high-earning highly educated migrants are less likely to respond to surveys. Survey methods that draw a sample from areas known to have a high concentration of migrants (such as the Pew survey) or from locations where migrants tend to congregate (such as the NIDI surveys) are especially likely to miss highly educated high-income individuals, who may be living in areas where there are fewer of their countrymen. 154 T A B L E 8 . Education Coef�cient as Controls Are Added: Log Remittances Australia Belgium France Germany Italy Japan Netherlands Spain Spain UK U.S. U.S. Pooled Pooled Variable LSIA IRSHS 2MO SOEP NIDI IDB CSR ENI NIDI BME NIS Pew Intensive Total University education Baseline 0.341* 0.433** 0.363 0.492 0.073 2 0.057 0.333** 0.093 0.430* 0.168 0.397* 2 0.199 0.249** 0.226** (0.145) (0.131) (0.211) (0.450) (0.171) (0.146) (0.116) (0.066) (0.202) (0.133) (0.169) (0.216) (0.060) (0.071) Income 0.237 0.243* 0.306 0.408 0.021 2 0.086 0.333** 0.040 0.367 0.097 0.023 2 0.278 0.143* 0.114 (0.138) (0.116) (0.203) (0.445) (0.165) (0.140) (0.116) (0.064) (0.200) (0.123) (0.168) (0.210) (0.058) (0.067) Familya 0.288* 0.258* 0.390 0.423 0.105 2 0.033 0.333** 0.092 0.495** 0.206 0.364* 2 0.253 0.246** 0.220** (0.139) (0.128) (0.207) (0.368) (0.178) (0.150) (0.116) (0.061) (0.187) (0.132) (0.166) (0.218) (0.057) (0.066) All 0.179 0.225 0.318 0.293 2 0.015 0.003 0.323** 0.054 0.409* 0.127 0.071 2 0.347 0.157** 0.118 (0.134) (0.118) (0.210) (0.309) (0.176) (0.138) (0.117) (0.059) (0.193) (0.123) (0.165) (0.206) (0.055) (0.063) Number of 958 317 713 184 545 690 648 3,966 761 993 1,118 514 11,392 9,038 observations Years of education THE WORLD BANK ECONOMIC REVIEW Baseline 0.0441* 0.0341 0.0224* 2 0.0085 2 0.0032 2 0.0040 0.0247* 0.0199** 0.0091 0.0548* 0.0329* 0.0369 0.0256** 0.0229** (0.0194) (0.0174) (0.0112) (0.0783) (0.0163) (0.0038) (0.0100) (0.0076) (0.0063) (0.0237) (0.0146) (0.0221) (0.0061) (0.0071) Income 0.0266 0.0103 0.0105 2 0.0387 2 0.0077 2 0.0041 0.0247* 0.0114 0.0021 0.0313 2 0.0008 0.0294 0.0135* 0.0112 (0.0199) (0.0164) (0.0115) (0.0770) (0.0164) (0.0048) (0.0100) (0.0075) (0.0062) (0.0220) (0.0126) (0.0216) (0.0053) (0.0062) Familya 0.0383* 0.0101 0.0344** 0.0053 0.0098 2 0.0042 0.0247* 0.0247** 0.0146* 0.0612** 0.0392** 0.0194 0.0272** 0.0243** (0.0187) (0.0167) (0.0111) (0.0649) (0.0168) (0.0037) (0.0100) (0.0072) (0.0066) (0.0235) (0.0144) (0.0231) (0.0060) (0.0070) All 0.0227 0.0060 0.0268* 2 0.0183 0.0009 2 0.0020 0.0274** 0.0179* 0.0086 0.0319 0.0172 0.0128 0.0169** 0.0139* (0.0193) (0.0165) (0.0124) (0.0511) (0.0159) (0.0050) 2 0.01 (0.0070) (0.0064) (0.0224) (0.0123) (0.0213) (0.0052) (0.0060) Number of 956 317 713 184 545 690 648 3,942 761 993 1,116 514 11,364 9,010 observations *Signi�cant at the 5 percent level ** signi�cant at the 1 percent level. Note: Baseline row includes only country of birth and dataset �xed effects. Income row adds working dummy and log income to baseline. Family row adds seven family member controls to baseline. All row is full speci�cation from table 3. Trimmed remittances greater than twice income. Pooled samples poststrati�ed by country and education. a. Includes household size, dummy variable if married, dummy variable if spouse is outside the country, number of children, number of children outside the country, number of parents, and number of parents outside the country Source: Authors’ analysis based on data described in the text. Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 Bollard, McKenzie, Morten, and Rapoport 155 only family variables to the baseline speci�cation reduces the coef�cient on uni- versity education by approximately 3 percent, but it remains highly signi�cant. However, if only income variables are added to the baseline speci�cation, the coef�cient becomes statistically insigni�cant, with approximately the same point value as the full speci�cation with the full set of covariates. I V. C O N C L U S I O N S The key advantage of this analysis over that in other papers in this literature (Faini 2007; Niimi and others 2008) is the ability to link the remittance decision of migrants with their education level and therefore directly answer the question of whether more educated migrants remit more. Cross-country macroeconomic analyses that relate the amount of remittances received at a country level to the share of migrants with tertiary education can at best tell us whether countries that send a larger share of highly skilled migrants receive Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 less or more remittances than countries that send fewer skilled migrants, without accounting for the other differences between countries that could underlie such a relationship. This new database on migrants allows direct examination of the relationship between education and remittance decisions. Results for the extensive margin (the decision to remit at all) and the intensive margin (the decision on how much to remit) combined show that, at least in this combined sample, more educated migrants remit signi�cantly more: migrants with a university degree remit $300 more yearly than migrants without one. Nonetheless, there is some heterogeneity across destination countries, with negative point estimates in a few of the surveys used—mainly in surveys that sample migrants only in areas where migrants cluster, thereby missing more educated, higher earning migrants who may live outside of the immigrant clusters. The data also allow analysis of several competing theoretical channels that help to explain this result. Differences in household composition between high and low skilled migrants do not explain the observed remittance behavior. One explanation may be that although low skilled migrants are more likely to have a spouse and children in the home country, they have larger families on average than do high skilled migrants and tend to live in larger households in the host country. In contrast, there is considerable support for an income effect as the dominant channel through which education operates. More educated migrants earn more money and therefore remit more than low skilled migrants. The article also highlights the clear limitations of existing microdata on remit- tances. While some basic information on migrants can be obtained from census microdata and government immigration records, there are no comparably reliable sources for remittances. The new database relies on specialized one-off surveys of migrants. Given the importance of remittances for many developing countries, it would be bene�cial for migrant-receiving countries to include ques- tions on remittances in their regular labor force or household budget surveys. 156 THE WORLD BANK ECONOMIC REVIEW This would be a �rst step to being able to analyze how remittance patterns change as countries pursue more skill-selective immigration policies. Policy debates on migration often raise concerns about the potential negative effects of the “brain drain� on developing countries. However, the main �nding that remittances increase with education illustrates one bene�cial dimension of high skilled migration for developing countries. High skilled migrants work in better jobs and earn more money than low skilled migrants and in turn send more money back home in remittance flows. This suggests that the fear that remittances will fall as the migrant skill level rises is not sup- ported by existing empirical evidence. REFERENCES Abbasi, Saif-Ur-Rehman Saif, and Arshad Hussain Hashmi. 2000. “Migrants Earning at Overseas Job Downloaded from wber.oxfordjournals.org at International Monetary Fund on August 12, 2011 and Extent of Remittances Transferred to their Families in Pakistan.� International Journal of Agriculture and Biology 2 (3): 222–25. Cox, Donald. 1987. “Motives for Private Transfers.� Journal of Political Economy 95 (3): 508–46. Cox, Donald, Zekeriya Eser, and Emmanuel Jimenez. 1998. “Motives for private transfers over the life cycle: An analytical framework and evidence for Peru.� Journal of Development Economics 55: 57–80. ´ de Docquier, Fre ´ ric, and Abdeslam Marfouk. 2005. “International Migration by Education Attainment, ¨ zden, and M. Schiff, 1990–2000.� In International Migration, Remittances and the Brain Drain, ed. C. O 151–99. New York: Palgrave Macmillan. Faini, Riccardo. 2007. “Remittances and the Brain Drain: Do more Skilled Migrants Remit More?� World Bank Economic Review 21 (2): 177–91. Groenewold, George, and Richard Bilsborrow. 2004. “Design of Samples for International Migration Surveys: Methodological Considerations, Practical Constraints and Lessons Learned from a Multi-Country Study in Africa and Europe.� Paper presented at the Population Association of America 2004 General Conference, Boston, MA, April 1–3. IDB (Inter-American Development Bank). 2005. “Survey of Brazilians and Peruvians in Japan� Survey commissioned by the Multilateral Investment Fund. Miotti, Luis, El Mouhoub Mouhoud, and Joel Oudinet. 2009. “Migrations and Determinants of Remittances to Southern Mediterranean Countries: When History Matters.� Paper presented at the 2nd Migration and Development Conference, Washington, DC, September 10– 11. Niimi, Yoko, C ¨ zden, and Maurice Schiff. Forthcoming. “Remittances and the Brain Drain: ¸ aglar O Skilled Migrants Do Remit Less.� Annales d’Economie et de Statistique. OECD (Organisation for Economic Co-operation and Development). 2007. Policy Coherence for Development 2007: Migration and Developing Countries. Paris: OECD. Rapoport, Hillel, and Frederic Docquier. 2006. “The Economics of Migrants’ Remittances.� In Handbook of the Economics of Giving, Altruism and Reciprocity, ed. S.-C. Kolm, and J. Mercier Ythier, 1135–98. Amsterdan: North-Holland. Robinson, Peter M. 1988. “Root-N Consistent Semiparametric Regression.� Econometrica 56: 931– 54. Siegel, Melissa. 2007. “Immigrant Integration and Remittance Channel Choice.� Working Paper 2007/ 09. Maastricht Graduate School of Governance. Stark, Oded. 1991. The Migration of Labor. Cambridge, MA: Basil Blackwell. World Bank. 2008. Migration and Remittances Factbook 2008. Washington, DC: World Bank. ———. 2009. “Migration and Development Brief No. 9.� http://siteresources.worldbank.org/ INTPROSPECTS/Resources/MD_Brief9_Mar2009.pdf. Accessed July 10, 2009. Forthcoming papers in THE WORLD BANK ECONOMIC REVIEW • Has India’s Economic Growth Become More Pro-Poor in the Wake of Economic Reforms? Gaurav Datt and Martin Ravallion • Are The Poverty Effects of Trade Policies Invisible? Monika Verma, Thomas Hertel, and Ernesto Valenzuela • Corruption and Confidence in Public Institutions: Evidence from a Global Survey Bianca Clausen, Aart Kraay, and Zsolt Nyiri • Agricultural Distortions in Sub-Saharan Africa: Trade and Welfare Indicators, 1961 to 2004 Johanna Croser and Kym Anderson • Thresholds in the Finance-Growth Nexus: A Cross-Country Analysis Hakan Yilmazkuday • The value of vocational education: High school type and labor market outcomes in Indonesia David Newhouse and Daniel Suryadarma • Disability and Poverty in Vietnam Daniel Mont and Nguyen Viet Cuong THE WORLD BANK 1818 H Street, NW Washington, DC 20433, USA World Wide Web: http://www.worldbank.org/ E-mail: wber@worldbank.org