75257 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