University of Groningen & University of Göttingen
Master Thesis
Destination Country Determinants of
International Immigration
Susanne Margarethe Wischnath
International Double Degree Program
M.Sc. International Economics and Business and M.A. International Economics
Student numbers: s2982110 (Groningen) and 21419692 (Göttingen)
Email: susanne.wischnath@gmail.com
Supervisor Co-assessor
Dr. Robbert Maseland Dr. Katharina Werner
Faculty of Economics and Business Faculty of Economic Sciences
University of Groningen University of Göttingen
Abstract
In the light of rising populations in developing countries and global inequalities, the pressures to migrate to developed countries will further increase, simplified through falling transportation and communication costs and existing migrant networks. This thesis studies annual immigration to 30 OECD countries between 1985 and 2012 to identify important time-varying destination country characteristics that attract immi-grants. I find that next to per capita income, immigrants respond to changes in the stock of foreigners, human capital, the unemployment rate, and social expenditures. The role of the age structure in the destination is ambiguous. The thesis serves as a basis for further research on the demand side of immigration and the development of migration policies.
Contents
1. Introduction 1
2. Literature Review 3
3. Methodology and Data 17
4. Results 30
5. Conclusion 38
References 42
A. Appendix 46
List of Figures
1. Stock of international migrants in millions, 1990-2015 . . . 3
2. International migrants relative to total population, 1990-2015 . . . 3
3. World Population Prospects: total population in millions . . . 4
4. World Population Prospects: potential support ratios (15-64/65+) . . . 5
5. Yearly immigration in thousands, 1985-2013 (IMD) . . . 21
6. Immigration to the OECD (aggregated immigration in thousands) . . . 22
7. MIPEX and immigration in 2012 . . . 27
8. Graphical illustration of fixed effects using the LSDV model . . . 34
A1. World Population Prospects: potential support ratios OECD countries
(15-64/65+) . . . 46
A2. Yearly immigration in thousands, 1985-2013 . . . 48
A3. Yearly immigration rates, 1985-2011, compared to the sample average . 50
List of Tables
1. Income differentials relative to the OECD average, 2011 . . . 24
2. Summary statistics . . . 28
3. Summary statistics for the individual countries . . . 29
4. Correlation matrix . . . 29
5. Comparison of aggregated and bilateral immigration flows: the role of income . . . 31
6. Destination country determinants of international immigration . . . 33
7. Country dummies: comparing columns (2) and (8) of Table 6 . . . 35
A1. Availability of Migration Data . . . 53
A2. Correlation matrix income measures, 828 observations . . . 53
A3. Comparison of income measures . . . 54
A4. Destination country determinants: alternative income measures . . . . 55
1. Introduction
Globalization is not only the flow of capital and goods between countries, but also in-ternational migration. In this sense, migration means more than the simple movement of production factors from one country to another and it is more complex than inter-national trade and investment. Migrants are not only workers who aim to maximize wages; they carry ideas and knowledge, increase demand for consumption goods and require public transfer services in the destination countries. The connections between countries are intensifying and expanding, which inevitably increases the awareness of inequality in living conditions and welfare. It is therefore comprehensible that people decide to emigrate to benefit from the existing differences (Martin, 2009).
Another factor contributing to more international mobility is population growth. At the end of 2015, more than 7.3 billion people lived in the world. The number is increasing – although very differently in different regions. While developed countries face overaging societies, the growth mainly takes place in developing countries where birth rates remain high, directly leading to an increase in the number of migrants (UN, 2015a). Immigration brings together people of different ethnic origins in the job market and in private life, engendering cultural exchange but also cultural issues. Governments in rich countries therefore introduce policies that control or restrict immigration to protect social security systems and the social and political stability within the countries. Furthermore, they need to support integration of those who arrive and find solutions to combat the causes for migration in the source countries. It is undoubted that the global migrant stock would be much higher if people were allowed to move and settle freely. It is also obvious that it will further increase in the future (Bodvarsson and Van den Berg, 2013). Identifying destination country characteristics that influence the number of immigrants can help to explain the attractiveness of one country over another and serve as an important basis for policy advice.
migration model by Sjaastad (1962) where migrants are seen as investors in human capital and develop new hypotheses. The destination country environment has certain qualities that attract or deter potential migrants, the so-called “push” and “stay away” factors. These factors might partly also reflect the destination countries’ demand for immigrants. So while the hypotheses stem from migration theory, they also contribute to a more demand-sided analysis of immigration flows. In general, immigration policies shall control and guide immigration and are hence also one main destination country determinant of immigration. Unfortunately, the effectiveness of these policies is hard to assess and there exists no common classification of the degree of the policy changes. In this regard, finding out more about general determinants can also promote the research on migration policy.
2. Literature Review
Global Demographic Change and International Migration Dynamics The population division of the United Nations recently published data on the trends in the international migrant stock (UN, 2015b). Figure 1 shows the increase from over 152 million migrants in 1990 to almost 244 million migrants in 2015 – the reason why the 21st century is often referred to as “the age of migration” (Castles and Miller, 2009). In 1960, the number of migrants ranged around 78 million, approximately half the size of 1990. To put the numbers into perspective, one needs to consider the overall population growth. Consequently, Figure 2 shows the stock of international migrants as a percentage of total population. Globally, the share has risen only slightly and is around 3 percent. However, there are strong differences between regions: in more developed regions, the share of migrants is much larger and increasing. This is caused by two factors: first, rich countries are more attractive destinations, which is also reflected in the higher total stock of migrants living there. Second, population growth in the developed regions is much slower and not as strong as the growth in the stock of migrants. When studying global migration movements it is therefore also relevant to comprehend global demographic change and population dynamics.
Figure 1: Stock of international migrants in millions, 1990-2015 50 100 150 200 250
Stock (in millions)
1990 1995 2000 2005 2010 2015
Year
World Developed regions
Developing regions
Figure 2: International migrants relative to total population, 1990-2015 2 4 6 8 10 12 Percent 1990 1995 2000 2005 2010 2015 Year
World Developed regions
Developing regions
regions (UN, 2013)1. Figure 3 demonstrates: even if the global percentage of
interna-tional migrants remains the same, the total stock of migrants will strongly increase. It also shows that population and migration dynamics are interlinked.
Figure 3: World Population Prospects: total population in millions
0 2 4 6 8 10 12
Population (in millions)
1950 2000 2050 2100
Year
World More developed regions
Less developed regions
In most OECD countries, a baby boom took place after World War II. Fertility levels began to decline around 1965 and fell even below the replacement level of 2.1 births per woman. After an economically favorable situation with a large labor force, the dependency rates are increasing, which leads to pressure on social welfare systems. Besides, life expectancy increased strongly, reaching 65 years in 2000. The increase is mostly related to improvements in medical care and health services, which led to declining child mortality rates and longer lives for old people (Bloom and Canning, 2004). The potential support ratio (PSR) of a country is the number of people between 15 and 64 per person aged 65 or above (UN, 2002). Figure 4 shows global PSRs based on United Nations projections (2015a, again assuming the medium fertility variant). It becomes clear that the PSR in the OECD countries is considerably lower than the world average, although the values are expected to converge within the twenty-first century.
1The Population Division of the United Nations Department of Economic and Social Affairs
Figure 4: World Population Prospects: potential support ratios (15-64/65+) 0 5 10 15 PSR 1950 2000 2050 2100 Year
World More developed regions
Less developed regions OECD average
Figure A1 in the appendix shows the PSRs of the OECD countries compared to the
OECD average, revealing large heterogeneity across countries2. In 2015, Japan was
affected most by overaging with the lowest PSR (2.3) of all countries. Most European countries also lie below the OECD average of 4.3, whereas the United States has a level above average (4.5). In the United States the fertility rate is higher than in Europe, where net migration rates are lower, too. This has implications for population growth and aging and the accompanying dependency ratios and PSRs. While immigration is not a major part of the classic demographic transition model, it is well a third channel that might affect the population structure of countries next to birth and death rates (Bloom and Canning, 2004).
Starting in the developed countries, the demographic transition took place in many developing countries and is expected to spread globally when improvements in health reach all countries (Bloom and Canning, 2004). However, the time difference in the transition processes has consequences for global migration flows. While the population sizes in developed countries shrink or stagnate, the number of people in developing countries continues to increase (Martin, 2009). Thus, there are on the one hand
coun-2Estonia, Greece, Slovenia and Turkey are not depicted in Figure A1 because the countries are
tries facing increasing dependency ratios and a lack of labor supply. On the other hand, the pressure to emigrate increases in developing countries when a large generation of young workers is faced with structural unemployment (Bloom and Canning, 2004). Besides population growth in developing countries, there are other reasons to expect that migration to developed countries will further rise in the future. . Inequalities between countries are one of the main pressures to migrate. As these inequalities – typically measured in income differences – continue, many people will be willing to leave their home to find a better life somewhere else. And in general, differences in economic, political and environmental conditions fuel the flow of migrants. The dynamics are facilitated through increases in mobility that arise from shrinking costs of communication and transportation and migrant networks that already exist in the destination countries (Skeldon, 2013). Policy makers and governments in destination countries face the challenge to implement policies that ensure the best possible outcome for the migrant, the source and the destination country.
The policies can be summarized in three broad fields. First, in order to reduce the pressures to emigrate, poor countries need support to further develop. When combatting the main causes of migration, the incentive to leave would diminish. Second, destination countries need to foster integration of foreigners to ensure social peace and political stability in the labor market and social life. Finally, being faced with large flows of immigrants, governments need to implement policies to control and restrict migration. In this regard, singling out the central determinants of past migration flows is important to develop useful methods for future projection and policy advice.
The Determinants of Migration
the potential migrant as opposed to negative “stay away” factors as discrimination or uncertainty. When push and pull factors are stronger than stay and stay away factors, the individual emigrates and global migration rates increase. Apart from these fac-tors, the migration decision is influenced by the costs that arise from resettlement and formal exit and entry barriers as visa regulations or migration quotas. To accurately model the migration decision, all these factors need to be taken into account, so that migration theory is not only consisting of economic theory but also relies on important contributions from other disciplines as sociology, psychology, demographics, or political economy (Bodvarsson and Van den Berg, 2013).
where Vij is the net present value of migrating from country i to country j:
Vij =
Z n
t=0
[Ej(t) Ei(t) Cj(t) + Ci(t)]e rtdt Iij (1)
An individual will move from home country i to foreign country j if Vij > 0. Ej(t)
and Ei(t) are the earnings in home and foreign, Cj(t) and Ci(t) represent the costs
of living. r is the discount rate to adjust future costs and benefits and n the time
horizon. The last element of the formula, Iij, captures all initial costs of moving from
one country to another. The time horizon adds age as a central component to the equation: the younger the individual, the longer the remaining working time and the more likely the migration to exploit the earning differential (Shields and Shields, 1989). Equation (1) provides a useful theoretical basis for empirical regression models. Empir-ical evidence on the determinants of international migration flows is relatively limited. Clark et al. (2007) state: “While the literature is long on examining the outcomes of immigration, it is surprisingly short on estimating the determinants of immigra-tion” (359). The shortage of research is partly explained by a lack of data in source and also destination countries. While there exist official numbers of migration inflows, illegal and return migration are often not captured. And even the official numbers differ regarding the measuring method and are often incomplete, which makes cross-country comparison difficult (Bodvarsson and Van den Berg, 2013). The dependent variable, migration, can take various forms, for example the migration rate or total im-migration. Following the gravity equation, bilateral migration flows are often modeled as a function of the distance between the two countries, the population sizes, the per capita incomes and a control vector that contains bilateral characteristics. Researchers’ choices of variables included in the regression vary and are affected by the availability of data and the research focus.
1985 and 2012. I hope to identify the main destination country characteristics without specifically controlling for source country and bilateral features. Further explanations about the exact methodology are described in the following section. In the past, the most important destination country feature has been per capita income, directly fol-lowing from the theoretical model. However, GDP per capita has been criticized for being an inappropriate measure for well-being (Costanza et al., 2009) and it might be interesting to take other indicators into account, too.
Examining the previous literature is useful to identify potentially relevant variables. When analyzing the determinants of migration, the existing literature mainly focuses on only one destination country (e.g. Clark et al., 2007) or a cross-sectional analysis (e.g. Grogger and Hanson, 2011), but there are also examples of panel data analysis with various destination countries (e.g. Mayda, 2010). A panel data approach has the advantage to identify general patterns. Because it is impossible to capture all relevant characteristics that explain for example why the United States are receiving more immigrants that Norway, one usually uses a fixed effects model, estimating how the change in immigration flows can be explained by changes in other variables. On the following pages, I therefore develop hypotheses that shall help detect the most important – time-varying – destination country determinants of immigration.
Income
As mentioned above, the first and most prominent determinant of international mi-gration is the difference in incomes. This follows from Equation (1) and also from the basic labor-flow models, where factor mobility, i.e. the mobility of capital, labor and
other production factors, is the reaction to differences in the rates of return3. When
income in the destination country rises, more people will want to move there. In em-pirical regressions usually a logarithmized form of the gravity equation is used. It is emphasized that the attractiveness of foreign consists of the differential in per capita incomes, and not the differential in GDPs (Lewer and Van den Berg, 2008). Authors have used different specifications of per capita income and confirmed the positive ef-fect of an increase in destination countries’ income on migration flows. Examples are Clark et al. (2007) who study immigration to the United States and use the ratio of
3Shields and Shields (1989) express the basic equation that regards migrants simply as suppliers
of labor as: Mij = ij(wj wi). Labor, here only seen as a factor of production, moves from i
to j as long as wj > wi, i.e. until wages are equalized. here captures barriers to migration
average, purchasing power parity-adjusted income in the source country relative to the
United States, Mayda (2010) who investigates immigration into 14 OECD countries4
by country of origin between 1980 and 1995 and uses the per worker GDP, purchasing power parity-adjusted, and Ortega and Peri (2013) who construct a dataset that
cov-ers migration flows from 120 source to 15 OECD countries5 from 1980 to 2006 and use
GDP per capita, purchasing power parity-adjusted as the main explanatory variable. This leads to the first hypothesis:
Hypothesis 1aThe higher the per capita income in destination country j,
the larger the immigration flows to j.
Theoretical models emphasize income differences and distance as the main determi-nants of immigration. An empirical model that studies total immigration thus lacks an important dimension: controlling for source country income and distance is impossible. Hatton and Williamson (2002) provide a solution when developing a simple migration model to explain the fundamental determinants of global migration. They introduce two education-adjusted relative income terms and compare country j to the world and to the region in which it is located. Including income relative to the region may at least to some extent capture the undeniable impact of distance on migration. In general, the income differentials are more accurate because they also consider income changes in other countries. The variables are expected to have a positive impact on immigra-tion: the richer country j compared to the world or the region, the more immigrants are attracted. Concentrating on the pure difference in incomes to reflect the relative economic attractiveness of country j, I build two income differentials without adjusting for education. The same has been done by Kugler et al. (2012) who also waive the education adjustment and relate income to the OECD average. I end up testing three variants of the income differential:
Hypothesis 1b The higher the per capita income in the destination
country j relative to the global, regional, or OECD average, the larger the immigration flows to j.
4Australia, Belgium, Canada, Denmark, France, Germany, Japan, Luxembourg, Netherlands,
Nor-way, Sweden, Switzerland, United Kingdom, United States
5Australia, Belgium, Canada, Denmark, Finland, Germany, Italy, Netherlands, New Zealand,
Stock of foreigners
The last part of Equation (1), Iij, captures all potential factors that make moving
from i to j more or less costly. One important non-economic factor is known as the “friends and relatives effect” (Pitfield, 1973:139). It stands for the former immigrants from the same source country who induce a reduction in the costs of migration. Compa-triots inform and support newcomers, which might also influence the well-being. While some bilateral models include the number of compatriots, this is impossible when total immigration is the dependent variable. Nevertheless, a large stock of immigrants may indicate an extensive network for migrants and therefore reduce in migration costs. It can also be seen as a pull factor that increases country j’s relative attractiveness. The positive relationship is proven by Hatton and Williamson (2002) and Kugler et al. (2012) who show that the percentage of foreign-born in the country has a positive effect on the net immigration rate. This leads to the next hypothesis:
Hypothesis 2The higher the share of international migrants in the
desti-nation country j, the larger the immigration flows to j. Human Capital
The first hypotheses follow directly from migration and human capital theory and form part of most empirical regressions. In the Sjaastad model, the earning differential is the main determinant for migration after adjusting for the differences in the cost of living. Beyond, researchers usually concentrate on source country determinants and the bilateral relation between source and destination country. Nevertheless, it might be helpful to develop a more multilayered framework of destination country characteristics to find other channels. As stated above, the migration decision is about maximizing the net present value of lifetime earnings. However, the theory has some shortcomings because it does not entail other, non-economic motives. These can be of social or political nature, involve differences in consumption opportunities or amenities or the decision to move might be a family instead of an individual decision (Bodvarsson and Van den Berg, 2013). From all additional push, pull, stay and stay away factors, I focus here on destination countries.
char-acteristics embedded in labor” (p. 85). The Penn World Table (PWT) offers data on an index of human capital per person that is based on years of schooling and returns to education (Feenstra et al., 2015). This makes it possible to study the differences between countries and the development of human capital within a country over time. The index is insofar a suitable supplement to the framework because differentials in hu-man capital might have an important magnetic effect on immigrants: especially highly educated people, i.e. people with a high amount of human capital, are willing to emi-grate to countries that are also rich in human capital to better exploit employment and education opportunities. The phenomenon of international migration of skilled workers is often referred to as “brain drain” (e.g. Docquier et al., 2007). And in general, higher human capital index can be interpreted as a sign for innovation and development and therefore attracts. Consequently, hypothesis 3 states:
Hypothesis 3 The higher the human capital index in the destination
country j, the larger the immigration flows to j. Macroeconomic Indicators
One indicator that measures the labor market performance of a country is its unem-ployment rate. It is straightforward that when deciding to leave the country of origin, a migrant needs to assess his chances in the foreign labor market as realistically as possible. One shortcoming of the Sjaastad model is that it assumes that the individ-ual directly obtains a job in country j with a probability of 100 percent. Based on Equation (1), migration would take place as long as wages are the same everywhere (after adjusting for living standards and costs of migration). The Harris-Todaro model (1970) provides a useful extension. Focusing on rural to urban migration, the authors consider that finding employment in the urban sector might take some time and the individual has no (or only low) income after moving. Uncertainty can easily be added to the mathematical model as in Equation (2) where p(t) is the probability that the individual will have a job in the destination country at time t.
Vij =
Z n
t=0
[p(t)Ej(t) Ei(t) Cj(t) + Ci(t)]e rtdt Iij (2)
labor in the destination country. It thus reflects both the capacity to integrate migrants in the labor market and the attractiveness of country j when an individual is assessing the migration decision. This leads to the fourth hypothesis:
Hypothesis 4 The lower the unemployment rate in the destination
country j, the larger the immigration flows to j. Age Structure
and give birth to more children, influences the dependency ratios (Bloom and Canning, 2004).
One of the first works discussing the idea of replacing births by migration was published by Lesthaeghe et al. in 1991 who compare fertility and immigration rates in Western Europe and the United States around the 1970s. After a huge inflow of guest workers in Western European countries in the 1960s, the numbers were declining again while remaining high in the United States. Lesthaeghe et al. project how many immigrants would be necessary to prevent the European countries’ population sizes from falling. Building upon this work, the Population Division of the United Nations defined the term “replacement migration” as the “international migration that would be needed to offset declines in the size of population of working age, as well as to offset the overall ageing of population” (UN, 2001: 93). The department calculates projections for sev-eral developed countries with below replacement fertility rates. The resulting numbers are extremely high, suggesting that migration can only be seen as a partial solution to population aging. Policies fostering integration need to be implemented, but also alter-native solutions as an increase in the retirement age and labor force participation, and reforms of benefits for the elderly. Coleman (2002) offers the most prominent response to the UN publication when studying population aging and replacement migration in the United Kingdom. He argues that mass migration only marginally influence the age structure but also the cultural environment of the country. The UN publication on replacement migration should hence not be understood as a call for lifting migration rates to unrealistically high levels but rather as an appeal to develop and implement a broader range of policies to reduce the pressure caused by population aging.
In this context, although being a controversial issue, it becomes interesting to assess whether the age structure of the destination country is also a determinant of interna-tional migration flows, reflecting the demand of the destination country to prevent its population from overaging. In my empirical regression I will test whether the PSRs of the OECD countries are related to immigration:
Hypothesis 5 The lower the potential support ratio in the destination
Policy Environment
Another very important explanatory variable is the immigration policy in the destina-tion country. The gaps between rich and poor countries make clear that the pressure to immigrate is much larger than the actual observed ex post migration (Hatton and Williamson, 2002). It is therefore necessary to include policy when explaining ex post migration flows and it is also clear that future flows will largely be determined by policy. Policies enter the Sjaastad model as an additional cost. The successful implementation of a tight immigration policy will reduce the number of immigrants. Nevertheless, it is complicated to find means to estimate the effectiveness of changes in immigration policy empirically, especially in a cross-country comparison.
Clark et al. (2007) focus on one destination country and integrate US immigration policies into the model. They differentiate several entry channels as changes regarding employment visas, quota for family members and refugee allowance. The authors in-clude 81 source countries from 1971 to 1988 and find that both individual incentives and policy constraints are highly significant in explaining immigration. Mayda (2010) is the first to study migration policies with multiple destination countries. She develops an index that documents migration policy changes: starting at zero, the variable in-creases by one if the country’s immigration policy became less restrictive and vice versa. Interacted with income in source and destination, it can be examined whether changes in the migration policy (that equally affect all source countries) have an effect on the push and pull factors. And indeed, Mayda finds that if migration policies become less protectionist, the effect of changes in income are stronger. Ortega and Peri (2013) build on the index on migration policies developed by Mayda and show that an increase in the entry tightness reduces immigration flows. They also include dummy variables that indicate whether source and destination countries adopted the Maastricht and Schen-gen Treaties, which had consequences for migration within and to the countries of the European Union. They find that the Maastricht Treaty dummy is positive and signifi-cant because it simplified inner-European migration. The Schengen Treaty dummy is negative and significant, because it did further simplify inner-European migration but also led to stricter border controls for those entering the Schengen area. Ortega and Peri also find that the dynamics change when only regarding flows within European
countries. So far, there exists no alternative index to account for immigration policies6.
It is straightforward that migration policies are the best indicator to account for the “demand side” of international migration. When implemented effectively, they are a sign for the destination countries’ handling with the migration pressure and reflect the capacity and the needs of a country. The approach by Mayda (2010) is a first attempt that serves as a base for future research. Some of the variables mentioned above might be good proxies for migration policy: both low PSRs and low unemployment rates also influence the demand for migrants and thus possibly also the tightness of migration policy. It would be interesting to more deeply study migration policy and how it is established as the product of political will and pressure and economic demand.
Not only migration policies shape migration flows – another potential determinant of immigration is a destination country’s policy environment in general. Kugler et al. (2012) add a measure for public good provision to account for the access to health and education and the attractiveness of the country. Although they do not find significant results when applying to the sample of OECD countries, I decided to test the variable. The argumentation is in accordance with Czaika and De Haas (2013) who claim that considering the overall policy environment is useful because some “non-migration” (p. 489) regulations and polices might also influence immigration, for example labor market or education policies. The provision of public goods by the government might have a pull effect on migrants, which could be regarded in the Sjaastard model as a reduction in the cost of living in the destination country or as a decrease in the costs of migration. This leads to the next hypothesis:
Hypothesis 6 The higher the social government expenditure in the
desti-nation country j, the larger the immigration flows to j.
Regarding migration policy, unfortunately there is no public index containing all immi-gration policy changes for OECD countries. In general, it is assumed that an increase in the tightness of entry significantly reduces immigration. When including the European treaties of Maastricht and Schengen, the main weakness of using total immigration be-comes clear: While moving inside the European Union bebe-comes easier, the Schengen
Agreement also results in more immigration restrictions for people from countries that are not participating. It is therefore conceivable that the effect is better detectable in a bilateral framework where it is possible to control for European origin countries. In general, assessing the effectiveness of migration policies is complicated. First, it is often unclear how to classify and measure the policy because there are large differences between the policy discourses, policies on paper and then the actually implemented policies. Second, the data available is very limited and when assessing the impact of migration restrictions, this seems to be even more severe than when estimating other determinants. Furthermore, is impossible to test for a substitution effect, i.e. to see whether migrants switch to other legal or even illegal channels because of missing data (Czaika and De Haas, 2013). Nevertheless, it will be interesting to study whether there exists a relationship between immigration policies and total immigration:
Hypothesis 7 The less tight the migration policies in the destination
country j, the larger the immigration flows to j.
The seven hypotheses provide a broad framework for estimating the main destination country determinants of international immigration. I have shown why integrating ad-ditional variables might reveal other important channels that have not been studied so far. The Sjaastad model serves as a basis for developing these new variables, combined with insights from Harris and Todaro (1970) who stress the importance of uncertainty regarding the outcome of migrating. Moreover, deviating from the theoretical models, the variables also reflect the demand side of immigration, i.e. the destination countries’ need for migrants and the capacity to integrate them into the labor market. In the next section, I will explain the methodology and data used for the empirical assessment.
3. Methodology and Data Methodology
mobility, population growth in poor countries and persisting income differences in the world it is important for destination country governments to gain a more in-depth insight about what determines immigration, why migrants might opt for country j rather than country k. Geographical distance and other bilateral characteristics are usually fixed and cannot be changed by the destination country’s government policy, the same is true (at least to a some extent) for the source country environment. The goal of this thesis is to go beyond income being the only destination country characteristic. Furthermore, the results can be informative regarding the role of the demand side in determining immigration in the context of the steadily increasing supply.
ln(Immigjt) = 0+ 1ln(Incomejt 1) + 2ln(Stockjt 1) + 3ln(HCjt 1)
+ 4Unempjt 1+ 5PSRjt 1+ 6GovExpjt 1
+ 7Maastrichtjt+ 8Schengenjt + 9TightEntryjt
+ jIj + tIt+ "jt
(3)
where j is the destination country and t is time. Immigjt is immigration to j
in t. Incomejt 1 is the per capita income; Stockjt 1 the stock of foreign born in
the country as a percentage of total population. HCjt 1 is the human capital
in-dex per person. Unempjt 1 is the unemployment rate, PSRjt 1 the potential
sup-port ratio. GovExpjt 1 measures the government expenditure on social services as
a percentage of GDP. Schengen and Maastricht are dummy variables that take the
value 1 after a country signed the EU treaties. TightEntryjt measures changes in
the tightness of migration policies. Finally, there are destination (Ij) and time
fixed effects (It). Based on the hypotheses derived in the previous section, I expect
that 1 > 0, 2 > 0, 3 > 0, 4 < 0, 5 < 0, 6 > 0, 7 0, 8 0, and 9 < 0.
I use a fixed effects model to control for unobserved heterogeneity. More precisely, I add dummy variables for each country (least squares dummy variable model = LSDV). This way, I concentrate on time-varying indicators in explaining changes in migration flows, but leave out factors that are country-specific and time-invariant as language or culture. There are several possibilities to estimate fixed effects models and while absorbing time fixed effects, I obtain results for the country dummies. This is quite informative because the dummies contain the effects that are particular to each country (Torres-Reyna, 2007).
countries’ incentives and are not as affected by migration policy changes. Therefore I
use the bilateral dataset by Ortega and Peri (2013) that is made available online7. The
dataset covers 15 OECD destination countries. Ortega and Peri (2013) also use a fixed
effects model8. Using the dataset I can compare the results of both specifications and
control for inner-European mobility. It will be interesting to see whether the results in the full bilateral sample are comparable to those using total flows. Regarding the demand side of immigration, total numbers might be the better choice because they reflect the countries efforts in controlling the total number of immigrants that might not be demonstrable in the flows between to countries. In general, I am convinced that my model is able to identify some new channels and can serve as a useful basis for more country-specific studies.
The sample consists of data for 30 OECD countries between 1985 and 2012. Below, I describe the data sources used and provide summary statistics of the main variables.
Data
Migration data
The data source for immigration is the International Migration Database (IMD) of the OECD (2015). From 1975 on, the database provides information on yearly inflows and outflows of foreign population into and out of the 34 OECD countries, although there are some gaps. The flows are either derived from population registers or residence permit data. I also look at the database of the United Nations Population Division
(2015c), which contains information on immigration to a number of countries9. In this
database, alternative ways to measure migration flows are listed: either by residence or citizenship and covering citizens, foreigners or both. It becomes clear that countries differ in the measurement method and comparing the two sources reveals exact matches in several years and divergence in others. In general, the criteria of citizenship covering foreigners corresponds best, although not all countries (e.g. Australia and the United
7see: http://giovanniperi.ucdavis.edu/data-and-codes.html
8First, the authors use origin and destination fixed effects, as well as origin-year fixed effects.
One regression goes even further: “[Column 9] includes a full set of origin-destination dummy variables, which absorb all time-invariant bilateral variables that affect migration flows. This is a very demanding specification as we absorb all bilateral-specific factors as well as all origin by time factors. Still we obtain that the destination income per capita plays a highly significant role in determining migration flows [...]” (62). The focus of the analysis is on the impact of changes in income and migration policies in the destination countries.
States) apply this method. Divergent and missing data is normally due to measurement differences and difficulties and reflects a main complication of empirical research on migration. While this can have consequences for cross-country comparison, it is not a major problem for my fixed effects regression model. Although the measurement methods might vary between countries, they do not change within countries over time. Table A1 in the appendix summarizes the data availability in the IMD for all 34 countries. Based on incompleteness and fragmentariness, I decide to exclude Estonia, Greece, Slovenia and Turkey. In a second sample I only consider countries where data is available from 1990 on, leading to further exclusion of Austria, Chile, Czech Republic, Iceland, Ireland, Israel, South Korea, Mexico, New Zealand, Poland, Portugal and Slovakia. The main sample contains 30 countries, the reduced version only 18.
Figure 5 shows annual immigration to three different countries to reflect the strong heterogeneity. In Belgium, immigration follows a relatively stable upward trend. In Spain, immigration remained very low until 2000, then it suddenly jumped to very high levels (up to 900,000 in 2007) but started to fall again in the aftermath of the global economic crisis. In the United States, a traditional immigration country, numbers have always been high, especially around 1990. During that time, some important reforms and legalizations took place. The last figure illustrates best the impact of immigration policies, studied in detail by Clark et al. (2007).
Figure 5: Yearly immigration in thousands, 1985-2013 (IMD)
40
60
80
100
120
Immigration (in thousands)
1985 1990 1995 2000 2005 2010 2015 Year Belgium 0 200 400 600 800 1000
Immigration (in thousands)
1985 1990 1995 2000 2005 2010 2015 Year Spain 500 1000 1500 2000
Immigration (in thousands)
1985 1990 1995 2000 2005 2010 2015
Year
United States
Figure A2 in the appendix shows immigration to all 30 countries in the sample, Figure A3 the immigration rates calculated using data from the 2015 version of the Penn World Table (PWT 8.1, Feenstra et al., 2015) on population. I also calculated the average
yearly immigration rate of the sample to allow for a more comparative analysis10.
10The average yearly immigration rate used in Figure A3 is the average of the migration rates of
Comparing Japan and Luxembourg shows that the latter – although receiving less migrants in absolute numbers – has immigration rates above the average while the other is clearly below. It is apparent that population sizes play a role for the total number of immigrants and I will control for the effect in the empirical estimation when using the migration rate as the dependent variable.
Figure 6 shows total immigration to the OECD. It confirms the upward trend, although with a lot of volatility. In 1985, around 1.7 million migrants entered the
OECD countries. In 2013, the number has increased to almost 5.7 million11. At the
same time total population in the OECD countries increased from 947 million in 1985 to 1.16 billion in 2011.
Figure 6: Immigration to the OECD (aggregated immigration in thousands)
2000
3000
4000
5000
6000
Immigration (in thousands)
1985 1990 1995 2000 2005 2010 2015
Year
By looking at the figures it becomes obvious that immigration does not simply follow income or demographic changes, as one would then expect monotone increases. Policies and other destination country conditions may also play a crucial role.
For the bilateral sample, Ortega and Peri (2013) use three sources for migration data. They use the data provided by the United Nations and fill missing values with IMD
11When interpreting these numbers, one has to consider that in the early years a lot of data is
data that reports bilateral flows only from 1998 onwards. Additionally, they take the data used in Mayda (2010) who also relied on OECD data. The merging of different sources illustrates quite well the difficulties of obtaining reliable data. The aggregated numbers from the Ortega and Peri sample do not correspond to the numbers I obtained but are usually higher. The correlation of the two variables is very high (⇢ = 0.9407), but might still lead to distortions in the empirical estimation.
Income data
Both theoretical and empirical models emphasize that differences in incomes are the main motive for emigration. PWT 8.1 contains the indicator “Expenditure-side real GDP at chained PPPs (in mil. 2005US$)” that can be used to compare relative living standards across countries and over time. Using population data from the same source, I calculate per capita incomes for all countries in the sample for the time period
1984-201112. To test the hypotheses related to income, I need per capita income and the
income differential of the country relative to the global, regional and OECD average. For the global income differentials, I calculate the average GDP per capita of all coun-tries available in PWT 8.1. It contains data for 143 councoun-tries; from 1990 on, 167 countries are covered. During 1984 and 2011 the average GDP per capita doubled from around 7,600US$ to more than 15,000US$, but the variation between countries is very high (in 2011, the GDP per capita in the Democratic Republic of the Congo was 400US$ compared to more than 134,040US$ in Qatar). To obtain the regional income differentials, I assign the countries in PWT 8.1 into regions according to the United
Nations geoscheme (UN Statistics Division, 2013)13. While in 2011 all countries (apart
from Mexico) lay above the global average (i.e. having an income differential above 1),
12PWT 8.1 contains data until 2011. Income enters the regression as a lagged variable, thus data
from 1984 on is needed.
13Oceania: Australia, New Zealand, Fiji; Europe: Belarus, Bulgaria, Czech Republic, Hungary,
studying the regional income differentials reveals that the Czech Republic, Hungary, Israel, Japan, South Korea, Mexico, Poland, Portugal, and Slovakia lay below the re-spective average income levels. The two measures are strongly correlated (⇢ = 0.8179), so the inclusion of both variables in the regression might lead to multicollinearity issues. I also calculate the income differential relative to the average income of OECD countries as in Kugler et al. (2012). The average per capita income was around 30,000US$ in 2011, varying from 12,800US$ in Mexico to 79,300US$ in Luxembourg. This measure is highly correlated with the more general global income differential (⇢ = 0.9874) and reveals more about the relative attractiveness of the countries included in the sample. Table A2 in the appendix shows the correlation matrix for all income measures. Table 1 shows the per capita incomes, PPP-adjusted, for the countries in the sample relative to the OECD average in 2011. An income differential below 1 indicates that the country is poorer in comparison to the OECD average and vice versa. With income being the main determinant of immigration, one would expect the countries on the right-hand side to be more attractive. However, the fixed effects model looks at changes in income, which is not captured by the differentials in Table 1.
Table 1: Income differentials relative to the OECD average, 2011
Countries below OECD average Countries above OECD average
Mexico 0.4 France 1.01
Chile 0.49 Iceland 1.01
Poland 0.58 United Kingdom 1.05
Hungary 0.61 Finland 1.07
Slovakia 0.68 Germany 1.09
Portugal 0.75 Canada 1.12
Czech Republic 0.75 Sweden 1.14
Israel 0.79 Belgium 1.15
New Zealand 0.86 Denmark 1.16
South Korea 0.89 Ireland 1.17
Spain 0.95 Austria 1.19 Italy 0.95 Netherlands 1.22 Japan 1 Australia 1.22 United States 1.37 Switzerland 1.45 Norway 1.64 Luxembourg 2.54
measure. The two income measures are positively correlated (⇢ = 0.8548), but the differences are large enough to lead to a distortion in the results. Therefore I decide to apply my income measure to the bilateral sample, too.
Immigrant stock
To account for the “friends and relatives” effect and the power of existing networks, I include the stock of foreigners within a country as a percentage of total population. Clearly this does not entail the composition of the stock, but I assume the variable to be positive and significant. Data is taken from the UN (2015d; available from 1990 on in five-year intervals) and from the World Bank database for 1985 (2016). To obtain data fitting my sample, I interpolate the values assigning 20 percent of the change between 5 years to each of the years in-between. The stock is then divided by the population size using PWT 8.1 data. In 2011, the average stock of foreigners relative to the population was around 12 percent, varying from 0.8 percent in Mexico to almost 35 percent in Luxembourg – those countries with the lowest and the highest relative incomes. The average migrant stock increased from 8.77 percent in 1985.
Human Capital Index
To measure differences in human capital, I use the human capital index per person from PWT 8.1. The index is based on years of schooling (Barro and Lee, 2013) and returns to education (Psacharopoulos, 1994). The average human capital index per person in 2011 was 3.12, ranging from 2.56 in Portugal to 3.62 in the United States. It increased from an average of 2.72 in 1984.
Unemployment rate
Data for unemployment comes from the International Monetary Fund (2016). It serves as a macroeconomic indicator that reflects the demand for (foreign) labor and the attractiveness of the country. In 2011, the average unemployment rate in the sample was 7.8 percent, ranging from 2.8 percent in Switzerland to 21.4 percent in Spain. In 1984, the average unemployment rate was slightly lower, around 7.6 percent.
Demographic data
population aged 65 or older and is a measure for the age structure within a society. If the hypothesis of some sort of replacement migration is true, i.e. aging societies have a higher demand or are more attractive destinations for migrants, the expected outcome is a negative estimate for the PSR: the higher the PSR, the lower the immigration flows. The average PSR in 2011 was 4.67, ranging from 2.69 in Japan to 10.77 in Mexico. In 1985, it was around 5.22 – but as above, many countries (as Mexico) are not considered here because of missing migration data.
Policy environment
To account for the public goods level I include the social expenditure in percentage of GDP. The data is made available by the OECD Social Expenditure Database (SOCX, 2016). The mean in 2011 is equal to 21.6 percent (ranging from 7.7 percent in Mexico to 31.4 percent in France), compared to 19.3 percent in 1984.
Migration policy
There does not yet exist a common indicator that measures immigration policies. Or-tega and Peri (2013) build on Mayda (2010) and study migration policy changes in twelve destination countries (Australia, Belgium, Canada, Switzerland, Germany, Den-mark, Spain, UK, Netherlands, Norway, Sweden, USA). Rather than to allocate scores at each point in time, they set the initial score equal to zero and only consider changes in laws as either tightening or lightening. I use the score regarding entry policies. Unfortunately the index is only available for a limited number of countries between 1980 and 2006. In 2006, migration policy was relatively tight in Denmark and the Netherlands (index equal to 1) and less tight in Germany and Sweden (index equal to -5), although this does not allow country comparison but only reflects the changes within the country compared to 1980. Figure A4 in the appendix shows how migration
policies changed between 1985 and 200614. Similar to the authors, I also construct
dummy variables for member countries of the European Union for the treaties that were important for inner-European mobility and migration to the EU (Maastricht and Schengen).
Another interesting migration policy indicator for future research is the MIPEX (Migrant Integration Policy Index), which measures how destination countries actively
14The index sets the initial value equal to zero in 1980. As my sample starts in 1985, some changes
promote the integration of immigrants based on different policy areas (MIPEX,
2015)15. Unfortunately, the index only exists since 2010, thus it cannot be
imple-mented into my regression. Also the scores change only in little steps, so here the fixed effects model might not be the best choice. Figure 7 shows the overall MIPEX score for 2012 and aggregated immigration in the same year (logarithmized). A slightly positive relation is observable.
Figure 7: MIPEX and immigration in 2012
AUS AUT BEL CAN CZE DNK FRA DEU HUN IRL ITA JPN KOR LUX NLD NZL NOR POL PRT SVK ESP SWE CHE GBR USA 8 10 12 14 Immigration (log) 40 50 60 70 80 MIPEX
Table 2 reports summary statistics for the variables used in the analysis. It shows the averages across the 30 countries and for the time period from 1985 to 2012, although data is not available for all countries at each point in time (see Table A1 in the appendix for the availability of migration data). Data from the PWT 8.1 is only available until 2011, but as all affected variables enter the regression lagged, the final sample ranges until 2012. On average, around 153.75 thousand individuals immigrated to each of the OECD countries every year, although the standard deviation is very high and values range from 1,353 (Iceland 2003) to 1.8 million (USA 1991). The average migrant stock is around 10 percent. Again the standard deviation is very large. The minimum of 0.51 percent of foreigners in the total population is in Mexico in 1997 (the first year in which the country is included in the sample) and the maximum of almost 35 percent
15Labor market mobility, education, political participation, access to nationality, family reunion,
is in Luxembourg 2011. On average, the PSR is 4.99, i.e. 5 people between 15 and 64 per individual above 65. The lowest PSR has Japan in 2012 (2.59), the highest Mexico in 1997 (12.52). Regarding the tightness of entry laws, most changes involved a lightening. Unfortunately, there are only 263 observations for the entry tightness variable.
Table 2: Summary statistics
Variable Mean Std. Dev. Min. Max. N
Immigration (in thousands) 153.75 240.71 1.35 1800 699
Population (in millions) 38.28 58.523 0.28 313.09 669
Income
Real GDP per capita in US$, PPP 27,639.26 10,379.33 9,706.78 82,904.16 669
Income Differential, World 2.526 0.863 0.835 5.747 699
Income Differential, Region 1.52 0.588 0.604 3.292 699
Income Differential, OECD 1.097 0.368 0.355 2.623 699
Migrant stock (in millions) 3.18 6.66 0.015 45.16 694
Migrant stock (in % of population) 10.19 8.20 0.51 34.96 664
Human capital index 2.97 0.3 2.1 3.62 669
Unemployment rate (in %) 7.26 3.93 0.46 24.8 698
Potential support ratio 4.99 1.42 2.59 12.52 699
Policy environment
Social expenditure (in % of GDP) 20.46 5.71 4 35.5 683
Maastricht 0.465 0.499 0 1 699
Schengen 0.351 0.477 0 1 699
Tightness of entry laws -0.624 1.894 -6 2 263
To illustrate the large country disparities, Table 3 contains averages across years for
some of the main variables for the individual countries in the sample16. It becomes
clear that the United States and Germany received on average the highest number of immigrants, Finland and Mexico the lowest. The countries with the largest populations are the United States and Japan, the smallest Luxembourg and Iceland.
Table 4 shows the correlation matrix of the independent variables. It can be seen that some variables show relatively high correlation, for example governments’ social expenditures and the PSR of the country (⇢ = 0.7584), income and the stock of immigrants (⇢ = 0.6798) or income and the unemployment rate (⇢ = 0.4213). It remains to be seen whether this leads to biased results.
16The averages refer to the observations that are in the final sample. If a country has missing data
Table 3: Summary statistics for the individual countries
Country N Immig Population Income Stock HC Unemp. PSR
(1) (2) (3) (4) (5) (6) (7) Australia 28 135.84 18.91 30,187 23.36 3.31 7 5.54 Austria 17 85.94 8.17 33,529 13.45 2.77 4.66 4.17 Belgium 28 66.76 10.20 28,068 8.85 2.93 8.24 4.11 Canada 28 217.58 30.14 30,639 17.54 3.17 8.31 5.59 Chile 13 52.03 16.37 12,466 1.66 2.88 8.73 7.67 Czech Rep. 19 33.51 10.30 20,106 2.65 3.37 6.71 4.95 Denmark 28 22.55 5.3 28,948 6.51 2.89 6 4.3 Finland 28 10.63 5.14 25,874 2.57 2.77 8.96 4.54 France 28 104.78 60.56 25,930 10.49 2.71 9.24 4.26 Germany 28 695.3 81.19 27,097 10.98 2.87 8.03 4.1 Hungary 25 21.54 10.21 14,310 3.44 3.08 7.5 4.53 Iceland 14 3.79 0.3 34,545 8.43 2.95 3.9 5.56 Ireland 19 42.43 4.01 32,789 11.56 3.16 8.47 6.25 Israel 15 30.14 6.61 25,475 28.79 3.19 10.23 6.1 Italy 15 200.65 57.76 26,000 4.65 2.64 9.11 3.97 Japan 28 27.73 124.50 27,567 1.24 3.08 3.71 4.41 South Korea 13 24.12 47.18 24,319 1.21 3.26 3.59 7.75 Luxembourg 28 11.52 0.43 54,745 31.19 2.82 3.09 4.90 Mexico 16 11.18 105.23 11,555 0.67 2.59 3.74 11.62 Netherlands 28 80.14 15.64 29,002 9.1 3.04 5.17 5.03 New Zealand 19 52.25 4 24,742 18.93 3.43 5.79 5.44 Norway 28 31.55 4.46 34,773 6.46 3.21 3.89 4.21 Poland 15 32.22 38.25 14,166 1.94 2.86 13.49 5.38 Portugal 21 34.07 10.38 19,334 6.37 2.42 7.18 4.05 Slovakia 20 77.53 5.41 15,303 2.09 3.14 14.64 6.02 Spain 27 27.12 41.15 21,814 5.73 2.63 16.8 4.47 Sweden 28 53.1 8.85 28,172 11.16 3.11 6.79 3.67 Switzerland 28 100.07 7.09 35,707 22.45 2.68 2.75 4.45 United Kingdom 28 261.52 58.85 26,516 8.35 2.68 7.47 4.13 United States 28 954.46 276.36 37,145 11.43 3.51 6.1 5.31
(1) Immigration in thousands; (2) Population in million; (3) GDP per capita in US$, PPP-adjusted; (4) Stock of foreigners as a percentage of population; (5) Human capital index; (6) Unemployment rate in percent; (7) Potential Support Ratio
Table 4: Correlation matrix
lnIncjt 1 lnStockjt 1 lnHCjt 1 Unempjt 1 lnPSRjt 1 SoExjt 1
4. Results
This section contains the regression results of the model in Equation (3). The sample consists of panel data for 30 OECD countries from 1985 to 2012. After removing
all missing values for immigration, I end up having 699 observations17. It became
evident that using total migration instead of bilateral data leaves out the possibility to control for source country and bilateral characteristics that have proved to be very relevant in both theoretical and empirical models. The fixed effects model allows concentrating on the importance of time-varying (and destination-country specific!) indicators in explaining changes in migration flows while leaving out factors that are country-specific and time-invariant. Simply pooling the data into a single regression and ignoring the panel structure would lead to misleading results as I would assume no correlation between the errors for each country. I hope that my work might be relevant to estimate future migration flows because the fixed effects will not change in the future: if one country attracts a lot of immigrants because of its geographical location or because of having a common language, the gravitational pull is likely to continue in the future. The focus of the empirical specification lies thus on within-destination-country changes over time. As stated above, I add dummy variables for each within-destination-country, which allows studying country specific factors, too. To allow for comparison, I also use data for bilateral migration flows made available by Ortega and Peri with more than 35,000 observations. To address heteroskedasticity, I apply robust standard errors. Table 5 contrasts the results from estimating variations of the regression equation and provides an empirical test of Hypothesis 1a. The availability of the Ortega and Peri dataset allows comparing the total with the bilateral immigration model. As expected, income has a positive and highly significant impact on immigration in all specifications, meaning that when income in a destination country rises, so does immigration. Column (1) shows the results for the whole sample of 30 OECD countries. Column (2) only considers the reduced sample of 18 OECD countries for which data is available for the entire time period from 1990 to 2012. The coefficients are very similar. In column (3), I use only those countries that appear in the bilateral sample, using data from 1985 to 2006. That way I am able to directly compare the results to the aggregated data of the bilateral sample in column (4). The coefficients
are quite different from each other, indicating that the differences in the migration data have a strong effect on the results.
Table 5: Comparison of aggregated and bilateral immigration flows: the role of income
Variable (1) (2) (3) (4) (5) (6)
my data reduced OECD15 O&P agg. bil. bil. (EU)
ln(Incomejt 1) 2.016*** 2.070*** 3.840*** 2.897*** 1.589*** 2.670*** (6.28) (5.13) (5.50) (5.52) (22.63) (16.24) C -8.985*** -9.531** -27.64*** -18.14*** -12.30*** -20.86*** (-2.71) (-2.30) (-3.87) (-3.39) (-17.10) (-12.48) N 699 498 318 312 36,060 3,842 R-squared 0.906 0.906 0.904 0.929 0.949 0.927 adj. R-squared 0.897 0.897 0.892 0.919 0.937 0.912 Fixed effects Year X X X X X X Destination X X X X X X Origin X X Origin-year X X Orig.-dest. X X
t statistics in parentheses * p<0.10, ** p<0.05, *** p<0.01; robust standard errors
Column (5) and (6) show the results for the bilateral sample, with a much larger number of observations. In column (6), I only look at inner-European migration and as expected, the coefficient for income is much larger. People tend to respond stronger to income changes when allowed to move freely within the Union. The result in column (5) compared to (4) indicates that looking at the aggregated numbers might overemphasize the impact of income on migration. This information should be kept in mind when discussing the results and potential shortcomings of the methodology.
Table 6 shows the main results when testing the hypotheses developed in section 2. Alternative destination country characteristics are added by and by. Column (1) shows the result when pooling the data without using fixed effects, ignoring the presence of unobserved heterogeneity. In column (2), destination country and time fixed effects are added and the explanatory power of the model increases a lot, indicating that the country differences are quite substantial. The result, which is also shown in column (1) of Table 5, indicates a large elasticity of migration to income: a 1 percent increase in per capita income in the destination country leads to a 2 percent increase in im-migration, on average ceteris paribus. This is higher than estimated by Ortega and
Peri (2013), but comparable to the results of Mayda (2010)18. The remaining columns
of Table 6 show that when adding additional variables to test the other hypotheses, the impact of income on migration decreases. In column (3), the migration stock is added. As expected, the variable is positive and significant and confirms Hypothesis 2: a higher share of foreigners in the destination country attracts further immigrants. Column (4) shows that the human capital per person in the destination country is a determinant of immigration flows confirming Hypothesis 3. The next column tests whether the unemployment rate is also a determinant of immigration, reflecting a lower demand on the labor market of the destination country. The variable is negative and significant as expected: if the unemployment rate increases by 1 percentage point
ce-teris paribus, on average immigration will be 4.47 percent lower19. When comparing
the point estimates it becomes clear that the coefficients change between columns and rather indicate general trends and no exact relationships. The point estimates should therefore be interpreted with caution and only capture tendencies. The PSR that enters in column (6) is not significant. The hypothesis that there is some degree of replacement migration is not confirmed. The PSR in a country changes only slowly, which is a reason why it might not be the best variable to include in the fixed effects regression (Torres-Reyna, 2007). Using five-year steps as done by other authors (e.g. Hatton and Williamson, 2002) might more appropriate in this regard. As the variable does not add new information, I exclude it in column (7) where I test the impact of social expenditure. As expected, the variable is positive and significant.
18Although her empirical specification is a bit different from mine: “[...] In other words, a 10%
increase in the host country’s per worker GDP implies a 20% increase in the emigration rate”. As stated above, Mayda uses the emigration rate of a country as a dependent variable and per worker GDP in source and destination country as the main explanatory variables.
Table 4 showed that the correlation between social expenditure and the PSR is rela-tively high (⇢ = 0.7459), so it could be interesting to further study the relationship between social expenditures, the age structure, and migration. In the eighth column of Table 6, I test the two European treaty dummy variables and Maastricht turns out to be very high and significant. The last column tests whether the migration policy index does lead to significant results in my sample. Reducing the sample to only 244 observations, I cannot confirm any impact of changes in the entry tightness.
One advantage of the least squares dummy variable model is that it allows for interpretation of the individual country dummies. Figure 8 shows a comparison of column (1) and (2) from Table 6. For the sake of clearness, I decided to not include all 30 countries but only six examples. The black line shows the result from estimating the equation without controlling for country and time fixed effects. The least squares dummy variable model is a good tool to demonstrate the intention of fixed effects: looking at the different country dummies it becomes clear that the effect of income on migration flows is mediated by differences across countries.
Figure 8: Graphical illustration of fixed effects using the LSDV model
8 10 12 14 ln(Immigration) 9 10 11 ln(Income) Australia France Iceland Mexico
Luxembourg United States Fitted values
that I took as an example demonstrate this quite well: while the United States are considerably above Australia (= the base country), Luxembourg, Iceland, and Mexico lie below – having considerably different income levels. France and Australia are very similar, i.e. the “France-dummy” is not significantly different from zero.
Table 7: Country dummies: comparing columns (2) and (8) of Table 6
Table 7 takes a closer look at two columns of Table 6, namely column (2) where only income is used as an explanatory variable (also in Figure 8) and column (8) that included other destination country characteristics. As mentioned before, the country dummies contain all the information about the 30 countries that explain why immigration flows are higher or lower compared to the base country Australia. The information give room for further, more country-specific research, which shall be shown by some examples. In general, comparing the two columns shows that after controlling for more time-varying factors, some of the unexplained differences between Australia and other countries vanish (see Austria, Belgium, Czech Republic, the Netherlands, Portugal and Slovakia). Nevertheless, other country dummies are still very large and significant and one can only speculate what causes the deviation. The results for Iceland, Luxembourg, Japan and the United States suggest that the population size of the destination country matters. I already mentioned this fact before when discussing immigration rates. Including population to the regression model does not lead to any significant results. Other interesting cases are Chile, Italy and Spain, where immigration is significantly higher. This might be due to the geographical location of the countries: they are simply “better accessible” than Norway or New Zealand. Additionally, Spain attracts many immigrants from Latin American countries because of the common language.
differential and per capita income would be interesting to study how the two measures develop over time. In general, the income differential has the advantage of taking other countries into account, too. The results hold when replacing the variable by the income differential relative to the world. Using the regional income differential changes the coefficients also only slightly. When testing regional and global income differentials, first both income differentials are positive and significant as expected but after adding the other controls, the global income differential loses its significance (see Table A4 in the appendix). Apart from that, the results are quite similar to those obtained in Table 6, but interestingly the variable PSR now is negative and significant – at least before including the social expenditure variable, again indicating multicollinearity. For the sake of completeness, I also regress the equation using the migration rate as the dependent variable. Although the magnitudes differ slightly, the coefficients and the significance levels are quite similar, so the general tendencies are still observable. Regarding the goodness of fit, the specification in Table 6 is slightly more convenient than using immigration rates. I prefer the specification because it better meets the research question: governments might rather be concerned about the total number of immigrants than about changes in the migration rate.
expected to be negative and significant in the full sample, which cannot be confirmed. This is surprising because Ortega and Peri (2013) did obtain significant results. To find an explanation, I replace the income variable by the one the authors used and the entry tightness variable becomes negative, but is still not significant. The results for the Treaty dummies are also quite inconsistent, only confirming the theory in columns (3), (4) and (6). By and large, it seems that transferring the aggregated to the bilateral model reveals some contradictions that might be related to multicollinearity and need to be studied in more detail.
To see whether the deviations stem from the method of measuring migration, I again compare the results to the total immigration – using both the aggregated bilateral flows and my data for the countries and years contained in the bilateral sample. And indeed, the human capital index is significant when using my data, but not in when using the aggregated bilateral flows as dependent variable. The social expenditure variable is positive and significant, although much larger in my sample. The income variable also varies a lot. The PSR is negative and significant, which shows that replacement migration could play a role when only looking at a subset of countries. It becomes clear that the accuracy of migration data is very important. Another potential explanation for the deviations in the bilateral sample is that some variables reflect the destination countries’ demand for immigrants that appears better in the aggregated data. All in all, further research on the determinants of migration policies and the links to the variables would be needed.
5. Conclusion
In this thesis, I empirically investigated the destination country determinants of inter-national immigration to 30 OECD countries. The focus of the analysis was on time-varying factors and their role in shaping migration flows. Deviating from the more common approach of studying bilateral immigration, I used total annual immigration as the dependent variable.
