THE RELATION BETWEEN
DEVELOPMENT ASSISTANCE
AND MIGRATION
EBM866B20
Master’s Thesis
by
Peter Geert van der Poel
S2414953
University of Groningen
Supervised by dr. Annika M. Mueller March 5th, 2019
Abstract
This paper analyses the relation between official development assistance and migration from Africa to Europe. To conduct the research, we construct a bilateral panel dataset from data acquired from the OECD and World Bank. The fixed effects model shows us a positive relation between official development assistance and migration flows. We also find that this relation increases during the European migration crisis.
2
1. INTRODUCTION
Migration is booming, in the recent decade the number of international migrants has grown exponentially, particularly in Europe1. With the refugee crisis of 2015 and the arrival of thousands of migrants on the southern borders of Europe2, the pressure on the European Union and its member states to find adequate solutions to manage the migration flows is rapidly growing. According to policymakers, official development assistance (ODA) is an essential part of the solution. They argue that ODA helps countries to develop and thus reducing the incentive to migrate. But does ODA really has a dampening effect on migration flows?
In theory, the effect of ODA on migration is ambiguous. Its net effect is subject to contrasting forces that are not clear cut (Berthélemy, 2009). ODA has no direct link to migration, but ODA is expected to have effect on determinants that have effect on migration flows. This paper identifies two important effects of ODA on migration, namely the income and network effect3. The first income effect of ODA on migration, is that ODA raises disposable income in the recipient country and higher incomes reduce the incentive to migrate. But additional wealth also releases budget constraints for migration, increasing the opportunity to migrate. Combining these two effects gives rise to the hump-shape pattern theory, first introduced by Venturini (1993). The hump-shape pattern theory predicts a positive relationship between GDP per capita and immigration for relatively low levels of GDP due to released budget constraints and a negative relationship for relatively higher levels of GDP per capita due to reduced incentive to migrate4. In addition to the income effect, we also identify a network effect, first introduced by Berthélemy (2009). The network effect shows that there is an association between ODA and migration flows, namely that increased contact due to the implementation of ODA projects increases the information available to potential migrants in the recipient countries, thus reducing the cost of migration as a result of an information advantage.
In this paper, we revisit the question of how ODA affects migration and we provide new empirical evidence on the relation between ODA and migration flows. We use the migration gravity equation constructed by Berthélemy (2009) as a starting point of our analyses, but we
1 According to the OECD, the number of immigrants in the OECD area has increased with 35% since 2000. 2 According to the International Organization for Migration (IOM), over 1 million migrants entered the EU in
2015.
3
make various extensions on his approach. First, we construct a fixed effects model, that incorporates a dummy variable for each country pair5 to prevent time-invariant omitted variables. Second, we add a polynomial term for the variable that represents the income effect, GNI per capita. With this inclusion we can test if the hump-shape pattern hypothesis also applies to our dataset. Finally, since we constructed a new dataset that includes the most recent migration crisis in 2015, we are able to add a year dummy for the crisis year 2015. With this extension, we are able to test how our results differ in crisis years in comparison to non-crisis years and we provide a first impression for policymakers how the relevant factors change in time of crisis.
We obtain evidence of a positive relationship between ODA and migration flows. In the crisis year, we observe that this relationship becomes stronger. We also obtain evidence that confirms the hump-shape pattern hypothesis, which suggests that higher levels of GNI per capita have an increasingly negative effect on migration flows.
In chapter 2, we will give a comprehensive overview of the relevant theory. In chapter 3, we explain the data and discuss potential endogeneity problems. In chapter 4, we present and interpret the results of our research. In chapter 5, we present a discussion and potential future research. In chapter 6, we will consolidate the results and draw our conclusion.
2. THEORY
2.1 Foreign policies on migration
After the Second World War, Western Europe experienced a high economic growth. Production increased rapidly in the first decades after the war, industrial manufacturing alone increased with 30 percent (Dietz, 2008). Social mobility increased and with it, the level of education in the working force increased. As a result, the willingness to take low status jobs, like agriculture construction and cleaning, decreased. Since the vacancies for low-end labor increased, Western European countries started importing, mostly temporary, labor to fulfill the demand for these jobs (Boyle, 1998). In the period up until 1974, when the oil crisis started, international
migration was considered a positive development due to its economic benefits
4
(Bonifazi, 2008), both from the country of origin and destination, since emigration helped to reduce unemployment in regions with lower productivity and income (Page Moch, 2003).
The oil crisis in the 1980s greatly reduced the demand for labor. To protect domestic employees, most Western European countries tried to implement policies that aimed to control and diminish migration. These policies had the opposite effect, since intended temporarily migrant workers increasingly settled permanently, as the risk of losing residence permission increased due to the new policies (Hansen, 2003).
With the introduction of the Schengen agreement (1985), which allowed free travel within the European Union for European citizens, foreign residents, most of which are coming from developing countries, have virtually doubled between 1980 and 1988 in most Western European countries (Faini, 1993). This tremendous flow of population brought migration policies to the top of the agenda. Policies differed significantly throughout Europe. In some countries, policies were set to favor social and economic integration. In other countries, instead, migration policy was set to prevent permanent residence. In all cases, no attention was paid to other policies that would have significant effects on migration. Migration cannot be analyzed in isolation from trade, debt and foreign aid policies (Stanton et al. 1992). Since the believed forecast stated that there would be no shortage for labor in Europe and labor demand would grow towards high skilled workers, immigration policies were mainly set to establish entry control (Salt 1992). Aid policies were set to induce growth in developing countries and therefore reduce the incentive to migrate, since classic theory suggests emigration as an ‘’investment’’ to secure a higher income abroad. The believe was that income growth in developing countries, due to ODA, should have a negative effect on migration.
2.2 Determinants of migration
5
However, the income effect is not as straight forward as it seems, recent literature has shown that there are some unintuitive exceptions to this assumption, which we will present in the following section.
2.2.1 Mobility transition
Venturini (1993) was the first to publish empirical evidence that income growth and emigration are positively related. Using data from Greece, Portugal, Spain and Turkey, he showed that income growth failed to reduce emigration, it actually increased. He introduced the hump-shape pattern hypothesis. A positive relationship between GDP per capita and immigration for relatively low levels of GDP and a negative relationship for relatively higher levels of GDP per capita. He argues that the existence of the hump-shaped pattern is due to the cost of migration, which reduces the possibility of migration for the poorest countries. This phenomenon is commonly referred to in the literature as the mobility transition and it incorporates a variety of factors. Migration costs consist of geographical distance (transportation costs) but also include non-monetary elements, for example linguistic or cultural distance. Graph 1 shows an example of the hump-shape pattern, presented by Berthélemy (2009) in a study based on Venturini’s hump-shape pattern hypothesis.
6 2.2.2 The network effect
A recent study of Berthélemy et al. (2009) identifies an additional effect of ODA on immigration, namely the network effect. They argue that bilateral aid improves the connection between donor and recipient country. The implemented aid projects provide receiving countries with additional information which reduces the cost of migration, thus increasing the bilateral migration flows. Epstein (2008) shows additional insights into the network effect. He shows that migration is accelerated after the first wave of immigrants. He argues that the first wave of immigrants reduces the asymmetric information between country of origin and destination, thus reducing the migration cost.
2.3 Cost of migration
There are many determinants of migration that are affected by ODA. Previously, we already mentioned the income and network effect. However, literature has identified multiple other determinants of migration that, in theory, are affected by ODA. We can categorize these effects into two groups, namely factors that increase the willingness to leave the country of origin, we identify these as push-effects. And factors that increase the willingness to migrate to a specific country of destination, pull-effects. We will briefly summarize the known determinants of migration to create a comprehensive overview.
2.3.1 Push-effects
7
Second are the environmental factors. There is extended empirical literature on the environmental effects on migration. First, a natural disaster can create a shock in the net income, which increases the incentive to migrate through the income channel. Although, in specific cases, this could also be identified as a pull-effect, since a natural disaster can destroy assets that decrease the opportunity to migrate. Lastly, the environmental uncertainty. If the environmental uncertainty increases, risk averse people will be more incentivized to choose for migration.
2.3.2 Pull-effects
The most important pull-effects are already incorporated into the income and network effect. However, there are additional factors that explain (bilateral) migration flows.
First geographical distance, which reduces/increases the cost of migration due to lower/higher transportation costs. High geographical distance also increases the mental cost of migration, because of the mental cost of being far from the country of origin (Swartz, 1973). The majority of other pull-effects include non-monetary elements. The most important one being linguistic distance. Adserà, and Pytliková (2015) find that linguistic similarities between country of origin and destination have explanatory power in international migration. They also find that the perceived difficulty of the language affects international migration and the ability to learn the native language quickly, significantly affects the success in the local labor market. They also note that linguistic proximity matters less when local linguistic networks are larger. Berthélemy (2009) identifies another non-monetary effect, namely colonial history. He shows that countries with a historical colonial bond have increased bilateral migration flows.
2.4 The European migration crisis
8
high influx is multidimensional. The IOM estimates that about 26% of the migrants that entered the EU in 2015 originate from Africa. the majority, about half, are Syrians fleeing the war6.
The migration crisis caused some serious political challenges in a policy area that has not been constructed for fast paced change (Collett, 2018). Because of the Dubliner regulation, the migrants have to apply for asylum in the first EU country they enter. As a result, the Southern European member states that form the Southern European border endured the majority of migration inflow.
National interest was increasingly important compared to a consolidated European response. The increasingly polarized political environment is partly to blame for this development (Park, 2015). As a result, there was little consensus on how the EU should respond to the crisis. Thus, instead of a comprehensive plan of the European Commission, most actions taken were basic support and ad-hoc measures.
3. RESEARCH FRAMEWORK
3.1 Research question
The main question of this paper is: how does ODA affect the migration from Africa towards Europe. We separate the total effect of ODA in the network effect and the income effect. To create a comprehensive conclusion, we add several pull- and push-effects identified by the existing literature, as well as the most important income and network effects to make sure our main variable of interest is not inflated due to omitted variable bias (OVB).
Besides our main question, we will also look into the European migration crisis of 2015. Since we constructed our own dataset with recent crisis date included, this paper is one of the first in the literature to incorporate the migration crisis in the discussion about the effects of ODA on migration.
6 We are aware Syria is not located in Africa, thus outside of the scope of this research. However, the high influx
9
3.2 Data
3.2.1 Data collection
To answer the research question, this paper will use a bilateral panel dataset constructed from a variety of different datasets. The majority of our data is provided by the Organization for Economic Co-operation and Development (OECD). The OECD collects a wide variety of data to support governments to set policies that increase the economic and social wellbeing of people around the world (OECD, 2018).
The other data source is the World Development Indicators (WDI) of the World Bank. The WDI is a compilation of relevant, high-quality, and internationally comparable statistics about global development and the fight against poverty (World Bank, 2018).
Our dataset consists of yearly observations from 2000 until 2017. We identified 51 African countries of origin and 26 European countries of destination. The data provided is of high quality and therefore we are confident it is sufficient for our research purposes.
3.2.2 Data limitations
There are some remarks about our dataset that are relevant for our research. Since the OECD depends on individual governments to provide the required data, there are large discrepancies between the reporting standards used by governments when compiling the datasets. Especially for migration flows, migration stocks and ODA flows. This causes problems when consolidating the datasets. Some governments recorded bi-annually and some didn’t record individual ODA or migration flows. Therefore, we have to drop 17 of the 26 countries of destination we initially incorporated into our dataset. Since the qualified countries are mainly the largest European countries, with the highest international reporting standards, we are confident that our dataset is still sufficient for our research purposes.
10
countries that provided less than 90 percent7 of the observations with any of the qualified European countries in any of the variables. Due to this requirement, we have to drop 26 of the 51 African countries from our dataset. Although this is a high number of countries, due to high polarization between the quality, we drop a relatively small part of our observations.
The last limitation of our dataset are the observations of 2017. Although OECD reports that the 2017 data would be provided in November 2018, many qualified countries haven’t provided the data of 2017 yet. Therefore, we decided to drop 2017 from the scope of our research.
3.2.3 Descriptive statistics
In table 1 we show the descriptive statistics of the variables that are included in the regressions, with additional explanations to increase the understanding of the data. Within some variables, there is a high disparity between European and African countries, therefore we added proxy variables to give a comprehensive view of the descriptive statistics8.
ODA is defined as government aid designed to promote the economic development and welfare of developing countries. Although most loans and credits are excluded, ‘’soft’’ loans (where the grant element is at least 25% of the total) are included (OECD, 2018). Therefore, it is possible to observe negative net value’s for individual years. The unit used for ODA in this paper is millions of US dollar, since the data is provided in millions of US dollar.
The quality of governance is measured with the WDI for political stability and absence of violence/terrorism. This indicator measures the perception of the likelihood of political instability and/or politically motivated violence, including terrorism. The value ranges from approximately -2.5 (weak) to 2.5 (strong) governance performance.
Migration flow is the bilateral flow of the number of migrants between country of origin and destination for the given year. While there is no formal legal definition of an international migrant, there is consensus that an international migrant is someone who changes his/her
7 Due to the high polarization of the data of African countries, picking a cut-off point between 80-90 percent
doesn’t affect the amount of qualifying countries.
8 Appendix C shows a balance test, to validate that our chosen variables are significantly different between
11
country of residence, irrespective of the reason for migration or legal status (United Nations). The migration stock (MS) presents the stock of foreign-born population in qualifying OECD countries, that is, the size of immigrant populations as measured in number of persons. Data on immigrant populations may be collected through registers, residence permits, statistical surveys, or censuses, depending on the country (OECD, 2018).
Gross national income (GNI) per capita is based on purchasing power parity (PPP). PPP GNI is converted to international dollars using purchasing power parity rates. An international dollar has the same purchasing power over GNI as a US dollar has in the United States. GNI is the sum of value added by all resident producers plus any product taxes (less subsidies) not included in the valuation of output plus net receipts of primary income (compensation of employees and property income) from abroad. Data are in current international dollars based on the 2011 International Comparison Program (ICP) round (World Bank, 2018). Using PPP adjusted data is beneficial for our model, since it accurately depicts the real difference between income in the country of origin and destination.
12
Table 1: descriptive statistics
Variable Mean Std.Dev. Min Max N Explanation
Total ODA
million $ 27.38 81.29 -97.09 2027 4092
Net total ODA received by qualified countries
Governance -0.41 1.11 -3.31 1.76 525 Index ratio for political stability
GovEU 0.87 0.47 -0.47 1.76 135 Governance index ratio countries of destination
GovAfrica -0.85 0.90 -3.31 1.20 390 Governance index ratio countries of origin
Migration flow 936 4272 0 84978 4067 Migration flows towards qualified countries
GNI 12266 15563 430 50500 612 PPP GNI per capita in all
qualified countries
GNI EU 37087 7195 21.450 50500 153 PPP GNI per capita countries of destination
GNI Africa 3197 3073 430 14870 459 PPP GNI per capita countries of origin
Dependency 77.37 19.25 44.94 109.86 612
Dependends, <15 and >65, per 100 working age, between 15-65, population
DepEU 51.28 3.37 44.94 60.05 153 Dependency ratio countries of destination
DepAfrica 85.22 14.63 48.49 109.85 459 Dependency ratio countries of origin
Migration stock 22543 129570 0 1.368.437 2090 Stock of migrants born in country
i and resident in country j
3.3 Model
To run our estimations, we constructed a bilateral panel dataset to create all possible country pairs between countries of origin and possible destinations. With 9 countries of destination and 25 countries of origin, this will result in 225 country pairs in total.
13
In equation 1, the dependent variable, Nij,t/Nii,t, shows the number of people choosing to migrate
from country of origin i to country of destination j at each period of time, denoted by Nij,t, over
the amount that decides to stay in their home country, denoted by Nii,t.
We start our model with the main variable of interest, total ODA. To make a comprehensive model, we incorporate variables that are proxies for the income effect and network effect. Together with a selection of important pull- and push-effects that have been identified by the existing literature.
To incorporate the income effect into our model, we use net income per capita in the country of destination at each period of time, denoted by wj,t, over the net income per capita in the
country of origin, denoted by wi,t. For the network effect we use the current stock of migrants
of country of origin already living in the specific country of destination, denoted by
Mig(ration)Stock ji,t-1 in the model. The remaining variables in our model are already specified in the descriptive statistics.
Our initial model thus looks as follows:
𝑙𝑛(𝑁ij,𝑡/𝑁ii,𝑡) = 𝛽1ln(TotODAi,t−1) + 𝛽2𝑙𝑛(𝑤j,𝑡/wi,𝑡) + 𝛽3(Governancei,t−1) +
𝛽4(Dependencyi,t−1) + 𝛽5ln(MigStockji,t−1) + 𝜀ij,t (1) 3.3.1 Endogeneity concerns
Endogeneity is present when an, observed or unobserved, variable that is not incorporated into our model, is correlated with variables we do include in our model. Endogeneity can arise due to several issues.
Classic measurement errors, suppose we get an incorrect measurement for our independent variable, wi,t=wi,t + vi,t, vi,t will be the measurement error both affecting wi,t and the error term,
14
governments. It is therefore unlikely that there are consistent measurement errors in our dataset causing correlation between our variables and the error term9.
Simultaneity, when two variables are codetermined with each variable affecting the other. Estimating one of the two variables by itself will cause endogeneity. It is possible that we experience some extent of simultaneity through the network effect. Since ODA affects the (bilateral) migration flows which in their turn could potentially affect ODA.
Omitted variable bias (OVB), the largest endogeneity concern for our model. OVB occurs when a model does not incorporate one or more relevant variables. Relevant variables meaning variables that significantly affect the left-hand side as well as the right-hand side. As a result, the estimated coefficients will be biased due to the correlation with variables not included. We most likely suffer some degree of OVB in our initial model. We summarized multiple push- and pull-effects in our theory that are not in our model due to unavailability of data or because they are outside our field of science. A linguistic factor is one of the omitted variables, a linguistic similarity between the country of origin and destination has been shown to have explanatory power on international migration. Since a high linguistic similarity lowers the cost of migration and increases acceptance in the foreign labor market (Adserà, and Pytliková, 2015). Another factor affecting bilateral migration flows is colonial history. Berthélemy (2009) shows that there is increased bilateral migration between countries with colonial history. Since Africa was almost completely colonized in the 19th and 20th century by European countries, this will affect some bilateral migration flows. All countries of destination in our dataset have a history with African colonization, although the Scandinavian activities were relatively small. As a result of the colonization, all countries of origin in our dataset have an official European language, that is a language accepted by the country’s government. We expect that many of the country pairs are, to different extent, positively affected by time-invariant omitted variables. We will adjust our model to account for time-invariant fixed effects.
𝑙𝑛(𝑁ij,𝑡/𝑁ii,𝑡) = 𝛽1ln(TotODAi,t−1) + 𝛽2ln(wj,𝑡/wi,𝑡) + 𝛽3(Governancei,t−1) +
𝛽4(Dependencyi,t−1) + 𝛽5ln(MigStockji,t−1) + 𝛽6(country pair) + 𝜀ij,t (2)
9 For example, Data on immigrant populations may be collected through registers, residence permits, statistical
15
We create a dummy variable for each country pair in our dataset, 225 in total, which consolidates all the variables that are not incorporated in our model. The dummy variable will differentiate between country pairs due to explained colonial history effects and different linguistic distances combined with all remaining unobserved effects.
Besides time-invariant omitted variables, our model, most likely, also suffers from time-variant omitted variables. The main set of time-variant factors that are not included into our model are climate effects. Since natural disasters happen infrequently and the perceived risk of global warming is increasingly incentivizing international migration within our timeframe of research. Theoretically is it possible to deal with the time-variant variables, that are not captured with the fixed effects model, with an instrumental variable model. We considered to include lagged ODA, however it is not completely certain that the exclusion restriction is satisfied. Since the lagged value is only allowed to affect migration flows through current ODA. But ODA is used as a tool to prevent anticipated migration in the future (Clemens, 2017). Therefore, the exclusion restriction is not satisfied.
Self-selection bias, self-selection bias arises in situations where unobserved characteristics causes people to self-select them into a group, causing abnormal conditions in that group. Self-selection bias of individuals into migration has been a key issue in individual level cross-section regressions exploring migration behavior. Some individuals, due to unobserved characteristics are more likely to migrate than others. If these characteristics correlate with our right-hand side variables, we will suffer OVB. However, since we do our analyses at the country level, self-selection bias is unlikely to be a problem unless it makes specific countries systematically different. But even if we experience systematic differences at the country level, our model takes these systematic differences into account by using country pair fixed effects10.
3.3.2 Multicollinearity concerns
We do expect multicollinearity problems with our variables, which happens when two independent variables are correlated. Multicollinearity inflates the variance which can decrease the explanatory power of the model. As stated in our theory, ODA is expected to increase the income per capita in the receiving country. Therefore, we need to check for multicollinearity
16
issues between the variables total ODA and the relative GNI per capita. To do this, we estimate the variance inflation factors (VIF) of our initial model. The VIF-test shows the effect of the correlation between different variables on the variance compared to no correlation. Table A.1 in the appendix shows the VIF-test results. The highest VIF in our model is 2.14, therefore we can conclude that multicollinearity is not an issue for our regressions.
3.4 Hypothesis
Our main variable of interest is total ODA, even though in existing literature the total effect of ODA on migration is primarily positive. We must note that the most recent conclusion is from 2009, almost a decade ago. In the meantime, there has been the European migration crisis. We hypothesize that the total effect will be nonzero, but we are inconclusive on whether we expect a positive or negative effect.
H0: β1=0 (3)
H1: β1≠0 (4)
17
4. RESULTS
In this chapter, we will show and discuss the results. First, we will look at our standard model, adjusted for fixed effects, and briefly interpret the estimates. Since we are using variables with different units of scale, we added a column with standardized coefficients for each regression. Standardized coefficients show us how much standard deviations our dependent variable will change per standard deviation in the independent variable. With standardized coefficients we can make a comprehensive conclusion on the relative effect of the independent variables on our dependent variable, migration flow.
Second, we will look deeper in our main variable of interest, total ODA. We will run our regression multiple times, each time adding a single variable. By doing this, we will get a deeper understanding of our coefficients for total ODA and the effect of incorporating the other variables on it.
Third, we will extend our model with an additional term for income, GNI2, because we expect a hump-shaped pattern in the relation between GNI and migration. By doing this, we can see if the hump-shape pattern also applies on our dataset and what its effect is on the power of our model.
18
4.1. Interpretation results
Table 2 shows us the results of our fixed effects model. In the second column, we see our results from the unstandardized standard regression. In the third column, we see the results with standardized coefficients. The dependent variable in our model is in log-transformed state. This means that the coefficients of governance and dependency will be interpreted by a one unit increase in the independent variable, which will result in a percent change in the dependent variable. Our other independent variables are all in log-transformed state, this will result in an elastic relationship with the dependent variable. This means that coefficients show the percentage change in our dependent variable if the independent variable increases 1%. Our main coefficient of interest, total ODA, is significant at the 5% level and has a positive sign. The coefficient for ODA is 3.74e-4, this means that for an increase in ODA of 1%, migration flows increase with 3.74e-4%. For the economic interpretations we hold all other variables that are included constant.
Table 2: results fixed effects model
Variable Coefficient Standardized Total ODA 3.74e-4**
(1.77e-4) 0.029** (0.014) GNI -0.54*** (0.12) -0.44*** (0.095) GOV -0.068* (0.037) -069* (0.037) DEP -0.018*** (0.0058) -0.36*** (0.11) MS -3.87e-8 (2.97e-7) -0.0053 (0.041)
Fe dummy Included Included
# observations 2570 2570 Breusch-pagan Chi2(5) prob>chi2 129.02 0.00 129.02 0.00
19
Both the relative GNI per capita and dependency ratio are significant at the 1% level. The coefficient for GNI per capita is -0.54. This means when the relative GNI increases, GNIi goes up or GNIj goes down, by 1%, migration flows decrease by 0.54%. The sign of the coefficient indicates that, for the whole dataset together, the average income is above the hump-shaped pattern peak shown by Berthélemy (2009). We will look further into this when we add a quadratic variable for GNI. The coefficient for dependency is -0.018. This means for a one unit increase in the dependency ratio, migration goes down by 0.018%. This result is in line with the theory, since more dependents per capita decreases, on average, the utility gain from migrating by increasing the cost of migration through a non-monetary social factor.
The quality of government is significant at the 10% level. The coefficient for government quality is -0.068. This means that if the perception of the likelihood of political instability and/or politically motivated violence decreases with one unit, thus ratio increases with one unit, the migration decreases with 0.068%. The sign of the coefficient is in line with the theory as well as with common sense. If the perceived quality goes up, the incentive and net utility gain of migration decreases.
20
4.2 Total ODA extension
Our main variable of interest, total ODA, affects the migration flows through the income and network effect. To establish accurate estimates for total ODA, we incorporated the most important factors for the income and network effect into our model. To get a deeper understanding to the effect of these variables on our model, we will perform a hierarchical regression analysis.
We start with only ODA in our model, we observe an insignificant coefficient with a relatively high standard error. When we add the relative GNI into our model, we directly observe a highly significant coefficient for GNI, but the extension doesn’t affect the coefficient for ODA drastically. This shows us, as well as any other estimation in this research, that the effect of GNI on migration is strong.
The next variable we add to our estimation is migration stock, which represents the network effect in our model. Even though we never observe a significant effect for migration stock in our research, which is probably due to the high standard error in our dataset, we do observe that the inclusion of migration stock increases the power of our model. With the inclusion of migration stock, our coefficient for total ODA changes from insignificant, to significant at the 5% level.
21
Table 3: results fixed effects model, variables added in succession.
Variable +ODA +GNI +MS +DEP +GOV Total ODA 9.07e-5
(9.62e-5) 8.67e-5 (9.69e-5) 4.03e-4** (1.86e-4) 3.87e-4** (1.82e-4) 3.74e-4** (1.77e-4) GNI -0.21*** (0.070) -0.37*** (0.88) -0.53*** (0.11) -0.54*** (0.12) MS 2.22e-7 (2.44e-07) 05,28e-8 (2.81e-7) -3,87e-8 (2.97e-7) DEP -0.019*** (0.0052) -0.018*** (0.0058) GOV -0.068* (0.037)
Fe dummy Included Included Included Included Included
# observations 3619 3619 2682 2682 2570 Breusch-pagan Chi2(n) prob>chi2 34.90 0.00 84.24 0.00 72.01 0.00 129.10 0.00 129.02 0.00
22
4.3 GNI per capita extension
GNI per capita is an important variable in our model, but our current model does not incorporate that we expect this variable to have a hump-shaped pattern. To test if we indeed experience a hump-shaped pattern we add a quadratic variable for GNI per capita in our model, GNI2. By doing this we maintain a linear regression, since the variable is quadratic but the coefficient is not, but we allow our model to curve.
We observe a significant negative coefficient for GNI2. This means that we, in line with our theory, have a concave curve in our regression. In other words, if GNI per capita increases, the effect of GNI per capita on the migration flow becomes increasingly negative.
Table 4: results fixed effects model
Variable Coefficient Standardized Total ODA 3.33e-4*
(1.70e-4) 0.026* (0.13) GNI -1.88*** (0.40) -1.53*** (0.33) GOV -0.056 (0.037) -0.056 (0.037) DEP -0.017*** (0.0059) -0.34*** (0.11) MS 1.62e-7 (2.63e-7) 0.022 (0.036) GNI2 -0.23*** (0.06) -0.99*** (0.27)
Fe dummy Included Included
# observations 2570 2570 Breusch-pagan Chi2(5) prob>chi2 151.13 0.00 151.13 0.00
Note: the robust standard errors are given between the parentheses. The dependent variable is migration flow, the number of migrants between country of origin and destination. The polynomial term, GNI2, is included to
23
4.4 European migration crisis
In this section we add a year dummy for the year 2015 for each variable. We observe some interesting results. First, we observe that the coefficient for GNI becomes closer to zero and, for the first time in our research, that the coefficient is not significant at the 1% level but at the 5% level. A logical explanation for this observation could be that in time of crisis, utility increase due to income increase is a relatively less important motivation for migration.
This reasoning is in line with the observations from our incorporated push-effects, governance and dependency. The coefficient for governance decreases, which means that a decrease in the perceived quality of governance has a larger effect on the migration flows. For dependency the coefficient decreases, which implies that in time of crisis, the number of dependents has less effect on the decision to migrate.
24
Table 5: results fixed effects model with crisis dummy
Variable Coefficient Standardized Total ODA 2,87e-4*
(1.61e-4) 0.023* (0.13) GNI -1.84*** (0.41) -1.50*** (0.33) GOV -0.059 (0.038) -0.060 (0.038) DEP -0.019*** (0.0060) -0.37*** (0.12) MS 1.71e-7 (2.56e-7) 0.023 (0.035) GNI2 -0.23*** (0.065) -0.97*** (0.028)
Total ODAcrisis 2.23e-3*** (6.67e-4) 0.028*** (0.0081) GNIcrisis -0.61** (0.27) -0.40** (0.17) GOVcrisis -0.21*** (0.059) -0.052*** (0.014) DEPcrisis -0.011** (0.0050) -0.19** (0.090) MScrisis -1.34e-6 (3.22e-7) -0.22 (0.0052) GNI2crisis -0.13*** (0.045) -0.26*** (0.91)
Fe dummy Included Included
# observations 2570 2570 Breusch-pagan Chi2(5) prob>chi2 167.20 0.00 167.20 0.00
25
5. DISCUSSION
So how will these results affect the ongoing debate on migration policies? Our results show that if you look at ODA and migration in isolation, the straightforward solution would be to reduce ODA to reduce migration flows. In reality, this solution is not feasible. ODA is established to achieve long-term humanitarian goals, for example the millennium goals set by the United Nations of which all our qualified countries are a member. Our results show that the network effect of ODA is the primary factor that affects migration flows. In the debate about migration control, consolidated ODA expenditure could mitigate this network effect. Aggregated ODA would imply shared responsibility for the resulting migration flows. In the current situation, the Dubliner agreement, in which migrants have to ask for asylum in the first country they enter, causes an inequality in migration flows between European countries, since the majority of the migrants enter Europe through the Southern border. Quantifying the effects of consolidated ODA on migration flows could improve the establishment of a joint plan for immigration reception and put the emphasis back on international cooperation and instead of national protection (Park, 2015).
26
6. CONCLUSION
Governments use ODA as a policy instrument to establish migration control. However, the total effect of ODA on migration is ambiguous. In this paper, we shed some new light on the effects of ODA on migration flows. We constructed a bilateral panel dataset consisting of 225 different country pairs from Africa and Europe.
We constructed a model that besides our main variable of interest, incorporates the most important pull- and push-effects as well as variables that represent an important income and network factor. By doing this, we increase the validity of our ODA coefficients by minimizing endogeneity concerns. We use a fixed effects model that eliminates any time-invariant omitted variables. Although we aim to eliminate all OVB, our model likely suffers to some degree of OVB due to time-variant omitted variables we are unable to incorporate into our model.
We identified two factors through which ODA effects the migration flows, the income and network effect. For the income effect, we see that GNI per capita is a very strong factor that affects the migration flows through the income effect. If we look at the standardized coefficients, we observe that, for all our regressions, GNI per capita captures almost the complete income effect. Furthermore, if we add the independent variables in succession, we observe that GNI per capita barely affects the coefficients for total ODA. Altogether, we can conclude that the income effect of ODA on migration flow is marginal at best.
Even though we conclude that the income effect of ODA on migration is marginal. We observe a significant positive relation between ODA and migration. From this we can conclude that there is a positive network effect of ODA on migration. We observe that adding our network effect proxy, migration stock, to our model increases the quality of our ODA coefficients even though we do not observe significant coefficients for migration stock. Furthermore, we observe that the included push-and pull-effects, dependency and governance, are significant and deflate the coefficients for ODA, thus increasing the accuracy of our estimations.
27
we observe that the coefficient for governance is amplified, the effect of a decrease in the perceived quality of the governance on migration flows increases. The effect of dependency dampens, the negative effect of an increase in the number of dependents decreases.
The coefficient of ODA is higher in the crisis year than during our whole timeframe. This means that the network effect of ODA is amplified during the crisis. These results are relevant for policymakers in the ongoing discussions about migration, migration control, bilateral and multilateral development assistance.
7. REFERENCES
Adserà, A., Pytliková, m., 2015. The role of languages in shaping international migration. Economic Journal 125, 49-81.
Beine, M., Bertoli, S., 2016. A practitioners’ guide to gravity models of international migration. The World Economy 39. 496-512.
Berthélemy, J.C., Beuran, M., Maurel, M., 2009. Aid and migration: substitutes or complements? World Development 37, 1589-1599.
Bonifazi, C., Okólski, M., Schroorl, J., Simon, P., 2008. International migration in Europe: New trends and new methods of analysis. IMISCOE research, 107-128.
Borjas, G.J., 1989. Economic theory and international migration. International Migration Review 23, 457-485.
Boyle, P., Halfacree, K., Robinson, V., 1998. Exploring contemporary migration. London: Routledge.
Clemens, M., Postel, H., 2017. Deterring Emigration with Foreign Aid: An Overview of Evidence from Low-Income Countries. Center of Global Development policy paper 119, 1-30.
28
Dietz, B., 2006. Aaussiedler in Germany: From smooth adaptation to tough integration. Paths of integration: Migrants in Western Europe, 116-136.
Epstein, S., 2008. Herd and network effects in migration decision-making. Journal of Ethnic and Migration Studies 34, 567-583.
Faini, R., Venturini, A., 1993. Trade, aid and migrations: some basic policy issues. European Economic Review 37, 435-442.
Faini, R., Venturini, A., 1994. Migration and growth: the experience of Southern Europe. CEPR Discussion Paper 964.
Hansen, R., 2003. Migration to Europe since 1945: Its history and its lessons. The Political Quarterly 74, 25-38.
The Organization for Economic Co-operation and Development, 2018. http://www.oecd.org/about, http://www.oecd.org/oda.
Page Moch, L., 2003, Moving Europeans: Migration in Western Europe since 1650. Bloomington: Indiana University Press.
Park, J., 2015. Europe’s Migration Crisis. Council on Foreing Relations. https://www.cfr.org/backgrounder/europes-migration-crisis
Pritchett, L., Summers, L.H., 1996. Wealthier is healthier. Journal of Human Resources 31, 841-868
Salt, J.,1992. The Future of International Labor Migration. International Migration Review 26, 1077–1111.
29
Stanton Russel, S., Teitelbaum, M., 1992. International migration and international trade. World Bank discussion paper 160.
Swartz, A., 1973. Interpreting the effect of distance on migration. Journal of political economy 81, 1153-1169.
World Bank, 2018. http://datatopics.worldbank.org/world-development-indicators/
Zelinsky, W., 1971. The hypothesis of the mobility transition. Geographical Review 61, 219-249.
Appendix A: results VIF test
Table A.1: results VIF test
Variable VIF 1/VIF
DEP 2.14 0.47
GNI 2.12 0.47
MigStock 1.19 0.84
ODA 1.14 0.88
30
Appendix B: countries list
Table B.1: qualified countries & unqualified countries
Origin Destination Origin Destination
Algeria Angola Burkina Faso Burundi Cameroon Côte d'Ivoire Egypt Ethiopia Ghana Kenya Malawi Mali Morocco Mozambique Namibia Nigeria Rwanda Senegal Sierra Leone South Africa Sudan Tanzania Uganda Zambia Zimbabwe Austria Denmark Finland France Germany Italy The Netherlands Spain Sweden Benin Central African Republic Chad Democratic Republic Congo Republic Congo Comoros Cabo Verde Djibouti Eritrea Eswatini Gabon Guinea Guinea-Bissau Equatorial Guinea Liberia Libya Lesotho Mauritania Mauritius Niger Somalia South Sudan Sao Tome and Principe Seychelles Togo Tunisia Belgium Czech Republic Estonia Greece Hungary Iceland Ireland Latvia Luxembourg Norway Poland Portugal Slovak Republic Slovenia Switzerland Turkey United Kingdom
31
Appendix C: balance table
Table C.1: descriptive statistics Variable Mean sample N sample Mean EU N EU Mean Africa N Africa Balancing tests (1) (2) (3) (4) (5) (6) (7) Governance -0.41 525 0.87 135 -0.85 390 1,13e-7*** (-7.46) GNI 12266 612 37087 153 3197 459 8.42e-5*** (-7.30) Dependency 77.37 612 51.28 153 85.22 459 1.10e-6*** (5.99)
