1
Corruption and Migrational Decision Making: A Gravity Model
Approach with in Depth Focus on Central Eastern Europe
University of Groningen Faculty of Economics and Business
Master Thesis International Economics and Business
Name Student: Steven Brandsma Student ID: S2579324
Student Email: S.J.Brandsma@student.rug.nl Date Thesis: 19-06-2018
Name Supervisor: R.K.J. Maseland Co-assessor: M. Rosta
ABSTRACT
Drawing from a framework frequently used in international trade economics, I examine the effect of corruption on migration decisions. More specifically, I distinguish between the average migrant and the high skilled worker, which allows for the investigation of the brain drain issue. In addition, a specific focus has been attached the Central and Eastern Europe due to the fact that brain drain is prevalent in this area and has been on the rise since the fall of the Soviet-Union. The results indicate that there is no sufficient evidence to state that corruption is a factor that affects the average migrant, let alone the high skilled individual.
2
1. Introduction
“International trade is 13 times more frequent than international migration” and “immigration is four times as frequent as emigration” are the words uttered by Clemens(2011) in his paper ‘Economics and Emigration: Trillion-Dollar Bills on the Sidewalk?’ when discussing the focus areas of the average economist. Nevertheless, over the years the number of migrants has been increasing and the individual mobility of people is greater than ever. Several factors are likely to be the driver behind this increase in migration flows, and several factors are likely to have an influence on the size of labor flows, their origins and their destinations. However, very few papers have looked into the effect corruption exercises on migration decisions.
In this paper I empirically try to prove that corruption in the country of origin negatively impacts the migration decisions of the citizens of said country of origin. In other words, decreases in the control of corruption level result in increases in the emigration rate of that country. My analysis uses stock migration data that consist of 193 origin countries and 20 OEC destination countries. Moreover, this data can distinguish between the level of education for each migrant which allows for further investigation of the brain drain issue. This phenomenon is especially present in Central and Eastern Europe, where highly educated individuals migrate to the Western world (Straubhaar & Wolburg, 1999; Docquier & Rapoport, 2012). This is why this paper will put a magnifying glass under this region, which is done by regressing the models with only CEE countries as nations of origin. The empirical method used in this paper rests on the usage of a gravity model, most commonly used in bilateral international trade models. However, due to the same structure of data, can be employed in international migration as well. This has been proven by Karemera, Oguledo, and Davis (2000)who conclude a gravity model is suitable for migrant flow analysis. Moreover, this same model has been applied in Mayda (2010). However, there have been two previous papers which have investigated the effect of corruption on migration decisions for both the average as the high skilled individual (Ariu & Squicciarin, 2013; Dimand, Krieger & Meierricks, 2013). Yet, these papers have their methodological drawbacks. They make use of a fixed-effects model, which considers country-pair fixed effects. In international migration, it is preferable to have country fixed effects. Moreover, they only consider the corruption in the country of origin. My paper is able to consider both the pull and push factors resulting from corruption, while at the same time considering country fixed effects instead of country-pair fixed effects. In addition, this paper specifically looks into the issue of brain drain in the CEE region, which has not been done in either papers.
The aim of this paper is therefore to answer four core research questions. Namely, whether corruption affects the migration decisions of (1) the average world migrant, (2) the skilled world migrant, (3) the average CEE citizen, and (4) the skilled CEE citizen.
3 significant effect on migration decisions.
Section 2 will extrapolate previous research with regards to this topic. This is followed up by an explanation of the used data in section 3 and the methodology is set out in section 4. Afterwards, the results of the regressions are expositioned in section 4. These results are further discussed in section 5, where I try to find explanations for the finding that we have seen in the previous section. Moreover, the limitations of the paper are discussed in section 6 and this paper ends with a conclusion in section 7.
2. Literature review
The core principle is that people will migrate once they assume that doing so will create benefits that outweigh the costs. This has also been noted by DaVanzo (1980), who describes that the most basic underlying premise of migration is that “an individual migrates in the expectation of being better off by doing so”. ‘Being better off’ ultimately aligns for a big part with income, however not all can be accounted to this. Mayda (2010) has studied the factors that determine migration decisions. One of the most important variables that have an influence on migration are the expected earnings. According to her, income opportunities in the destination country significantly increase the size of emigration rates. She also already found this in Mayda (2005), thus confirming her findings. This seems to be in line with Cebula (2005), who researched the topic of migration within the US. According to him, gross state in-migration is an increasing function of income per capita. Also Clark, Hatton, and Williamson (2007) and Kennan, and Walker (2011) seem to support this, as they conclude that relative incomes matter.
Other factors besides income play a role in migration decisions. Lewer (2008) note that current emigration decisions are correlated with earlier immigrant flows. The interpretation for this observation is that emigrants incur costs when adapting to a new society. By moving to places in which other original country members have already settled, adaption-costs are mitigated by the presence by the earlier immigrants. Clark (2007) notes that there might be another reason, besides the costs-lowering theory, for the fact that emigrants follow previous flows. They state that this partially reflects the stance of immigration policy that encourages family reunion, which results in a powerful cumulative effect. According to Zawodny (1997), welfare does not act as an omitted variable in this case. It could potentially be the case that the immigrants all go to the same place because of that specific area having high welfare benefits. Instead the author finds that the immigrants that are already located in high welfare areas are in fact the source of the following migrants. This finding seems to point into the direction that income does not per definition have to be the biggest factor in determining emigration flows, but instead cultural factors are of importance as well. Lastly, Borjas (1987) comments that inequality in the origin and destination economies affect the size of migration rates. Thus, even though income factors are definitely important, cultural and other economic factors definitely play a role in the development of migrational decisions.
4 are relatively abundant in high skilled labor. As a result, the poor countries will produce products which are manual labor intensive, and the rich countries focus on products which are human capital intensive. Taking these facts in mind, one can deduct that high skilled workers are not incentivized to migrate to rich countries, since their labor is more valuable due to their rarity in their own country. As a result, theory would predict that high skilled people in poor countries are not eager to migrate at all. This seems to be confirmed by O’Rourke (2003). He finds that, line with the expected findings resulting from the Heckscher-Ohlin model, beings high-skilled in a poor nation has a negative impact on pro-globalization sentiment and thus migration.
Heckscher-Ohlin seems to predict that high skilled laborers in poor countries are dis-incentivized to migrate. However, empirics appear to somewhat differ from economic theory. Brain drain, the emigration of highly skilled people, seems to be a problem in the Central and Eastern-European (CEE) region. This proofs to be problematic, since it could potentially hamper economic growth in de CEE nations. This is found by Wong and Yip (1999) who do indeed find that that brain drain hampers economic growth, and thus should be limited by the government of the brain drain suffering nation. They note that brain drain is hurtful for non-emigrant and non-high skilled in the country. Thus, it has adverse effects on unskilled workers that remain in the country of origin due to the fact that high skilled migration lowers the wages of the remaining citizens. This can partially be explained by the Rybczynski proposition, which has been built on neo-classical economics. The theorem argues that ‘an increase in the supply in a factor of production results in an increase in the output of the final good that uses this factor of production relatively intensively’ (Van Marrewijk, 2007). The opposite of this proposition effectively happening when brain drain occurs. The input of (human) capital intensive goods is declining. As a result its output will decline, relative to goods that use low skilled labor as input, and the brain drain suffering nation will be locked into the production of goods that mostly use low skilled labor.
5 which Kazlauskienė and Rinkevičius (2006) contribute to ‘the state of the academic system’. Lastly, the authors conclude that a country’s macroeconomic status and governmental policy are able to push people away. Respondents that named this factor as an emigration cause were most likely the economically driven. This thus implies that citizens leave the country searching for higher wages, which has already been described previously.
From these findings it is not difficult to see how corruption can amplify these mechanisms that drive migration. The second, socio-economic factors, are in line with the status inconsistency theory. This theory can partially explain why high skilled laborers are incentivized to migrate to less corrupt countries. The argument states that it is possible for individuals to exhibit some status characteristics that rank relatively high and relatively low at the same time, hence they are not consistent with each other. One or some aspects are lagging behind on others. Thus, let us transpose this theory to a highly educated worker in a poor and corrupt country. In terms of education, this individual is highly ranked. However, it could be the case that his job he currently holds is ranked lowly in terms of status. This could be due to the fact that in corrupt systems, elite positions are closed off to outsiders and thus only open for the inner-circle. Without corruption, the individual might have been able to get the better job and hence join the elite of the country. The inconsistency between the educational status and economic/political status might drive this highly educated worker to migrate, since in other countries those type of jobs he seeks might be attainable. High skilled individuals could potentially be unhappy about their social status as a result of corruption. Moreover, corruption could also negatively affect the academic environment. Ineffective spending of budget for personal gain in de education sector is an example of this, or the hiring of political friends in high positions in the educational environment. As a result, the level of education will deteriorate and the push factor will take effect, stimulating emigration desires for high skilled citizens. In conclusion, corruption could lay at the foundation of the push factors that make citizens, and in particular high skilled workers, migrate.
Alternatively, high skilled people are more incentivized to migrate is because they occupy jobs which are hampered more by corrupt practices than other career paths. For instance, a regular farmer will not feel the effects of corruption as hard as an economist. From this leads that the negative utility a person receives from corruption is larger for high than for low skilled people. As a result, high skilled people are more likely to migrate due to the higher incurred costs.
6 Their first regression consists of a pooled OLS estimation. The authors find that corruption has a positive significant effect on migration. However, they only consider the corruption levels of the country of origin and not that of the destination country. Nevertheless, they find that corruption does indeed push citizens away. This holds true for both the average worker as well as for the high skilled individuals. However, it is also noted that ‘the marginal effect of corruption on skilled migration tends to be approximately three to four times higher than its effect on average migration’. Next to the OLS estimate they also regress a fixed effects model. This is a fixed effects model with country-pair fixed effects, since time invariant variables between the nations, like distance, are being omitted. When looking at the fixed-effects estimate, the authors conclude that the corruption variable is only significant for high skilled workers. According to them, this strengthens their believe that the decision of high skilled workers to emigrate is partially dependent on the level of corruption in the home country.
Ariu and Squicciarini (2013) is the other paper that specifically studies the effect of corruption on migration decisions by the average citizen and the highly skilled professional. Their sample includes 123 countries for the period 1990-2000. According to the authors of this paper, corruption might affect migration through two channels. First of which is the most straightforward hypothesis. Namely, corruption acts as a push factor, which causes high skilled people to move to less corrupt countries where employment is meritocratic. The second hypothesis states can be both defined as a pull and push factor. Namely, high skilled laborers are likely not attracted to markets in which access to high-paying and prestigious jobs is determined by political affiliations or nepotism. The dependent variable in this study is defined as ‘’the net flow of highly skilled migrants weighted by population’’. However, they do not specify whether they mean the population in the country of origin or in the destination. In contrast to other papers that look at migration decisions, Ariu and Squicciarini (2013)make use of flow data instead of stock data. The methodological approach that these authors do differently is that they regress corruption to inflows of people as well as outflows. However, their measurement of corruption is the same as in Dimand et al. (2013). Namely, only the corruption levels in the country of origin are taken into account. In terms of the results, the authors find that ‘the higher the corruption of the country, the lower the immigration of foreign skilled workers’ and that ‘skilled people are more likely to move abroad if their country of origin is highly corrupt’.
7 the latter would be a much better approach, since it effectively measures the percentage of a country that migrates to a certain country.
My paper aims to fill multiple gaps in the literature. First of which is the timeframe. Whereas Dimand et alstudy the effect of corruption on migration during the period 1985-2000, this paper will include a later time period. Namely, data on migration is available for the period 1980-2010 for 20 destination countries. In addition, the timeframe 2000-2010 has had some interesting developments with regards to migration. First of all, migration has been booming in the more recent years. People are more mobile than ever and this is visible when looking at the migration data. Secondly, the collapse of the Soviet Union and its effects are only becoming visible starting around 1990 at its earliest. The most notable change that we are currently seeing is the brain drain occurring in the CEE region, in which the high skilled workers move to mainly Western-European countries. In addition, many of these countries have joined the European Union in 2004 and Romania and Bulgaria joined in 2007. This has increased labor mobility in Eastern Europe significantly, and thus the moving costs to migrate have declined due to this decline in relative distance. Since CEE countries have some problems with corruption in the public sector then it might, according to theory, be an underlying cause for the observed brain drain.
Besides the fact that the time frame fills a gap in the literature, it is also more complete in its methodological approach. As mentioned before, Ariu and Squicciariniand Dimand et al only consider the level of corruption in the country of origin. This paper also has the advantage that it considers the GDP per capita and the corruption in both the origin and destination country. The added benefit is that it can distinguish between the pushing effect of corruption in the country of origin, as well as the pulling effect of the destination country. Moreover, this paper makes use of two estimation methods to verify its robustness. The first and main regression is a gravity model, which allows for the use of country fixed effects instead of country-pair fixed effects. This type of fixed effects are generally preferred when looking at migration. However that does not mean that country-fixed effects are per definition wrong by any means. The second model that has been used is a country-pair fixed effects estimation. This is in line with Mayda (2005; 2010). Even though she does not research the effect of corruption, her approach is looking for significant variables that affect migration decisions is useful. In that paper, the author uses a gravity model that includes country fixed effects and an additional country-pair fixed effects model, which are the same models which will be used in this paper. Lastly, the dependent variable in my paper will be defined as the stock migrants relative to the population in the country of origin, as opposed to the population in the destination country.
All in all concluded, literature seems to point into the direction that corruption might have a positive relationship with migration. Thus, it might act as a pushing factor if control of corruption at home has a negative relationship with migration or as a pull factor if control of corruption in the destination country has a positive relationship with migration. This leads to the following hypotheses:
8
H2: Corruption limits the professional academic realization of high skilled workers. In addition, status inconsistency theory dictates that high skilled workers seek to migrate in order to maximize their utility. As a result, the expectation is that high skilled workers are more inclined to migrate as corruption increases, and to a bigger degree than the average person because they have more to gain and are more mobile.
H3: Corruption acts as a deterrent for citizens in their satisfaction in a country. It limits their opportunities in said nation. As a result, corruption in the country of origin acts as a pushing factor, stimulating the amount of emigrants in the country. Moreover, low levels of corruption in destination countries act as a pull factor for migrants. The magnitude of the effects is expected to be higher in Eastern-Europe specifically, since Europeans have lower costs of mobility due to the European Union. Moreover, on average they are richer than the average world citizen and thus have more transportation opportunities.
H4: Corruption limits the professional academic realization of high skilled workers. In addition, status inconsistency theory dictates that high skilled workers seek to migrate in order to maximize their utility. As a result, the expectation is that high skilled workers are more inclined to migrate as corruption increases, and to a bigger degree than the average person because they have more to gain and are more mobile. The magnitude of the effects is expected to be higher in the Eastern-Europe specifically, since Europeans have lower costs of mobility due to the European Union. Moreover, on average they are richer than the average world citizen and thus have more transportation opportunities.
3. Data and methodology
3.1 Data sources and sample info
The data on migration has been gathered from The Institute for Employment research, for which Brücker, Capuano, and Marfouk (2013) have to be accredited for. Their database lists the level of stock immigrants in 20 OECD destination countries1 which are originating from 191 source countries2 and covers the time period 1980-2010 with five year intervals. The requirement for someone to be considered an immigrant is that they are foreign born. This implies that naturalized citizens are still counted towards the migrant stock. Legislation to acquire citizenship of destination countries differ across nations. Therefore the requirement of being foreign born as a criteria removes the problem of inconsistent measurements across destinations. One drawback in this database is that the data for each country is retrieved from a country’s own statistical office, which means that there might be slight variations in definitions across countries. However this problem is per definition unovercomable, unless a researcher has established his own public database with the same methodology across nations and has perfect knowledge of the country’s migrational in- and outflow, which is not the case.
There are some countries who ‘shapeshifted’ over time. For instance, Germany used to be split in East- and West-Germany in the earlier years of the time period which the sample
1 Australia, Austria, Canada, Chile, Denmark, Finland, France, Germany, Greece, Ireland, Luxembourg, the
Netherlands, New Zealand, Norway, Portugal, Spain, Switzerland, Sweden, United Kingdom and the United States
9 data covers. However, this has been dealt with. For example, the data for Germany and Yemen has been entirely aggregated over the period. In addition, there are also countries which did not exist during the entire period of 1980-2010. Think of some former Soviet-Union, colonies that gained independence during this time frame (Belize), or nations that split up during this time (East-Timor). In these cases, the authors estimated the “immigrant stock from each post-secession origin by multiplying the total migration stock of the pre-post-secession state by the gender and skill-specific share of the post-secession country population over the total pre-secession country migration stock” (Brücker et al).
Migrants are grouped by their level of education and are assigned to either being low, medium, or high skilled. This allows for researching differences across educational levels and thus investigate the effect of corruption on brain drain. The first group are people who have had lower secondary, primary and no schooling. The second group are people with high-school leaving certificate or equivalent, and the high skilled people have a higher than high-school leaving certificate or equivalent. In order to not skew the data towards low skilled people, since young people are more often lower educated because they are still learning, an age limit has been imposed. Migrants only show up in the database if they are at least 25 years or older.
Three ‘countries’ from the dataset have been left out of the dataset that is being used in this paper. Those identified places are the West Bank, Holy See and Unknown. Since these areas do not have data for the other variables that are used in this paper, they are deemed unusable.
Lastly, there seems to be a problem with the reliability of data on earlier years which has not been mentioned by the authors. For instance, in the years 1980, 1985 and 1990 it seems that the number of male Colombians in Germany was exactly 0 in all of these periods. However, in 1995 the amount rises to 1579. These numbers seem non-plausible to be correct. As a result, some parts of the data had to be transformed. To overcome this problem, zero values have been replaced by missing values if in a certain year the amount of low, medium and high skilled people from country x to country y all have a zero value. Thus, zero values remain possible while also preventing zero values that should be missing values to skew the data.
The preferred data for corruption would have been retrieved from the International Country Risk Guide, which is the database used in Dimand et al as well as in Ariu and Squicciarini. Unfortunately this database is locked behind a paywall and therefore not accessible for this paper. Instead the main corruption variable has been gathered from the World Bank database, which they define as the control of corruption. Thus, high scores indicate that the control of corruption is high and corruption itself is low. According to the World Bank, the variable ‘captures perceptions of the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as "capture" of the state by elites and private interests’3. This variable has come about by combining a range of other variables which measure different sorts of corruption. Therefore this variable is suitable to measure the general effect of corruption on migration.
The values of the control of corruption variable range between -2.5 to 2.5. This is due to the fact that the data has been standardized. However, the standardization has been implemented relative to the entire dataset. The list of countries in this paper is slightly different.
10 For instance, Serbia and Montenegro have been combined, since the migration data list them both as one country. Therefore the average of both countries have been taken for the corruption value. Also, because of the differences in the total sample and the sample in this paper, the data has been re-standardized. The Z-score scaling method has been used in order to make the data in the set have equal means and standard deviations. Thus, using the formula:
(1) 𝛧𝑐𝑡=
(𝑋𝑐− 𝜇)
𝜎 , 𝑤𝑖𝑡ℎ 𝑍 𝑏𝑒𝑖𝑛𝑔 𝑡ℎ𝑒 𝑛𝑒𝑤 𝑠𝑐𝑜𝑟𝑒 𝑓𝑜𝑟 𝑐𝑜𝑢𝑛𝑡𝑟𝑦 𝑐 𝑖𝑛 𝑡𝑖𝑚𝑒 𝑡, 𝑋 𝑡ℎ𝑒 𝑟𝑎𝑤 𝑠𝑐𝑜𝑟𝑒 𝑜𝑓 𝑐𝑜𝑢𝑛𝑡𝑟𝑦 𝑐, 𝜇 𝑡ℎ𝑒 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑟𝑎𝑤 𝑠𝑐𝑜𝑟𝑒 𝑜𝑓 𝑡ℎ𝑎𝑡 𝑦𝑒𝑎𝑟, 𝑎𝑛𝑑 𝜎 𝑎𝑠 𝑡ℎ𝑒 𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑡ℎ𝑎𝑡 𝑦𝑒𝑎𝑟.
The biggest drawback for the use of this dataset is the fact it only starts in 1996, while we have migration observations starting as early as 1980. This implies that several years will be lost in which the effects of corruption on migration could be tested, since the corresponding corruption values are unavailable.
The second type of corruption data is a dataset that is most commonly used, next to the World Bank data set, when looking at overall corruption levels in the public sector. The data has been retrieved from Transparency International their yearly reports and is titled the corruption perceptions index.
The same drawback exists for this variable as well as for the World Bank control of corruption data. Namely, the data availability in earlier years is limited. The data starts pouring in for a select group of about 40 countries for the year 1985. However, as we advance further in time, more scores are becoming available for more countries. In this dataset as well, a higher score indicates that a country is less corrupt. The same procedure has been applied to this data as with the primary corruption data. Namely, the data has been standardized according to formula (1). The construction of the variable itself is a composite index that is based on a combination of surveys, assessments of corruption from 13 different sources, and scores and ranks countries based on how corrupt a country’s public sector is perceived to be. Thus, these sources combined create the variable which we perceive as the corruption perceptions index. The expectation is that the results of the regressions with the World Bank (WB) control of corruption data do not give significantly different results than the robustness check regression which uses the data from Transparency International (TI). This expectation can be supported with a statement by Jain (2001), who concludes that ‘correlation between various measures tends to be very high’ when looking into different types of corruption measurement methods.
11 should not be a problem since they are only small countries4 which should not influence the results by significant margins. The natural logarithm of the variable has been taken, since it is a continuous variable and it takes care of outliers5.
Data on distances between countries, common languages and former colonies all have been retrieved from the CEPII database. For the distances between countries, the amount of kilometers between respective capitals has been taken, since this would be a better measurement than border distance in the case of common borders. A dummy variable has been used to account for common languages. Both countries need to have 1 or more of the same official language for the dummy variable to turn to 1. The colony variable is also a dummy variable. It denotes whether two countries have had a colonial relationship with each other, or whether both countries once belonged to the same country. In addition, this dummy variable also turns to 1 when the two countries used to be the same nation. However, an additional fixed effects model has also been run, which includes country-pair fixed effects (and can be found in the appendix). In that model, these three time-invariant variables will be omitted due to the nature of the model.
A variable that measures whether two nations co-exists at the same time within the European Union has been implemented. Its value will denote a 1 if the country pair are in the European Union together at time t, since within this economic region free transport of goods, services and people is available. As a result, it will lower the costs to migrate. The country in the pair with the latest accession date will determine the change of the dummy value from 0 to 1. For instance, Hungary joined in 2004, while Romania joined the EU only in 2007. As a result, the dummy variable will display a value of 0 for the years 1980-2005, and will only turn to a 1 for the year 2010.
As previously mentioned, the migration data only counts people with the age of at least 25 years old. Therefore, the sending as well as the receiving country population variables should contain only people aged 25 and older, since they are of influence on the amount of 25+ people migrating. The data source for this is the World Bank. More specifically, the Barro-Lee estimates of people aged 25 years and older. However, there is not a direct measurement readily available that shows us the absolute number of people aged 25+ in country x at time t. What is available however, is the absolute number of people between each age bracket for both genders separately. Therefore the numbers have been gathered for each gender and age bracket separately and simply been added together.
3.2 Model building
In order to test the two stated hypotheses using the previously discussed data, I will make use of the slightly differing methodological approaches. The first one of which is able to find whether there is evidence that corruption in the origin country has a positive relationship with the relative amount of migrants from the country of origin. On the other hand, it could also be the case that corruption does not only work as a pushing factor, but could instead act as a pulling factor if the level of corruption is low in the destination country. Indeed, it could also be the case that both forces are at play at the same time. Therefore the following expression has been constructed with is able to answer H1:
12 (2) 𝑆𝑡𝑜𝑐𝑘𝑀𝑖𝑔𝑟𝑎𝑛𝑡𝑠𝑜,𝑑,𝑡
𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑜,𝑡 = α + 𝛽1𝐺𝐷𝑃𝑜,𝑡+ 𝛽2𝐺𝐷𝑃𝑑,𝑡+ 𝛽3𝐶𝑜𝑟𝑟𝑢𝑝𝑡𝑖𝑜𝑛𝑜,𝑡+
𝛽4𝐶𝑜𝑟𝑟𝑢𝑝𝑡𝑖𝑜𝑛𝑑,𝑡+ 𝛽5𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑜,𝑑+ 𝛽6𝐿𝑎𝑛𝑔𝑢𝑎𝑔𝑒𝑑𝑢𝑚𝑚𝑦𝑜,𝑑+ 𝛽7𝐶𝑜𝑙𝑜𝑛𝑦/ 𝐶𝑜𝑢𝑛𝑡𝑟𝑦𝑑𝑢𝑚𝑚𝑦𝑜,𝑑+ 𝛽8𝐸𝑈𝑑𝑢𝑚𝑚𝑦𝑜,𝑑,𝑡 + 𝛽9𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑑,𝑡+ 𝜀𝑖,𝑡
In this equation, the nominator of the dependent variable is the amount of stock migrants from origin country o, to destination country d, in the year t. This will in turn be divided by the denominator of the dependent variable, the population of the same origin country o and the equivalent year t. Thus, the result is that the dependent variable is a measurement of relative migration. It only grows of the increase in relative migrants is higher than the level of relative country of origin population. This is the same approach as Mayda uses, but not the same as Dimand et al. The latter divides by the population in the destination country. The explanatory variables containα, which functions as the constant. Furthermore, 𝛽1 and 𝛽2 capture the effect of income per capita on migration decisions at time t. 𝛽3 and 𝛽4 are the main variables of interest. They try to capture the effects of corruption. However, keep in mind that this variable signifies the control of corruption. Higher values are indications that actual corruption is lower. 𝛽5 indicates the distance in kilometres between the capital of the origin and destination country, 𝛽6 is a dummy variable that displays a 1 if there is an official common language between the origin and destination country, 𝛽7 displays a value of 1 if both counties once had a colonial relationship or have belonged to the same country and 𝛽8 is a dummy variable that displays a value of 1 if both the origin and destination country were part of the European union at time t. Lastly, 𝛽9 captures the effect of a larger population pool in the destination country, since it partially affects the migrant absorption capabilities.
According to the model, I expect the signs to be as follows: 𝛽1 < 0, 𝛽2 > 0, 𝛽3 < 0, 𝛽4 > 0, 𝛽5 < 0, 𝛽6 > 0, 𝛽7 > 0, 𝛽8 > 0, 𝛽9 > 0. 𝛽1 and is expected to be negative since a lower level of GDP per capita in the country of origin incentivises people to migrate, thus creating a negative relationship. The same holds when the relationship is reversed. When GDP per capita rises in the country of origin, people are more likely to opt for staying as opposed to migrating. 𝛽3 is assumed to be negative, since the variable indicates control of corruption. Lower levels of control of corruption incentivises people to migrate, which creates the negative relationship. For 𝛽2 and 𝛽4 the exact opposites hold true, since they deal with the destination country. Higher levels of GDP per capita and control of corruption in the destination country are stimulus for people to immigrate into that country, thus creating a positive relationship. Also, higher control of corruption in the destination country is expected to have a positive influence on migration since it is an attractive feature for migrants. 𝛽5 is assumed to be negative, since migration will be higher as distances are smaller. Lower distances decrease the moving costs (Berger & Blomquist, 1992). The three dummy variables are fairly straightforward, since country-pairs sharing those same characteristics should enable higher migration rates. Lastly, 𝛽9 is assumed positive, since a larger destination country can absorb more migrants ceteris paribus.
13 (3) 𝑆𝑡𝑜𝑐𝑘𝑆𝑘𝑖𝑙𝑙𝑒𝑑𝑀𝑖𝑔𝑟𝑎𝑛𝑡𝑠𝑜,𝑑,𝑡
𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑜,𝑡 = α + 𝛽1𝐺𝐷𝑃𝑜,𝑡+ 𝛽2𝐺𝐷𝑃𝑑,𝑡+ 𝛽3𝐶𝑜𝑟𝑟𝑢𝑝𝑡𝑖𝑜𝑛𝑜,𝑡+
𝛽4𝐶𝑜𝑟𝑟𝑢𝑝𝑡𝑖𝑜𝑛𝑑,𝑡+ 𝛽5𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑜,𝑑+ 𝛽6𝐿𝑎𝑛𝑔𝑢𝑎𝑔𝑒𝑑𝑢𝑚𝑚𝑦𝑜,𝑑+ 𝛽7𝐶𝑜𝑙𝑜𝑛𝑦/ 𝐶𝑜𝑢𝑛𝑡𝑟𝑦𝑑𝑢𝑚𝑚𝑦𝑜,𝑑+ 𝛽8𝐸𝑈𝑑𝑢𝑚𝑚𝑦𝑜,𝑑,𝑡 + 𝛽9𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑑,𝑡+ 𝜀𝑖,𝑡
In this equation, the dependent variable is the amount of skilled people from country o moving to country d at time t. This number then is divided by the amount of people in the country of origin. These aforementioned formulas will change slightly when looking more in depth into Eastern-European countries6 and its brain drain. Interaction variables will be used in these models. They are a combination of the EU dummy and corruption variables. In addition, the colony or country dummy variable will not be included since this is not relevant when looking specifically at the countries in the CEE region. This can be characterized by the following regression which will be able to answer H3:
(4) 𝑆𝑡𝑜𝑐𝑘𝑀𝑖𝑔𝑟𝑎𝑛𝑡𝑠𝑜,𝑑,𝑡
𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑜,𝑡 = α + 𝛽1𝐺𝐷𝑃𝑜,𝑡+ 𝛽2𝐺𝐷𝑃𝑑,𝑡+ 𝛽3𝐶𝑜𝑟𝑟𝑢𝑝𝑡𝑖𝑜𝑛𝑜,𝑡+
𝛽4𝐶𝑜𝑟𝑟𝑢𝑝𝑡𝑖𝑜𝑛𝑑,𝑡+ 𝛽5𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑜,𝑑+ 𝛽6𝐸𝑈𝑑𝑢𝑚𝑚𝑦𝑜,𝑑,𝑡 + 𝛽7𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑑,𝑡+ 𝛽8𝐸𝑈𝑥𝐶𝑜𝑟𝑟𝑢𝑝𝑡𝑖𝑜𝑛𝑜,𝑡+ 𝛽9𝐸𝑈𝑥𝐶𝑜𝑟𝑟𝑢𝑝𝑡𝑖𝑜𝑛𝑑,𝑡+ 𝜀𝑖,𝑡
From this model not much changes when looking at high skilled citizens within the European Union. The only real difference lies again in the change in dependent variable. Instead of the nominator in this variable displaying the stock migrants, it now shows the stock of skilled migrants. The reason why the language dummy variable is excluded lies in the fact that for this sample, the value is always 0 and thus not useful. This leads to the last model which will give the answer:
(5) 𝑆𝑡𝑜𝑐𝑘𝑆𝑘𝑖𝑙𝑙𝑒𝑑𝑀𝑖𝑔𝑟𝑎𝑛𝑡𝑠𝑜,𝑑,𝑡
𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑜,𝑡 = α + 𝛽1𝐺𝐷𝑃𝑜,𝑡+ 𝛽2𝐺𝐷𝑃𝑑,𝑡+ 𝛽3𝐶𝑜𝑟𝑟𝑢𝑝𝑡𝑖𝑜𝑛𝑜,𝑡+
𝛽4𝐶𝑜𝑟𝑟𝑢𝑝𝑡𝑖𝑜𝑛𝑑,𝑡+ 𝛽5𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒𝑜,𝑑+ 𝛽6𝐸𝑈𝑑𝑢𝑚𝑚𝑦𝑜,𝑑,𝑡 + 𝛽7𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑑,𝑡+ 𝛽8𝐸𝑈𝑥𝐶𝑜𝑟𝑟𝑢𝑝𝑡𝑖𝑜𝑛𝑜,𝑡+ 𝛽9𝐸𝑈𝑥𝐶𝑜𝑟𝑟𝑢𝑝𝑡𝑖𝑜𝑛𝑑,𝑡+ 𝜀𝑖,𝑡
As is indicated, there are two additional interactional variables when compared to the first two models. The intuition of this moderator variable is to show the change in migration when corruption in internal within the EU. The expected signs are quite straightforward. In these cases, 𝛽8 is expected to have a negative sign, and 𝛽9 has a positive expected sign.
Table 1 depicts the summary statistics of the data used in this paper. In total there are 3840 country pairs, which over 7 years, make a total amount of 26,880 observations. The amount of observations for the two dependent variables, average migration and skilled migration, are however lower than 26,880 due to the fact that there are some missing values for earlier years for some countries. Moreover, the way in which the migration data has been handled with respect to zero values, as explained in section 3.1, has reduced the amount of actual observations. One other aspect that has to be highlighted is the notion that the World Bank control of corruption data has more observations due to the fact that the dataset has more observations starting 1995. What is quite striking however is that the mean for the corruption
6 Albania, Belarus, Bosnia and Herzegovina, Bulgaria, Croatia, Czech Republic, Estonia, Hungary, Latvia,
14
Table 1 – Summary statistics
(1) (2) (3) (4) (5)
VARIABLES Observations Mean S.D. Minimum Maximum
Year 26,880 1995 10.00 1980 2010
Country-pair indentifier 26,880 1921 1109 1 3840
Average Migration 19,829 0.00468 0.0276 2.48e-08 0.703
Skilled Migration 19,829 0.00143 0.00846 0 0.246
Per capita income origin (log) 23,100 3.704 0.561 2.197 5.105
Per capita income destination (log) Corruption origin (WB) 26,880 14,860 4.339 -0.014 0.254 0.995 3.535 -1.677 4.957 2.458 Corruption destination (WB) 15,360 1.798 0.484 -0.0140 2.458
Corruption origin (TI) 11,206 -0.00835 0.988 -1.884 2.563
Corruption destination (TI) 22,464 1.284 0.724 -0.740 2.518
Distance 26,180 7,488 4,515 59.62 19,586
Language 26,180 0.144 0.351 0 1
Colony or country 26,180 0.0452 0.208 0 1
EU 26,880 0.0472 0.212 0 1
Population destination 26,880 2.178e+07 3.825e+07 250,646 2.030e+08
destination variable is much higher for the World Bank data than opposed to the Transparency International database, despite the fact that both the data have been standardized.
15
Table 2 – Matrix of correlations
Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)
(1) Average Migration 1.000
(2) Skilled Migration 0.891 1.000
(3) Per capita income origin (log) 0.034 0.047 1.000
(4) Per capita income destination (log) 0.030 0.048 -0.001 1.000
(5) Corruption origin (WB) 0.057 0.065 0.439 -0.001 1.000
(6) Corruption destination (WB) -0.014 -0.015 -0.000 0.429 -0.003 1.000
(7) Corruption origin (TI) 0.032 0.039 0.386 -0.002 0.852 -0.004 1.000
(8) Corruption destination (TI) 0.017 0.024 -0.000 0.381 -0.002 0.958 -0.005 1.000
(9) Distance -0.067 -0.058 0.040 -0.177 -0.056 0.080 -0.061 0.114 1.000
(10) Language 0.229 0.247 0.040 0.018 0.034 0.036 0.033 0.078 0.098 1.000
(11) Colony or country 0.177 0.110 0.028 -0.038 0.027 -0.047 0.026 -0.037 -0.011 0.341 1.000
(12) EU 0.006 -0.009 0.229 0.006 0.385 -0.050 0.372 -0.067 -0.346 -0.045 0.013 1.000
16
3.3 Diagnostic tests
Because a second panel dataset is used, it is not appropriate to assume that there is no systematic difference in coefficients. The Hausman test is used to analyse which model, the fixed or random effects, is best in estimating the effects on migration. Even though it is quite clear that the fixed effects model will outperform the random effects model due to the nature of data. Nevertheless, it is wise to skill test whether this indeed holds true. As expected, in all regressions the p-values are zero or just slightly above zero7. Thus, the hypothesis that there is no systematic difference in coefficients has been rejected and the fixed effects mode is to be used.
Furthermore, due to the nature of the data it is quite likely that the data is not homoskedatic. This is due to the fact that every country-pair in the sample is different and thus not all random variables in the sequence or vector have the same finite variance and are thus heteroskedastic. To test whether the data is indeed heteroskedastic, a modified Wald test for groupwise heteroskedasticity in fixed effect regression model has been used. In all regressions there is a significant indication that there is indeed heteroskedasticity present8.
Lastly, a test to detect the presence of serial-correlation has been performed. The test used for this is the Wooldridge test which is for autocorrelation, which is specifically for panel data. In all but one regression there was a significant suspicion of autocorrelation9. This has thus been corrected for.
These findings indicate two things. First, the cluster option has to be used in the regressions on order to correct for heteroskedasticity and autocorrelation. Secondly, these findings are in favour for the gravity model.
3.4 The model
As mentioned before, this paper will makes use of two different models. The first of which is a gravity model10. Gravity models are most often when modeling international trade. However, they also see use in models that focus on migration. This is mainly to the fact that both are structurally quite similar to each other. In both models mostly consist of panel data. In addition, double sided explanatory variables are being used. For instance, in trade models GDP in country
a as well as GDP in country b are considered explanatory variables. The same could be applied
to migration models, in which either factor could be considered a pull or push factor. Moreover, such a gravity model has a big advantage over a fixed effects model. In a fixed effects model it is not possible to consider time-invariant variables, since they are being omitted due to the fact that they are collinear with the fixed effects themselves. This means that per definition it is impossible to include country fixed effects, since they are time invariant, and thus the only fixed effects that can be considered are country-pair fixed effects. However, in migration and trade models it is preferable to have destination fixed effects Lewer (2008). Moreover, Mayda (2010) argues that migration quota’s matter, which are not available for this paper. Therefore having destination fixed effects are critical to have. The gravity model allows for the usage of the panel
7 See appendix table 2 8 See appendix table 3 9 See appendix table 4
17 structure, while at the same time include origin and destination fixed effects. Also, this model assumes that heteroskedasticity and is present. This is not a drawback, but instead a strength, since all regressions do indeed exhibit heteroskedasticity. Another option besides the country-pair fixed effect model would be a pooled-OLS estimation. However, I have favored the country-pair fixed effects model over the pooled OLS model. Reason for this is that the pooled OLS regression is basically a middle of the road approach, which lies between a random and fixed effects model. Half is explained outside and half inside the model. I concluded it is more useful to use a country-pair fixed effect model as a second regression because of the previously listed arguments.
4. Empirical Results
Table 3 presents the results from equations 2, 3, 4, and 5. In this table, the gravity model has been used to estimate the models. Thus, they include country fixed and time fixed effects. The first regression in the table estimates equation (2), in which the effect of corruption is measured for the average worker in the world.
4.1 Gravity model
18 Table 3
Migrational Decisions Gravity Model Regression
(1) (2) (3) (4)
Variables Average Migration Skilled Migration Average Migration CEE Skilled Migration CEE
Per capita income origin 0.2486 0.2853 3.0151*** 2.0415***
(0.1601) (0.1749) (0.6827) (0.5658)
Per capita income destination 2.4462** 0.9296 3.4227* 1.6243
(0.9975) (1.0198) (0.6827) (2.0032) Corruption origin 0.0298 0.0416 -0.2370 -0.1646 (0.0363) (0.0424) (0.1641) (0.1470) Corruption destination 0.0070 -0.0702 -0.3068 0.2769 (0.0931) (0.0891) (0.2413) (0.2001) Distance -0.00017*** -0.00016*** -0.00194*** -0.00129*** (0.00002) (0.00001) (0.00042) (0.00034) Language 1.5478*** 1.5237*** (0.2021) (0.2142) Colony or country 1.9362*** 1.9187*** (0.1586) (0.1550) Population destination
9.34e-09*** 2.70e-09 3.58e-09 1.93e-09
(2.86e-09) (2.58e-09) (5.88e-09) (5.69e-09)
EU -0.0395 0.0107 -0.0572 -0.2359 (0.1885) (0.1676) (0.5282) (0.6467) EU x corruption origin -0.8493* -0.9746** (0.4959) (0.4877) EU x corruption destination 0.1321 0.3396 (0.2920) (0.3248) Constant -16.928*** -12.006*** -18.771** -10.622 (4.1436) (4.4953) (9.4722) (8.0985)
Country Fixed Effects YES YES YES YES
Year Fixed Effects YES YES YES YES
Observations 12020 12020 1301 1300
19 Moving onto regression (3) and (4) in table 3, which only consider CEE-counties as countries of origin, we see a slight change in the result compared to the earlier regressions. The average migrant in the CEE is somewhat comparable to the average migrant in the world. Per capita in income in the destination country seems to have a positive effect on the migration rate. However, the effect is bigger than for the world migrant. Namely, a 10% increase in the level of per worker GDP in the destination country increases emigration by 0.342 emigrants per 100,000 individuals of the origin country’s population. Thus, it seems that income is more detrimental to a citizen in the CEE region than for a random person in the world in its decision making process to migrate. Another noteworthy point is that the GDP in the home country is positively and significantly related to migration, even though we would have expected the sign to be negative. Once again, this might have to deal with the poverty trap theory which will be discussed later on in the following section. Moreover, the skilled migrant in the CEE is not much different than the average migrant in the CEE region. The only real difference lies in the magnitudes of the GDP per capita variables, which are lower for the skilled migrants. This also holds true for the distance variable. In both cases, for the world and the CEE region, its coefficient is lower for their respective high skilled people which suggests that distance is less of an issue for higher educated. Lastly, model (3) and (4) include two interaction variables, namely the interaction between EU and corruption. For both the average and high skilled, the interaction term between corruption in the country of origin and EU is negatively and significantly related to migration. This implies that the change in the effect of corruption is different when migration happens between two EU countries. Corruption might indeed have an effect on migration, albeit being it through an interaction. For the average and skilled citizen in the CEE region, a decrease in the control of corruption of one standard deviation in the origin country will result in an increase of the emigration rate when migration happens between EU countries by 0.8493 and 0.9746 respectively.
4.2 Fixed effects model
The effects from the fixed effects model are somewhat similar. There are however some differences. Table 4 denotes the same models, however, they have been ran in a fixed-effect model which considers country-pair fixed effects. For the average world migrant (1), corruption nor GDP seem to have significant effects on migration decisions. For the skilled migrant however (2), it seems that corruption in the destination country has a negative effect on migration, which is quite unexpected. Lastly, the fixed effects model estimates that GDP matters for high skilled CEE country members, however not in the expected way since the signs are switched.
20
Table 4
Migrational Decisions Country-Pair Fixed Effects
Regression
(1) (2) (3) (4)
Variables Average Migration Skilled Migration Average Migration CEE Skilled Migration CEE
Per capita income origin -0.00005 0.00005 0.00162 0.00095*
(0.00063) (0.00030) (0.00201) (0.00053)
Per capita income destination -0.00119 -0.00019 -0.00173 -0.00218*
0.00077 (0.00041) (0.00220) (0.00124) Corruption origin 0.00021 0.00002 0.00150* 0.00068 0.00034 (0.00019) (0.00074) (0.00042) Corruption destination -0.00018 -0.00056*** -0.00182 -0.00063 0.00025 (0.00015) (0.00128) (0.00042)
Distance (omitted) (omitted) (omitted) (omitted)
Language (omitted) (omitted) (omitted) (omitted)
Colony or country (omitted) (omitted)
Population destination
3.30e-10*** 1.94e-10 3.85e-10* 2.98e-10**
8.24e-11 (4.04e-11) (2.07e-10) (1.27e-10)
EU 0.00051 0.00017 0.00015 -0.00059 0.00033 (0.00014) (0.00179) (0.00056) EU x corruption origin -0.00139 -0.00035 (0.00114) (0.00030) EU x corruption destination 0.00031 0.00041* (0.00061) (0.00024) Constant 0.00209 -0.00168** -0.00039 0.00149 0.00167 (0.00086) (0.00843) (0.00253)
Country-pair Fixed Effects YES YES YES YES
Year Fixed Effects YES YES YES YES
Observations 12,020 12,020 1,301 1,301
21
5. Discussion and implications
As extrapolated in the previous section, corruption generally does not have a significant effect on migration. This is in contrast to Dimand et al and Ariu and Squicciarini, who did in fact find that corruption tends to drive people away. However, there is one small difference that might be the cause for such a different result. Namely, they used the stock migrant and divided them by the population in the destination country, whereas this paper looked at the stock migrant relative to the population in the country of origin. Moreover, this paper made use of PPP adjusted GDP, contrasting the previous papers that looked into the effect of corruption on migration decisions.
One of the more surprising observations that has been made is the fact that, in the case of CEE citizens, the per capita income in the country of origin is positively and significantly related to migration. However, this could potentially explained by the poverty constraint argument. Increases in the GDP in the country of origin enable people to migrate. When people are poor, they are unable to save money since it is dedicated to staying alive. However, when people get richer they are becoming able to make more savings. As a result, this money can be allocated to migrate since migration is not costless. This might thus explain the positive relationship between GDP per capita in the home country and the migration rate. The same theory might be applicable to why the magnitude for the positive relationship between GDP per capita in the origin country and the migration rate differs for average and high skilled in the CEE region. The coefficient is higher for the average worker. Within the framework of the poverty constraint framework, one might argue that the higher skilled are per definition richer due to their educational level. Either their educational level gave them better jobs which makes them richer, or they were initially rich which allowed them to get a proper education. However, the way the causality does not matter for this purpose. Either way, they are more likely to be richer and therefore already have saving. These savings allows them to migrate easier, and thus additional money, in terms of GDP per capita in the country of origin, benefits them less due to diminishing marginal returns. Because after a certain amount of money, you don’t per definition need even more to be able to migrate.
In the gravity model it became apparent that the per capita income in the destination country has a positive and significant effect on the migration decision for the average migrant but not for the skilled migrant. Moreover, the coefficients are higher for the average migrant. This seems to indicate that income is less of a reason to migrate for the skilled migrant. One possible explanation might be that the average migrant has more to gain in terms of utility stemming from monetary sources. For instance, the average migrant gains $1000 by migrating, while the high skilled person gains $1200. However, the utility gain for the average migrant is higher due to the diminishing marginal return effects of money. The high skilled person was already better off than the average person.
22 realization. This would also be in line with Vizi (1993), who notes that the atmosphere must be changed to make it more favorable for intellectual work.
The core purpose of this paper was however to focus on the effects of corruption on migration decisions. However in the gravity model it became apparent that the control of corruption variables never turned significant, except for once in the fixed effects model. In the regression that looked into the average migrant in the CEE region, it became evident that corruption in the country of origin does have a significant effect on migration. The expected sign for the variable was negative, however the empirically tested sign is positive. Firstly, I would like to state that my preference goes to the gravity model, in which there was no such significant effect to be found. However, the fixed effect model fids a significant relationship which requires a theoretical explanation. One answer for this puzzle might be found in educational accessibility in the country of origin. It is possible that education is more accessible to privileged individuals. The people in power, which likely are the corrupt individuals, have access to education. From this follows that high skilled individuals do not want to migrate from their countries to the non-corrupt countries, since they reap the benefits of corruption in their home country due to them being part of the elite. Another explanation could be found in the theory by Heckscher-Ohlin. According to this theory, people move where their marginal productivity is the highest since in a perfect world the marginal productivity equals the wages. In poor countries, which are more likely to be corrupt, high skilled labor is relatively more productive due to the abundance of low skilled workers. As a result, the returns are higher in the country of origin and thus there is an incentive to stay instead of migrate. Corruption thus acts as a proxy variable. However, it can be said that the hypotheses could not be proven. There is no supportive evidence for the assumption that corruption does push people away and incentivizes them to migrate. This applies to the average person as well as for the high skilled citizen. With regards to the third and fourth hypotheses, neither of them could be proven as well. The migration decisions of citizens from CEE nations are not affected by corruption in the home or destination country. However, the only significant effect with regards to migration that was found in the gravity model is that a decrease in the control of corruption results in an increase of the emigration rate when migration happens between EU countries. However, this might not be enough to prove H3 and H4. Lastly, cultural factors seem to always play a big role. In every regression in the gravity model, they are significant. All in all concluded, there is no sufficient evidence to support the four hypotheses and corruption minimization is not likely to prevent brain drain.
6. Limitations
23 complete corruption database, like the International Country Risk Guide, would have been more useful but was unfortunately unavailable to me. In addition, more sophisticated data on GDP per capita would have been useful. Having different wages for each type of citizen (based on educational level) would have increased the strength of the models. Data on the ease of immigration would have been useful, since migration quota’s matter (Mayda, 2010).
As for reverse causality, it is possible that the GDP per capita is interacting with migration. Immigrant inflows are likely to decrease wages in the destination country and outflows are likely to increase the wages in the country of origin (Clemens, 2011).
7. Conclusion
In this paper, I have empirically investigated whether corruption acts as a pulling or pushing factor in international trade for the average person as well as for high skilled individuals. In addition, this investigation has given the CEE region special attention with regards to this issue, since they are currently experiencing a brain drain. To test this, I have made use of a gravity model which includes country fixed effects, as well as a country-pair fixed effects model. I have not found evidence that supports my hypotheses that corruption in the country of origin is an incentive for people to migrate to less corrupt nations. This applies to the average migrant as well as to the high skilled individual. These same conclusions can be reached when looking specifically at the CEE region. However, there is one interaction in which corruption does have a significant effect on migration. This is the case, a decrease/increase in the control of corruption in the country of origin will result in an increase/decrease of the emigration rate when migration happens between EU countries. However, this is not enough to accept H3 and H4 as true, since the control of corruption variables themselves are non-significant in the gravity model.
As for future research, the attention can be shifted to other area’s that could potentially be affecting migration, and brain drain in particular. One could think of investigating whether the theory of Kazlauskienė and Rinkevičius (2011) empirically holds when considering that their surveys indicated that the professional academic environment in foreign countries attracts high skilled individuals. However, the entire topic on corruption has not been finalized yet. This paper specifically focused on corruption that stems from using public power for private gain. There might be other forms of corruption that have an impact on migration decisions.
Even though I have failed to find significant results to answer my research questions positively, that does not mean that this research has been futile. In fact, I believe it is a valid contribution to existing literature. Indeed, corruption is not a main force which drives brain drain. However, this study shows that governments should focus on other aspects in their policies in order to prevent brain drain from occurring in their country. Hence, valuable resources could be allocated elsewhere. Naturally this does not mean that preventing corruption is a fruitless task, but it does not help in preventing brain drain.
References
Ariu, A., & Squicciarini, M. P. 2013. The balance of brains—corruption and migration.
24 Beine, M., Docquier, F., & Rapoport, H. 2008. Brain drain and human capital formation in developing countries: Winners and losers. The Economic Journal, 118: 631-652.
Berger, M. C., & Blomquist, G.C. 1992. Mobility and destination in migration decisions: The roles of earnings, quality of life, and housing prices. Journal of Housing Economics, 2: 37-59.
Borjas, G. J. 1987. Self-selection and the earnings of migrants. Working paper no. 2248, National Bureau of Economic Research, Cambridge, MA.
Brücker, H., Capuano, S., & Marfouk, A. 2013. Education, gender and international
migration: Insights from a panel dataset 1980-2010. Methodology report, Institute for
Employment Research, Nuremberg.
Brücker, H., Capuano, S., & Marfouk, A. 2013. Education, gender and international migration: insights from a panel-dataset 1980-2010, mimeo.
Cebula, R. J. 2005. Internal migration determinants: Recent evidence. International
Advances in Economic Research, 11(3): 267-274.
Clark, X., Hatton, T. J., & Williamson, J. G. 2007. Explaining US immigration, 1971–1998.
The Review of Economics and Statistics, 89(2): 359-373.
Clemens, M. A. 2011. Economics and emigration: Trillion-dollar bills on the sidewalk?
Journal of Economic Perspectives, 25(3): 83-106.
DaVanzo, J. 1980. Micro economic approaches to studying migration decisions. Working paper 1201-Nichd, The Rand Corporation, Santa Monica, CF.
Defoort, C., & Rogers, G. 2008. Long-term trends in international migration: An analysis of the six main receiving countries. Population, 63(2): 285-317.
Dimand, E., Krieger, T., & Meierrieks, D. 2013. The effect of corruption on migration, 1985– 2000. Applied Economics Letters, 20(13): 1270-1274.
Docquier, F., & Rapoport, H. 2012. Globalization, brain drain, and development. Journal of
Economic Literature, 50(3): 681-730.
Horvat, V. 2004. Brain drain: Threat to successful transition in South East Europe? Southeast
European Politics, 5(1): 76-93.
Jain, A. K. 2001. Corruption: A review. Journal of Economic Surveys, 15(1): 71-121. Karemera, D., Iwuagwu Oguledo, V., & Davis, B. 2000. A gravity model analysis of international migration to North America. Applied Economics, 32:13: 1745-1755. Kazlauskienė, A., & Rinkevičius, L. 2006. Lithuanian “brain drain” causes: Push and pull factors. Engineering Economics, 46(1): 27-37.
25 Lewer, J. J., & van den Berg, H. 2008. A gravity model of immigration. Economics Letters, 99(1): 164- 167.
Massey, D. S., Arango, J., Hugo, G., Kouaouci, A., Pellegrino, A., & Taylor, J. E. Theories of international migration: A review and appraisal. 1993. Population and Development
Review, 19(3): 431-466.
Mayda, A. M. 2005. International migration: A panel data analysis of economic and
non-economic determinants. Discussion paper series, IZA DP No. 1590, Bonn.
Mayda, A. M. 2010. International migration: a panel data analysis of the determinants of bilateral flows. Journal of Population Economics, 23(4): 1249-1274.
O’Rourke, K.H., 2003. Heckscher-Ohlin theory and individual attitudes towards
globalization. Working paper 9872, National Bureau of Economic Research, Cambridge,
MA.
Stark, O., & Taylor, J. E. 1991. Migration incentives, migration types: The role of relative deprivation. The Economic Journal, 101(408): 1163-1178.
Straubhaar, T., & Wolburg, M. R. 1999. Brain drain and brain gain in Europe. Jahrbücher
für Nationalökonomie und Statistik, 218(5-6): 574-604.
Van Marrewijk, C. 2007. International economics. Oxford: Oxford University Press. Wong, K. Y., & Yip, C. K. 1999. Education, economic growth, and brain drain. Journal of
Economic Dynamics and Control, 23(5-6): 699-726
Zavodny, M. 1997. Welfare and the locational choices of new immigrants. Economic
26
Appendix
Appendix table 1
Afghanistan Djibouti Liberia St. Vincent & the Grenadines
Albania Dominica Libya Samoa
Algeria Dominican Republic Liechtenstein San Marino
Andorra Ecuador Lithuania Sao Tome and Principe
Angola Egypt, Arab Rep. Luxembourg Saudi Arabia
Antigua and Barbuda El Salvador Macedonia, FYR Senegal Argentina Equatorial Guinea Madagascar Serbia and Montenegro
Armenia Eritrea Malawi Seychelles
Australia Estonia Malaysia Sierra Leone
Austria Ethiopia Maldives Singapore
Azerbaijan Fiji Mali Slovak Republic
Bahamas, The Finland Malta Slovenia
Bahrain France Marshall Islands Solomon Islands
Bangladesh Gabon Mauritania Somalia
Barbados Gambia, The Mauritius South Africa
Belarus Georgia Mexico Spain
Belgium Germany Micronesia, Fed. Sri Lanka
Belize Ghana Moldova Sudan
Benin Greece Monaco Suriname
Bhutan Grenada Mongolia Swaziland
Bolivia Guatemala Morocco Sweden
Bosnia and Herzegovina Guinea Mozambique Switzerland
Botswana Guinea-Bissau Myanmar Syrian Arab Republic
Brazil Guyana Namibia Taiwan, China
Brunei Darussalam Haiti Nauru Tajikistan
Bulgaria Honduras Nepal Tanzania
Burkina Faso Hungary Netherlands Thailand
Burundi Iceland New Zealand Timor-Leste
Cambodia India Nicaragua Togo
Cameroon Indonesia Niger Tonga
Canada Iran, Islamic Rep. Nigeria Trinidad and Tobago
Cape Verde Iraq Norway Tunisia
Central African Republic Ireland Oman Turkey
Chad Israel Pakistan Turkmenistan
Chile Italy Palau Tuvalu
China Jamaica Panama Uganda
Hong Kong SAR, China Japan Papua New Guinea Ukraine Macao SAR, China Jordan Paraguay United Arab Emirates
Colombia Kazakhstan Peru United Kingdom
Comoros Kenya Philippines United States
Congo, Dem. Rep. Kiribati Poland Uruguay
Congo, Rep. Korea, Rep. Portugal Uzbekistan
Costa Rica Kuwait Qatar Vanuatu
Cote d'Ivoire Kyrgyz Republic Romania Venezuela, RB
Croatia Lao PDR Russian Federation Vietnam
Cuba Latvia Rwanda Yemen, Rep.
Cyprus Lebanon St. Kitts and Nevis Zambia
Czech Republic Lesotho St. Lucia Zimbabwe
27
Appendix figure 1
