Corruption as a Push and Pull Factor of Migration Flows Evidence from European Countries

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


Introduction
The European economic landscape has undergone major transformations since the early 1990s through international migration.During this recent period, the volume of migration in the European Union (EU) has more than doubled, with the number of foreign-born residents in EU countries reaching 60 million in 2019.This represents 12% of the EU resident population and over 23% of the stock of global migrants, while the EU population accounts for only less than 7% of the total world population.Among the current stock of the foreign born in the EU, more than 21 million individuals (or 35% of the total) are migrants from other EU member countries. 1  What motivates migration decisions?Individual rational actors decide to migrate because a cost-benefit calculation leads them to expect a positive net return -both monetary and non-monetary -from movement (Sjaastad 1962, Schwartz 1973, Todaro and Maruszko 1987).
In this microeconomic model of individual choice, where both push and pull constructs shape migration patterns, potential migrants move to where the expected discounted net returns are greatest (Lee 1966, Borjas 1987). 2 Then, because of its large economic and social costs, corruption could be an important determinant affecting migration decisions.In particular, does a high level of corruption in a country promote emigration and deter immigration?In order to make progress on this important question, in this paper we use a gravity model specification to investigate whether corruption could be identified as a push factor and/or as a pull factor shaping migration patterns.
Using data on bilateral migration flows across EU28 and EFTA countries between 2008 and 2018, we show that four different measures of corruption -both subjective and objective indicators -are strongly linked with migration flows.Across all the specifications considered in this paper, corruption is found to act both as push factor and as a pull factor for migration decisions.A 1-unit increase in the corruption level in the origin country is associated with an 11 percent increase in the outflow of migrants, while the same increase in the destination country is associated with a 10% decline in in-migration.
1 Of these intra-EU migrants, 11 million (or 18%) come from the EU15: Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, Spain, Sweden, and the United Kingdom.Another 10 million (or 17%), up from 4 million in 2000, come from the NMS13 (the New Member Sates of the EU): Bulgaria, Croatia, Cyprus, Czechia, Estonia, Hungary, Latvia, Lithuania, Malta, Poland, Romania, the Slovak Republic, and Slovenia.An additional 10 million (or 17%) are migrants from other non-EU countries in the western Balkans, Turkiye, Eastern Europe, and the European Free Trade Association (EFTA).
2 Direct tests support this microeconomic model of individual choice in migratory responses.Graves and Waldman (1991) show that the elderly retire in counties where public goods are capitalized more into wages than into land prices.Kahn (2000) finds migration into counties with improving air quality.Banzhaf and Walsh (2008) assess the impact of toxicity-weighted emissions and air quality on the composition of the population.
Comprehensive research on the nexus between corruption and migration is in its infancy and conclusive evidence remains to-date very scarce.This paper speaks to this new body of research, which could be split into two blocks.
The first strand of literature is interested in the role of corruption as a factor explaining migration patterns.It finds that corruption has both direct and indirect effects.A high level of corruption in a country significantly discourages immigration, as it is associated with weaker and more volatile economic conditions and more job insecurity (Poprawe 2015).
More corruption is also associated with higher emigration rates in sending countries, with a particularly strong effect for high-skill migrants (Morano Foadi 2006, Clausen et al. 2011, Ahmad and Arjumand 2015, Auer and Tjaden 2020).Multiple channels are behind this effect.For example, higher corruption levels have a negative impact on the quality of local institutions, which further leads to lower levels of trust from the public, weaker economic security, fewer job opportunities, and a lower quality of life.Corruption has a stronger impact on high-skill migrants because it erodes the meritocratic structure and it reduces the returns to education (Ariu and Squicciarini 2013, Dimant et al. 2013, Cooray and Schneider 2015). 3he second strand of the literature analyzes the impact of migrants on the corruption levels in both the home country and in the destination country.By increasing the demand for political accountability and a higher quality of local institutions, migration lowers the overall level of corruption in the origin country (Batista andVicente 2011, Abdih et al. 2012).For example, students obtaining tertiary education abroad and in more democratic countries can reduce corruption in the home country (Spilimbergo 2009, Beine and Sekkat 2013, Ferreras 2013).As for the impact on the destination country, immigration from corruption-ridden countries tends to boost corruption levels (Dimant et al. 2015).This is in contrast with general migration, which is found to have a positive impact on the institutional quality of the destination country, with reductions in overall corruption levels in countries with a high economic freedom (Clark et al. 2015, Bologna Pavlik et al. 2019).
To further investigate the role of corruption both as a pull factor and as a push factor of migration decisions, we use a broad array of -subjective and objective -corruption measures combined with bilateral migration flow data, exploiting both cross-sectional and time dimensions.Having access to a panel dataset on both bilateral migration flows and corruption levels is a contribution of this paper, and it improves on the most closely related studies that exist on this topic.In particular, both Dimant et al. (2013) and Cooray and Schneider (2015) consider net migration rates, without being able to disentangle whether corruption should be interpreted as a push factor or as a pull factor.On the other hand, the bilateral migration data included in Poprawe (2015) lack the time dimension, with information on 230 countries only for the year 2000. 4Although this paper improves on the existing literature by having access to a panel dataset, it may potentially still suffer from reverse causality.
Because of this, the analysis remains about correlation, and all the conclusions of this paper should not be interpreted as causal statements.
Lastly, by controlling for many drivers besides corruption levels, this paper speaks to the large body of research interested in quantifying the role of the different determinants of migration flows.A key driver of migration found in the literature is the presence of networks of previous migrants of the same nationality in the destination country, with cultural distance affecting migration decisions (Clark et al. 2007, Bauernschuster et al. 2014, Falck et al. 2012, Falck et al. 2018, Krieger et al. 2018).Especially relevant for low-skill migration, these diasporas explain the majority of variation in migration flows (Beine et al. 2011).By sharing a common language and culture, the economic and psychological costs associated with migration are greatly reduced.Migrant networks provide an essential service to newer waves of migrants during their process of job search, improving labor market outcomes in the destination country (Edin et al. 2003, Munshi 2003).From the perspective of the destination country, good education and health systems tend to attract migrants (Geis et al. 2013).This is in contrast with the significantly less clear role of social safety nets, which only seem to be salient for migrants from the least developed countries (Pedersen et al. 2008).From the perspective of sending regions, local amenities (e.g., security and public services) play an important role in shaping the intentions to emigrate (Dustmann and Okatenko 2014).
The remainder of the paper is organized as follows.Section 2 describes the dataset assembled for this study.Section 3 presents the empirical results.Section 4 concludes.

Data
We have built a novel dataset on corruption levels and bilateral migration flows for EU28 and EFTA countries between 2008 and 2018.Exploiting both a cross-section and a time dimension is the main contribution of this study.First, bilateral migration flow data (instead of net migration rates) allow us to disentangle the overall effect of corruption on migration into two distinct parts: push (encouraging emigration) and pull (discouraging immigration) factors.We also exploit the time dimension to include fixed effects absorbing the impact of unobserved and time-invariant factors.

Corruption
One definition of corruption could be "[t]he abuse of entrusted authority for illicit gain" (Norwegian Agency for Development Cooperation 2008).This could include both subjective and objective measures.For instance, corruption might range from bureaucratic corruption to a broader notion that further includes nepotism, patronage networks, fraud, conflicts of interest, bribery, embezzlement, extortion, favoritism, and political corruption (Johnston 2005).Given the complexity of capturing this variable, there exists a lot of debate on how to appropriately measure it.
In this paper, we try to capture the different elements of corruption by considering more than one measure.In particular, the dependent variable is based on three subjective indicators of corruption: i) the Corruption Perception Index (CPI); ii) the Control of Corruption Governance Metric (WGI CC); and iii) the Government Effectiveness Governance Metric (WGI GE).We also consider another measure of corruption: the International Country Risk Guide (ICRG).As shown in Hamilton and Hammer (2018), the survey-based indicators (in particular, the CPI and the WGI CC indicator) are the most valid measures of overall corruption in many country contexts.However, and in line with the existing literature on corruption, Hamilton and Hammer (2018) suggest to cross-check the results when using one measure of corruption with the use of additional indicators, as significant differences exist in the construction of these measures. 5So far, no consensus on the optimal measure of corruption has been reached.
The CPI, an annual index published by Transparency International since 1995, defines corruption as the misuse of "entrusted power for private benefit" (Transparency International, 2011).It ranks countries by the perceived levels of public sector corruption, as determined by expert assessments and opinion surveys. 6In particular, the CPI is a composite index based on a combination of surveys and assessments from 13 different sources, with each individual indicator of corruption standardized to have the same weight in the aggregate score. 7As the CPI measures only public sector corruption, two additional subjective 5 Besides ICRG, other objective measures of corruption exploit variations in conviction rates (Fiorino et al. 2012, Hill 2003) or press reports (Rehren 1996).However, these methods remain mostly unsystematic, thus leading to validity and reliability problems (Morris 2008, p. 390).
6 As mentioned by Eurostat: "As there is no meaningful way to assess absolute levels of corruption in countries or territories on the basis of hard empirical data, capturing perceptions of corruption of those in a position to offer assessments of public sector corruption is so far the most reliable method of comparing relative corruption levels across countries." 7 The main weakness of this measure is that the number of sources used to construct it varies over time, especially pre-2012.However, because the CPI uses sources from the last three years preceding each publication of the index, it is still possible to use the CPI year average to make comparisons over short periods of time (Persson and Tabellini 2003).There is a very strong significant correlation between the CPI and other proxies for corruption: black market activity and an overabundance of regulation (Wilhelm 2002), indicators are considered.
Compared to the CPI, the WGI CC, which is published by the World Bank Group, is a broader measure of public sector corruption.It measures both political and general bureaucratic corruption. 8The WGI CC is updated yearly and it ranges from -2.5 (most corrupt/least effective) to 2.5 (least corrupt/most effective).It is based on 30 individual data sources, selected to include the views of citizens, business owners, academics, and experts drawn from the public, private, and NGO sectors from across the globe.Then, the WGI GE measures non-elected public sector corruption.To the extent that bureaucrats behave somewhat independently from elected officials, the WGI GE captures corruption in the bureaucracy, rather than political corruption.
Lastly, we consider one objective indicator of corruption: the index issued by ICRG.It refers to the financial corruption associated with conducting business (e.g., bribes) as well as other forms of political corruption (e.g., excessive patronage, nepotism, and close ties between politics and business).9 As these four indicators use different scales, we have re-scaled them to range from 0 (least corrupt) to 10 (most corrupt).Summary statistics are presented in Table 1.10As shown in Figure 1, there is a substantial difference in corruption levels across European countries.
The southeastern region of the European continent is characterized by more corruption levels (Bulgaria and Romania at 5.6, Greece at 5.4, and Italy at 5.1) when compared to the Nordic region (Finland at 0.7, Denmark at 0.8, and Sweden at 1.0).

Migration
We focus on bilateral migration flow data across EU28 and EFTA countries between 2008 and 2018.These data are from Eurostat.Across the sample, countries differ both in terms of inflows and outflows of migrants, as shown in Figure 2. Larger migration outflows are observed in Eastern (Romania and Poland), Central (Germany and France), and Southern (Italy, Spain, and Portugal) European countries.The countries with the largest inflows are the U.K, Spain, Italy, and Switzerland.11as well as business regulation and the public perceptions of corruption (Treisman 2007).
In Figure 3, we begin to show that migration flows between origin and destination countries could be associated with corruption levels.Although most country-pairs bilateral migration is relatively low, high migration outflows are observed for high levels of corruption in sending countries, while much lower outflows are observed from countries that experience low levels of corruption.Preliminary analysis also suggests that higher migration flows are observed from relatively more corrupt to relatively less corrupt countries.Corruption appears to be both a push and a pull factor for migration decisions.

Control variables
The control variables that we use in the analysis are summarized in Table 1.
First, following the literature on gravity models, we measure the geographical distance between countries by including a dummy variable equal to one when countries share a common border, as well as indicators for the difference in time zones, the distance (weighted by population), and the area of the country.To control for cultural and economic distances, we include variables measuring whether countries share a common language, a common religion, and a common currency.All these variables come from the CEPII Gravity Database.
Second, to control for differences in income levels, labor markets, and openness, we include GDP per capita levels, inflation rates, and the importance of agriculture and services in the local economy.These variables come from the World Bank World Development Indicators.
Third, to measure differences between countries in legal barriers to mobility, employment legislation, and the generosity of the welfare system, we consider OECD measures on: i) employment protection legislation; ii) product market regulation; iii) social expenditure; and iv) the right to work. 1212 The OECD employment protection legislation is a synthetic indicator of the strictness of regulation on the dismissal of workers on regular contracts and the hiring of workers on temporary contracts.It ranges from 0 to 6, where 6 measures the strictest regulation.Then, a larger number indicates more job security.The OECD product market regulation measures the degree to which policies promote or inhibit competition in markets.It ranges from 0 to 6, where 6 captures policies that most inhibit competition.The OECD social expenditure indicator is presented as a share of GDP, and it includes social expenditure on old age, survivors, incapacity-related benefits, health, family, active labor market programmes, unemployment, housing, and other social policy areas.When looking at 2010 data on migration policies from the Immigration Policies in Comparison (IMPIC) database (Helbling et al. 2017), the previous indicators correlate in the expected direction.For example, social expenditure tends to be higher in countries with less restrictive labor migration policies (e.g., on work permit validity, labor market tests, loss of employment, or age limits).

Corruption levels and migration decisions
We develop a specification built around a gravity model (Anderson andVan Wincoop 2003, Poprawe 2015).As a starting point, we assess the relationship between each of the four different corruption indicators (the three subjective measures -CPI, WGI CC, and WGI GE -and the objective indicator -ICRG) and bilateral migration flows, with the inclusion of only year fixed effects.The specification is shown in Eq. 1 and the results are presented in Table 2.
Across all the four columns of Table 2, the results remain statistically significant.As mentioned in Section 2.1, all four indicators of corruption have been re-scaled on a 0-10 scale, with 10 indicating the most corrupt scenario.An increase in the corruption indicator by 1 unit in the sending country o is associated with a rise in migration from o to d of between 9% and 29%.On the other hand, a higher level of corruption in the receiving country d is associated with a fall in migration from o to d of between 34% and 44%.According to these preliminary results, corruption seems to act both as a push factor and as a pull factor of migration decisions, with the latter effect being twice as large.Migrants seem to respond more to the expected levels of corruption faced in the destination country.
As geographical, cultural, and economic distances between countries might have an effect both on the decision to migrate and on level of corruption, we consider a follow-up specification that includes variables indicating whether countries o and d lie on a contiguous border, and whether they share a common language, a common religion, and a common currency.Additionally, we include the area of the country, the difference in time zones, and the distance (weighted by population).This augments Eq. 1 with the inclusion of the vector Gravity o,d,t , as shown in Eq. 2.
The results are presented in Table 3. Reassuringly, all the estimates on the corruption indicators maintain their sign and significance: an increase in corruption in the origin country o (resp., the destination country d) is associated with a rise (resp., fall) in migration flows of between 13% and 28% (resp., 39% and 67%). 13The pull dimension of corruption is now 13 The included gravity controls enter the equation with the expected sign.In particular, a common border, a common language, a common religion, and a common currency are all found to increase migration flows between o and d.On the other hand, as countries lie further apart and more time zones separate them (both are proxies for migration costs), migration flows are reduced.Lastly, larger countries are more likely to be positively associated with patterns of both emigration and immigration.As there might be about three and a half times larger than the push factor. 14Table A2 estimates a specification using the bilateral difference (between o and d) in corruption levels.The coefficients show a positive relationship between the gap in corruption and the bilateral flow that occurs from o to d.
As omitted variable bias may be a cause for concern, we include additional factors that might have a simultaneous effect on both corruption levels and migration flows.These include: GDP per capita, inflation rates, the importance of agriculture and services in the economy, and also whether citizens of country o have the right to work in country d.Reassuringly, the sign and the significance of the estimated coefficients on the four corruption indicators remain unchanged.However, as shown in Table 4, both push and pull factors have smaller magnitudes following the inclusion of these additional controls: the former is estimated to be between 3% and 24%, while the latter is between 8% and 13%. 15Averaging the results across the four columns: a 1-unit increase in the corruption level in the origin country is associated with an 11% increase in out-migration.The same 1-unit increase in the destination country is associated with a 10% decline in in-migration. 16  Lastly, to control for differences in legal frameworks and regulations between countries, we include indicators on the employment legislation, product market regulation, and social expenditure.Results are presented in Table 5.All the estimates on the corruption indicators remain broadly unchanged: an increase in the corruption level in the origin country o (resp., the destination country d) is associated with a rise (resp., fall) in migration flows of between 18% and 33% (resp., 27% and 53%).

Conclusions
Policies to shape migration flows have become more important in a growing number of countries, especially those facing a brain drain.As such, policy makers have been looking at ways to encourage native citizens to stay, while attracting foreign workers.By looking interactions between the gravity control variables and the corruption indicators, in Table A3 we estimate a specification that explicitly includes the interaction between each control variable and the bilateral difference in the corruption indicator.
14 Table A1 further includes country fixed effects in the specifications of Eq. 1 and Eq. 2. Results remain broadly unchanged.
15 Legal barriers to mobility also matter substantially in the migration decision: removing all work restrictions between two countries is associated with a 172% increase in bilateral migration flows.As the dependent variables are log-transformed, to obtain the exact impact on the right to work variable, the coefficient β must also be transformed according to the formula: e β − 1. Taking the average value (0.28) across columns (1) to (4), the coefficient is 1.72. 16These results are in line with the existing literature.For example, Poprawe (2015) finds that a 1-unit increase in the corruption level is associated with a 22% increase in out-migration and a 14% decline in in-migration.
at four separate measures of corruption, the findings in this paper suggest that corruption can act both as a push factor and as a pull factor for migration decisions.Then, tackling corruption may be a viable policy option for reducing outward migration flows, as well as encouraging foreign immigration.Our results open the door to several important questions.
What are the most effective ways to design anti-corruption policies that are able to retain local workers and attract foreign workers?Is it corruption in general -or a particular dimension of corruption -that affects migration flows?Does the effect of a reduction in corruption levels vary between different types of workers?The map presents the average level of corruption across the four indicators of corruptions considered in this paper, between 2008 and 2018.The indicators are the Corruption Perception Index -CPI, the Control of Corruption -WGI CC -Governance Metric, the Government Effectiveness -WGI GE -Governance Metric and the International Country Risk Guide -ICRG.These four indicators have been re-scaled such that they range from 0 (least corrupt) to 10 (most corrupt).Then, an increase in these indices corresponds to a rise in corruption levels.0;2,500 2,500;5,000 5,000;10,000 10,000;50,000 50,000;100,000 100,000+ (b).Average migration inflows, 2008-2018 0;2,500 2,500;5,000 5,000;10,000 10,000;50,000 50,000;100,000 100,000+ No data

→
The figure identifies each observation in the sample according to a three-dimensional pattern, with the migration flow from origin to destination countries on the y-axis, corruption in the country of origin on the x-axis, and corruption in the destination country on the z-axis.A Appendix: Additional material

Figure 3 :
Figure 3: Migration flows and corruption levels in origin and destination country Triplets of corruptions in o and d, and migration o to d, 2008-2018

Table 1 :
Summary statisticsWGI CC and WGI GE are the Worldwide Governance Indicators Control of Corruption and the Worldwide Governance Indicators Government Effectiveness, respectively.CPI is the Corruption Perception Index and ICRG is the International Country Risk Guide.

Table 2 :
Year fixed effects

Table 3 :
Geographical, cultural, and economic distances

Table 4 :
Controlling for additional factors

Table 5 :
Legal regulations and the welfare stateThe additional gravity controls are those included in Table3.Robust standard errors clustered by county in parenthesis.***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.