Uncovering the Effect of Immigration on European
Employment Levels
-
An evaluation of the relation between increasing immigration and EU-28
employment levels at varying levels of aggregation
Matthijs Dekker S2759004
m.h.m.dekker@student.rug.nl University of Groningen Faculty of Economics and Business
Supervisor: K.M. Wacker, PhD
Abstract
Using data from the European Labor Force Survey (EULFS) this thesis provides a first empirical analysis of the way immigration affects employment in the EU-28 at varying levels of aggregation. The factor proportions approach provides the reader with a framework to theorize on this effect of immigration on employment. Problems arising in the empirical estimation are discussed, and how to appropriately address them. The results suggest that the effect brought about by immigration is heavily dependent on the level of aggregation at which the relation is estimated, and which (sub)section of the population is considered. Overall, the effect of immigration on employment is positive in its most robust setting. A 10 percentage points increase in the immigrant share is associated with a 0.98 percentage point increase in employment levels at the NUTS-2 level. Surprisingly, the effect is even more pronounced at the NUTS-0 level. In this case, a 10 percentage points increase in the immigrant share is associated with a 6.33 percentage point increase in employment levels. Differentiation between education levels reveals that those who enjoyed lower education tend to suffer more from an immigrant influx compared to those who enjoyed higher education at the NUTS-2 level. At the NUTS-0 level the overall relationship (and the corresponding conclusion) switches sign. Theory has a hard time explaining this phenomenon. Differentiating immigrants on the basis of their origin reveals that EU-28 immigrants appear to have a more positive effect on employment than non-EU-28 immigrants. This effect is even more pronounced at the NUTS-0 level compared to the NUTS-2 level. Yet, these results lack the desired statistical significance. Taken as a whole, this thesis stresses the general idea that immigration is an intricate phenomenon and its effect cannot be pinned down so easily.
Table of Contents
1. Introduction ... 3
2. Literature Review ... 5
The Theoretical Impact of Immigration ... 5
The National Skill-Cell Approach ... 6
A Spatial Approach ... 8
Implications for the Research Question and Methodology ... 11
3. Data ... 12
Description of the Dataset ... 12
Variables ... 13
Data Characteristics ... 14
4. Methodology ... 17
Ordinary Least Squares ... 17
First-Differences ... 18
Instrumental Variable ... 18
5. Results ... 19
6. Discussion and Conclusion ... 22
7. References ... 24
1. Introduction
The European Union (EU-28) has traditionally been an attractive destination for immigrants. This can easily be inferred from the number of immigrants it hosts. While globally around 200 million people, or 3% of the world, live outside their country of birth, within the EU-28 this number lays around 23 million people, or 7.0% of the EU-28 (Clemens, 2011). Besides that, the relative share of immigrants appears to be on a continuous rise (Figure 1). Although every immigrant has a unique story accompanied by varying reasons to migrate, a common theme amongst them seems to be the desire for a better life. More specifically, economically motivated migrants often seek to fulfill their desire for a better life by means of better employment opportunities and significant wage gains. The empirics indeed suggest that, in general, migrants enjoy a static wage gain upon moving (e.g. Clemens et al., 2009; McKenzie et al, 2010). Moreover, the remaining population of the sending countries, the ones who stay behind, often benefit as well. It is found that the groups who compete most directly with emigrants experience the largest relative gains (e.g. Elsner, 2012; Dustmann et al., 2012; Aydemir and Borjas, 2007; Bouton et al., 2011).
The effect on the receiving countries appears more intricate to answer. The literature largely fails to reach a consensus. Results differ from consistently negative (e.g. Borjas, 2003; Aydemir and Borjas, 2007), to modest (Card, 1990; Dustmann et al., 2016), and sometimes even positive (Dustmann et al., 2005; Foged and Peri, 2016) effects from immigration on labor market outcomes (wages and employment). Although a vast body of literature does not seem to agree on the effect of immigration, the European Union seems to be moving towards more restrictive immigration policies. Right-wing populist-nationalist parties in continental Europe are on the rise (Rodrik, 2018). Zaslove (2006) argues the rise of these radical right-wing parties have been instrumental in passing more restrictive immigration policy, limiting the flow of immigrants and the ability of non-EU labor to live, work, and settle permanently in either Austria or Italy. Goodwin and Milazzo (2017) even suggest that the sentiments regarding control over immigration was a key predictor of the vote for Brexit. Thus, the increasing presence of immigrants appears to have ignited a heated political debate regarding immigration. Politicians and the people are seeking to protect cultural identities, the
availability of jobs, and welfare state provisions.
From a purely economic point of view, this leaves one wondering what the employment effect of the rising share of immigrant stocks is on the receiving EU-28 countries. Although there are ample examples of case studies investigating this relation1, studies on the European Union
in its entirety remain absent. My study will add to the literature by providing an overview of the employment effect of immigration on EU-28 countries. Following the long-standing tradition of Card (1990), I will appropriate a spatial approach. This approach evaluates the
total effect of immigration on labor market outcomes and provides a meaningful and policy
relevant parameter. It is also considered to be robust to concerns introduced by skill
downgrading and takes labor and capital complementarities into account. However, since it is argued that expanding the geographic area covered in the analyses, results in the coefficient on the change in immigrant/native ratios to become less positive or more negative, I will evaluate its effects on two different levels of aggregation. Eurostat constructed the NUTS (Nomenclature of territorial units for statistics) classification for dividing up the economic territory of the EU. This classification is used to denote the level of aggregation. I resort to NUTS-0 (which overlaps with national borders) and NUTS-2 (basic regions for the
application of regional policies) to test whether different levels of aggregation affect results.
Research Question: What is the effect of increasing immigration on employment levels in EU-28 countries at the NUTS-2 and NUTS-0 level?
The European Labor Force Survey (EULFS) combined with Eurostat Population Data
provides me with the necessary information to answer this research question. My results show that immigration has a positive effect on employment in its most robust setting. A 10
percentage points increase in the immigrant share is associated with a 0.98 percentage point increase in employment levels at the NUTS-2 level. Surprisingly, the effect is even more pronounced at the NUTS-0 level. In this case, a 10 percentage points increase in the immigrant share is associated with a 6.33 percentage point increase in employment levels. Comparing the effect of immigration among education levels reveals that those with lower education suffer more from an influx of immigrants compared to those who enjoyed higher education at the NUTS-2 level. At the NUTS-0 level the overall relationship changes sign. Now, immigration appears to hit those who enjoyed higher education while the ones that have gained medium or lower educational attainment benefit. This largely contradicts theory and appears intricate to explain. Differentiating between immigrants from EU-28 countries and non-EU-28 countries reveals that immigrants from EU-28 countries have a more positive effect on employment than non-EU-28 immigrants. This effect is even more pronounced at the NUTS-0 level compared to the NUTS-2 level. However, this result does lack the desired statistical significance. The general conclusion would be that the effect of immigration on employment is heavily dependent on the level of aggregation and which (sub)section of the population is considered.
1 For instance, Dustmann et al. (2005) investigate the impact of immigration on the British labor market. They find no strong
This thesis is structured as follows. Section 2 provides an overview of the literature. It introduces a basic supply and demand framework to think about the potential effect of
immigration. Thereafter, it evaluates the different methodological strategies that are employed to uncover the effect of immigration on labor market outcomes in receiving countries. Section 3 elaborates on the appropriated dataset. Section 4 presents the applied methodological
strategy. Section 5 presents the results of the various regression specifications. Section 6 concludes and touches upon the limitations of this research. Some suggestions for future research are made as well.
2. Literature Review
The most common approach to conceptualize the impact of immigration on labor market outcomes is what we refer to as the factor proportions approach. This approach is appealing since it only requires the elasticities of substitution between skill groups, making it a
parsimonious way to tackle a rather complex issue. This complexity is illustrated by the fact that the vast body of literature describing the effect of immigration on wages or employment has been unable to reach a consensus. The sign and the size of the effect of immigration on labor market outcomes tends to differ with the empirical strategy that one applies. In general, there are roughly two methods by which one can estimate the effect of immigration on labor-market outcomes: The spatial approach, or the national skill-cell approach. This section will evaluate the theoretical impact of immigration by an evaluation of the factor proportions approach. Hereafter, the national skill-cell approach and the spatial approach are analyzed. I will draw on several examples and its corresponding strengths and weaknesses. Lastly, the implications for my research question and methodology are discussed.
The Theoretical Impact of Immigration
The factor proportions approach underpins most studies in this literature. The key insight is that immigrants change the relative abundance of different skill groups in the economy. Consider for instance an immediate inflow of immigrants in Figure 2.
Relative number of low-skilled to high-skilled workers 𝑂" Pre-immigration Labor supply Post-immigration Labor supply R el at ive la bo r m ar ke t ou tc om es ( 𝑂# ) of low -s ki lle d to hi gh -s ki lle d w or ke rs
Figure 2: The theoretical effect of an inflow of low-skilled workers, source: World Bank (2018). Demand
𝑂$
𝑛"= 𝑛𝑎𝑡𝑖𝑣𝑒𝑠 𝑛$= 𝑛𝑎𝑡𝑖𝑣𝑒𝑠 + 𝑖𝑚𝑚𝑖𝑔𝑟𝑎𝑛𝑡𝑠
It is often argued that immigrants are less skilled than native-born workers and therefore their arrival will increase the relative supply of low-skilled workers to high-skilled workers (𝑛" → 𝑛$). That, in turn, will change the relative wages or employment opportunities across skill groups. In this case, an increase in the supply of skilled labor is likely to depress low-skilled wages (or employment levels) relative to high-low-skilled workers, increasing inequality (𝑂" → 𝑂$).
The single most important variable in this model is the elasticity of substitution, given by the inverse slope of the demand curve. It depicts the relationship between the skill composition of labor demand and the relative labor market outcome ratios. Specifically, an elasticity of S tells us that a 1% increase in the ratio of low- to high-skilled labor, due to immigration, will
decrease the ratio of low- to high-skilled wages or employment levels by 1/S%. Or more formally:
𝑆 = – 𝑛$− 𝑛" 𝑂$− 𝑂" .
(1)
So, for example, assume the elasticity of substitution between low- and high-skilled workers equals 1.5 for European workers. In that case, an inflow of immigrants, increasing the relative labor supply of skilled workers by 3%, will decrease the wage difference between low-skilled and high-low-skilled workers by 4.5%. By the same rhetoric, wages can be substituted by employment to evaluate the impact of immigration on relative employment of low-skilled to high-skilled workers. Of course, in reality, this relationship is infinitely times more intricate. For instance, this model assumes a perfectly inelastic supply of labor. In reality, native workers will respond to the inflow of immigrants, moving to different industries, regions, or even in or out of the labor force. Yet, this simple model serves as a useful tool to think about the potential effect of immigration evaluated in the literature (World Bank, 2018).
The National Skill-Cell Approach
To estimate the impact of immigration on native workers, the national skill-cell approach categorizes native-born workers and immigrants into different skill groups. Instead of just differentiating between high- and low-skilled workers, this strand of literature tries to capture the complex interaction between many different worker characteristics. Educational
attainment, job tenure, and age structure are often appropriated characteristics. Then elasticities of substitution, being the inverse of the parameter of interest 𝜃89:;;, across skill
groups are estimated using variation over time at the national level. Finally, these papers typically compare the actual supply of workers in particular skill groups to those that would have prevailed in the absence of immigration. The papers, thus, simulate the change in labor market outcomes of native-born workers on the basis of estimates for the elasticity of substitution between skill groups. A typical estimation equation takes the following form2:
2 Elsner (2012), for instance, reports the following baseline estimating equation 𝑤
=>?: = 𝛿𝑚=>?+ 𝑋=>?: 𝛽 +
𝑦𝑒𝑎𝑟 + 𝑒𝑑𝑢𝑐 + 𝑒𝑥𝑝𝑒𝑟 + 𝜀=>?: to explore the effect of emigration on Lithuanian wages. He extends this baseline
∆𝑙𝑜𝑔𝑂;Q?= 𝜃89:;;∆𝐼𝑀;Q?+ ∆𝜋?+ (𝑠; × ∆𝜋?) + U𝑥Q × ∆𝜋?V + 𝜀;Q? , (2)
where ∆𝑙𝑜𝑔𝑂;Q? denotes the change in native wages or employment in education group 𝑙, experience group 𝑗 at time 𝑡, and ∆𝐼𝑀;Q? denotes the education-experience specific
immigration shock, defined as the difference in the ratio of immigrants to all labor in each education-experience group 𝑙𝑗 between two time periods. The variables 𝑠; , 𝑥Q , and 𝜋? are vectors of education, experience, and time fixed effects. The inclusion of time fixed effects in first differences absorbs the overall immigration shock. Any effects of immigration common to all education and experience groups are therefore differenced out. The education-time fixed effects capture, in addition to differential time trends by education unrelated to immigration, differences in immigration shocks across education groups. Any effects of immigration common to all experience groups within education groups are therefore likewise differenced out. The inclusion of experience-time fixed effects, in turn, soaks up the experience-specific immigration shock, in addition to allowing for differential time trends by experience unrelated to immigration. The parameter 𝜃89:;; therefore identifies the relative effect of immigration by
experience within a skill-cell (Dustmann et al., 2016).
These studies tend to find negative effects for natives in response to immigration3. Borjas
(2003) can be considered as a pioneer in using this approach. His analysis uses data drawn from the 1960-1990 U.S. Decennial Censuses, as well as the 1998-2001 Current Population Survey (CPS), and assumes that workers with the same education but different levels of work experience participate in a national labor market and are not perfect substitutes. Defining these skill groups in terms of educational attainment and work experience introduces a great deal of variation in the data. In some years, the influx of immigrants with a particular level of schooling mainly affects younger workers, in other years it mainly affects older workers. Borjas consistently indicates that immigration reduces the wage and labor supply of
competing native workers, as suggested by the simplest textbook model of a competitive labor market. Put differently, according to Borjas (2003), a 10% supply shock (i.e. an immigrant flow that increases the number of workers in the skill group by 10%) reduces weekly earnings by about 4%.
The appeal of the national skill-cell approach is the fact that it uses national level data. This acknowledges the idea that native workers respond to immigration-induced changes in outcomes, so that the immigrant/native ratio overstates the immigration-induced increase in supply in any locale obscuring the effect of immigration on native wages and employment (Borjas et al., 1996). In other words, in smaller geographic areas native workers might relocate due to the immigrant influx resulting in high levels of immigration accompanied by high employment levels. Only when the geographic area covered expands, these displaced native workers are included into the analysis, which depresses the overall employment level due to immigration.
However, the approach used in Borjas’ (2003) work cannot be considered to be free of criticism. Card (2012) might be one of Borjas’ most well-known critics. His comment is threefold. First, Borjas (2003) holds capital fixed. This assumption appears crucial. If
immigration increases aggregate labor supply by 10% (as it did in the United States between 1980 and 2000) and capital is fixed, average wages would be expected to fall by about 4%. If capital can adjust, however, the effect on average wages is approximately zero. Importantly, Ottaviano and Peri (2011) show that the trend in the US capital-labor ratio was similar in the 1980-2000 period as in earlier decades. This and other evidence, including the remarkable inflows of capital to the United States in the past decade, suggest that the assumption of fixed capital for analyzing the long-run effects of ongoing immigration inflows is unreasonable. Second, Borjas (2003) assumes there are four distinct education groups in the labor market (high school dropouts, high school graduates, people with some college, and college
graduates) whereas a long tradition in labor economics (e.g. Katz and Murphy, 1992) suggest to assume there are only two: high-school equivalents and college equivalents. The
delineation of high school dropouts as a separate skill group has extremely important implications for the potential effect of immigration on low-skilled natives in the United States. This is because US immigrants have about the same fraction of high school equivalents as the native population (63% versus 59%), but a much higher fraction of dropouts (31% versus 11%). In a four-education-group model the relatively high fraction of immigrant dropouts distorts the overall share of dropouts in the economy and lowers their wage relative to other groups. In a two-group model, however, what matters is the share of high school equivalent labor, which has been largely unaffected by immigration. Third, Borjas (2003) assumes that immigrant and native workers with the same education are perfect substitutes. Even a modest degree of imperfect substitutability makes a significant difference in the implied impacts of immigration on native wages. Once this imperfect substitution is taken into account, calculations suggest most native groups have actually gained from immigrant inflows (Ottaviano and Peri, 2011). Dustmann and Preston (2011) further explore this argument. They also argue that the perfect substitutability within skill-cells assumes that immigrants and natives can be assigned to education-experience cells, based on their observed characteristics even though immigrants typically downgrade upon arrival, as illustrated, for the case of the UK in a paper by Dustmann et al. (2008). Thus, immigrants may compete with natives at other parts of the skill distribution than where they have been assigned to, based on their observed characteristics. As a consequence, estimates of the elasticity of substitution between groups may be biased. Since this is a key parameter, it may significantly influence the results.
A Spatial Approach
effect of immigration on native labor market outcomes (Borjas et al., 1996). In many of this type of studies that exploit spatial variation in immigrant inflows, the log wage or
employment changes of natives in education group 𝑙 and experience group 𝑗 in region 𝑟 are related to the total region-specific immigration shock ∆𝐼𝑀X? (defined as the ratio of all
immigrants entering the region to all natives in that region), while controlling for nation-wide education-experience specific time trends (𝑠;Q × ∆𝜋?)4:
∆𝑙𝑜𝑔𝑂;QX?= 𝜃;Q8YZ?:Z;∆𝐼𝑀X?+ 𝑠;Q × ∆𝜋? + ∆𝜀;QX? . (3)
In the case of two time periods and two regions A and B, the parameter 𝜃;Q8YZ?:Z; equals a difference-in-difference estimator where differences are taken over time and across regions. Provided that region B, otherwise identical to region A, did not experience a certain influx of immigrants and is not indirectly affected by the immigration shock in region A (e.g. through emigration of natives), this parameter identifies the total effect of immigration on labor market outcomes of a particular skill group (Dustmann et al., 2016).
Studies exploiting the spatial approach often report modest estimates that fluctuate from negative to positive depending on the skill group that is studied5. Card’s (1990) seminal work
on the impact of the Mariel boatlift on the Miami labor market is a perfect example of the pure spatial approach. Card appropriates the sudden inflow of 125,000 Cuban immigrants in Miami. Their arrival was the consequence of an unlikely sequence of events culminating in Castro’s declaration on April 20, 1980, that Cubans wishing to emigrate to the United States were free to leave the port of Mariel. As a result 50% of the Mariel immigrants settled permanently in Miami. This resulted in a 7% increase of the Miami labor force, and the percentage increase in labor supply to less-skilled occupations and industries was even greater because most of the immigrants were relatively unskilled. He appropriated this influx to estimate wages for each non-Cuban worker in Miami, and sorted the sample from 1979 to 1985 into quartiles on the basis of predicted wage rates. A comparison between quartiles reveals that the Mariel influx appears to have had virtually no effect on the wages or
unemployment rates of less-skilled workers, even among Cubans who had immigrated earlier. This can be considered as rather surprising since an increase in the supply of labor is usually expected to have some significant effects on wages and especially employment rates.
An European case study regards the work of Dustmann et al. (2005) on the impact of immigration on the British labor market. Given the fact that British migration history,
settlement patterns, and the type of immigrants significantly differ from its US counterpart, it would be wrong to infer from other studies the possible effects of immigration on the British labor market. Using data from the British Labor Force Survey from 1983 to 2000, the authors also find no strong evidence that immigration has overall effects on aggregate employment, participation, unemployment and wages but they find some differences according to
education. An important reason is that the overall skill distribution of immigration is
4 Dustmann (2005) appropriates 𝑂
:?= 𝛼\+ 𝛼"𝜋:?+ 𝛼$𝑙𝑛𝑛:?+ 𝛼]𝑎:?+ 𝜆?_+ 𝜇:_+ 𝑢:?_ to evaluate the effect
of immigration on British labor market outcomes (employment, participation, and wages).
relatively similar to those of the native born workforce. In relation to Figure 2, this means that the post-immigration labor supply curve does not shift outwards, minimizing its effects on wages and employment.
An advantage of the area analyses is that it is straightforward to use an instrument to predict for immigration flows. Instruments are used since the correlation between immigrant inflows and labor market outcomes does likely not reflect a causal relationship. A valid instrument induces changes in immigrant flows but has no effect on native wages and employment. In practice, studies typically predict the current distribution of immigration flows by using the historical distribution of immigrants across local labor markets. The underlying justification is that immigrants will be attracted to settle where there are existing networks of individuals with the same cultural and linguistic backgrounds. These pre-existing immigrant
concentrations are unlikely to be correlated with current labor market outcomes if measured with a sufficient time lag, since existing concentrations are determined not by current labor market conditions, but by historical settlement patterns of previous immigrants (Dustmann et al., 2005)6. Moreover, since this approach does not attempt to distinguish between skill-cells,
it takes into account complementarities across skill-cells and across capital and labor.
Likewise, this approach is robust to the potential bias that is introduced by skill downgrading (Dustmann et al., 2016).
Yet, also the spatial approach applied by Card (1990) has its shortcomings. Lewis (2004), for instance, investigated two possible explanations for the absence of any wage and employment effects. First, the authors suggests the possibility that following the boatlift Miami increased its production of unskilled-intensive manufactured goods, allowing it to “export” the impact of the boatlift. Second, another explanation would be that Miami adapted to the boatlift by implementing new skill-complementary technologies more slowly than they otherwise would have. Using a confidential microdata version of the Annual Surveys of Manufactures, he shows that following the boatlift, Miami’s relative output of different manufacturing industries trended similarly to other cities with similar pre-boatlift trends in manufacturing mix. The response of industry mix to the boatlift therefore appears to be small. Supporting the second type of adjustment, utilization of Cuban labor by Miami’s industries rose
proportionately to the supply increase generated by the boatlift. Moreover, post-boatlift computer use at work was lower in Miami than in other cities with similar levels of computer-based employment before the event. This suggest that the boatlift induced Miami’s industries to employ more unskilled-intensive production technologies resulting in the absorption of most of the immigrant labor. Bodvarsson et al. (2008) explores a different possible reason. He argues that immigrants as consumers contribute to the demand for their services. Modelling an economy where workers spend their wages on a locally produced goods, the author tests this statement using Wacziarg’s (1999) Channel Transmission methodology. CPS data on workers in nine different labor markets and Survey of Buying Power data on retail spending
6 Dustmann et al. (2005) e.g. exploit three and four period lags to solve the potential problems introduced by
by consumers in Miami and four comparison cities reveal strong evidence that the Mariel Boatlift augmented labor demand.
Implications for the Research Question and Methodology
It is clear that both the spatial approach and the national skill-cell approach have its corresponding pros and cons. To uncover what the effect of immigration is on EU-28 employment, it is desirable to exploit the advantages of both approaches whilst avoiding the shortcomings. In that sense, it appears appropriate to deploy the spatial approach. This since the nature of the shortcomings in the national skill-cell approach appear worrisome. They are clear cases of misspecification. For instance, the potential bias that is introduced by skill-downgrading is the result of assigning immigrants and natives to a certain skill-cell based on worker characteristics. While it is true that these workers might have the same capabilities, it is unlikely that they are perfect substitutes. Consider for instance a Polish neurosurgeon that moves to the Netherlands. Although he might have all the necessary qualities to perform the same type of surgery in the Netherlands as he did in Poland, he might lack the necessary official documents or language skills for him to fit in the labor market. The result, the neurosurgeon now works as a nurse. In reality, he thus competes with workers at the lower end of the skill-distribution. Unfortunately, it is not possible to correct for this
misspecification. The spatial approach does not suffer from this type of misspecifications since it evaluates the total effect of immigration on employment.
I also do attempt to correct for shortcomings of the spatial approach by deviating from its typical research design. Firstly, the spatial approach usually focuses on smaller geographical areas (e.g. cities and municipalities). Advocates of the national skill-cell approach argue that this may result in an upwards biased estimate for the effect of immigration. Native workers can respond to the immigration-induced increase in labor supply, obscuring its effect on wages and employment in any locale. Expanding the geographic area under analysis allows for the inclusion of these potentially displaced workers. Thus, a larger geographic area might be necessary to uncover the true effect of immigration. To acknowledge this idea, I will assess the effect of immigration both on the NUTS-2 and NUTS-0 level. NUTS-2 regions represent basic regions for the application of regional policies, NUTS-0 regions are typically larger and coincide with national borders. Second, normally the pure spatial approach compares two regions A and B. In this case the parameter 𝜃8YZ?:Z; illustrates the causal effect of
immigration in region A, provided that region B, otherwise identical to region A, did not experience a certain influx of immigrants and is not indirectly affected by the immigration shock in region A. This assumption appears somewhat unrealistic. In my case there is no valid region B that can act as a counterfactual. I thus restrict the analysis to the effect of
3. Data
This section familiarizes the reader with the overall characteristics of my dataset. I will elaborate on the origin of the data, motivate my timespan and the areas under analysis.
Thereafter, the variables of interest are defined. Lastly, some graphical and summary statistics lay the foundation for the supporting hypotheses that are of interest besides the main research question.
Description of the Dataset
In order to answer the question what the effect of increasing immigrant is on employment in EU-28 countries, I appropriate data from the European Labor Force Survey (EULFS)
supplemented by Eurostat Population Data. The EULFS is constructed by Eurostat as of 1983. It contains data for all EU-28 member states as well as Iceland, Norway, Switzerland,
Montenegro, North Macedonia, Serbia, and Turkey. In general, data for individual countries are available depending on their accession date. The EULFS is conducted by the national statistical institutes across Europe and are centrally processed by Eurostat, making
harmonized data at the European level available. The EULFS is a large household sample survey providing yearly results on labor participation of people aged 15 and over as well as on persons outside the labor force. Persons carrying out obligatory military or community service are not included in the target group of the survey, as is also the case for persons in
institutions/collective households.
From this large household survey I focus on the timespan 2006-2018. This since 2006 is the first year for which immigration data is available for all EU-28 member states. 2018 is the most recent year for which data on immigration is available. During these years the amount of immigrants rose from 19.8 million to 28.9 million throughout the EU-28. This results in a 46% increase in the number of immigrants and a rise in the immigrants share from 6.0% to 8.8%. This should be considered as sizeable and might be sufficient to produce interesting results.
The areas under analysis regard all EU-28 member states. Although Eurostat also provides data on four candidate countries and three EFTA members, I restrict the analysis to member states. This allows my results to be interpreted as a policy recommendation at the EU-level. I will evaluate the effect of immigration on employment levels at the NUTS-2 and the NUTS-0 level. The NUTS classification (Nomenclature of territorial units for statistics) is a
Variables
In the empirical analysis, I focus on 𝐸𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡, defined as the proportion of the
population with the age 15-64 that is employed. This variable is available both at the NUTS-2 and NUTS-0 level. Normally, the literature isolates the effect of immigration on the native population. The amount of native working persons would thus be scaled by the native population with the age 15-64. However, the EULFS does not allow me to filter immigrants out of the employment rate. This has some implications for the interpretability of my results. Consider the positive effect of an inflow of immigrants on 𝐸𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡. It implies we experience an increase in the overall employment rate. This means that there is a possibility that all immigrants find jobs while some native workers are displaced resulting in a net increase in the employment rate. Likewise, it could be that following immigration more natives find themselves among the employed while immigrants fail to find jobs resulting in a net increase in employment. In other words, it is not possible to say anything about the potential re-sorting effect of immigration. Besides that, it would clearly also be interesting to evaluate the effect of immigration on other labor market outcomes (e.g. monthly earnings, yearly earnings). Unfortunately, the EULFS does not contain the necessary data. Although this somewhat limits the scope of this thesis, it does not take away the relevancy of the current research question.
I attempt to explain variation in 𝐸𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 by including the explanatory variables 𝐼𝑚𝑚𝑖𝑔𝑟𝑎𝑡𝑖𝑜𝑛, 𝐴𝑔𝑒, 𝑎𝑛𝑑 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛. The single most important explanatory variable is 𝐼𝑚𝑚𝑖𝑔𝑟𝑎𝑡𝑖𝑜𝑛, made up by the amount of immigrants in region 𝑟 scaled by the population with the age 15-64 in that particular area. This variable is available at the NUTS-2 and NUTS-0 level. Besides that, I include 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 and 𝐴𝑔𝑒 data at the national level (due to limited availability of the data) to control for nation-wide education-experience specific trends. The 𝐴𝑔𝑒 variable consists out of the amount of people in the age group 15-29 and 30-64 scaled by the population of age 15-30-64. 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 consists out of number of persons having gained a certain educational attainment (low/medium/high) as a ratio of the population of age 15-64. Clearly, these three variables are not the only factors explaining variation in 𝐸𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡. For instance, ambition and a person’s proximity to companies might
significantly influence employment opportunities. However, I decided to restrict my analysis to these widely adopted explanatory variables in labor economics. Other factors appeared difficult to construct, being faced with severe data limitations.
potential disapproval of the research proposal, it does not seem feasible to gain access to such microdata.
Data Characteristics
Before introducing the supporting hypotheses it is instructive to be informed about the general relation between employment and immigration. More specifically, the association between the proportion of the population with the age 15-64 that is employed and the amount of
immigrants in region 𝑟 scaled by the population with the age 15-64 in that particular area at varying levels of aggregation. From the simple scatterplots in Figure 3 one can see that this association appears rather similar at the NUTS-0 and NUTS-2 level: there is great variation in immigration and employment. There appears a modest positive correlation between the two variables. Moreover, high immigration ratios are concentrated within a few regions. Table 1 lists the top five regions housing the highest shares of immigrants. As can be seen, the highest values for immigration originate from Luxembourg (where the NUTS-0 and NUTS-2 level completely overlap). Besides that, the NUTS-2 level immigration ratios almost always supersede those at the NUTS-0 level. Meaning that even within countries immigration tends to be concentrated within a few regions.
2018
NUTS-0 NUTS-2
Region Immigration Region Immigration
Luxembourg 0.514 Luxembourg (LX) 0.514
Cyprus 0.194 Mayotte (FR) 0.477
Malta 0.190 Guyane (FR) 0.360
Austria 0.176 Région de
Bruxelles-Capitale (BG) 0.356
Ireland 0.157 Wien (AU) 0.324
.5 .6 .7 .8 E m pl oym ent 0 .1 .2 .3 .4 .5 Immigration .3 .4 .5 .6 .7 .8 E m pl oym ent 0 .1 .2 .3 .4 .5 Immigration
Table 1: The five regions hosting the highest concentration of immigration at varying levels of aggregation.
NUTS-0 NUTS-2
In general one might thus expect an overall positive relation between immigration and employment. However, this is not necessarily the case. The factor proportions approach has taught us that it is not the concentration of immigrants in certain regions that necessarily influences employment levels. An inflow of immigrants that exactly mirrors the skill
distribution of natives is expected to leave the relative labor supply unchanged. Resulting in no effect on labor market outcomes (see Figure 2). It is those immigrant inflows that
significantly alter the skill-distribution within a particular region that are expected to have an effect on employment levels. As mentioned before, I cannot explicitly differentiate between immigrants and natives and their accompanied skills or characteristics. The only available information is on the origin of immigrants (EU-28 or non-EU-28) and their sex (male or female). I can, however, based on the mean values of several samples within the dataset, hypothesize on the expected effect of immigration on certain population groups.
Table 2 displays summary statistics of the population and immigrants based on their
corresponding available characteristics. The summary statistics are presented at the NUTS-2 and NUTS-0 level. The time interval spans from 2006 to 2018 for all displayed variables. As expected, mean variable values do not differ much with a varying level of aggregation. Small discrepancies may be due to different weights in the aggregation process. In general, these seem negligible.
NUTS-2 NUTS-0
Mean deviation Standard Mean deviation Standard
Age Age 15-29 to population .273 .022 .278 .024 Age 30-64 to population .725 .023 .721 .024 Education Low education to population .284 .107 .277 .120 Medium education to population .455 .110 .471 .125 High education to population .250 .069 .246 .074 Number of observations 3,459 364
As mentioned, a concern of not appropriating national level data, is the possibility of
obscuring the potential negative effect of immigration on employment. It is argued that as the geographic area covered expands, the coefficient on the change in immigrant/native ratios tends to become less positive or more negative. A way to control for this would be estimating the regression analysis at several levels of aggregation. Thus, I will run all regressions at the NUTS-2 and NUTS-0 level.
Hypothesis 1: expanding the geographic area under investigation from NUTS-2 to NUTS-0 will result in a less positive or more negative effect of immigration on employment.
Besides that, it is interesting to infer whether the potential effect of immigration is equally distributed amongst education levels. As depicted in Table 2, education levels seem to follow somewhat of an inversed U-curve among the employed. Meaning that employment is highest among those who enjoyed medium education and employment falls with lower or higher education levels. Unfortunately, I do not have any information on the educational attainment of immigrants. However, it is often argued that immigrants generally have lower education levels. Following the reasoning of the factor proportions approach, one would expect that the lower education groups are less positively or more negatively affected by immigration than those who enjoyed higher education.
Hypothesis 2: employment levels for the lower education groups are less positively or more negatively affected by immigration than the higher education group.
Moreover, the effect brought about by immigration is likely to depend on the immigrant’s characteristics. The EULFS allows me to differentiate between the origin of immigrants. Although the EU has strong immigration restrictions for individuals who live outside the EU, there are slightly more non-EU-28 immigrants (4.5%) compared to EU-28 immigrants (4.1%). It is likely that EU-28 immigrants follow a skill-distribution that more closely resembles that of the total population than non-EU-28 immigrants. Therefore, one would expect EU-28
immigrants to have a more positive or less negative effect on employment than non-EU-28 immigrants7.
Hypothesis 3: EU-28 immigrants are expected to have a less negative or more positive effect on employment levels compared to non-EU-28 immigrants.
4. Methodology
This section elaborates on the applied methodological strategy. Problems arising in the empirical estimation are discussed, and how to appropriately address them. This results in the evaluation of the OLS estimator, a difference estimator, and the IV estimator in differences. It is explained why the IV estimator in differences is likely to produce most robust results. Ordinary Least Squares
OLS acts as a useful starting point in exploring the desired methodological setting. It provides a basic and straightforward point of reference. With OLS, the effect of immigration on
employment can interpreted as the period-by-period cross-sectional correlation between immigration shares and employment levels. This would result in the following regression equation:
𝐸𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡=8X?= 𝛽\+ 𝛽"8YZ?:Z;𝐼𝑚𝑚𝑖𝑔𝑟𝑎𝑡𝑖𝑜𝑛cX?+ 𝛽$𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛8?+ 𝛽]𝐴𝑔𝑒:?+ 𝜀=8X? , (4)
where 𝐸𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡=8X? is the amount of employed people in the age group 15-64 years relative to the population aged 15-64 years of sex 𝑔, educational attainment 𝑠
(low/medium/high), in region 𝑟, at time 𝑡. 𝐼𝑚𝑚𝑖𝑔𝑟𝑎𝑡𝑖𝑜𝑛cX? reports the ratio of immigrants to
the population in the age group 15-64 years, of origin 𝑜 (Total immigrants; EU-28
immigrants; non-EU-28 immigrants) in region 𝑟, at time 𝑡. 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛8? equals the amount of people in a certain education group (low/medium/high) as a ratio of the population at time 𝑡. Similarly, 𝐴𝑔𝑒:? is the amount of people in an age group (15-29/30-64) to the population at time 𝑡. 𝛽"8YZ?:Z; is the parameter of interest, representing the total effect of immigration on employment. This equation is evaluated at the NUTS-0 and NUTS-2 level while both 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛8? and 𝐴𝑔𝑒:? are only reported at the NUTS-0 level.
Although hailed for its simplicity, the OLS estimator 𝛽"8YZ?:Z; can only be considered as the best linear unbiased estimator if the following conditions are fulfilled: the expected value for the error term is zero for all observations (𝐸(𝜀:) = 0), the errors are homoskedastic (𝐸(𝜀:$ =
𝜎f$) = 𝑐𝑜𝑛𝑠𝑡𝑎𝑛𝑡) for all observations, the error terms are uncorrelated (𝐶𝑜𝑣U𝜀
:, 𝜀QV =
7 One could also argue that EU-28 immigrants are probably closer substitutes to native workers, having a more
0 𝑤ℎ𝑒𝑟𝑒 𝑖 ≠ 𝑗), and the explanatory variables are uncorrelated with the error term
(𝐶𝑜𝑣(𝑥:, 𝜀:) = 0). In other words, the Gauss Markov theorem has to be met. However, since I am dealing with time series data, it is likely that this setting suffers from endogeneity. In other words, there is correlation between the explanatory variable and the error term. There may be unobserved region and time fixed effects. This can be problematic since these time and region fixed effects are attributed to the included explanatory variable resulting in a biased estimator. That is to say, the regression specification suffers from an omitted variable bias.
First-Differences
The first-difference approach addresses the issue of omitted region fixed effects by taking simple differences. Any effects of immigration common to all regions are therefore
differenced out. Time fixed effects are accounted for by including a full set of year dummies. The dummies absorb any time fixed effects that would otherwise have been wrongfully attributed to the explanatory variable. This results in the following regression equation:
∆𝐸𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡=8X?= 𝛽"8YZ?:Z;∆𝐼𝑚𝑚𝑖𝑔𝑟𝑎𝑡𝑖𝑜𝑛cX?+ 𝛽$∆𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛8?+ 𝛽]∆𝐴𝑔𝑒:?+ 𝜷𝟒𝑌𝑒𝑎𝑟?+ 𝜀=8X? , (5)
where ∆ denotes the change of a particular variable from time period 𝑡 − 1 to time period 𝑡 and 𝑌𝑒𝑎𝑟? is a vector of time dummies. Notwithstanding the fact that this regression form is
more robust, it is still possible that it suffers from simultaneity. It is clear that immigrants may affect employment levels. However, high employment rates may also attract the attention of immigrants. Thus, the flow of causality may run both ways.
Instrumental Variable
A method to deal with simultaneity is the instrumental variable approach. As described earlier, one advantage of the spatial approach is that it is straightforward to use an instrument to predict for immigration flows. Namely, historical immigration flows. This instrument induces changes in contemporary immigration but has little effect on current employment. In my case, I will appropriate four period lags of immigration flows to control for the potential problems introduced by simultaneity8. In practice this instrument solves the issue of
simultaneity by means of a two stage least squares regression. In the first stage I use the four period lag of immigration flows to predict current immigration flows.
∆𝐼𝑚𝑚𝚤𝑔𝑟𝑎𝑡𝚤𝑜𝑛o cX?= 𝛽p"∆𝐼𝑚𝑚𝑖𝑔𝑟𝑎𝑡𝑖𝑜𝑛cX?qr+ 𝛽p$∆𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛8?+ 𝛽p]∆𝐴𝑔𝑒:?+ 𝜷s𝟒𝑌𝑒𝑎𝑟?+ 𝜀cX? . (6)
In the second stage the predicted values for current immigration flows are used to estimate employment levels. Since the four period lags for immigration flows are assumed to be exogenous, the predicted values for the current immigration flows will be uncorrelated with
8 It has to be noted that different lags produce vastly different second stage results. However, it is only the four
the error term. This allows for the consistent estimation of the effect of immigration on
employment levels. More formally, this results in the follow second stage regression equation:
∆𝐸𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡=8X?= 𝛽"8YZ?:Z;∆𝐼𝑚𝑚𝚤𝑔𝑟𝑎𝑡𝚤𝑜𝑛o cX?+ 𝛽$∆𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛8?+ 𝛽]∆𝐴𝑔𝑒:?+ 𝜷𝟒𝑌𝑒𝑎𝑟?+ 𝜀=8X? . (7)
Combining the first-difference approach with my desired instrument will thus produce most robust results. Eicher Huber White covariance estimators are used to control for the existence of heteroskedasticity9.
5. Results
Table 3 presents a series of estimates of the overall effect of immigration on employment. I reported results using the OLS estimator, a difference estimator, and the IV estimator in differences. The OLS estimator reveals that a 1 percentage point increase in immigration raises employment levels by 0.03 percentage points at the NUTS-2 level. As expected, this estimate depresses as the geographic area under investigation expands. However, this effect is far from significantly different from zero at conventional levels. Since this regression
specification likely suffers from an omitted variable bias it would be wrong to derive any strong conclusions from the results.
The first difference approach supplemented with year dummies addresses the omitted variable bias. Immigration is now associated with a decrease in employment at the NUTS-2 level. Apparently, the time trends and region characteristics disguise the negative within region correlation between immigration and employment. This negative relation disappears at the NUTS-0 level as we see an inflated estimate for immigration, although this result can be considered as insignificant. A change of the estimation strategy thus has a significant impact on the size and sign of the estimates. Besides that, we now perceive that a larger geographic area is associated with a more positive effect from immigration on employment. This largely contradicts theory and appears intricate to explain. However, as this specification is still prone to simultaneity, one has to be cautious with the interpretation of the results.
The instrumental variable approach attempts to solve this simultaneity issue. The results reveal that immigration has a modest positive effect on employment at the NUTS-2 level. The negative estimate produced by the first-difference approach may thus be driven by correlation between current immigration data and the error term10. At the NUTS-0 level, the positive
effect of immigration becomes more pronounced. Again, this contradicts theory as a larger
9 A Breusch-Pagan test reveals that the data is indeed subject heteroskedasticity. I also tried to control for
potential clusters within the standard errors. However, this leads to insignificant results.
10 It could also be the case that the positive IV-estimate is driven by its deviating sample composition. Since I
t statistics in parentheses * p < .10, ** p < .05, *** p < .01
geographic area under analysis produces more positive estimates. I also report the first stage F-statistic (Table 3) and the first stage regression results (Appendix, Table 7). This reveals that four year lagged immigration data is indeed an efficient instrument. As expected, there is a positive and significant relation between current immigration and historical immigration data. The IV-results can, thus, be considered as most robust and its corresponding results form the basis for my conclusions.
From Table 3 we would infer that overall immigration is expected to have a positive effect on employment levels at both levels of aggregation. However, since it is widely believed that immigrants have lower educational attainment, I would expect that those who enjoyed lower
OLS First Difference IV
Level of
aggregation: NUTS - 2 NUTS - 0 NUTS - 2 NUTS - 0 NUTS - 2 NUTS - 0 Dependent
variable: Employment Employment Employment Employment Employment Employment
Immigration 0.0342* -0.0584 -0.0781*** 0.0262 0.0980* 0.633** (1.78) (-1.55) (-5.43) (0.86) (1.71) (2.39) Age 15-29 2.545*** 1.287* -2.715*** -1.711*** -2.716*** -2.761*** (13.26) (1.83) (-4.35) (-3.25) (-3.37) (-4.74) Age 30-64 1.723*** 1.110 -2.841*** -1.198** -2.628*** -2.851*** (9.56) (1.59) (-4.78) (-2.31) (-3.41) (-3.82) Low education -1.043*** -0.579** -0.0950 -0.154 0.199 1.542** (-10.71) (-2.57) (-0.59) (-0.92) (0.71) (2.21) Medium education -0.687*** -0.470** 0.216 -0.131 0.606*** 1.624** (-7.09) (-2.11) (1.40) (-0.80) (2.66) (2.19) High education -0.824*** -0.196 0.505*** 0.325* 0.839*** 1.720*** (-7.85) (-0.83) (2.83) (1.85) (3.47) (2.95) Constant -0.452** -0.0677 (-2.56) (-0.10) N 2676 319 2459 275 1970 194 𝑅$ 0.38 0.23 0.11 0.36 - -
Region f.e. No No Yes Yes Yes Yes
Time f.e. No No Yes Yes Yes Yes
Robust s.e. No No No No Yes Yes
First stage
F-statistic - - - - 120.74 10.88
education would be disproportionally hurt from an immigrant influx. Table 4 compares the effects of immigration on employment among varying levels of education. At the NUTS-2 level results meet expectations. Immigration is associated with a decrease in employment for those with lower and medium education, while those with high education benefit. At the NUTS-0 level the exact opposite seems to be the case. Immigration is now associated with an increase in employment for lower and medium education levels while those who enjoyed high education suffer. In this case, estimating the relationship between immigration and
employment at a higher level of aggregation does not result in inflated estimates but in a clear switch in signs and its corresponding conclusion.
t statistics in parentheses * p < .10, ** p < .05, *** p < .01
Besides that, it is interesting to see whether the overall effect of immigration is mostly driven by EU-28 immigrants or non-EU-28 immigrants. In general, one would expect EU-28
immigrants to follow more closely the skill-distribution of natives compared to non-EU-28 immigrants. Thus, I would expect the association between EU-28 immigrants and
employment to be more positive or less negative than that between non-EU-28 immigration and employment. Table 5 depicts the effect of EU-28 immigration versus non-EU-28
IV
Level of
Aggregation: NUTS-2 NUTS-0
Dependent variable: Low education employment Medium education employment High education employment Low education employment Medium education employment High education employment Immigration -0.222*** -0.268*** 0.581*** 0.262** 0.430*** -0.0551** (-4.34) (-4.66) (7.00) (2.45) (2.64) (-2.02) Age 15-29 0.0410 2.533*** -5.727*** -1.073*** -2.051*** -0.0800 (0.06) (5.39) (-6.88) (-4.30) (-5.74) (-0.68) Age 30-64 0.562 2.671*** -6.194*** -1.125*** -2.115*** -0.0350 (0.92) (6.86) (-9.12) (-3.51) (-4.57) (-0.27) Low education 0.150 -0.821*** 1.436*** 1.289*** 1.040** -0.130* (0.56) (-3.53) (2.92) (4.57) (2.38) (-1.72) Medium education -0.311 0.132 1.352*** 0.656** 1.759*** -0.138* (-1.27) (0.66) (2.99) (2.19) (3.79) (-1.75) High education 0.0136 -0.228 1.650*** 0.676*** 0.947*** 0.766*** (0.05) (-1.03) (3.34) (2.91) (2.60) (11.97) N 1970 1970 1970 236 236 236
Region f.e. Yes Yes Yes Yes Yes Yes
Time f.e. Yes Yes Yes Yes Yes Yes
Robust s.e. Yes Yes Yes Yes Yes Yes
First stage
F-statistic 120.74 120.74 120.74 10.88 10.88 10.88
immigration on employment. In this case I appropriate a four period lag on EU-28
immigration data to predict current EU-28 immigration, and a four period lag on non-EU-28 immigration data to predict current non-EU-28 immigration data. Again, first stage results can be found in the Appendix (Table 8). Unfortunately, the results in Table 5 lack statistical significance. At the NUTS-2 level we do perceive that EU-28 immigration has a more positive effect on immigration than non-EU-28 immigration. At the NUTS-0 level this same line of reasoning holds. However, again, we see inflated estimates which refutes theory.
t statistics in parentheses * p < .10, ** p < .05, *** p < .01
6. Discussion and Conclusion
This thesis provides a first analysis on how immigration affects employment in EU-28 countries. EULFS and Eurostat Population Data from 2006 to 2018 is combined to estimate this relation. The spatial approach provides the basis for the methodological strategy. To control for the potential issues introduced by an omitted variable bias and simultaneity, I use the IV approach in-differences including a full set of year dummies. This allows me to produce most robust results considering the given data limitations. If anything, the results explain why the literature largely fails to reach a consensus on the effect of immigration on receiving countries. Results appear to be significantly influenced by the level of aggregation at which the relation is estimated, and which (sub)section of the population is considered.
IV
Level of
Aggregation: NUTS-2 NUTS-0
Dependent
variable: Employment Employment Employment Employment EU-28 Immigration 0.0196 2.239 (0.13) (1.19) Non-EU-28 immigration -0.00667 0.244 (-0.02) (0.91) Age 15-29 -1.863 -2.492*** -1.798*** -5.318 (-1.14) (-3.42) (-4.72) (-1.62) Age 30-64 -1.867 -2.784*** -1.244*** -5.098 (-1.16) (-3.78) (-3.28) (-1.47) Low education 0.119 0.184 -0.190 4.209 (0.36) (0.58) (-1.01) (1.13) Medium education 0.373 0.515* -0.152 4.450 (1.46) (1.88) (-0.87) (1.12) High education 0.559 0.722*** 0.274 3.936 (1.51) (2.89) (1.36) (1.27) N 2313 2064 315 215 F-statistic 11.65 70.58 24.56 1.56
Firstly, the effect of immigration on total employment is examined. It appears that overall, immigration has a positive effect on employment. It has to be noted that since I evaluate the effect of immigration on employment among the total population, it is impossible to state anything about a potential re-sorting effect. An increase in immigration is associated with a rise in overall employment levels, yet it is unknown whether this is caused by the native population reaching a higher level of employment or most immigrants succeed in findings jobs (or a mixture). This positive relation between immigration and employment can be found at varying levels of aggregation: at NUTS-2 and NUTS-0 level. It is suggested that as the geographic area under analysis expands, the effect of immigration on employment depresses. This since a larger geographic area allows for the inclusion of potential relocations of
displaced workers. So, one would expect that the effect of immigration on employment is more positive at the NUTS-2 level than at the NUTS-0 level. However, the results imply the exact opposite. Theory provides no explanation for this finding.
Secondly, it is assessed how the effect of immigration is distributed among several (sub)sections of the population. The data allows me to differentiate the population on the basis of three education levels: low, medium, or high. Again, this relation is estimated at the NUTS-2 and NUTS-0 level. Since it is argued that immigrants generally have lower education levels, it follows from the factor proportions approach that the lower education groups are less positively or more negatively affected by immigration than the higher education group. At the NUTS-2 level, results coincide with theory. Those who enjoyed lower or medium education are hurt by immigration, while those who have higher educational attainment benefit. The effect of immigration thus appears to be unevenly distributed among education levels. At the NUTS-0 level one would expect the same conclusion but with depressed estimates. However, the results show a clear sign switch. Now immigration is associated with increasing
employment for those who enjoyed lower or medium education and those who attained higher education are hurt. Again, a puzzling finding for which theory provides no answer. It does nonetheless illustrate that conclusions regarding the effect of immigration hinge upon the level of aggregation and the (sub)section of the population that is considered.
Thirdly, it is reviewed whether immigrant characteristics influence the effect of immigration on employment. The data allows for differentiation between EU-28 and non-EU-28
immigrants. Thus, the effect of EU-28 immigration versus non-EU-28 immigration on employment is evaluated at the NUTS-2 and NUTS-0 level. One could argue that it is likely that EU-28 immigrants follow a skill-distribution that more closely resembles that of the native population compared to non-EU-28 immigrants. Therefore, it is expected that EU-28 immigrants have a more positive or less negative effect on employment than non-EU-28 immigrants. The results indicate that this seems to be the case. This idea is even more pronounced at the NUTS-0 level compared to the NUTS-2 level. Yet, the results lack the desired statistical significance. Hence, it would be inappropriate to base any conclusions on these results.
about a potential re-sorting effect. The results do suggest that the current migration policy has been beneficial in achieving high employment levels. From an economic point of view, this finding is rather interesting since higher employment levels can contribute to reaching the full potential of an economy. Yet, digging deep into the relation between immigration and
employment uncovers that depending on the level of aggregation the results would plea for compensating differing (sub)sections of the population (if one feels that those hurt by immigration should be compensated). Immigration is not beneficial to anyone. This partly explains the discontent that seems to be growing regarding increasing immigrant inflows. As a result, EU policymakers have to tread a fine line between achieving high levels of
employment and the concerns of the native population.
Having gained the insight that the level of aggregation at which the relation between immigration and employment is estimated and which (sub)section of the population is
considered significantly influences my results, it would be interesting to reconsider the results from previous research. For instance, consider the results from the likes of Card (1990), although his results consistently report that immigration had virtually no effect on the wages or unemployment rates of less-skilled workers, one might have reached vastly different conclusions had the estimation taken place at a higher level of aggregation. Similarly, if one considered the effect of immigration on a different skill group than less-skilled workers, results might have taken a different form. My thesis uncovers the complexity of the relation between immigration and labor market outcomes. This leaves one wondering if
acknowledging this complexity in previous research would yield conclusions similar to those in this thesis. If so, it would help shape our understanding of the effects brought about by immigration.
Besides that, for this specific research it would be interesting to gain access to the EULFS microdata. This would allow for the exploration of the effect of immigration on the native population. Resulting in a more clear cut policy recommendation regarding migration. In addition, it would be interesting to see what the effect of immigration is on other labor market outcomes (e.g. weekly earnings, yearly earnings). This would paint a more complete picture on the effect of immigration. For instance, can we consider immigration to be desirable if it boosts employment levels yet leads to significant wage decreases that outweighs the positive employment effect? Lastly, the Age and Education variables are constructed at the national level. Having access to NUTS-2 level data on these matters would elevate the estimation precision.
7. References
Aydemir, A., & Borjas, G. J. (2007). Cross-Country Variation in the Impact of International Migration: Canada, Mexico, and the United States. Journal of the European Economic
Association, 5(4), 663–708. https://doi.org/10.1162/jeea.2007.5.4.663
Bodvarsson, Ö. B., Van den Berg, H. F., & Lewer, J. J. (2008). Measuring immigration’s effects on labor demand: A reexamination of the Mariel Boatlift. Labour Economics, 15(4), 560–574. https://doi.org/10.1016/j.labeco.2008.04.003
Borjas, G. J. (2003). The Labor Demand Curve is Downward Sloping: Reexamining the Impact of Immigration on the Labor Market. The Quarterly Journal of Economics, 118(4), 1335–1374. https://doi.org/10.1162/003355303322552810
Bouton, L., Paul, S., & Tiongson, E. R. (2011). The Impact of Emigration on Source Country Wages: Evidence from the Republic of Moldova. Policy Research Working Papers.
https://doi.org/10.1596/1813-9450-5764
Card, D. (1990). The Impact of the Mariel Boatlift on the Miami Labor Market. ILR
Review, 43(2), 245–257. https://doi.org/10.1177/001979399004300205
Card, D. (2012). Comment: The Elusive Search For Negative Wage Impacts Of Immigration. Journal of the European Economic Association, 10(1), 211–215. https://doi.org/10.1111/j.1542-4774.2011.01057.x
Clemens, M. A. (2011). Economics and Emigration: Trillion-Dollar Bills on the Sidewalk? Journal of Economic Perspectives, 25(3), 83–106.
https://doi.org/10.1257/jep.25.3.83
Clemens, M. A., Montenegro, C. E., & Pritchett, L. (2009). The Place Premium: Wage Differences for Identical Workers Across the U.S. Border. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.1211427
Dustmann, C., Glitz, A., & Frattini, T. (2008). The labour market impact of immigration. Oxford Review of Economic Policy, 24(3), 477–494.
https://doi.org/10.1093/oxrep/grn024
Dustmann, Christian, Fabbri, F., & Preston, I. (2005). The Impact of Immigration on the British Labour Market. The Economic Journal, 115(507), F324–F341.
https://doi.org/10.1111/j.1468-0297.2005.01038.x
Dustmann, Christian, Frattini, T., & Rosso, A. (2015). The Effect of Emigration from Poland on Polish Wages. The Scandinavian Journal of Economics, 117(2), 522–564.
https://doi.org/10.1111/sjoe.12102
Dustmann, Christian, Schönberg, U., & Stuhler, J. (2016). The Impact of Immigration: Why Do Studies Reach Such Different Results? Journal of Economic Perspectives, 30(4), 31–56. https://doi.org/10.1257/jep.30.4.31
Elsner, B. (2012). Does emigration benefit the stayers? Evidence from EU enlargement. Journal of Population Economics, 26(2), 531–553.
https://doi.org/10.1007/s00148-012-0452-6
Foged, M., & Peri, G. (2016). Immigrants’ Effect on Native Workers: New Analysis on Longitudinal Data. American Economic Journal: Applied Economics, 8(2), 1–34. https://doi.org/10.1257/app.20150114
Goodwin, M., & Milazzo, C. (2017). Taking back control? Investigating the role of immigration in the 2016 vote for Brexit. The British Journal of Politics and International
Katz, L. F., & Murphy, K. M. (1992). Changes in Relative Wages, 1963-1987: Supply and Demand Factors. The Quarterly Journal of Economics, 107(1), 35–78.
https://doi.org/10.2307/2118323
Lewis, E. G. (2004). How Did the Miami Labor Market Absorb the Mariel Immigrants? SSRN
Electronic Journal. https://doi.org/10.2139/ssrn.502504
McKenzie, D., Stillman, S., & Gibson, J. (2010). How Important is Selection? Experimental VS. Non-Experimental Measures of the Income Gains from Migration. Journal of the
European Economic Association, 8(4), 913–945.
https://doi.org/10.1111/j.1542-4774.2010.tb00544.x
Ottaviano, G. I. P., & Peri, G. (2011). Rethinking the Effect of Immigration on Wages. Journal of the European Economic Association, 10(1), 152–197. https://doi.org/10.1111/j.1542-4774.2011.01052.x
Rodrik, D. (2018). Populism and the economics of globalization. Journal of International
Business Policy, 1(1–2), 12–33. https://doi.org/10.1057/s42214-018-0001-4
Wacziarg, R. (1999). Measuring the Dynamic Gains from Trade. Policy Research Working
Papers. https://doi.org/10.1596/1813-9450-2001
World Bank. (2018). Moving for Prosperity: Global Migration and Labor Markets. Policy
Research Rapport. https://doi.org/10.1596/978-1-4648-1281-1
Zaslove, A. (2004). Closing the door? The ideology and impact of radical right populism on immigration policy in Austria and Italy. Journal of Political Ideologies, 9(1), 99–118. https://doi.org/10.1080/1356931032000167490
8. Appendix
NUTS-0 NUTS-2
Austria Abruzzo Martinique
Belgium Åland Mayotte
Bulgaria Alentejo Mazowiecki regionalny
Croatia Algarve Mecklenburg-Vorpommern
Cyprus Alsace Mellersta Norrland
Czechia Anatoliki Makedonia, Thraki Merseyside
Denmark Andalucía Midi-Pyrénées
Estonia Aquitaine Midtjylland
Finland Aragón Mittelfranken
France Área Metropolitana de Lisboa Molise
Germany Arnsberg Moravskoslezsko
Greece Attiki Münster
Hungary Auvergne Niederbayern
Ireland Basilicata Niederösterreich
Italy Basse-Normandie Noord-Brabant
Latvia Bedfordshire and Hertfordshire Noord-Holland Lithuania Berkshire, Buckinghamshire and Oxfordshire Nord-Est
Luxembourg Berlin Nord-Est
Malta Border, Midland and Western Nord-Pas-de-Calais
Poland Brandenburg Nordjylland
Portugal Bratislavský kraj Norra Mellansverige
Romania Braunschweig Norte
Slovakia Bremen North Eastern Scotland
Slovenia Bretagne North Yorkshire
Spain Bucuresti - Ilfov Northern and Western
Sweden Budapest Northern Ireland (UK)
United Kingdom Burgenland (AT) Northumberland and Tyne and Wear
Calabria Notio Aigaio
Campania Nyugat-Dunántúl
Canarias (ES) Oberbayern
Cantabria Oberfranken
Castilla y León Oberösterreich
Castilla-la Mancha Oberpfalz
Cataluña Opolskie
Centre - Val de Loire Östra Mellansverige Centro (PT) Outer London - East and North East
Centru Outer London - South
Champagne-Ardenne Outer London - West and North West
Chemnitz Overijssel
Cheshire Övre Norrland
Ciudad Autónoma de Ceuta (ES) País Vasco Ciudad Autónoma de Melilla (ES) Pays-de-la-Loire
Comunidad de Madrid Peloponnisos
Comunidad Foral de Navarra Pest
Comunidad Valenciana Picardie
Cornwall and Isles of Scilly Piemonte
Corse Podkarpackie
Cumbria Podlaskie
Darmstadt Pohjois- ja Itä-Suomi
Dél-Alföld Poitou-Charentes
Dél-Dunántúl Pomorskie
Derbyshire and Nottinghamshire Praha
Detmold Principado de Asturias
Devon Prov. Antwerpen
Dolnoslaskie Prov. Brabant wallon Dorset and Somerset Prov. Hainaut
Drenthe Prov. Liège
Dresden Prov. Limburg (BE)
Düsseldorf Prov. Luxembourg (BE)
Dytiki Ellada Prov. Namur
Dytiki Makedonia Prov. Oost-Vlaanderen East Anglia Prov. Vlaams-Brabant East Wales Prov. West-Vlaanderen East Yorkshire and Northern
Lincolnshire Provence-Alpes-Côte d'Azur Eastern and Midland Provincia Autonoma di Bolzano/Bozen
Eastern Scotland Provincia Autonoma di Trento
Eesti Puglia
