Emigration and Source Country Wages:
Evidence from Lithuania in the Post-Crisis Years
Master ThesisMSc International Economics and Business, University of Groningen MA International Economy and Business, Corvinus University of Budapest
Abstract
The consistent emigration from Lithuania provides an opportunity to investigate how emigration influences the wages of non-migrants after the financial crisis. In this paper, I apply the skill cell technique by merging work permit and census data from Ireland and the UK with survey data from Lithuania. This approach shows that a one percent increase of the skill cell emigration rate corresponds to an average increase in the real wage of 1.3% between 2011-2016. Furthermore, this paper found different results for gender after applying a large set of controls and model specifications. The finding is only robust for women and is therefore at odds with the results of previous research.
Keywords Emigration, Sourcing country wages, European Union labour mobility
18-06-2018 Roy J.E. Germain
R.J.E.Germain@student.rug.nl S2610957
University of Groningen, Faculty of Economics and Business Supervisor: Dr. Milena Nikolova
1 I Introduction
The enlargement of the European Union (EU) in 2004 led to a large shift of labour across Europe. As a result, labourers emigrated from the new Eastern European member states and went to countries in Western Europe. For example, due to emigration, Lithuania’s workforce diminished by approximately nine percent in three years (Elsner, 2013a). Similarly, Poland’s workforce decreased with two percent due to migration in the years 1998-2007 (Dustmann, Frattini, & Rosso, 2015). Although most of these migrants were temporary, substantial amounts of Lithuanians emigrated permanently.1
The effects of migration have already been researched with respect to the impact of immigration on the receiving countries’ labour market, concerning: wages, employment and welfare benefits. However, there is another side of the coin: emigration can have detrimental impacts on the sourcing countries’ labour market in terms of wages and employment. Moreover, it has been found that migration flows have an uneven distribution in terms of educational attainment and age. The outcome in the sourcing labour market will therefore be heterogeneous, depending on the respective socio-economic group. High relative emigration rates will create the so-called ‘winners and losers’ of increased liberalization and globalization of labour markets, especially within a European perspective.
This paper investigates the effects of emigration on the source country wages. More specifically, it will investigate the effect of migration from Lithuania to Ireland and the UK in the post-crisis years (2011-2016). The method heavily relies on Elsner’s (2013a) approach but investigates the post-crisis and post-enlargement years and uses a similar dataset. Similar to Elsner, this paper uses the Irish census as a measure of the skill distribution of emigrated Lithuanians to both Ireland and the UK. Unlike Elsner, the wages of ‘those left behind’ in Lithuania are taken from the Income and Living Condition Survey (ILCS) of the Lithuanian statistical office instead of the Household Budget Survey (HBS). The advantage of the ILCS is that it is measured annually and therefore allows to capture more variation than the HBS of Lithuania, which is compiled every four years.
1 Since 1990, Lithuania has permanently lost 18% of its residents due to emigration. Recently, there has been a resurge in
2
The aforementioned skill distribution of Lithuanians in Ireland is used to disaggregate the emigration flows per skill cell—gender, education and experience—which subsequently calculated emigration rates for the respective skill cells in Lithuania. Using the skill cell approach, this paper found that wages, on average, increased by 1.3% for a one percent increase in the emigration rate in the period 2011-2016 for women. This result is robust under a large set of controls which account for, among others, personal characteristics, different returns to education and experience, geography, time, and labour demand factors. The relationship of male wages and emigration is not significant in any of the models.
The rest of this paper is structured as follows: section II deals with the literature, section III describes the data, the methodology is shown in section IV and the results which follow are in section V. Finally, some conclusions are drawn in section VI.
II Literature review
Emigration and wages in origin countries
Immigration is investigated extensively in terms of the effects for the receiving countries but the effect on sourcing countries with respect to wages and income is left relatively unexplored (Clemens, 2011; Kerr & Kerr, 2011).2 It is clear to many that the net global gains from reducing labour mobility barriers are large. However, the literature is not precise on the exact figure as it is highly dependent on the assumptions made and, thus, can be different on a case-by-case basis (Clemens, 2011; Moses & Letnes, 2004).3
One could think of several reasons why it has been left relatively open. First, net emigration countries are often economically less developed than net immigration countries. Therefore, the economic impact of migration from these countries may be limited due to the economic position of these countries in the world. Second, the data needed to properly assess the effect on sourcing countries are difficult to find, since most countries do not keep detailed statistics on the characteristics of individuals who emigrate. The data, if available at all, are usually aggregated and therefore fail to provide complete statistics on the individual level. However, more recently, there were several attempts to quantify the effect of emigration on the sourcing economy. More precisely,
2 Notable exceptions included contributions to large “brain drain/gain” literature. This will be discussed later on in the literature
review. For a review of the literature on immigration, consult appendix A2.
3 Another crucial element is whether the researcher is a strong opponent or advocate of international labour migration. For
3
the effect of emigration on wages of stayers has been investigated for Mexico, Lithuania and Poland.
Initially, research started with the effects of Mexican emigration to the US on Mexican natives’ wages. Most notably, it is Hanson (2003) who is one of the first to document the relationship between emigration and the wage structure in the origin. First, Hanson analyses the effects of trade and foreign direct investment (FDI) liberalization on the domestic wage structure of Mexico and finds that this has led to an increase in the skill premium for higher skilled workers. Second, and more relevant, are the effects of the North American Free Trade Agreement (NAFTA) which also included a liberalization of migration between members. According to Hanson, regions which have higher rates of migration to the US experienced stronger wage growth than lower-migration regions. This wage growth was driven by a combination of exposure to lower-migration, trade and FDI with the US.4 Additionally, Hanson finds that higher skilled workers in the origin are more likely to benefit from higher wages due to a more liberalized economy and the emigration that follows.
Re-using a regional approach, Hanson (2005) investigates the labour market outcomes for Mexican workers between 1990 and 2000 for both high and low emigration states. The data show that there was a positive self-selection of migrants who have high earnings potential (i.e. education; since the returns for education are higher in the US than in Mexico). This confirms the earlier findings of Hanson (2003).5 In high emigration states, wages increased between six and nine
percent in 1990-2000 relative to low emigration states. This is the first empirical research which exclusively investigated the effects of emigration on the wages in the sourcing country at the regional level.
On an aggregate level, Mishra (2007) looks at the effects of emigration on national wages in Mexico for the period 1970-2000. Apart from the aggregation, Mishra also incorporates differences between schooling and experience groups in the labour force of the origin (Borjas, 2003). This is interesting since earlier research indicated that, for Mexico, there are significantly different outcomes for the likelihood to emigrate and the wage structure of different skill levels on a regional level for Mexico (Hanson, 2003; Hanson, 2005). Mishra reports similar findings as the
4 Therefore, clearly exposing the influence of NAFTA on Mexican wages by investigating both trade and FDI at the same time. 5 Using different data sources, Fernández-Huertas (2011) found negative self-selection for emigration from Mexico to the US and
4
aforementioned literature has found for Mexico: the outflow of workers in 1970-2000 increased the wages of the average Mexican worker by eight percent. More important is the result that more educated/experienced workers have higher average earnings due to emigration than less educated/experienced workers. Again, this seems to confirm the fact that there are different effects for educational attainment and work experience.
Similarly, research by Aydemir and Borjas (2007) quantifies the effect that migration (both immigration and emigration) can have on the labour supply and, therefore, wages. The authors state that the migration flows reduced wage dispersion in Canada but was diverging wages in the US and Mexico. The motive is that the composition of the migration in terms of skill distribution had a detrimental effect on the labour supply and wages. Emigration was examined only for Mexico, but the results are in line with what the above-mentioned literature has found: wages increased relatively for the middle of the skill distribution, whereas it decreased for the lower skilled. This is due to the distribution of skills of emigrants from Mexico, as it is heavily skewed towards high school graduates and ‘some college education’ (Aydemir & Borjas, 2007). Quantitatively, Aydemir and Borjas have found that a ten percent shift in labour supply (due to immigration) leads to a three percent to four percent decrease in wages. Similar elasticity is found for emigration from Mexico, but this differs on educational attainment. Nevertheless, wages increased between five and approximately ten percent depending on the assumptions made and the distinction between short and long term.
The enlargement of the EU in 2004 led to some interesting new research on the effects of emigration on wages using this natural experiment. Prior to the enlargement, it was expected that there will be ‘massive’ inflows of labour from the east. Therefore, the EU allowed old member states to enforce labour mobility restrictions up until 2011 (Galgóczi, Leschke & Watt, 2011).6 The
new Eastern European member states7 experienced a large outflow of workers to a select group of countries, which allowed researchers to quantify the effects thereof.
Most recent is the research of Dustmann, Frattini and Rosso (2015) who measure the effect of emigration on Polish wages. Dustmann et al. use the extensive Polish Labour Force Survey, which allows the measurement of the effects of emigration on individuals within the household and
6 i.e. Transitional provisions. See Fic et al. (2011) for a detailed study on EU enlargement, labour mobility and the effect of
transitional restrictions.
5
their respective wages.8 In short, the authors measure the changes in wages for households of which
one of the household members was working abroad in the 1998-2007 period. Their main results indicate that average wages increased in the 1998-2007 period but, specifically, it increased wages for middle and higher skilled workers with respect to lower skilled workers. Again, the data show that the emigration of Polish workers consisted mostly of medium skilled workers, and the econometric results reflect this in terms of wage increases. These findings are consistent with what has been found in the previous literature (Hanson, 2003; Hanson, 2005; Mishra, 2007).
Another country which experienced massive outflows of labour after the enlargement is Lithuania. As with Dustmann et al. (2015) the enlargement of the EU allows researchers to conduct a natural experiment, due to the exogenous regulatory shock of increased labour mobility. Most relevant to this paper is Elsner (2013a) (and to lesser extent, Elsner, 2013b), who investigates the impact of emigration using the same approach as Mishra (2007) but uses the UK and Ireland as the main destination countries for Lithuanians.9 Using the aforementioned skill group clusters— education and work experience—Elsner (2013a) combines several data sources in order to construct the emigration flows from Lithuania to Ireland and the UK. The results are clear: a 10% increase in emigration leads to a 6.6% increase in average wages. However, differentiating between men and women reveals that it is only significant for men. The underlying reason behind the different results for gender are not exposed by Elsner. However, there are some indications that the explanation lies within the emigration flows. Additionally, the domestic labour market could also bear some answers to this question. Moreover, these findings remain robust after using a large amount of control variables which include (among others): time, geography, education and experience.
Summarizing, the effects of emigration on sourcing countries’ wages in the short run is clear: emigration leads to higher wages. Additionally, increases in wages were higher in groups (i.e. education, age, gender, and experience) which experienced the largest relative outflows. These findings hold on household, regional and national levels.
8 That is, those who are ‘left behind’ in the household with respect to emigration. The impact on these household members goes
via several spheres (i.e. social, economic and demographical) and follows several causal lines and is therefore hard to quantify. For a general overview of the implications for the left behind, consult Démurger (2015).
9 Due to the transitional restrictions, most emigrants went to the UK and Ireland. At the same time, Spain, Portugal, and Norway
6
Brain gain
On the one hand, there are arguments which point to the beneficial effects of the emigration of high educated individuals; it increases investment in human capital by the non-migrants via education. Alternatively, the outward migration of highly educated individuals poses as a loss of human capital for the sourcing country. Hence, on average, it is possible that the average level of education in the sourcing country increases. This is due to the fact that demand for high skilled labour increases after the most educated leave. Thus, it can be beneficial for the economy to ‘export’ human capital (Beine, Docquier, & Rapoport, 2001; Stark, Helmenstein, & Prskawetz, 1997). Beine et al. quantify the theoretical conditions for brain gain to occur and find that these settings are closely linked to reality. Additional benefits can arise via the so-called positive network externalities which link receiving and sourcing countries via migration (Docquier & Rapoport, 2012).
More related to this paper are Mayr and Peri (2009), who investigate the brain gain effect specifically for intra EU migration with focus on migration from eastern to western Europe. Using similar emigrations rates after the fall of the Berlin wall, Mayr and Peri estimate that emigration increased average schooling with approximately one year in Eastern Europe. Additionally, Mayr and Peri put forth the argument that workers can return to their sourcing country which mitigates the negative effects of emigration.10 Return migration has been qualitatively investigated by Tung and Lazarova (2006), who point out that returning emigrants have small, if none at all, problems with readjusting to the sourcing country, thereby confirming the aforementioned argument of Mayr and Peri.
Explicit research regarding brain drain for Lithuania is so far limited to Kazlauskienė and Rinkevičius (2006), who investigate push and pull factors with respect to skilled emigration. In short, Kazlauskienė and Rinkevičius found that pull factors dominate the push factors for Lithuania.11 For example, the socio-economic status of those who emigrated in their dataset was self-characterised as satisfactory which, combined with the recent economic development, supports
10 Additionally, it is often the case that returning migrants bring new skills upon return and thereby further mitigating the negative
effect of emigration. It is also argued that some countries are learning centres where it is easier for individuals to obtain skills, either on job or at the educational institutions (Dustmann, Fadlon, & Weiss, 2011). Dustmann et al., also show that brain gain is dependent on the transferability of foreign experience to the origin country. This transferability influences the likelihood of return migration. See also Dustmann and Weiss (2007) on other motives for return migration.
11 Push factors can be any negative factor that is present in the sourcing economy. Pull factors are variables which attract
7
the notion of dominant pull factors. In other words, although opportunities in Lithuania grew over time, the possibilities abroad are simply better.12
Finally, research on remittances is also relevant to this research. The positive impact of remittances on those ‘left behind’ is the fact that it increases the income and therefore eases the budget constraint vis-à-vis the monetary inflows (Niimi, Ozden, & Schiff, 2010). This effect has been denominated as the income effect. Alternatively, household members might be persuaded to reduce their labour supply or shift towards unpaid activities using the remittances—the substitution effect. There is some evidence which points towards the disruptive effects of remittances on sourcing country labour supply. Niimi et al. find, in the case of Mexico, that males shift from formal to informal employment, but their labour supply remains constant. Women, however, do reduce their labour supply in response to remittances. This confirms earlier findings of Amuedo-Dorantes and Pozo (2006), who find similar results with respect to labour supply in response to remittances in Mexico.13
Theory on the effects of emigration on wages in the origin country
Emigration has a clear theoretical implication for the origin country’s labour market: it is a reduction in the labour supply. From a basic labour market model, we can expect that the labour supply curve will retract all else equal (Boeri & Van Ours, 2013). In a perfect competitive labour market, wages will adjust vis-à-vis the reduction in labour supply.14
The response of wages to changes in labour supply depend on the elasticity of labour demand. If labour demand is elastic, decreases in labour supply will lead to a relative modest increase in wage but a sharp fall in employment. Conversely, inelastic labour demand will lead to large increases in wages and a modest decrease in employment. Both cases are shown in figure 1: the first panel shows elastic labour demand and the second panel displays inelastic labour demand. The response of wages is exactly what the literature on both immigration and emigration tries to model: the changes of wages with respect to a shock in labour supply.
12 For example, the economic situation is no longer a major push factor for emigration (The Lithuania Tribune, 2017).
Alternatively, certain psychological factors can influence the emigration decision of the individual. Lithuanian emigrants are found to be more risk taking and open to experience than non-emigrants (Seibokaite, Endriulaitiene, & Marksaityte, 2009)
13 Comparable effects with respect to labour supply—in the origin—are found for other countries; see Binzel & Assaad (2011)
for Egypt, Rodriguez and Tiongson (2001) for the Philippines, Jadotte (2009) for Haiti. An overview of the literature is provided by Adams (2011). There is an ample amount of literature devoted to identifying the factors which have an influence on the amount of remittances sent (Dustmann & Mestres, 2010; Niimi, Ozden, & Schiff, 2010).
8
Figure 1 Supply shock to the labour market
Notes: The figure displays the competitive aggregate labour market in response to a contraction of labour supply. Elastic labour demand is shown in the first panel, whereas inelastic labour demand is shown in the second panel. Source: author’s own calculations
Based on the literature, there is another important factor which should be taken into account regarding the response of wages. As the aforementioned theory is based on aggregate labour demand and supply, it should be noted that there are differences on wages depending on the skill composition of that group (Aydemir & Borjas, 2007; Hanson, 2005; Mishra, 2007). Basically, it is important to take into account the relative distribution of skills in the sourcing country. The composition of the emigration flow is therefore detrimental for the effect on wages, as this can change the distribution of skills in a country.15 If outflows decrease the relative size of the higher educated, they decrease the labour supply—of that respective skill group—and therefore lead to increases in wage. Conversely, other groups will increase in relative size and can experience wage decreases.16 This result follows from the empirical finding that emigration leads to an unequal distribution of wage increases, skewed relatively towards higher educated individuals.
Finally, the important factor of complementarity and substitution should be taken into account. In general, it has been found that highly-skilled immigrants are complements for native labourers, whereas lowly-skilled immigrants are imperfect substitutes. From this, it is also expected that high skilled emigration will lead to wage increases, in the origin, for highly-skilled jobs, but to wage decreases for lower-skilled occupations due to complementarity (Borjas, 2014; Zou & Pieretti, 2007). As a result, we can expect that there will be different results based on skill complementarity between skill groups in the origin, depending on the composition of emigration.
9
Emigration choice
Theoretically, when an individual moves from one country to another, a very basic labour migration model predicts the inverse shift of the labour supply curve from the sending to the receiving country. 17 Using a modified methodological approach from the immigration research, it is possible to investigate the effects on the sending country.
The literature highlights the distinction between why and where people migrate. These two decisions are imperative in explaining the composition of the emigration flows and their respective destinations. In addition, combining the decision making of the individual with the sourcing labour market helps in formulating the expected results on wages of each group.
The selection of migrants is dealt with using the Roy (1951) model. The model is extended in Borjas (2014) and assumes that earnings in origin and destination depend on one single factor and that labour is perfectly mobile between origin and destination. Using this assumption, the wages are modelled follows:
𝐿𝑜𝑔𝑊𝑘1 = 𝑎1+ 𝑟1𝑠, (1.1)
𝐿𝑜𝑔𝑊𝑘2 = 𝑎2+ 𝑟2𝑠. (1.2)
Where wages are given by a constant α, a skill s and a return to skills r in the two countries k1 and k2. Basically, since 𝑟1 ≠ 𝑟2 and 𝑎1 ≠ 𝑎2 both wage curves will have different constants and
coefficients. For example, if 𝑟1 > 𝑟2 and 𝑎1 < 𝑎2, there will be a point s* where 𝐿𝑜𝑔𝑊𝑘1 > 𝐿𝑜𝑔𝑊𝑘2 , which will lead to emigration of individuals with 𝑠 > 𝑠∗. This type of selection is
denoted as positive selection, since higher educated individuals will emigrate. Negative selection, on the other hand, occurs when the foreign wage curve is above the home wage curve for the area below s* and, consequently, lower educated emigration will occur. Figure 2 graphically displays both negative and positive selection of immigrants.
17 That is, if immigration leads to lower wages in a group with characteristics Ω in the receiving country, it can be expected that it
10
Figure 2 Selection of migrants
Notes: The graph display positive and negative selection of migrants depending on the slope and constant of the wage curves. Source: Borjas (2014, p. 18)
Also relevant is the alternative specification of the Roy (1951) model by Grogger and Hanson (2011) which is discussed in Borjas (2014). This model is used to describe the phenomenon that emigration flows, in general, have above average education. Following Borjas (2014), wages are now given as:
𝐿𝑜𝑔𝑊𝑘𝑢 = 𝑎
𝑘, (2.1)
𝐿𝑜𝑔𝑊𝑘𝑠 = 𝑎𝑘+ 𝑟𝑘, (2.2)
where s refers to skilled labour and u to unskilled labour for country k and r is again the skill premium. The main difference with equation (1) is the difference between skilled and unskilled labour: unskilled labour does not gain a skill premium from migration. Introducing fixed cost of emigration Chk (from home h to country k) and individuals maximize utility𝑈ℎ𝑘𝑗 for skill group j (unskilled or skilled):
11
where β > 0 and the worker maximizes its utility by evaluating all possible destinations and incorporates its own skill and costs. Using the log odds of emigrating from home h to country k for skill group j: 𝑙𝑜𝑔 𝑝ℎ𝑘 𝑗 1 − 𝑝ℎ𝑘𝑗 = 𝛽(𝑊𝑘 𝑗− 𝑊 ℎ𝑗) − 𝛽𝐶ℎ𝑘, (2.4)
where 𝑝ℎ𝑘𝑗 is the fraction of the workforce in skill group j which emigrates from h to country k. From this, it follows that group j is more likely to migrate if the difference between the wages in country k and home are larger—given that costs do not change. The final step is to introduce the difference between skilled and unskilled wages and their respective emigration probabilities:
𝑙𝑜𝑔 𝑝ℎ𝑘
𝑠
1 − 𝑝ℎ𝑘𝑠 − 𝑙𝑜𝑔
𝑝ℎ𝑘𝑢
1 − 𝑝ℎ𝑘𝑢 = 𝛽[(𝑊𝑘𝑠− 𝑊ℎ𝑠) − (𝑊𝑘𝑢− 𝑊ℎ𝑢)]. (2.5)
Since costs are assumed to be equal across skill groups, it does not appear in equation (2.5). The equation shows that positive selection occurs as long as the absolute difference between skilled (𝑊𝑘𝑠− 𝑊
ℎ𝑠) and unskilled (𝑊𝑘𝑢− 𝑊ℎ𝑢) labour is positive. In other words, positive selection
occurs when the difference between home and foreign skilled wages is larger than the difference between home and foreign unskilled wages. This result is interesting and relevant since it predicts that emigration will be more educated even if 𝑟𝑘 ≤ 𝑟ℎ. The absolute wage difference ensures that (𝑊𝑘𝑠− 𝑊
𝑘𝑢) > (𝑊ℎ𝑠− 𝑊ℎ𝑢), which basically states that if the wage gap between foreign and home
is large enough, return to education can be lower in foreign than in home country.
From equation (2.5), individuals are more likely to emigrate if the wages are higher in the foreign country with respect to home wages and the costs of emigrating are small (i.e. the enlargement of the EU which is used in Dustmann et al. (2015) and Elsner (2013a). Although this model does not give a prediction about the age of migrants, it is evident from the life cycle model that younger migrants are more likely to migrate due to a longer working life.18
18In short, the life cycle model calculates the net present value of emigration, using a personal discount rate, differences between
12
The theoretical model and the literature allow for several testable hypotheses. Firstly, the aforementioned model predicts that emigration flows from Lithuania to Western Europe will consists of relatively higher educated individuals. Additionally, the model states that the emigration will consist of relatively young individuals. Secondly, the effects on labour market are straightforward: higher emigration rates per skill group should lead to higher wages in that respective group in the sourcing country (Aydemir & Borjas, 2007; Dustmann et al., 2015; Elsner, 2013a, 2013b; Hanson, 2003; Hanson, 2005; Mishra, 2007).19
III Data
To analyse the effect of emigration on the origin countries wages, a dataset with information on both emigrants and the left behind would be ideal. Unfortunately, this information is not readily available.20 Therefore, this paper relies—and builds—on the methods of Mishra (2007) and Elsner (2013a): merging individual level data from Lithuania with aggregate data from Ireland and the UK as destination countries. More specifically, the data are closely aligned with those in Elsner’s since this concerns the same sending and receiving countries but a different period.
The following data sources are used: The Lithuanian ILCS is used to calculate the skill cells and wages in Lithuania. Census data from Ireland is used to construct the skill cells of Lithuanian’s abroad. Work permit data from both the UK and Ireland are used to determine the emigration rate from Lithuania.
Lithuanian Income and Living Conditions survey
The ILCS is an annual national representative survey, which is available at both the individual and household level based on face-to-face interviews. It contains information on: income (gross and net), age, education, gender, region, and more individual characteristics related to work and housing.21 The variable cash income from employment gives the (gross) annual labour income of the individual. There earnings are deflated by the Harmonized Consumption Price Index from
19Additionally, there may be difference in terms of skill complementarity between emigrants and the left behind. Strong
complementarity between the two may increase the socio-economic effects of emigration in the origin country (Docquier, Özden, & Peri, 2010; Zou & Pieretti, 2007). See also the discussion of the immigration literature in appendix A2 which also covers complementarity—and substitution—between immigrants and natives.
20 With the exception of Dustmann et al. (2015), which uses the Polish Labour Force survey which, until recently, disclosed
whether one of the household members were (working) abroad. Unfortunately, this data is no longer available in the most recent surveys.
21 Unfortunately, the survey does not report information on whether a household member was living abroad. Also, it is not
13
Eurostat, and the income from 2011-2014 is converted to Euros using the pegged rate of ERM II.22
The ILCS, which is used as the main dataset, is summarized in table 1 below.
Table 1 Summary table Lithuanian Income and Living Condition Survey 2011-2016
2011 % 2012 % 2013 % 2014 % 2015 % 2016 % Men 5780 46.3 5833 46.0 5416 46.1 5479 46.1 5011 45.5 4920 45.1 Women 6712 53.7 6844 54.0 6340 53.9 6419 54.0 6004 54.5 5985 54.9 100 100 100 100 100 100 Lower secondary Average 17.9 16.4 16.6 13.0 13.2 12.7 Men 20.4 19.3 19.3 14.2 14.2 13.5 Women 15.8 14.0 14.4 11.9 12.4 12.0 Upper secondary Average 56.6 56.6 55.6 59.6 59.6 59.9 Men 58.0 58.2 57.6 62.8 62.6 62.5 Women 55.3 55.2 53.9 56.7 57.0 57.7 Third-level Average 25.5 27.0 27.7 27.5 27.2 27.4 Men 21.6 22.6 23.2 22.9 23.2 24.0 Women 28.9 30.8 31.8 31.4 30.6 30.3 100 100 100 100 100 100 Age <20 17.8 16.8 15.8 15.0 16.6 17.2 20-29 9.7 9.6 8.9 9.2 8.4 8.0 30-39 8.9 8.5 8.1 7.6 7.8 8.3 40-49 15.1 14.7 14.0 13.7 14.6 14.5 50< 48.4 50.4 53.1 54.5 52.6 52.1 100 100 100 100 100 100 Real Annual earnings Average € 5508 € 5722 € 5997 € 6368 € 6613 € 7179 Men € 5638 € 5906 € 6348 € 6862 € 7194 € 7773 Women € 5392 € 5546 € 5660 € 5892 € 6047 € 6624
Notes: The table shows summary statistics of the ILCS. Education groups: lower secondary (10 years or less schooling), upper secondary (more than 10 years but did not finish 3rd level), third-level (at least 15 years of schooling and a BSc. equivalent). With
respect to education, the first column gives average attainment of the whole dataset that year (i.e. 18% of the Lithuanian workforce had a lower secondary degree in 2011) whereas 20% of the male workforce had a lower secondary degree in 2011. Age distribution concerns the whole dataset. The real annual individual labour earnings are deflated by HCPI and converted to Euro using the pegged ERM-II rate. Source: author’s calculations based on the ILCS
Table 1 excludes individuals who have zero or negative income. Since gross cash income from employment is used as annual (individual) labour earnings, those who are unemployed are excluded in the real annual earnings statistic. Additionally, it concerns earnings before taxes and
22 I.e. 3,4528, the pegged rate against the Euro. Lithuania became a member of the EMU on the first of January 2015. Thus, the
14
transfers from the government and excludes household transfers such as remittances.23 The
educational attainment is changing over time: on average, Lithuanians became more educated. The percentage of people with a lower secondary education decreased in relative percentages whereas upper and third-level education increased.24 The distribution of age shows the aging demographics of Lithuania as it shifts from towards older strata.
Real annual earnings are computed as real mean income for both genders in 2011-2016 using gross annual income from employment. Table 1 shows that, on average, income increased substantially in the period 2011-2016 for both genders. The gap between genders increased relatively in 2011 and 2016. Nevertheless, the statistics show a significant25 increase in income between 2011 and 2016, taking into account that these figures might be underreported. 26
Irish Census
The Irish census is used to determine the skill distribution of Lithuanians residing in Ireland and the UK. The Minnesota Population Center (2018) has provided a microsample of the 2011 census, which will be used to calculate the emigration shares for each respective skill cell.27
Table 2 reveals that the Lithuanian population in Ireland has a relatively similar distribution of gender. The distribution of educational attainment is slightly different as lower secondary scores decreases, whereas upper secondary attainment increases with respect to the origin. Also, the data confirms that emigrants were on average younger than stayers. These descriptive results are in line with what the theory predicts; migration favours younger cohorts and the higher educated.28 The distribution of migrants according to gender, education and work experience in Ireland will be used
23 Ex post the financial crisis, remittances accounted for approximately 3.5% of GDP. See the Migration in Numbers project for a
detailed overview.
24 A simple test of significance reveals that average education between 2011 and 2016 increased: as the differences are significant
at the 1% level for all three educational levels. This has three potential explanations i) older, lower educated workers left the workforce; ii) the methodology of the survey allowed for more detailed educational attainment from the 2014 survey onwards which led to an improved identification of educational attainment; iii) enrolment and completion rates increased over time.
For example; the gross graduation rate of individuals with tertiary education increased for both genders over time. Secondary gross enrolment increased primarily for males in 2010-2016. See UNESCO data for the Sustainable Development Goals Lithuania (2018) for more detailed data on educational attainment and completion rates. Clearly, there are several factors at play that can explain this phenomenon but identifying the specific factor goes beyond the scope of this paper and is left for future research.
25 The differences for income are significant: intermediate differences between the years 2011-2016 are significant at the 1%
level, whereas both men and women experienced significant wage increases between 2011 and 2016 at the 1% level.
26 A brief comparison between the computed average annual income and the monthly earnings by the statistical office reveals that
the data is indeed underreported. However, the data from the statistical office includes individual enterprises and average earnings include more than only employee income (as in the ICLS).
27 At the time of writing, the 2016 census has not been made public yet.
15
to calculate the composition of emigration flows from Lithuania to the UK and Ireland. The number of emigrants will be taken from the work permit data which will be dealt with in the next section.
Table 2 IPUMS 10% micro sample of Lithuanians in Ireland 2011
Observations % Gender Men 1604 46.9 Women 1818 53.1 Education Lower secondary 422 15.5 Upper secondary 1629 60.0 Third-Level 666 24.5 Age <20 574 16.8 20-29 1152 33.7 30-39 962 28.1 40-49 434 12.7 50< 300 8.8
Notes: The table shows summary statistics for the relevant variables from the Irish census. The micro sample contains 3422 observations which reflects a total of approximately 34220 Lithuanians in Ireland in 2011. Education groups: lower secondary (10 years or less schooling), upper secondary (more than 10 years but did not finish 3rd level), third-level (at least 15 years of schooling
and a BSc. equivalent). Percentages of education and age distribution are aggregated at average levels. Source: IPUMS and author’s own calculations
Work permit data
The work permit data are taken from both the UK and Ireland in order to construct the most realistic measure of the emigration from Lithuania. Quantitatively, the amount of work permits issued to Lithuanians abroad can be viewed in figure 3. The work permit data measure how many individuals obtain an insurance number to legally work in Ireland or the UK. These numbers are issued once in a lifetime and therefore only count first-time emigration to the UK and Ireland from Lithuania.29 The advantage of using these data is that they only account for labour immigration and thereby exclude those who are not in the workforce. Unfortunately, the work permit data do not provide
29 Circular migration between Lithuania and Ireland and the UK is not counted in the data. Additionally, the work permit data
16
any characteristics about the individual who applied for a permit. Nevertheless, the data show that emigration from Lithuania to the UK and Ireland decreased substantially after 2010.30
The validity of the work permit data is checked via the European Migration Network (EMN) which provides data on migration to and from Lithuania via the Migration in Numbers (2018), a joint project of six institutions.31 The EMN data confirm the use of the work permit data as a proxy for emigration from Lithuania.32
Figure 3 Work permit data from the UK and Ireland
Notes: Lithuanian immigration to the UK and Ireland by work permits issued. Irish data is taken from the PPSN allocation in Ireland and the UK data is taken from the National insurance number (NINo) allocation. Sources: Central Statistics Office Ireland, Department of Work and Pensions UK and author’s own calculations.
30 Emigration to Ireland has been decreasing prior to 2010. Maximum annual emigration flows were recorded around 2008-2009
and rapidly decreasing thereafter. Probable cause is the extent to which Ireland and the UK were affected by the financial crisis both in speed and severity. Alternatively, the transitional provisions were lifted for all ‘old member states’ in 2011, as Germany and Austria were the final members to remove restrictions for Lithuanian labour. This led to increasing numbers of Lithuanians migrating to countries other than the UK and Ireland ex post the financial crisis. See appendix A4 for detailed description of emigration from Lithuania in the period 2010-2016.
31 Ministry of the Interior, Statistics Lithuania, Migration department Lithuania, Lithuanian Labour Exchange, European
Migration Network (EMN) and the International Organization for Migration (IOM) Vilnius Office.
32In summary: the table shows the declining emigration to both the UK and Ireland but both countries remain the main
destinations for emigrants from Lithuania. Relatively, the UK receives 46% of the emigration and Ireland 9% in the period 2010-2016. Germany and Norway became more important destinations recently: both attracting 8% of the migration flows. The complete table on emigration from Lithuania in the years 2010-2016 can be found in appendix A4.
17
Emigration share
In order to obtain the emigration per skill cell, several datasets need to be combined. The skill distribution of Lithuanian emigrants can only be retrieved from the Irish census. The number of migrants is taken from the aforementioned work permit data from both Ireland and the UK. The work permit data cover approximately 55% of the emigration from Lithuania, as the UK and Ireland are the two main destinations. Ideally, one would use the skill distribution of both Ireland and the UK separately but the UK census is not publicly available. Nevertheless, using the Irish census as an approximation for Lithuanians in the UK is justified, since migrants in both countries have similar skill distributions (Elsner, 2013b).
Because the surveys do not directly report work experience, the following variable is used as a proxy:
𝑒𝑥𝑝𝑒𝑟𝑖𝑒𝑛𝑐𝑒 = 𝑎𝑔𝑒 – 𝑒𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 – 6. (3)
Where the years for education are 10 years for lower secondary, 12 for upper secondary, and 15 years for third-level.33
The number of emigrants to Ireland and the UK is calculated as follows
𝐸𝑔ℎ𝑥𝑡 = 𝐼
𝑔ℎ𝑥𝑡 (𝑁𝐼𝑁𝑂𝑡+ 𝑃𝑃𝑆𝑁𝑡). (4)
Where 𝐸𝑔ℎ𝑥𝑡 is the number of emigrants from Lithuania at time t (2011-2016) for gender (g), education (h), and experience (x) cell. 𝐼𝑔ℎ𝑥𝑡 refers to the fraction of the Lithuanians in the Irish census at time t for gender (g), education (h), and experience (x) cell. Education is grouped at the aforementioned lower secondary, upper secondary and third level34; experience is grouped in five- year intervals. This creates 48 cells (2 genders x 3 education groups x 8 experience groups).35 NINOt and PPSNt are the number of work permits issued to Lithuanians at time t—the former referring to work permits in the UK and the latter to Ireland.
33 This approach is standard in the literature that uses or discusses the skill cell approach or labour economics in general (Borjas,
2003; Elsner, 2013a; Hanson, 2005; Mishra, 2007).
34As a reminder for the reader: lower secondary (10 years or less schooling), upper secondary (more than 10 years but did not
finish 3rd level), third-level (at least 15 years of schooling and a BSc. equivalent).
35Further differentiating between educational attainments would be optimal, but this is currently limited by the 2011 ILCS (survey
18
The emigration rate per skill cell and year is calculated by dividing the emigrated Lithuanians to Ireland and the UK by the Lithuanians at the origin in that respective skill group:
𝑒𝑔ℎ𝑥𝑡 = 𝐸𝑔ℎ𝑥𝑡
∑ 𝜑𝑡 𝑔ℎ𝑥𝑖𝑡 𝑖
. (5)
Where 𝑒𝑔ℎ𝑥𝑡 is the emigration rate per skill group and year (t = 2011-2016) which is constructed as
the emigrants per skill cell (𝐸𝑔ℎ𝑥𝑡 ) divided by the total population in that respective skill cell in
Lithuania (∑𝑡𝑖𝜑𝑔ℎ𝑥𝑖𝑡 ). In this case, the population is constructed as the sum of all the individual weights of workers i in the ILCS at time t for skill group ghx. For an overview of the skill cells in Lithuania over time, consult appendix A5.
IV Methodology
I estimate a basic labour supply and demand model that features a constant downward sloping labour demand curve. To investigate the effect of emigration on average wages, the methodology of Borjas (2003) is applied to Lithuania for the period 2011-2016.
The empirical setting is a regression of individual i’s wages in Lithuania on the emigration rate in that respective skill cell from a pooled cross-sectional model:
𝑤𝑔ℎ𝑥𝑖𝑡 = 𝛿𝑒
𝑔ℎ𝑥𝑖𝑡 + 𝑦𝑒𝑎𝑟 + 𝑋𝑔ℎ𝑥′𝑖𝑡 𝛽 + 𝜀𝑔ℎ𝑥𝑖𝑡 (6)
Where 𝑤𝑔ℎ𝑥𝑖𝑡 is the log real wage of individual i in skill group ghx (gender, education and experience
respectively) in year t (2011-2016). Emigration rate per skill group is captured by 𝑒𝑔ℎ𝑥𝑖𝑡 . Year is a dummy, which controls for each year (2011-2016). 𝑋𝑔ℎ𝑥′𝑖𝑡 is a vector of individual characteristics: gender, education (three levels), experience (eight groups), region (ten regions)36, urban and fulltime employment are dummies, and age is a continuous variable. The error term is 𝜀𝑔ℎ𝑥𝑖𝑡 . Standard errors are clustered at gender, education, and experience level since the variable of interest, emigration rate is at the same level. This creates 48 clusters in total (2 genders x 3
36 That is, the ten counties according to the Lithuanian administrative system: Alytus, Kaunas, Klaipėda, Marijampolė, Panevėžys,
19
educational levels x 8 experience groups). Since log real wages are defined at gender, education and experience level, equation (6) uses variation across gender, education and experience groups.
To control for different returns to education and experience, interaction variables are included in the extended regressions: education * experience, year * education, year * experience. This should control that i) different educational levels have different returns to experience; ii) there are differences for returns to both education and experience throughout time (Juhn, Murphy, & Pierce, 1993; Psacharopoulos, 1989). After applying these controls, the variable of interest, 𝛿, measures the average effect of emigration in skill group ghx on the wages within that same skill group ghx (Elsner, 2013a).
V Empirical Analysis
The estimation of equation (6) is given by table 3.
Table 3 Baseline estimation
Log real wages in Lithuania at gender, education and experience level Baseline model
Emigration rate 1.419* [0.707] Men 0.267*** [0.025] Upper secondary 0.455*** [0.054] Third level 1.079*** [0.065] Age 0.037** [0.015] Constant 5.446*** [0.271] Observations 25976 Adjusted R2 0.253 Controls Experience Yes Region Yes Year Yes Urban Yes
20
The initial estimation shows that, on average, real wages of Lithuanians increased by 1.4% for a one percent increase in the emigration rate in the respective skill cell. The baseline equation (6) controls for education level, work experience, year, age, region, urbanization, and gender. As expected, higher education corresponds with higher wages. On average, the real wages for upper secondary are 0.4% higher, whereas third level wages have 1% higher wages compared to lower secondary. Additionally, women, on average, earn 0.27% less than males. This confirms the aforementioned earnings gap in table i. Additionally, the age variable shows that older individuals earn on average more than younger individuals.
Table 4 extended regressions
Log real wage in Lithuania at gender, education and experience level
1 2 3 4 5 6 7 8 9 Emigration rate 1.448** 1.294** 1.076 1.259* 0.409 1.148 -0.479 0.336 -0.549 [0.709] [0.606] [0.725] [0.648] [0.584] [0.774] [0.674] [0.592] [0.702] Constant 19.5 19.55 20.24 18.71 18.31 19.15 19.23 17.45 18.18 [15.02] [17.84] [17.83] [18.31] [17.91] [18.29] [17.89] [18.38] [18.36] Observations 25976 25976 25976 25976 25976 25976 25976 25976 25976 Adjusted R2 0.253 0.432 0.432 0.433 0.435 0.433 0.435 0.435 0.435 Controls Gender, Edu, Exp, Age Region, Year, Urban
Yes Yes Yes Yes Yes Yes Yes Yes Yes
FDI, Export, Region*Year, Unemployment
Yes Yes Yes Yes Yes Yes Yes Yes Yes
Fulltime No Yes Yes Yes Yes Yes Yes Yes Yes
Edu*Year No No Yes No No Yes Yes No Yes
Exp*Year No No No Yes No Yes No Yes Yes
Edu*Exp No No No No Yes No Yes Yes Yes
Notes: Robust standard errors clustered at gender, education and experience (48 clusters in total) in brackets. Age is a continuous variable. Dummies for gender, education, experience, regions, year, and whether the individual lives in an urban area are included in the first row of controls. Regions are interacted with time. FDI and Export are per capita on a regional level. Unemployment is on a regional level at both aggregated and gender level. Fulltime is a dummy capturing fulltime employment versus part-time. Education and experience are interacted with time and each other in models 3-9. *** p<0.01, ** p<0.05, * p<0.1. Source: Author’s own calculations
21
Different levels of regional development can create substantial differences in income and wages within a country. Accordingly, external demand shocks are controlled for in model (1) which adds FDI and exports per capita on a regional level. Also, unemployment on a regional level on both aggregate and gender is included in model (1) since unemployment is negatively related to wages. Higher levels of exports and FDI can shift demand and can therefore increase wages. Additionally, regions are interacted with time to capture difference across regions and throughout time. However, as argued previously, the empirical literature is not concise on the effects. Thus, it is no surprise there is only a negligible effect on the effect of emigration on wages.37
The distinction between fulltime and part-time is added in model (2). A closer look at the result reveals that full-time employment explains a large share of the premium on emigration. The magnitude drops to 1.2%, but remains significant.
Model (3) controls for differences in returns to education per year by interacting educational levels with time. The association between emigration and wages becomes insignificant.
To control for different earnings to experience profiles across time, experience groups are interacted with time in model (4). The regressor of interest is significant at the 10% level and the magnitude has dropped to 1.25%.
Differences between combinations of experience and education profiles are controlled by the interaction of all combinations of experience and education groups in model (5). Evidently, this replicates the skill cell structure of this research as there are 24 possible combinations of this variable (3 education groups x 8 experience groups). Emigration become insignificant in this model.
The models 6-9 allow for different combinations of the interaction of education, experience, and year. Emigration remains insignificant in all models. The final model (9) includes all controls which allows only for variation within skill cell, as all other factors are controlled for (Borjas, 2003). In the full model specification, there is no significance for aggregate effects. However, it is likely that there are differences between genders in terms of magnitude and significance with respect to emigration rate and wages. Elsner (2013a) found results which indicate that the effects are significant for men, but not for women in the period 2004-2008. To disaggregate the influence of emigration on wages by gender, table 4 is replicated but regresses gender separately.
22
Table 5 extended regressions per gender
Log real wage in Lithuania at gender, education and experience level
1 2 3 4 5 6 7 8 9 Men Emigration rate 0.3 0.42 0.368 0.384 -0.0935 0.675 -0.676 -0.183 -0.494 [0.872] [0.784] [0.937] [0.857] [0.890] [0.979] [1.136] [0.931] [1.173] Observations 12461 12461 12461 12461 12461 12461 12461 12461 12461 Adjusted R2 0.239 0.411 0.411 0.411 0.414 0.411 0.414 0.414 0.414 Women Emigration rate 3.269*** 2.516*** 2.034*** 2.423*** 2.810*** 1.844** 1.637 2.784*** 1.345** [0.738] [0.605] [0.710] [0.643] [0.881] [0.748] [0.980] [0.676] [0.557] Observations 13515 13515 13515 13515 13515 13515 13515 13515 13515 Adjusted R2 0.286 0.462 0.463 0.462 0.465 0.463 0.466 0.466 0.466 Controls
Exp, Edu, Age Region, Year, Urban
Yes Yes Yes Yes Yes Yes Yes Yes Yes
FDI, Export, Region*Year, Unemployment
Yes Yes Yes Yes Yes Yes Yes Yes Yes
Fulltime No Yes Yes Yes Yes Yes Yes Yes Yes
Edu*Year No No Yes No No Yes Yes No Yes
Exp*Year No No No Yes No Yes No Yes Yes
Edu*Exp No No No No Yes No Yes Yes Yes
Notes: The table shows the regressions per gender. Robust standard errors clustered at gender, education and experience (48 clusters in total) in brackets. Age is a continuous variable. Dummies for gender, education, experience, regions, year, and whether the individual lives in an urban area are included the first row of controls. Regions are interacted with time. FDI and Export are per capita on a regional level. Unemployment is on a regional level at both aggregated and gender level. Fulltime is a dummy capturing fulltime employment versus part-time. Education and experience are interacted with time and each other in models 3-9. *** p<0.01, ** p<0.05, * p<0.1. Source: Author’s own calculations
Table 5 is a re-estimation of table 4 but shows the regressions for men and women separately. Column (1) already shows the stark difference between genders: a one percent increase in emigration rate has no significant impact on the wages of men, but is associated with a 3.2% increase in wages for women in Lithuania. The underlying reason for this difference is not entirely clear. The data however, point to the fact that women, on average, were higher educated than men and could face more exposure to emigration. Additionally, there are some results which indicate that women earn higher premiums on education than men (Hazans, 2003). Finally, it is possible that occupational sorting is responsible for a part of the difference between men and women.38
38 For example, assume that women are relatively more employed in an occupation which is exposed to emigration. Hence, this
23
Following the same method as in table 4, interaction controls for education, experience, and time enter the models independently and jointly in models 3-9. In short, emigration remains insignificant with respect to the wages of men in Lithuania in all models. The estimation of the wages of women with respect to emigration remains significant throughout the model—except for model (7)—and is in line with what would be expected by both the literature and the theory. Most importantly, the results for women remain significant in the full specification in model (9). For a one percent increase in emigration, wages were 1.34% higher for women in Lithuania in the period 2011-2016.
The aforementioned results provide some interesting implications for both the literature and policy. In terms of the literature, the results confirm that emigration does in fact—in the short term—corresponds with higher wages based on the skill cell approach. Identifying variation by gender, education, and experience shows that wages for women—on average—were 1.34% higher for a one percent increase in the emigration rate in the full model specification. For men however, the effect of emigration on wages is insignificant in all models.
More interesting is the fact that these results are in line with Elsner (2013a); higher emigration rates lead to higher wages in Lithuania. Conversely, the results are only applicable to women in the time period 2011-2016, whereas earlier results were only significant for men. The findings are also comparable with Dustmann et al. (2015) which is also in the context of the Central Eastern European (CEE) region. Dustmann et al. found that exposure to emigration leads to higher wages on a regional level, which were mostly applicable to higher educated household members that were left behind. Alternatively, lower educated individuals could experience wage decreases in the context of emigration. Furthermore, the results are consistent with similar approaches that were discussed previously such as Mishra (2007) and Hanson (2003; 2005) for Mexico.
Limitations
24
restriction of the experience group to eight groups instead of ten. Moreover, the work permit data does not take re-migration or circular migration into account. This can lead to an underestimation of the true emigration flows to both Ireland and the UK. The alignment of the sending and receiving country data on migration should help to alleviate this issue. Additionally, the results are only applicable in the short run, but do point to the possible effects which might be present in the long term. Finally, remittances play an important role in any research investigating migration and are also relevant for Lithuania. These data, however, cannot be incorporated in the model since it is not available at a regional or individual level.
Although Lithuania became an EU member in 2004, full access to the European labour market was only granted in 2011, when the restriction on labour mobility from EU8 to EU1539 were lifted. This can clearly be seen in the number of work permits issued to Lithuanians in the UK and Ireland—it dropped significantly after 2011 as the European labour market became more liberalized. The lifting of these restrictions, however, do not infer exogeneity of the migration flows, since Ireland and the UK opened their labour markets in 2004. The emigration from Lithuania to mainly the UK persisted throughout time, as flows only started to decrease after the aforementioned labour market mobility liberalization.
Another area which is related to this research is the presence of push and pull factors. The presence of push factors (i.e. low wages, high unemployment) creates the potential of reverse causality. However, although wages are on the lower end of the distribution in Europe, wage growth in Lithuania is substantial throughout the analysis.40 Furthermore, an analysis of the push
and pull factors for Lithuanian migration revealed that pull factors dominate the push factors (Kazlauskienė & Rinkevičius, 2006). The returns to higher education are found to be higher in the UK and Ireland than in Lithuania, which is shown in the data as relatively more higher educated Lithuanians emigrate.
Since this paper estimates the impact of changes in labour supply, fluctuations in labour demand should be accounted for. Potential shocks of labour demand can lead to an over- or underestimation of the effect of emigration on wages in Lithuania. To control for shocks to labour
39 Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain,
Sweden and the UK.
40 For example, see Bank of Lithuania (2014, 2017) for a more detailed analysis of macroeconomic indicators. Note that wage
25
demand, FDI, exports per capita and unemployment on a regional level are used to control for exogenous shocks to labour demand and, consequently, wage.41 Additionally, capital stocks might
be used to control for adjustments due to decreases in labour supply. However, since it is a short run analysis of emigration on wages, such adjustments are not applicable.42 Nevertheless, including regional controls, unemployment, FDI, and exports per capita should indirectly control for capital adjustments in the short term.
The self-selection of migrants is another potential issue when investigating the consequences of migration on wages in the origin. Negative self-selection will bias the results upwards since it draws individuals from the lower end of the income distribution. As a result, average wages at the origin will increase.43 Conversely, positive self-selection will bias the results downwards. The data concerning emigrated Lithuanians already demonstrated that Lithuanians abroad were slightly higher educated than the workforce in the origin. The differences however, are small. Moreover, the fact that native workers in the UK and Ireland are, on average, more skilled than Lithuanian workers makes it arguable that migrants are required to have more skills than domestic Lithuanian workers.44 Therefore, the presence of self-selection will only lead to an underestimation of the effect of emigration on wages.
Additionally, another point of concern regards the selection across education and experience groups. Again, the data showed that Lithuanians in Ireland were younger and therefore have less work experience. These aforementioned selection issues are dealt with in the model via the dummies and the interaction effects. Finally, the model cannot account for students who emigrated to Ireland to study. However, the statistics from Ireland are only relevant for the highest level of education that is completed. Also, the high income and cost of living differences between Ireland and Lithuania creates barriers for Lithuanians to study in Ireland. This makes it less likely that the flow of students between Lithuania and Ireland will influence the results.
41 The literature on FDI and wages is not concise in terms of effects. For example, see Earle, Telegdy and Antal (2012) compared
to Onaran and Stockhammer (2008). FDI can either increase wages vis-à-vis labour demand or supress wages via its bargaining power. However, it still can be used to control—to some extent— for shifts in labour demand.
42 In the long run, the capital market becomes relevant as it can adjust to migration flows. The literature is divided on the elasticity
of the supply of capital. Often it is assumed to be either fully elastic or imperfectly elastic with respect to shocks by immigration (Docquier, Özden, & Peri, 2010; Dustmann, Glitz, & Frattini, 2008). On the other hand, literature which is investigating the short run impact of migration on wages often assumes that capital—in the short run—cannot adjust to the shocks to the labour market vis-à-vis the migration flows.
43 This is a pure mechanical effect by simply removing lower income individuals from the income distribution, average wage
increases all else equal.
44 The fact that workers need to be fluent in a second language is the most basic ‘additional’ skill required, apart from the ability
26
Finally, to identify the exogeneity of migration flows, the shift share instrument is often used (Card, 2001). This strategy employs the historical variation in immigrant inflows to address endogeneity of locational choice of migrants. This method employs past settlement behaviour of immigrants to explain variation in immigration today. However, since the population of Lithuanians in Ireland a priori EU enlargement was non-existing, using such an instrument is not an option in this research. Also, the validity of the shift share instrument has recently attracted some critique; see Jaeger et al. (2018) for a thorough and critical examination of the instrument.
VI Conclusion
This paper has studied the effect of emigration on sourcing countries wages for Lithuania in the period 2011-2016. In a perfect labour market, a contraction of labour supply should result in higher wages. By using both sourcing and receiving country data, it is possible to exploit differences in skill cell emigration rates. These differences are used to investigate the variation between education, experience and gender with respect to wages in Lithuania.
Empirically, the estimation shows that wages increased, on average, by 1.3 percent in response to a one percent increase in the emigration rate. This effect is robust and significant for a large amount of controls and model specifications for women in Lithuania. The relationship of men with respect to emigration and wages is insignificant. Nevertheless, the estimation provides an economically relevant coefficient as it reflects changes in the short term for the Lithuanian labour market. Interestingly, the results show larger magnitudes than previous research and they are only applicable to female labour market participants in the short term (Aydemir & Borjas, 2007; Dustmann et al., 2015; Elsner, 2013a; Mishra, 2007).
27
countries in the CEE are likely to experience similar increases in wages vis-à-vis emigration— albeit less extreme than the case of Lithuania. For labour abundant prospective EU members, these results are an indication for possible labour market outcomes ex post accession.
Future research should analyse whether the same results can be found for other countries, as the current literature on the effect of migration on origin countries wages is quite limited. Other net emigration countries could be investigated since these are most likely to replicate the results of Lithuania. Additionally, this paper fails to provide evidence for the underlying factor which explains the different results for men and women in Lithuania. A more thorough investigation of this phenomenon might prove interesting for future research.
Acknowledgement
28 VII References
Adams Jr, R. H. (2011). Evaluating the economic impact of international remittances on developing countries using household surveys: A literature review. Journal of Development Studies, 47(6), 809–828.
Amuedo-Dorantes, C., & Pozo, S. (2006). Migration, Remittances, and Male and Female Employment Patterns. The American Economic Review, 96(2), 222–226.
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.
Bank of Lithuania. (2014). Lithuanian Economic Review.
Bank of Lithuania. (2017, September 20). Economic growth to accelerate, unemployment to fall, wage growth to outpace price increases. Retrieved 22 April 2018, from
https://www.lb.lt/en/news/economic-growth-to-accelerate-unemployment-to-fall-wage-growth-to-outpace-price-increases
Beine, M., Docquier, F., & Rapoport, H. (2001). Brain drain and economic growth: theory and evidence. Journal of Development Economics, 64(1), 275–289.
Binzel, C., & Assaad, R. (2011). Egyptian men working abroad: Labour supply responses by the women left behind. Labour Economics, 18, S98–S114.
https://doi.org/10.1016/j.labeco.2011.03.002
Boeri, T., & Van Ours, J. (2013). The economics of imperfect labor markets. Princeton University Press.
