1

Immigrant Language Fluency in the Low-skilled Labor Market∗ Ana Damas de Matos† March 2016

Abstract The author investigates the return to being a native speaker for immigrant men in the low-skilled labor market. To do so, the author uses longitudinal linked employer-employee data to compare the two main recent immigrant groups in Portugal: Brazilians, who are Portuguese native speakers, and Eastern Europeans, who are not. The author shows that both wage level and wage growth of the two groups are similar. To better understand this surprising result, the author studies two mechanisms through which language fluency may lead to higher wages: sorting across occupations and across firms. Brazilians do sort into occupations that require greater language skills. However, this does not translate into a wage premium. There is considerable workplace segregation in Portugal, but Brazilians are not less segregated from natives than Eastern Europeans. Altogether, the evidence in this paper suggests that language skills are not a major driver of economic assimilation in the low-skilled labor market.

∗

I thank Daniel Parent, Baris Kaymak and participants of the Montreal Applied Micro Group for comments, and Miguel Portela for invaluable support with the data. I am grateful to the Portuguese Ministry of Employment, Statistics Department, for access to the data. † HEC Montr´eal, 3000 Chemin de la Cˆote-Sainte-Catherine, Montr´eal, Canada. E-mail: [email protected].

2

Introduction Fluency in the host country’s language has been shown to ease the labor market integration of immigrants. In particular, most of the literature finds a large wage premium to host country language fluency for immigrants (See Chiswick and Miller [2014] for a review of the literature). This effect is at least partially driven by highly educated immigrants (Chiswick and Miller [2003]) or immigrants that work in high-skilled occupations (Berman et al. [2003]). However, in many countries the majority of immigrants work in low-skilled occupations and may, consequently, not benefit from large returns to language fluency. Despite this fact, the evidence on the role of host country language fluency in the assimilation of immigrants in the low-skilled labor market is scarce.1 This paper attempts to fill this gap by comparing the two main immigrant groups from the recent immigration wave to Portugal: Eastern Europeans and Brazilians. Brazilians are Portuguese native speakers whereas Eastern Europeans in most cases have no contact with the Portuguese language before migration. The vast majority of immigrants of both groups work in low-skilled occupations. I use longitudinal linked employer-employee data covering virtually all workers in the Portuguese private sector to follow the two immigrant groups in the first years in the labor market. The data set contains accurate information on wages, as well as detailed information on the occupations and firms in which immigrants work. I start by estimating the wage gap and wage growth of Brazilians and Eastern Europeans relative to natives to identify whether there are any differences in mean wages and wage growth between the two groups which could be linked to language fluency. Furthermore, immigrants with greater language fluency may sort into occupations with greater language requirements, as well as into less segregated firms, which may positively impact their wages. Hence, I investigate the extent to which the sorting of immigrants from Eastern Europe and Brazil across occupations and firms explains differences in mean wages and how occupational and firm mobility is correlated with the estimated wage growth of the two immigrant groups. To further investigate how immigrant language fluency is related to occupational sorting, I use a measure of occupational language requirements from the O*NET data set for 1

Berman et al. [2003] find that improved language fluency only leads to higher wages in high-skilled occupations in Israel, whereas Kossoudji [1988] finds that for Hispanic immigrants in the United States there is a premium for language fluency in low-skilled occupations.

3 the United States. This measure allows me to assess whether Eastern Europeans work in occupations with lower language requirements than Brazilians upon arrival and with time in Portugal. Finally, I study the workplace segregation of the two immigrants groups. Previous work has shown that immigrants work disproportionately with other immigrants (Andersson et al. [2014], ˚ Aslund and Skans [2010]), and in particular immigrants speaking the same language (Hellerstein and Neumark [2008], Glitz [2014]). The fact that Brazilians have the same mother tongue than natives allows me to discriminate between different theories explaining immigrant workplace concentration.

1

Context and Data

1.1

Brazilian and Eastern European Immigrants

Portugal, like the other Southern European countries, was mainly an emigration country until the mid-1990s. In the late 1990s and early 2000s, Portugal received an unprecedented wave of immigration. Several factors contributed to the inflow of immigrants to Portugal such as good economic performance in the 1990s and a large number of important infrastructure works, such as the Universal Exhibition in Lisbon in 1998, and the construction of the TransTagus Bridge and the metropolitan network in Oporto. The two main groups of immigrants from this immigration wave to Portugal came from Eastern European countries,2 mainly from the Ukraine, and from Brazil. An overview of immigration in Portugal may be found at OECD [2008]. This paper focuses on the comparison between the labor market integration of Eastern Europeans and Brazilians.3 Brazilians are the largest immigrant community in Portugal, followed by the Ukrainian community in 2012, according to the Portuguese Immigration Borders Service (SEF [2013]). Both groups of immigrants are economic immigrants. In fact, a recent nationwide representative survey of immigrants in Portugal shows that the main reason to choose Portugal as destination country was job opportunities for both 2

Eastern Europe is defined in the text as Slovakia, Poland, The Czech Republic, Hungary, Slovenia, Latvia, Estonia, Lithuania, Romania, Russia, Moldova, Ukraine, and former Yugoslavia. 3 This paper focuses on the immigration wave from the late 1990s and early 2000s. For that reason, I do not consider immigrants from former Portuguese colonies in Africa (Cape Verde, Angola, Mozambique, Guinea Bissau), since the immigration wave from these countries to Portugal started in the 1960s. Furthermore, although Portuguese is an official language in these countries, most immigrants speak mainly Portuguese creole.

4 groups (Malheiros and Esteves [2013]). A main difference between these two groups is that Brazilians are Portuguese native speakers and have cultural ties with Portugal, whereas Eastern Europeans do not have any contact with the Portuguese language nor culture before migration. Differences in the ability to speak the mother tongue is not the only difference between the two immigrant groups. Eastern European immigrants are on average older and more educated than Brazilian immigrants. The results in this paper show that the wage level and growth of the two immigrant groups are similar despite the difference in language skills. One interpretation, the one put forth in the paper, is that the returns to language skills for immigrants are low in the low-skilled labor market. An alternative interpretation is that the language advantage of the Brazilians is offset by some other advantage of Eastern Europeans with education being the main candidate. As shown later, the decomposition of the initial mean wage and mean wage growth into industry, occupation, and firm effects is similar for both groups. Therefore, language and education effects would need to offset each other both across and within industries, occupations, and firms– a scenario which seems unlikely. There is also evidence that returns to education and experience acquired in the home country is low for immigrants (Friedberg (2000), Cohen-Goldner and Eckstein (2008), Ferrer and Riddell (2008) and Oreopoulos (2011)). For these reasons, the alternative interpretation of off-setting skills is not a compelling one.

1.2

Data

The empirical analysis uses mainly linked employer-employee data for Portugal (‘Quadros de Pessoal’ or QP) which allows following immigrants over time in the Portuguese labor market. Additionally, I use information on language requirements by occupation from the O*NET data set for the United States. I merge this information with the Portuguese data to correlate the sorting of immigrants across occupations with the occupational language requirements. 1.2.1

Quadros de Pessoal

The main data set used for the analysis is the Quadros de Pessoal (QP), a linked employeremployee panel covering virtually all wage earners in the Portuguese economy, with the exception of the public service and domestic workers. The data contain detailed infor-

5 mation on firms and workers and makes it possible to follow immigrants over time in the Portuguese labor market. The data is well suited for the analysis for several reasons:4 First, the data contain information on worker nationality, which is crucial to identify immigrants from the two regions of origin. Immigrants are defined as foreign citizens in the paper.5 Second, the longitudinal dimension of the data makes it possible to track immigrants through their first years in the Portuguese labor market. I focus on immigrants from the new immigration wave who first appear in the data in 2002 and later, and follow them until 2009. Years of experience in the labor market are defined as years since the first time the immigrant has a job in the formal sector. The QP, like other administrative data sets, covers only the formal sector.6 Third, the data contains detailed information on the workers’ occupation, industry and employer over time. I use the information on the occupation and industry, both at the three-digit level. This information makes it possible to compare the assimilation of the two groups of immigrants in detail. 1.2.2

O*NET data

I use information on language requirements of occupations from the Occupational Information Network data set for the United States, O*NET version 9.7 The data set contains detailed information on the tasks and skills used in each occupation in the Standard Occupational Classification System (SOC) in the United States. In particular, the data set contains information on the importance of English and the level of English required in each occupation. The information comes from surveys to workers in each occupation. The questions asked are the following: • How important is knowledge of the English language to the performance of your current job? 4

More details on the construction of the panel are presented in Appendix A. In the short run, naturalization is not a concern. Immigrants need at least six years of legal residence to be able to apply for Portuguese citizenship. 6 Estimations of the size of the Portuguese informal sector show that it is slightly larger than the EU average, Schneider [2002]. The authors estimate that the share of the informal economy with respect to GNP around the year 2000 is 18% for Western European OECD countries, 23% for Portugal, as well as for Spain and Belgium. The share is highest for Italy 27% and Greece 29%. 7 For more information on the O*NET data: http://online.onetcenter.org. 5

6 The answer to this question varies from 1 (not important) to 5 (extremely important). • What level of English language knowledge is needed to perform your current job? The answer to this question varies from 1 to 7. An example given in the questionnaire to help respondents is: 2 “write a thank you note”; 6“teach a college English class”. The score is set to 0 for individuals who reply that language is “not important” in the previous question. Chiswick and Miller [2010] and Chiswick and Miller [2013] used these two variables from the O*NET data in the context of language fluency of immigrants in the United States. Data from the O*NET has also been matched to data sets from other countries. For example, recent research on immigrant assimilation in Canada uses information on occupational skill requirements from the O*NET merged to Canadian data (Adser`a and Ferrer [2014]; Adser`a and Ferrer [2014]; Imai et al. [2014]). Another example is AmuedoDorantes and De La Rica [2011] that uses information on job task requirements from the O*NET matched to the Spanish labor force survey to study the task specialization of immigrants and natives. I match the O*NET language scores for the two questions above to the occupational classification in the Portuguese data. To the best of my knowledge, no other paper has used the information on language in the O*NET for a country other than the United States. To match the SOC in the O*NET data to the Portuguese classification, I first match the SOC to the U.S. 2000 Census codes, and then match these with the ISCO 88 using the crosswalk, which is also used in Amuedo-Dorantes and De La Rica [2011], and made available by the Center for Longitudinal Studies in the United Kingdom. The Portuguese occupational classification in the data (CNP 94) is the same as the ISCO 88, except for two occupations at the three digits, in which no immigrant works. The assumption I am making by matching the language scores from the O*NET to the Portuguese data is that the importance of language for a given occupation in the U.S. SOC is the same as that in the corresponding occupation in the CNP. In Appendix B, I provide a listing of the occupations in the Portuguese classification at the 2-digit level and their importance of language score and level of language score.

7

1.3

First Descriptives

Selected population means from the QP are presented in Table 1. All means for immigrants are calculated in the first year upon entry in the labor market. Women represent less than 30% of immigrant observations in the data. Immigrant women in Portugal often get jobs as domestic workers, who are not covered in the data. The sample of immigrant women in the QP is selected and small. Consequently, I restrict the analysis to men. The analysis takes into account approximately 30,000 Eastern European men and 24,000 Brazilian men. The larger number of workers, and observations, in the data for Eastern Europeans translates the fact that the wave of Brazilian immigration is slightly more recent than the wave from Eastern Europe. Immigrants are younger than the native population, and immigrants from Brazil are younger than immigrants from Eastern Europe. Immigrant men are very concentrated in a small number of industries and occupations. Upon entry in the labor market, 47% of immigrants from Eastern Europe and 36% of immigrants from Brazil work in construction. The differences between origin groups are also noticeable: 20% of Eastern Europeans work in manufacturing compared with 9% of Brazilians, and 20% of Brazilians work in hotels and restaurants compared with only 4% of Eastern Europeans. Immigrants from both groups are over-represented in low-skilled occupations such as craft and related trade workers and elementary occupations.8 Figure 1 shows the distribution of the O*NET importance of language scores for both groups of immigrants upon entry in the labor market and for all Portuguese men in the data. A similar figure for the level of language scores is presented in Appendix B. The distribution of the language scores for natives has a higher mean and a larger standard deviation than that of immigrants. The occupations in which Brazilians work in require a greater language proficiency than those in which Eastern Europeans work in. The mean score of the importance of language is 2.43 for Brazilians and 2.26 for Eastern Europeans, compared with 2.61 for natives. The mean scores for the level of language are 2.10 and 1.92 for Brazilians and Eastern Europeans, relative to 2.36 for natives. The language scores are also more dispersed for Brazilians than Eastern Europeans. The standard deviation of the score of the importance of language is 0.46 for Brazilians and 0.37 for Eastern Europeans, compared with 0.67 for natives. Similarly, the standard deviations for 8

I refer to low-skilled occupations in contrast with high-skilled occupations which are occupations classified 1 to 3 in the ISCO 88, that is 1 senior government officials, corporate and general manager ; 2 professionals and scientists; 3 technicians and associate professionals.

8 the level of language are 0.55 and 0.42 for Brazilians and Eastern Europeans, compared with 0.80 for natives. The lowest dispersion of scores for immigrants relative to natives is consistent with the fact that immigrants work in specific occupations in the Portuguese labor market as documented in the previous table.

2

Wages

In this section, I describe the initial wage level and the wage growth in the first years in the host country for both immigrant groups and how they correlate with the immigrants’ sorting across industries, occupations and firms. Despite differences in characteristics, in particular mother tongue, Eastern Europeans and Brazilians have similar initial mean wage and mean wage growth.

2.1

Framework

To describe the wages of the two immigrant groups relative to natives, I use a fairly standard human capital earnings function which has often been used in the immigrant assimilation literature (Chiswick [1978], Borjas [1985], Baker and Benjamin [1994], Lubotsky [2007], among many others): ln(wit ) = α + β E Easti + β B Brazi + δ ysmit + δ E ysmit ∗ Easti + Xit γ + θt + ϵit (1) where i represents the individual and t the calendar year. The model is estimated for native, Eastern European and Brazilian men. The dependent variable is the log of real hourly wages expressed in 2002 euros. Easti is a dummy variable for Eastern European immigrant, and Brazi is a dummy variable for Brazilian immigrant. ysmit is years since migration, or more precisely in this context, the time spent in the host country labor market. Since, the analysis covers immigrant entry cohorts in a relatively short period (2002 to 2009), the estimations have only a linear trend in ysm. Results with a quadratic function in ysm are presented in Appendix C. θt are calendar year fixed effects and the vector X represents the control variables. Age dummies are included in all specifications. Region, industry, occupation and firm fixed effects are introduced in different steps of the analysis. The returns to observable characteristics, γ, are assumed to be the same

9 for immigrants and natives, as are time effects, θt . Although these assumptions are restrictive (LaLonde and Topel [1992]; Green and Worswick [2012]), this framework is a useful tool to describe the wages of immigrants relative to natives. βˆB estimates the wage gap, and δˆ the wage growth, of Brazilian immigrants relative to natives. The wage gap and wage growth of Eastern Europeans relative to natives is measured similarly by βˆE and δˆ + δˆE . A main concern in the literature when estimating the mean wage growth of immigrants is selective out-migration.10 The comparison of the wage growth of the two immigrant groups may be misleading if their out-migration patterns are different. I introduce individual fixed effects in the estimations of the wage regressions as a way to alleviate this concern. Individual fixed effects account for all time-invariant characteristics in equation (1), in particular the years of education and unobserved individual characteristics, which may differ across the two groups. Differences in wage growth driven by unobserved characteristics between the two groups, however, are not taken into account by the individual fixed effects. 9

2.2

Results

Table 2 presents the OLS estimates from equation (1) with different sets of controls. Relative to natives of the same age, the initial wage gap of Brazilians (Eastern Europeans) is approximately 25% (29%) (column 1).11 This difference in the wage gap is explained by the age composition of the two groups of immigrants. Immigrants from Eastern Europe are on average older than immigrants from Brazil and are compared with older natives, who earn on average higher wages. In fact, the initial wage gap is approximately 31% for Brazilians and 32% for Eastern Europeans when age is not taken into account. The estimations without controlling for age are presented in Appendix C. The determinants of the wage gap are also similar for immigrants of the two groups. In fact, approximately 60% of the wage gap of both groups relative to natives is accounted 9

I do not control for immigrant entry cohort effects as in much of the literature (e.g. Borjas [1985] or Lubotsky [2007]) since the period in the analysis is short. Furthermore, the estimations of the wage catch up are done with individual fixed effects that also take into account the cohort effects. Results for each entry cohort separately are available upon request. 10 Dustmann and G¨orlach [2014] provide a review of the literature on selective out-migration and its effect on the estimation of immigrant wage assimilation. 11 The value of the variable ysm is 1 in the first year. The initial wage gap for Brazilians is −0.25 = exp(−0.307 + 0.0126) − 1.

10 for by their sorting across industries, occupations and regions, all measured at the threedigit level (column 2). Within narrow regions, occupations, and industries, immigrants sort also into low paying firms. The wage gap controlling for firm fixed effects is 5% for Brazilians and 9% for Eastern Europeans (column 3). Although firm heterogeneity is an important determinant of the wage gap for both groups, it accounts for more of the wage gap of Brazilians than that of Eastern Europeans. The fact that immigrants work in lower paying firms is in line with evidence for Canada (Aydemir and Skuterud [2008], Pendakur and Woodcock [2010]) and previous evidence for Portugal (Carneiro et al. [2012]). The estimations above do not control for individual characteristics apart from age. In Appendix C, I present the same estimations introducing the years of education.12 The conclusions remain the same. Despite differences in observed and unobserved individual characteristics across the two groups, the wage gap and its determinants are similar for Brazilians and Eastern Europeans. Not only is the wage gap and its determinants similar for Brazilians and Eastern Europeans, but their wage catch-up to natives is also similar. Table 2 presents estimations of the wage catch-up with different sets of controls estimated by OLS (columns 1 to 3) and with individual fixed effects (columns 4 to 6). I will focus on the estimations controlling for individual fixed effects to mitigate the concerns on attrition as mentioned above. Both groups of immigrants experience significant wage growth in the first years in the labor market relative to natives of the same age. The wage catch-up is estimated at approximately 1.2 percentage points for Brazilians and 1.4 for Eastern Europeans (column 4). Most of the wage catch-up takes place within regions, occupations and industries for both groups. The estimated wage growth does not change by much once these control variables are introduced. Approximately 85% of the wage catch up for Brazilians and 89% for Eastern Europeans occurs within regions, industries and occupations measured at the three-digit level (column 5). Moving to better paying firms within industries and regions is an important determinant of the immigrant wage catch-up for both groups. The estimated wage catch up of Brazilians decreases by one third (from 1 percentage point to 0.6 percentage point), and that of Eastern Europeans by 28% (from 1.2 percentage points 12

In the QP, the information on education is reported by employers. Bender and von Wachter [2006] mention that the educational data reported by establishments in German administrative data tends to reflect required rather than actual education. I document that this may lead to an underestimation of the years of education of immigrants in Appendix E. Therefore, my preferred specifications do not control for years of education.

11 to 0.8 percentage point), when firm fixed effects are added to the estimation (column 6).13 The results show that firm fixed effects are relevant in explaining the immigrant wage gap and wage catch-up. Firm fixed effects do not merely capture the fact that firms with a high share of immigrants pay lower wages. I present alternative specifications in Table C.4 in the Appendix in which I replace the firm fixed effects with the individual’s exposure to immigrants, that is the share of the individual’s co-workers who are immigrants. The wage gap and the wage growth of immigrants is correlated with the share of immigrant co-workers in the cross-section. Nevertheless, the firm fixed effects account for more of the wage gap and of the wage growth than the immigrant exposure variable. This result indicates that other unobserved firm characteristics are relevant in explaining immigrant assimilation.

2.3

Link to the literature

A large literature estimates the OLS returns to language fluency in an augmented human capital framework similar to the one in equation (1).14 In particular, several studies use a dichotomous language fluency variable, as Chiswick and Repetto [2000], Budr´ıa and Swedberg [2012], Miranda and Zhu [2013].15 In the above estimations, Eastern Europeans may be considered as having a language fluency of 0, since they are not native speakers, and Brazilians as having a language fluency of 1. Given that Brazilians are native speakers, one would expect that they earn a wage premium in the Portuguese labor market. However, the results show that the two groups of immigrants have similar initial mean wage. Furthermore, if Eastern Europeans improve their language fluency with time in the host country, their wages would grow faster than the wages of Brazilians. The wage growth of the two immigrant groups is similar. I interpret the results as evidence that the returns to language fluency are not high in the low-skilled labor market. Most studies focus on traditional anglophone destination countries where the estimated returns to language fluency by OLS are high: 10 to 20% in the United States and Australia 13

The estimation in column 6 has both individual and firm fixed effects and is performed using the algorithm presented in Guimaraes and Portugal [2010] for OLS estimation with two high dimensional fixed effects. 14 The estimating equation may be written as: ln(wit ) = βimmigi + δysmit + ζlanguageit + Zi ϕ + Xit γ + θt + ϵit 15 The variable is set to 1 for immigrants who can write a simple letter in Hebrew; to immigrants who report speaking Spanish very well; to immigrants for whom English is the first language.

12 and 20 to 30% in Canada (Chiswick and Miller [2014]). However, these estimations consider immigrants in both the high-skilled and the low-skilled labor market. Kossoudji [1988] estimates the premium to language fluency for different types of occupations in the United States. She finds evidence of a language premium in the low-skilled labor market for Hispanics (the largest non-native speaking immigrant group in the United States) but not for Asians. The results for Portugal are consistent with recent evidence for Spain. Spain has a similar immigration context to Portugal, given that it is a recent immigration country and that part of the immigrants come from Latin America and are native speakers. Few studies exist on the role of language fluency of immigrants in Southern Europe. Budr´ıa and Swedberg [2012] finds in OLS specifications only a 5% wage premium for immigrants who are fluent in Spanish. Furthermore, Izquierdo et al. [2009] show also for Spain that the wage catch-up is similar for South-American immigrants (Spanish native speakers) and European immigrants from outside of the EU15 (non-native speakers).17 16

3

Occupational Sorting

Differences in language fluency between immigrant groups may lead to differences in occupational sorting, which in turn may lead to differences in wages. In this section, I take a closer look at how immigrants from Eastern Europe and Brazil sort into occupations with different language requirements, initially and in the first years in the labor market, and how this sorting relates to the wage gap and the wage growth. Figure 1 in the descriptive statistics shows that immigrants from Brazil work initially in occupations with higher language requirements in terms of importance and level of language, as defined in the O*NET data, than Eastern Europeans.18 However, the estimated initial wage gap is similar for the two groups (Table 2). To understand this perhaps puzzling fact, I look at the correlation between language requirements and hourly wages 16

Other host countries for which there is quite some literature on the language fluency of immigrants and their labor market outcomes are the United Kingdom (Dustmann and Fabbri [2003]), Israel (Berman et al. [2003]) and Germany (Dustmann and Van Soest [2002]). 17 The estimated wage catch-up of immigrants in Spain is also similar to the one estimated above. Izquierdo et al. [2009] find a ten percentage point wage catch up in the first 6 years for immigrants arrived in the 2001-2005 period. 18 The sorting of immigrants with greater language skills into occupations with higher language requirements has been documented for high-skilled occupations in the United States in Chiswick and Taengnoi [2007].

13 at the occupational level. Figure 2 shows the relationship between the mean occupational real hourly wage and the O*NET importance of language score for immigrants of the two groups compared with natives. For natives the data corresponds to the year of 2002. For immigrants, the data is for the first year in the Portuguese labor market. Each bubble represents an occupation at the three-digit level. The size of each bubble is proportional to the number of individuals working in the occupation. The same figures for the level of language score are similar and presented in Appendix B. The figures provide an explanation for why the initial wages of Brazilians and Eastern Europeans are similar despite Brazilians sorting into occupations with greater language requirements. In fact, all immigrants are concentrated in low-wage occupations with relatively low language requirements. For these occupations, the relationship between the mean occupational wage and the occupational language requirements is relatively flat. The relationship becomes steeper for occupations with greater language requirements. In the first years in the labor market, there are no significant changes in the occupational sorting of immigrants of both groups. Figure B.3 in the Appendix presents the correlation between occupational language requirements and mean wages in the first year, and after three and five years in the labor market. All figures are similar. These figures are consistent with the results in the previous section that show that the immigrant wage catch-up occurs mainly within narrowly defined occupations and industries. Chiswick and Miller [2013] show that the language fluency wage premium is driven by immigrants with greater language skills working in occupations with higher mean wages. The results for Portugal are not inconsistent with this finding. The difference may be that Chiswick and Miller [2013] consider immigrants in the United States who are distributed across the whole range of occupations, whereas Eastern Europeans and Brazilians are concentrated in the low-skilled occupations in Portugal. The results imply also that greater language fluency is not strongly correlated with higher wages within occupation. In fact, the initial mean wage of Brazilians is only two percentage points higher within occupation than that of Eastern Europeans (column 2 of Table 2). The fact that language fluency does not translate into higher wages in low-skilled occupations is consistent with evidence for Israel. Berman et al. [2003] study the impact of improved language fluency for immigrants from the former Soviet Union using a survey

14 targeted at immigrants in four specific occupations, two high-skilled (software and technicians) and two low-skilled occupations (construction workers and gas station attendants). They find that there is a complementarity between language fluency and occupational skill. Immigrants in the high-skilled occupations benefit from improved language fluency, whereas immigrants in the low-skilled occupations do not. My paper generalizes this result by considering all low-skilled occupations in the private sector.

4

Workplace segregation

Previous research has shown that immigrants tend to cluster in the same workplaces, and in particular in low-paying firms (Aydemir and Skuterud [2008], Pendakur and Woodcock [2010], Carneiro et al. [2012]). Section 2 shows that immigrants both from Brazil and Eastern Europe work in low-paying firms, and that this accounts for a large share of their wage gap. One of the reasons why immigrants may work with other immigrants, and in particular immigrants of the same origin, is language. In fact, it may be optimal for immigrants to work with others who speak the same language. In this section, I explore the workplace segregation of the two immigrant groups and show that despite Brazilians speaking the host country language they are not less segregated from natives than Eastern Europeans. Following ˚ Aslund and Skans [2009] and ˚ Aslund and Skans [2010], I use the overexposure 19 ratio as a measure of segregation. The overexposure ratio of group A to group B is actual exposure rate defined as : overexposure ratio = expected . The actual exposure rate is the exposure rate average share of co-workers belonging to group B for workers of group A. The expected exposure rate is the average propensity of belonging to group B among the co-workers of individuals in group A, conditional on characteristics. The characteristics considered in the analysis are the distribution of workers across regions, industries and occupations. For example, an own-group overexposure ratio of two for Brazilians would mean that the average Brazilian worker has twice as many Brazilian co-workers as would be expected if workers were randomly assigned to firms holding their distribution across regions, industries and occupations fixed. I calculate the own-group overexposure ratio for Brazilians and Eastern Europeans, as well as the exposure of Brazilians to Eastern Europeans and vice-versa. 19 ˚

Aslund and Skans [2009] define the overexposure odds ratio, whereas ˚ Aslund and Skans [2010] use the overexposure ratio.

15 To more easily compare the workplace segregation of the two immigrant groups to the literature, I also calculate another segregation measure: the index of co-worker segregation as in Hellerstein and Neumark [2008]. The observed index of co-worker segregation of group A relative to natives is defined (using the same notation than in Hellerstein and O O Neumark [2008]) as ICS O = HH − WHO . HH is the observed isolation index, that is, the mean share of co-workers of group A who also belong to group A. WHO is the observed exposure index, that is the mean share of co-workers of natives who belong to group A. An observed index of co-worker segregation of 100 would mean complete segregation and an index of 0 would mean no segregation. To be meaningful, the observed co-worker segregation index is compared to the random segregation index which would be observed if workers were randomly allocated to firms holding their distribution across regions, industries, and occupations fixed.20 The effective co-worker segregation index is defined O −ICS R as: ICS = ICS 100, where ICS O is the observed segregation index and ICS R is 100−ICS R the random segregation index. Table 3 presents the overexposure ratios, as well as the index of co-worker segregation in 2007. The own-group exposure rate of Brazilians and that of Eastern Europeans is the same: 22.8.21 The overexposure ratio is larger for Brazilians than Eastern Europeans. Brazilians (Eastern Europeans) have 3.7 (2.7) times more Brazilian (Eastern European) co-workers than a random assignment of workers to firms would predict holding fixed the distribution of workers across regions, industries, and occupations. Moreover, immigrants from the two groups sort into separate firms. The overexposure ratio of Brazilians to Eastern Europeans and vice versa is 1. Brazilians (Eastern Europeans) are not more exposed to Eastern Europeans (Brazilians) than a random assignment of workers to firms would lead to. Turning to the results on the index of co-worker segregation, the observed index of co-worker segregation is similar for Brazilians and Eastern Europeans. It is 22.2 for Brazilians and 24.0 for Eastern Europeans. Part of the observed co-worker segregation for both immigrant groups is driven by their segregation across regions, industries and 20

The random index of co-worker segregation is calculated using 30 simulations. In each simulation, I assign workers to firms randomly holding the number and the size of firms fixed, as well as the distribution of workers across industries, regions, and occupations. An index of co-worker segregation is calculated for each simulation. The average of the co-worker segregation indices is the random co-worker segregation index. 21 Carneiro et al. [2012] shows that immigrants from different origins are similarly distributed across firms with different shares of immigrants in Portugal. However, they do not take into account the fact that different groups of immigrants work in different industries, occupations, and firms of different sizes.

16 occupations. The effective co-worker segregation indices are 18.2 for Brazilians and 15.6 for Eastern Europeans. Thus, Eastern Europeans are if anything less segregated than Brazilians. The results for the other years are presented in Tables D.1 and D.2 in the Appendix. Figure D.1, also in the Appendix, shows the own-group overexposure ratio for different entry cohorts of immigrants with time spent in the labor market. Despite some differences between the two groups (e.g. the segregation index is increasing for Brazilians in the first years of the analysis; the own-group over-exposure rate is decreasing for Brazilians with time in the host country but stable for Eastern Europeans), the results are consistent with the results on wages presented in section 2. Firm heterogeneity has a similar effect on the wage gap and wage growth of immigrants of the two groups. The observed co-worker segregation indices are larger than the ones observed for immigrants in Germany or Sweden. Glitz [2014] calculates an index of 18.1 in 2008 in Germany, and 14.6 in 2002 in Sweden.22 The indices are nevertheless lower than that for Hispanics in the United States in 2000. Hellerstein and Neumark [2008] study ethnic segregation in the United States and find an index of unconditional co-worker segregation of 34.9 for Hispanics.23 Hellerstein and Neumark [2008] show that the segregation between Hispanics and whites is driven by differences in language proficiency. Hispanics with low English fluency work mainly with other Hispanics with low English language fluency. The fact that Brazilians are not less segregated from natives than Eastern Europeans speaks to the mechanisms driving workplace segregation. Hellerstein and Neumark [2008], ˚ Aslund and Skans [2010], and Glitz [2014] present evidence that immigrants speaking the same language tend to work together. Two different mechanisms could drive this effect: language related productivity spillovers or job search networks. The fact that Brazilians are as segregated from natives as Eastern Europeans is not consistent with productivity spillovers being the main driver of immigrant workplace segregation. However, it is consistent with job search networks driving workplace segregation. Dustmann et al. [2011] show that referrals from workers of the same origin are an important mechanism in explaining immigrant segregation in Germany. The index for Sweden is calculated in Glitz [2014] based on the results in ˚ Aslund and Skans [2010]. The effective co-worker segregation index is 19.8 when taking into account the distribution of workers across metropolitan areas. 22

23

17

Conclusion A main difference between the two main recent immigrant groups in Portugal is that Brazilians are Portuguese native-speakers, whereas Eastern Europeans share neither cultural nor linguistic ties with the host country. This paper shows that the economic assimilation patterns of both immigrants groups are nevertheless remarkably similar. The initial wage gap is similar for both groups, as is a wage catch-up of over one percentage point per year in the first years. The vast majority of immigrants of both groups work in low-skilled occupations and there is no substantial occupational mobility over time. Most of the wage catch-up occurs within low paying occupations and industries. While native-speaking immigrants sort initially into occupations with greater language requirements, this does not translate into higher wages. Higher language requirements do not seem to be correlated with higher mean occupational wages in low-skilled occupations. Immigrants from Brazil and Eastern Europe sort into different firms. Immigrants from both groups have a higher share of co-workers from their region of origin than would be predicted by a random assignment of workers to firms. This result is not driven by the immigrants’ sorting across occupations, industries, regions, and years. Although Brazilians share a common mother tongue with natives, they are not less segregated from natives across workplaces than Eastern Europeans. This result is not consistent with language driven productivity spillovers driving workplace segregation.

18

References Al´ıcia Adser`a and Ana M Ferrer. The effect of linguistic proximity on the occupational assimilation of immigrant men. CLSRN Working Paper, (144), 2014. Al´ıcia Adser`a and Ana M Ferrer. The myth of immigrant women as secondary workers: Evidence from canada. The American Economic Review, 104(5):360–364, 2014. Catalina Amuedo-Dorantes and Sara De La Rica. Complements or substitutes? task specialization by gender and nativity in Spain. Labour Economics, 18(5):697–707, 2011. Fredrik Andersson, Monica Garcia-Perez, John Haltiwanger, Kristin McCue, and Seth Sanders. Workplace concentration of immigrants. Demography, 51(6):2281–2306, 2014. Olof ˚ Aslund and Oskar Nordstr¨om Skans. How to measure segregation conditional on the distribution of covariates. Journal of Population Economics, 22(4):971–981, 2009. Olof ˚ Aslund and Oskar Nordstr¨om Skans. Will I see you at work? ethnic workplace segregation in Sweden, 1985–2002. Industrial & Labor Relations Review, 63(3):471– 493, 2010. Abdurrahman Aydemir and Mikal Skuterud. The immigrant wage differential within and across establishments. Industrial & Labor Relations Review, 61(3):334–352, 2008. Michael Baker and Dwayne Benjamin. The performance of immigrants in the Canadian labor market. Journal of Labor Economics, 12(3):369–405, 1994. Stefan Bender and Till von Wachter. In the right place at the wrong time: The role of firms and luck in young workers’ careers. American Economic Review, 96(5):1679–1705, 2006. Eli Berman, Kevin Lang, and Erez Siniver. Language-skill complementarity: returns to immigrant language acquisition. Labour Economics, 10(3):265–290, 2003. George J Borjas. Assimilation, changes in cohort quality, and the earnings of immigrants. Journal of labor Economics, 3(4):463–489, 1985. Santiago Budr´ıa and Pablo Swedberg. The impact of language proficiency on immigrants’ earnings in Spain. IZA Discussion Paper, (6957), 2012.

19 Anabela Carneiro, Nat´ercia Fortuna, and Jos´e Varej˜ao. Immigrants at new destinations: how they fare and why. Journal of Population Economics, 25(3):1165–1185, 2012. Barry R Chiswick. The effect of americanization on the earnings of foreign-born men. Journal of Political Economy, 86(5):897–921, 1978. Barry R Chiswick and Paul W Miller. The complementarity of language and other human capital: Immigrant earnings in Canada. Economics of Education Review, 22(5):469– 480, 2003. Barry R Chiswick and Paul W Miller. Occupational language requirements and the value of english in the US labor market. Journal of Population Economics, 23(1):353–372, 2010. Barry R Chiswick and Paul W Miller. The impact of surplus skills on earnings: Extending the over-education model to language proficiency. Economics of Education Review, 36: 263–275, 2013. Barry R Chiswick and Paul W Miller. International migration and the economics of language. Handbook of the Economics of International Migration, 1A: The Immigrants, 1:211, 2014. Barry R Chiswick and Gaston Repetto. Immigrant adjustment in israel: literacy and fluency in Hebrew and earnings. IZA Discussion Paper, (177), 2000. Barry R Chiswick and Sarinda Taengnoi. Occupational choice of high skilled immigrants in the united states. International Migration, 45(5):3–34, 2007. Christian Dustmann and Francesca Fabbri. Language proficiency and labour market performance of immigrants in the UK. The Economic Journal, 113(489):695–717, 2003. Christian Dustmann and Joseph-Simon G¨orlach. Selective outmigration and the estimation of immigrants’ earnings profiles. Technical report, CESifo Working Paper, 2014. Christian Dustmann and Arthur Van Soest. Language and the earnings of immigrants. Industrial & Labor Relations Review, 55(3):473–492, 2002. Christian Dustmann, Albrecht Glitz, and Uta Schnberg. Job search networks and ethnic segregation in the workplace. IZA Discussion Paper, (5777), 2011.

20 Albrecht Glitz. Ethnic segregation in Germany. Labour Economics, 29:28–40, 2014. David A Green and Christopher Worswick. Immigrant earnings profiles in the presence of human capital investment: measuring cohort and macro effects. Labour Economics, 19(2):241–259, 2012. Paulo Guimaraes and Pedro Portugal. A simple feasible procedure to fit models with high-dimensional fixed effects. Stata Journal, 10(4):628, 2010. Judith K Hellerstein and David Neumark. Workplace segregation in the united states: Race, ethnicity, and skill. The Review of Economics and Statistics, 90(3):459–477, 2008. Susumu Imai, Derek Stacey, Casey Warman, et al. From engineer to taxi driver? language proficiency and the occupational skills of immigrants. Ryerson Working Papers, 2014. Mario Izquierdo, Aitor Lacuesta, and Raquel Vegas. Assimilation of immigrants in spain: A longitudinal analysis. Labour Economics, 16(6):669–678, 2009. Sherrie A Kossoudji. English language ability and the labor market opportunities of hispanic and east asian immigrant men. Journal of Labor Economics, pages 205–228, 1988. Robert J LaLonde and Robert H Topel. The assimilation of immigrants in the US labor market. In Immigration and the workforce: Economic consequences for the United States and source areas, pages 67–92. University of Chicago Press, 1992. Darren Lubotsky. Chutes or ladders? a longitudinal analysis of immigrant earnings. Journal of Political Economy, 115(5):820–867, 2007. Jorge Malheiros and Alina Esteves. Diagn´ostico da popula¸c˜ ao imigrante em Portugal. Desafios e potencialidades. Alto Comissariado para a Imigra¸ca˜o e o Di´alogo Intercultural, 2013. Alfonso Miranda and Yu Zhu. English deficiency and the native–immigrant wage gap. Economics Letters, 118(1):38–41, 2013. OECD. Jobs for Immigrants (Vol.2): Labour Market Integration in Belgium, France, the Netherlands and Portugal. OECD, 2008.

21 Krishna Pendakur and Simon Woodcock. Glass ceilings or glass doors? wage disparity within and between firms. Journal of Business & Economic Statistics, 28(1):181–189, 2010. Friedrich Schneider. Size and measurement of the informal economy in 110 countries. In Workshop of Australian National Tax Centre, ANU, Canberra, 2002. SEF. Relat´orio de Imigra¸c˜ao, fronteiras e asilo 2012. Servi¸co de Estrangeiros e Fronteiras, 2013.

22

0

5

10

Percent 15

20

25

30

Figure 1: Distribution of O*NET importance of language scores Brazilians

1

2 3 4 ONET score importance of language

5

0

5

10

Percent 15 20

25

30

Eastern Europeans

1

2 3 4 ONET score importance of language

5

23

0

5

10

Percent 15

20

25

30

Natives

1

2 3 4 ONET score importance of language

5

Notes: The figures represent the distribution of workers according to the O*NET importance of language scores in their occupation. The data for natives is for 2002 and for Eastern Europeans and Brazilians for their first year in the data. Source: Quadros de Pessoal, 2002 to 2009

24

Figure 2: O*NET importance of language scores and mean wage by occupation for immigrants compared with natives Brazilians

Eastern Europeans

25

Natives

Notes: The figures represent the correlation between the mean real wages and the O*NET importance of language score for natives in 2002 and for immigrants of the two groups in the first year in the data. Each observation represents an occupation at the 3-digit level, and is weighted by the number of workers in the occupation. The dotted line is a local mean smoother using an Epanechnikov kernel function. Source: Quadros de Pessoal, 2002 to 2009 O*NET data set

26 Table 1: Population Selected Means EastEur. 33

Brazil 30

Natives 39

By Region Alentejo Algarve Centro Lisboa Norte

7.85 19.96 22.69 34.92 14.58

6.10 11.03 14.78 55.77 12.31

5.53 3.67 22.24 29.01 39.54

By Industry Manufacturing Construction Wholesale and retail trade Hotels and restaurants Transport, storage and communication Real estate, renting and business act. Other community social and personal service act.

19.69 47.39 8.14 3.76 4.91 13.30 1.53

9.14 36.50 12.18 19.97 4.23 14.01 2.64

31.33 18.21 19.09 3.86 8.48 8.67 2.26

By Occupation Clerks and related workers Service workers, shop and market sales workers Craft and related trades workers Plant and machine operators and assemblers Elementary occupations

2.47 3.51 41.02 11.20 36.59

4.28 22.17 32.09 7.84 27.77

11.27 8.46 32.02 17.00 10.96

Number of Workers Number of Observations

29 792 84 726

24 282 58 396

1 648 062 7 277 494

Age

Notes: This table shows the mean age for immigrants from Eastern Europe and Brazil and their distribution by region, industry, and occupation, all upon entry in the labor market. Immigrants are compared with all native men. All the differences in means between groups are statistically significant. Source: Quadros de Pessoal, 2002 to 2009.

27

Table 2: Wages of Brazilians and Eastern-Europeans relative to natives

Brazil EastEur. ysm EastEur*ysm

Age dummies Indiv.FE Year FE Region FE Industry FE Occupation FE Firm FE N R2

OLS (1) (2) -0.307*** -0.124*** (0.003) (0.002) -0.361*** -0.144*** (0.002) (0.002) 0.013*** 0.011*** (0.001) (0.001) -0.002* -0.003*** (0.001) (0.001)

(3) -0.055*** (0.002) -0.096*** (0.002) 0.002*** (0.001) 0.000 (0.001)

yes no yes no no no no

yes no yes yes yes yes no

yes no yes

7408503 0.087

7408503 0.602

(4)

FE (5)

(6)

0.012*** 0.010*** 0.006*** (0.001) (0.001) (0.000) 0.002** 0.002*** 0.002*** (0.001) (0.001) (0.000)

yes yes

yes yes yes no no no no

yes yes yes yes yes yes no

yes yes yes

yes yes

7408503 0.783

7408503 0.119

7408503 0.143

7408503 0.945

Notes: Robust standard errors are in parentheses. The dependent variable is the log hourly wage expressed in 2002 euros. *** p<0.01, ** p<0.05, * p<0.1 Source: Quadros de Pessoal, 2002 to 2009.

28

Table 3: Immigrant own-group (and other group) exposure Panel A Own-group exposure Exposure Expected exposure Over-exposure ratio Other group exposure Exposure Expected exposure Over-exposure ratio

Brazil

East

22.8 6.1 3.7

22.8 8.4 2.7

4.4 4.2 1.0

3.6 3.4 1.0

Panel B Brazil Observed Segregation Isolation Index 22.8 Exposure Index 0.6 Observed Segregation Index 22.2 Random Segregation Isolation Index Exposure Index Random Segregation Index Effective Segregation Standard Deviation

East 24.7 0.7 24.0

5.4 0.8 4.6

7.4 1.0 6.4

18.2 0.002

15.6 0.002

Notes: Panel A presents the exposure, expected exposure and over-exposure ratio of immigrants to others in the same immigrant group and to the other group as defined in the text. Panel B presents the co-worker segregation indices of immigrants of the two groups relative to natives as defined in the text. The over-exposure ratio and the random segregation index are conditional on the workers’ distribution across regions, industries, and occupations, all measured at the 3-digit level. Source: Quadros de Pessoal, 2007.