Regional labour market effects of immigration on low-skilled workers: the case of Sweden 1990–2003

Introduction

Rising figures on foreign labour supply have been part of the general development in the EU15, where, according to the European Labour Force Survey, the average share of immigrants in the labour force has increased from 10.7 to 16.9% between 1995 and 2018. Sweden is no exception. On the contrary, within the European Union (EU), Sweden stands out as one of the most prominent receivers of refugees, and one of the major changes on the Swedish labour market over the past 30 years has consequently been the growing share of foreign workers. By the mid-1990s, 8.8% of the working-age population was foreign born, while the figure today is 23.6% of the total labour force. The expansion of the labour force over this period, caused by an increase in the number of foreign born of working age from 0.5 million to 1.5 million, has been the major driver behind expansion of the total labour supply in Sweden (SCB, 2019a, 2019b).

The labour market consequences of immigration for hosting communities have become a major interest among policymakers and researchers. The impact of immigration on the labour market has been controversial. This especially concerns the extent to which the inflow of migrants helps or harms less-skilled natives. Studies on immigration in the USA have shown different findings. While Card (2009), for example, found only minor effects on wages, Borjas (2003) showed that immigration is harmful for natives' wages. In turn, Ottaviani and Peri (2012) demonstrated positive wage effects for natives.

Studies on Europe equally have presented mixed findings. In their study on the EU15 and the European Economic Area (EEA) countries in the 1980 and 1990s, Angrist and Kugler (2003) showed how immigration had a harmful effect on natives' employment rates, especially among the younger parts of the workforce. In their follow-up study on the EU15 in the 1990s and early 2000s, Amuri and Peri (2011) found less evidence for reduced employment rates among natives. Concerning the Spanish case, González and Ortega (2011) showed that unskilled immigrants did not impact the employment or wages of unskilled natives, while a study on France revealed that rising immigration actually helped improve wages and the employment rate among natives (Ortega and Verdugo, 2014). From the UK, there is evidence for a negative impact of new immigrants on previous immigration cohorts, but less of an effect on natives (Manacorda et al., 2012). In turn, a study on foreign labour inflow in German border regions identified adverse effects on natives' employment rate and wages (Dustmann et al., 2017). Similar results were found for Norway, where natives' wages in particular were responsive to inflow from neighbouring countries, but less so to inflow from developing countries (Bratsberg et al., 2014). Hence, the international literature provides rather mixed results.

As shown in a recent literature survey (Edo, 2019), Swedish evidence on the impact of immigration has been rather limited, given the influx of immigrants, and one cannot automatically assume that findings from other countries are generalisable to all countries, as institutional differences influence the extent to which the labour market is affected by immigration (Andersson et al., 2019). Sweden is one of the countries with the most notable growth in its foreign-born population during the past 30 years. In particular, refugee immigration is responsible for a growing part of the population, as Sweden is among the countries accepting the largest number of refugees in proportion to population, both during the more recent Syrian war and during the Balkan war of the 1990s.

The aim of the present paper is to examine the effects of immigration on the labour market outcomes of low-educated natives (i.e. residents without a university diploma). For this purpose, we focus on the case of Sweden in the 1990s – a period with a major influx of refugee immigration following the Balkan wars. Our reason for analysing the impact on low-educated natives is that they are the group of native workers most likely to be directly affected by an increasing exogenous labour supply. Moreover, recent studies have demonstrated the role of complementarity within education groups as well as specialisation of native workers in response to immigration (Ottaviano and Peri, 2012; Peri, 2012). The present study will help broaden and deepen our understanding of such mechanisms by examining groups less likely to benefit, such as less-educated natives.

While there have been manifold studies on the relationship between the arrival of immigrants and the labour market outcomes of native workers, our ability to convincingly measure the causal effects of immigrants on native workers is limited (Foged and Peri, 2015). Considering the measurement and selection problems associated with this exercise, we use longitudinal matched employer–employee data for Sweden for the period 1990–2003. The rationale for this period is twofold. First, following the Balkan wars, Sweden experienced a rapid increase in refugees from the former Yugoslavia. Compared to Foged and Peri (2015), who used a policy reform that redistributed immigrants in Denmark as an identification strategy, we use real inflows to select regional labour markets (or homogenous commuting areas: FA regions) that received the largest shares of refugees from 1993 onwards. Focusing on the regional dimension is imperative, as national figures, employed by most previous studies, cannot discern the local variation in supply change or its potential effect. Second, while the absorption of immigrants into the labour market is usually facilitated in periods of capital expansion (cf. Tabellini, 2020), this particular period coincided with a deep recession with rising unemployment figures (Eriksson and Hane-Weijman, 2017). Despite this, Ekberg (2016) showed exceptional labour market integration among Bosnian immigrants during this period, something that has not been seen for other immigration groups in other periods. Hence, compared to the period of relative growth during 2015/2016, when economic conditions were relatively more favourable, the period analysed here is characterised by a relatively low labour demand. With these data, we are not only able to identify the exact time (year) of immigrant inflows, but also follow them across time and space, and control for both short- and long-term effects of immigrants on low-skilled natives' labour market outcomes.

Our difference-in-differences estimations indicate three main findings. First, similar to other studies, our analysis shows that low-educated native workers respond to the arrival of immigrants with an increase in real wages. Second, although employment prospects in general worsened for low-skilled workers in most regions during this period, this is not attributable to the regions experiencing the largest foreign labour supply shock. Third, there are indications that low-skilled natives in immigration-rich regions are more likely to change workplace, particularly in combination with moving upwards in the wage distribution.

The remainder of the paper is organised as follows. Section 2 presents the theoretical arguments for how immigration affects low-educated natives. Section 3 describes the data, variables and empirical model. Section 4 presents the results, and section 5 presents the conclusions.

Conceptual framework

The impact of immigration on earnings and employment is often studied within the framework of standard economic models. Starting from Roy's (1951) model of occupational self-selection, it could be argued that optimal matching of skills and occupations cannot be readily assumed, as sorting into different labour market segments depends on the distribution of skills, the correlation of skills in the population and the technology available to apply the skills in a productive manner. When applying the Roy model to the case of immigrant earnings in the USA, Borjas (1987) showed that the expected income of immigrants depended on the economic and political conditions in their country of origin, hence not only the skills of each respective migrant. Thus, western Europeans immigrating to the USA would do better than those from developing countries, even if individual skills were equal. Given that immigrants originate from relatively economically buoyant countries, the supply of low-skilled or unskilled immigration is likely to harm the labour market outcomes of less-skilled natives (cf., Borjas, 1987). The argument holds that an increasing supply of labour due to immigration will cause a wage reduction among native labour competing for the same jobs immigrants are seeking, especially if immigrants are supplied in a closed labour market (Borjas, 2006).

Immigration is also often thought to change the skill composition of the workforce. If immigrants have a very different skill composition than natives, which could be the case, especially in relation to a large influx of refugees, the impact of immigrants will be heterogonous across skill groups (Borjas, 2003; Ottaviano and Peri, 2012). Drawing from a competitive labour model, it is expected that immigration will lower the wages of native workers who are most similar to the immigrants due to the supply shock, while increasing the wages of groups of workers most different from the immigrants (Dustmann et al., 2016). The latter group will benefit from the improved complementary of skills, raising their marginal productivity. In turn, immigration is not expected to impact natives if the skill distribution is perfectly similar, as only labour supply will increase, keeping the relative supply of skills unchanged.

In the long run, however, the heterogeneity or homogeneity of skill composition is not expected to have an impact on mean wages in the economy (Borjas, 2003). For the native population, there will still be “winners” and “losers” as long as the skill distribution of immigrants plays a role in creating complementarity effects for some and substitution effects for others (Borjas, 2014). Nor will the accumulation of capital to the pre-immigration capital/labour ratio resort the initial wage polarisation as long as the degree of complementarity of the real capital stock is the same for both low- and high-skilled workers.

The harmful effects are commonly associated with a reduction in wages. But, other outcomes are also likely to occur. One expected response among competing (native) workers to the influx of immigrants is to move to different regions seemingly less affected by a supply shock (Aydemir and Borjas, 2011; Borjas, 2006). Borjas (2006) even suggested that native migration flows diffuse the impact of immigration across the entire economy, hence making native internal migration decisions sensitive to immigrant-induced increases in labour supply and wages. Borjas further demonstrated that native migration explains about 40% of the gap in wage effects. The labour market competition model, thus, suggests that immigration increases the supply of a certain type of labour and that, depending on the skill sets of the immigrants and natives, this can affect not only wages but also employment and mobility. In the short run, the higher inflow of immigrants should lower the outcomes of competing workers and increase those of complementary workers (Aydemir and Borjas, 2011; Edo, 2015).

The standard models of immigration-induced labour market impacts have more recently been extended, showing that low-skilled immigration could have much less impact on low-skilled natives than previously anticipated. Akin to Roy's (1951) early notion on the role of technology, one line of argument is that firms, by changing their technology in response to immigration, mitigate the impact on native workers. This is because firms will adjust their technology to take advantage of the relatively more abundant supply of skills following immigration. By allowing capital to be complementary to high skills and supplementary to low skills, Lewis (2011) showed that firms will respond to immigration by investing in capital following an inflow of skilled workers and disinvest after an inflow of low-skilled workers. Wages will then return to pre-immigration levels when the capital stock has adjusted for the change in skill composition.

Immigrants with similar education and working experience, however, may not be perfect substitutes for native workers. Better communication abilities, cultural and social capital specific to a national context and discriminating norms tend to give native workers a competitive advantage compared to immigrants with seemingly similar skills. In theory, natives will then specialise in high-communication-intensive tasks, while immigrants will specialise low-communication-intensive work (Foged and Peri, 2015; Peri and Sparber, 2009). Any negative effect of wages would then arise among previous immigrants who are substitutes for new waves of immigrants (Manacorda et al., 2014).

The extensions from the standard model allow for an adjustment to immigration shocks through two channels, first, through investments in new capital and, second, through the occupational mobility of natives. Assuming that capital and high-skilled labour are complements, only the latter effect is expected to increase the wages and employment of natives, such as low-skilled workers taking advantage of language or other context-dependent skills.

A number of approaches are being exploited to argue for the various findings on natives' labour market outcomes. Borjas (2003) invented an approach that measures the same skill sets for natives and immigrants to enable comparisons within the same skill group. Altonji and Card (1991) used a spatial variation approach to capture the total immigrant flow across regions and examine the effect of immigration. Card (2001) later developed a variation on the spatial variation approach by allowing variation in the composition of immigrant inflows by education and region. The skill approach and the mixed approach seek to identify a relative wage/employment effect of immigration, whereas the spatial approach seeks to quantify the total wage/employment effect of immigration. Both approaches are sensitive to the assumption that foreign and native labour with the same human capital are placed in the same cell. There is a risk of measurement error concerning education and skills, but also a risk of immigrants being downgraded upon arrival, leading to misclassification. As recommended by Dustman et al. (2016), we apply the spatial approach here by assessing the total effect of immigration on a specific group of native workers (the low educated). This is done in an attempt to understand whether groups with similar characteristics are affected differently depending on whether the regional labour market is subject to a sudden supply shock of immigrants.

Data

The data used in the present study are from a matched employer–employee database covering the entire Swedish population. The database contains annual reports on firms (size, profitability, location, etc.) and on individuals (place of birth, year of immigration, age, employment status, educational level, real wages, etc.). The relational characteristics of the database allow us to link data on individuals to their workplaces and the specific industries and regions in which they work.

From this database, we focus on low-skilled Swedes selected according to the following criteria: first, selected individuals who were between 18 and 64 years old (potential labour force or working-age population, who were not attending school between 1990 and 2003). With this population of potential labour force, we further restricted our sample to workers between 20 and 50 years in 1990 to avoid right-censoring and restricting new workers from entering the group. A person will continuously be in the sample unless (s)he becomes incapacitated, retires, leaves Sweden (no longer in the Swedish tax registers) or dies during the 1990–2003 period. This restriction prevents noise caused by the entry and exit of individuals. Furthermore, we restricted our analysis of outcomes to a panel of Swedish-born individuals without a university education (three years or longer university education). As argued above, we believe that this group of natives are those who are most likely to be at immediate risk of competing with immigrants. All of these selections give a total of 1,033,028 individuals whom we can observe for the period 1990–2003 (note that, due to the factors mentioned above, we do not have a perfectly balanced panel with a total of year-person observations summing up to 11,058,364). Figure 1A displays the share of low-educated natives by region. The three metropolitan regions (Stockholm in the East, Göteborg in the West and Malmö in the South) and other large regional centres have the lowest percentages of low-educated natives, while more sparsely populated regions in the East have higher shares.

There are several options for defining the regional shares of refugees. If we use official registers, we cannot know in which region immigrants were first distributed when seeking asylum but only where they obtained their first employment. Hence, the potential relocation between distributed location and later selected location cannot be assessed. To account for this shortcoming, we have calculated the share of Yugoslavian-born residents in each region using the 2018 official population statistics from Statistics Sweden. In so doing, we assume that the regions with high shares in 2018 were also the regions in which many Yugoslavians found employment and settled during the 1990s. Compared to using micro-data to ascribe each person a region based on their first registry in Sweden, we thereby reduce the effect of spatial relocation immediately after being assigned a residence permit. Ekberg (1995, 2011) showed that the reallocation effect levels off after a few years, after which time immigrants are just as likely to move as same-age natives.

Variables

Empirical model

The empirical model employed in the empirical analysis is a difference-in-differences model (cf. Card and Kreuger, 1994), which is a statistical technique that mimics an experimental research design by studying the differential effects of a treatment on a treatment group vs a control group. Difference-in-differences calculates the effects of a treatment (explanatory variables) on an outcome (response variable) by comparing the average change over time in the outcome variable for the treatment group, compared to the average change over time for the control group. Essentially what difference-in-differences does in the present case is to take the average change in the outcome variable (real wage, employment, mobility) of the control group from 1990 to 2003 and the average change in the same outcome variable of the treatment group from 1990 to 2003 and highlight the differences between the groups defined above. Consequently, we are able to estimate the average changes in the respective outcome variable for low-educated native workers in regions characterised by low shares of Yugoslavian refugees compared to those for low-educated native workers in regions characterised by high shares of refugees before and after the supply shock in 1993. The idea behind difference-in-differences estimation is that it allows us to control for many other factors that might cause endogeneity. An alternative would be to employ propensity score matching to identify the “statistical twin” of each low-skilled native in high-immigration regions. However, as our treatment group indicator is the region of work, and not the individual per se, it is difficult to find observable regional indicators that can effectively explain why a certain labour market region receives high- or low-shares of refugees. Hence, like Card (2001) did, we resort to a spatial difference-in-differences approach as follows:

Although income is the sole continuous variable, while employment and mobility are binary variables, we applied an ordinary least squares (OLS) regression on all variables. Hence, employment and the two mobility variables are linear probability models. Although the standard approach would be to use logit or probit functions on these variables, the main rationale for using this type of model is that it facilitates comparison across models and allows interactions, something that is generally difficult in a logit or probit setting (see Mood, 2010; Wooldridge, 2002, for extensive discussions on the pros and cons of linear probability models).

We also used the panel structure of the dataset in a fixed-effect estimation on natives' labour market outcomes. An important feature of this approach is that it uses the panel structure of the data by accounting for within transformation and for unobserved heterogeneity among individual workers (e.g. unobserved ability or aspiration). The problem with this particular model is that it cannot handle time-invariant variables (i.e. treatment group), and hence, it was only estimated as a robustness check.

Results

Over the years, Sweden has been a destination for economic and political migrants. With the exception of Nordic immigrants, all other immigration categories have consistently shown the tendency to increase, especially immigrants from other parts of Europe. The inflow of Yugoslavians (Serbia, Croatia, Montenegro, Kosovo, Slovenia, Bosnia, Herzegovina, Macedonia) to Sweden began with the migrant work agreement signed between Sweden and the Yugoslav government in the 1960s to deal with the severe labour shortage in primarily the manufacturing sectors. The Balkan war in the early 1990s saw another influx to Sweden, first, with the arrival of refugees and, second, with family reunification. In all, about 110,000 persons from the former Yugoslavia applied for asylum during this period. Despite this unpreceded influx of refugees to Sweden, which coincided with a deep recession, as a group these immigrants did relatively well on the labour market, showing high employment rates (Ekberg, 2016). For this reason, the potential competition for available jobs intensified, which may have had adverse effects for low-skilled natives.

Similar to what is shown in Table 1, Figures 2 and 3 map the means of each dependent variable (left) before the treatment period (1990–1992) and the respective change before and after treatment (1993 onwards) to the right. The highest incomes for low-educated workers are found in Stockholm (East) and Gothenburg (West), but also in the mining regions in the very north. Moreover, employment rates are far higher in, and around, the three largest regions in the south, and due to the ongoing recession, the employment figures for low-educated workers also worsened in all regions, but less severely so in the southern parts of Sweden (many of which belong to the treatment group of regions). This regional pattern of wages and employment can be contrasted with mobility, as general mobility is mainly higher in the northern part of the country and upward wage mobility more evenly distributed.

Figure 4 shows the employment shares for all natives, irrespective of educational level (A), and all low-educated native workers (B) before the supply shock in 1993 and after the supply shock 1993–2003 (treatment period), based on their position in the wage distribution (five groups equal in size, low to high, left to right). While the distribution of natives overall is largely unaffected over time (A), we find signs of a moderate upgrade for low-educated natives. Although the share is increasing in all income groups, this is particularly true for income Groups 2, 3 and 4.

So, while the descriptive evidence indicates a slight upward trend in the share of low-educated natives' relative position in the wage distribution after the supply shock, we still have not assessed whether this is the case in a multivariate setting. Consequently, we turn our attention to Table 2. The main variable of interest in this table is the interaction between treatment period and treatment group (coefficient β3 in equation (1)). Models 1, 2, 3 and 4, respectively, indicate the association between immigration and low-educated native workers' real wages (1), employment (2), mobility (3) and mobility into higher income job (4).

Controlling for individual, sectorial and regional attributes, the estimates show that low-educated workers in immigration-rich regions (treatment group) have positive wage development, but also that the treatment period in general is associated with increasing incomes. Concerning employment (model 2), the treatment period that also is associated with a recession has a general negative association with employment. This is expected, as the recession in general influenced the working prospects of particularly low-skilled workers (Eriksson and Hane-Weijman, 2017). Our data also support the idea that low-educated natives are more likely to obtain gainful employment in more immigration-exposed regions. However, we cannot find any evidence to show that low-skilled native workers respond to the inflow of low-skilled immigrants by changing workplace (model 3), as the interaction term is insignificant at the same time as the main effects of both treatment period and group are significantly negative. However, for the low-skilled natives who do change workplace, this mobility could of course both be due to complementary as well as substitution effects, as some native workers either are crowded out and need to look for new employment or move to new, better paid jobs. The former process is confirmed in model 4, as the upward mobility of workers is less likely to occur during the treatment period in general, but particularly so in the treatment regions.

Our findings showing that the income development of low-educated natives is relatively favourable after the supply shock, which is in contrast to previous findings identifying negative (e.g. Borjas, 2003; Dustmann et al., 2017; Angrist and Kugler, 2003) or negligible (e.g. González and Ortega, 2011; Manacorda et al., 2012) effects of immigration on natives. In turn, we find support for previous findings, suggesting that low-skilled natives transition into high-wage jobs (e.g. Foged and Peri, 2015; Peri and Sparber, 2009; Ortega and Verdugo, 2014). One reason could be that native workers have specialised more in communication-intensive tasks than have immigrants, as suggested by Foged and Peri (2015) and Peri and Sparber (2009).

Conclusions

The aim of the present paper was to assess how immigration affects low-educated natives' real wages, employment and income mobility. Using Swedish longitudinal data, we analysed the immigration of refugees from the former Yugoslavia following the Balkan war in the early 1990s. Although this coincided with a period of recession, previous studies have identified the remarkable labour market success of these immigrants relative to other immigration waves (Ekberg, 2016). However, we have very limited knowledge about whether this relative labour market success also implied a crowding-out effect of native workers, as the scarcity of jobs was relatively low.

Despite the general assumption that immigration imposes an adverse effect on native workers' wage and employment opportunities, our most robust finding was that low-educated native workers responded to the inflow of immigrants with an increase in real wages. This was, however, mainly driven by a slow, but steady, increase in real wages after the deep recession. Nevertheless, this finding does support the argument that the arrival of immigrants might cause, in the short to medium term, a downward pressure on real wages and employment of natives with competing skills (e.g. Aydemir and Borjas, 2011; Edo, 2015). Second, our analysis shows no clear evidence of a significant decline or reduction in employment of low-educated natives. On the contrary, the year-on-year analysis shows that natives respond to immigration with an increase in employment. Third, low-educated natives in immigration-rich regions are more likely to change job. Whether this mobility leads to upgrading or crowding out depends on the regional context. In smaller regions, job mobility seems to be associated with a lower probability of moving upwards in the wage distribution (despite a general positive effect of real wages). By contrast, in larger regions, this mobility is associated with upward mobility into a higher-income group. Hence, from a regional perspective, this supports the claim that in economies with relative capital adjustment (expansion), upgrading is more likely, while the pressure on real wages increases in less expanding markets (Tabellini, 2020). Hence, in expanding markets, low-educated natives are more likely to respond to the arrival of immigrants by leaving their former job (e.g. Foged and Peri, 2015; Peri and Sparber, 2009). In effect, the arrival of immigrants is likely to force low-skilled natives to upgrade their skill portfolio.

The findings of the present study have some implications for policy. First, apart from immigration fuelling economic growth, it must also be noted that the arrival of immigrants has an overall positive effect on native workers in the long term. Therefore, the findings presented here refute some of the rhetoric on more recent years' waves of refugee immigration, according to which the arrival of refugees only induces negative effects on real wages and heighten risk of unemployment. Second, if the arrival of immigrants is likely to induce negative effects on real wages and heighten risk of unemployment among low-educated native workers in the short term, as suggested in other studies (e.g. Aydemir and Borjas, 2011; Borjas, 2003), policies should aim at income redistribution to curtail the effects of wage fall and risk of unemployment rather than at impeding immigration. Finally, as immigration is part of human existence, migration policies ought to be reformed with the intent of encouraging higher education or skill specialisation to shorten the period of short-term wage fall and risk of unemployment. This is particularly important because recent studies suggest that the correlation of an expanding low-wage sector and refugee immigration is the highest in countries with low potential for social mobility and low rates of investment (Andersson et al., 2019).

In summary, we have presented a detailed analysis of the regional effects of immigration on low-skilled native workers. This is important because the proportion of immigrants is seldom evenly distributed across a nation. It is, therefore, not very likely that a low-skilled native worker will respond to a supply shock of immigrants in another part of the country. Instead, these effects tend to be more local. The present analyses have, in this regard, presented a more comprehensive picture of the impacts of immigration by tracking the labour market effects on natives' real wages, employment and job mobility. We hope that future analyses of the impact of immigration will consider multiple labour market outcomes and both short- and long-term effects.