The role of labour market integration in migrants’ decisions about family reunification: a comparative study of Polish migrants in Norway, Sweden, and the UK

With about 2.7 million nationals residing elsewhere in the European Economic Area, Poland was the second largest country of origin of all intra-European migrants in 2018. After the country’s accession to the European Union in 2004, some Polish migrants have brought abroad their families, while others continue living alone and commuting to visit their families in Poland. Based on a sample of 1153 survey respondents in three European contexts—Norway, Sweden, and the United Kingdom—this article investigates from a comparative perspective how Polish migrants’ family reunification status is related to their integration on the labour market. Results show that only in the Scandinavian labour markets, and not in the British, is migrants’ economic integration decisive for family reunification. This article argues that in contexts which offer broader employment opportunities for family migrants, who tend to be women, such as in the UK between 2004 and 2016, the lead migrant’s labour market integration may not be crucial for family reunification. The findings may be of special interest for migration policymakers in the Scandinavian countries working towards greater gender equality outcomes among its immigrant populations.


Rationale and research question
The successive enlargements of the European Union [EU] in 2004, 2007, and 2013 resulted in a community of 28 member states with large differences in the levels of economic development between countries (Eurostat 2019a). Notable disparities in job availability, wages, and living costs across the member states, combined with the EU freedom of movement, have contributed to increased labour migration within the EU and the European Economic Area [EEA]. With about 2.7 million Polish nationals residing elsewhere in the EU/EEA 1 in 2018, Poland was the second, after Romania, most significant country of origin of all intra-European migrants (Eurostat 2019b). Increasingly more and more Polish migrants are settling abroad together with their families, either by emigrating together, through family formation in the immigration country, or via reunification with existing family members (White 2017;Friberg 2012). Whereas some migrants decide to bring their families to their new countries, others continue living alone and commuting to their families in Poland. Drawing on the rational choice perspective in household migration decision-making (Lundberg and Pollak 2003) and looking at two distinct groups of Polish migrants (reunited vs. non-reunited) in three European contexts-Norway, Sweden, and the United Kingdom [UK]-this article investigates to which extent migrants' labour market integration is conducive to reunification with their partners in the destination country.

Study contexts
Featuring some important differences and similarities in their immigration policy, experienced economic growth, functioning of their labour markets and welfare systems, Norway, Sweden, and the UK make a compelling case for studying labour market integration and family reunification of Europe's second largest internal migrant group. Firstly, Norway, Sweden, and the UK are members of the EEA which provides the legal framework for free movement for their respective citizens. The conditions for both the labour and family migration of EEA nationals within the area are therefore very different from those of third-country nationals. Hence, it is reasonable to conceptualise intra-European migration as a largely unrestricted migration. Notwithstanding, when Poland joined the EU in 2004, some "old" EEA member states adopted different transitional provisions vis-à-vis immigration from the "new" EU member states in Central and Eastern Europe [CEE]. Of the three countries in this study only Sweden chose to apply Community rules for labour mobility which implied no restrictions on the employment and immigration from the new EU member states (Boeri and Brücker 2005). Conversely, Norway introduced compulsory work permits for a five-year transitional period whereas the UK launched the Workers Registration Scheme 2 responsible for issuing eligibility certificates for the workers from the EU members states in CEE (Home Office 2004a in Boeri and Brücker 2005). Notwithstanding, immigration from Poland responded to such different transitional provisions in a counterintuitive way. Despite their more restrictive transitional provisions, Norway and the UK experienced an unprecedented increase of the number of Polish-born residents between 2004 and 2017: from 7000 to 97,000 in Norway (Statistics Norway 2019) and from 75,000 to 908,000 in the UK (ONS 2019). The growth in Sweden, on the contrary, was less spectacular with its Polish-born population merely doubling from 42,000 to 89,000 in the same period (Statistics Sweden 2018).
Secondly, work being the main reason for the post-2004 immigration from Poland, the dynamics and regulations of the national labour markets at both the origin and destination are without a doubt essential for understanding these migration flows. The robust economic growth which Norway and the UK experienced throughout most of the 1990s and 2000s resulted in an expansion of the labour-intensive job markets for migrants, such as construction, manufacturing, and services (Friberg and Midtbøen 2019).
The Swedish economy, on the contrary, stagnated throughout most of the 1990s and experienced a far more moderate development in the early 2000s (Gerdes and Wadensjö 2008), which arguably left fewer jobs for migrants from the new EU member states. Furthermore, the varying degree of regularisation of labour markets in the three countries of study may have played an important role for the dynamics of Polish labour migration and family reunification after 2004. On the one hand, Norway and Sweden feature highly regularised labour markets with strong trade unions, industry-specific minimum wage negotiations, high levels of job security, but, at the same time, more restricted access to permanent employment. On the other hand, the British labour market is known for its less regulated recruitment and firing practices, higher labour rotation, and very meagre National Minimum and National Living wages. Given that this article's aim is to test whether the place of residence of migrant's spouse/partner is related to their economic integration, it is worthwhile to compare contexts which differ in terms of their labour markets.
Finally, the three countries exhibit notable differences with regards to how their social protection systems operate. Both Norway and Sweden exhibit similar features of the socio-democratic welfare state model (Esping-Andersen 1990), such as the universal membership, generous coverage of social risks, heavily subsidised education and early childcare system. On the contrary, the more liberal welfare state regime in the UK is characterized by means-tested social benefits, more limited access to childcare support, etc. In addition, the rapidly changing political landscape in the post-Brexit UK, and the associated uncertainty among intra-European migrants, makes the UK context especially interesting for analysis.

Theoretical and empirical perspectives on family reunification
Family and migration decision-making Family migration, both as a field of research and an area of migration policies, has been of a particular interest to both migration scholars and policymakers. Issues such as skills composition, employment rates, and fertility trends of the family migrants, and (inter)marriage choices and educational attainment of their descendants have been especially relevant due to their implications for the host countries' labour market, public social spending, and society in general. Research into these domains has therefore produced a large body of literature and knowledge which arguably reflects the prevalence of the host countries' nation-state perspective. With an exception of some classical (e.g. Borjas and Bronars 1991) and more recent studies (Baizán et al. 2014), it is, however, less common for research on family migration to make migrants' decision-making visà-vis family reunification its central research question. Consequently, there is a need for more knowledge about the socio-economic processes that shape migrants' decisions about family reunification and how family reunification in turn shapes the migrants' socio-economic situation.
Notwithstanding, the general literature on migration decision-making can provide relevant hypotheses in relation to this question. The two main theoretical perspectives on migration decision-making, both anchored in the rational-choice paradigm, part from each other in how they regard the role of the family in the mobility decisionmaking. Whereas neoclassical microeconomics emphasises the individual perspective and views migration exclusively as a strategy aiming at maximizing individual net benefits (Todaro 1980), studies within the New Economics of Labour Migration [NELM] school of thought shift the focus to the meso level and conceive migration as a strategy aiming at maximizing household's income or utility (Sandell 1977;Mincer 1978;Stark 1991).
The unrestricted nature of inter-state migration in the U.S. American context provided rich data for studies that pioneered the field of couple and household migration decision-making during the 1960s-1970s. Thus, from the neoclassical microeconomic perspective, single persons were expected to migrate if the move would increase their individual utility, most commonly conceptualised as income. Couples, on the contrary, appeared less mobile compared to single persons, which was attributed to concerns over couple integrity (Mincer 1978;Stark 1988;Clerge et al. 2017). However, not only the marital status of (potential) migrants, but also their work status at the place of origin was found to be an important factor when family migration was concerned (Sandell 1977;Mincer 1978;Nisic and Melzer 2016). Hence, Mincer (1978) observed that dual-income families were more reluctant than families with only one income to migrate because of a potential loss of the other spouse's income. He observed that couples would not necessarily migrate even if the move was likely to increase one of the spouses' utility, resulting in what he called a "tied stayer" situation. At the same time, some couples migrated despite the loss in one of the spouses' utility (as a rule, that of the wife), resulting in what is known as a "tied mover" or "trailing wife" phenomenon (Mincer 1978). Further, Sandell (1977) noticed that the longer the spouse worked at his/her current job, the less likely the family was to migrate.
However, not only the current work status of the spouses and their job tenure in the home country was found to influence family migration, but more so their current income and the expected change in it due to migration (Sandell 1977). Because of uneven gendered structures of labour markets in sending and receiving contexts, household migration was found to result in different outcomes for men and women. Hence, most of the research on household migration from the U.S. American context in the 1960s-1970s largely agreed that family relocation usually improved the husband's but not the wife's earnings (Sandell 1977;Mincer 1978;Stark 1988). Similarly, a more recent study by Nisic and Melzer (2016) shows that, despite a full or partial loss of wife's income, more egalitarian dual-income East German couples tend to migrate to West Germany, where the gender gap in earnings is larger, more often than less egalitarian West German couples move within the region. They argue that the economic structure of West Germany, more favourable for men, enabled East German men to overcompensate for the income losses of their female partners and, thus, dominate the mobility decision-making process. This appeals to the need to acknowledge the importance of host countries' different labour market structures and institutions for family migration and reunification. Cohen and Kogan (2007) argued that the greater regularisation of the labour market and the more generous welfare state in Germany compared to Israel can explain why Jewish immigrants from former Soviet Union there lag behind those in Israel in terms of both their initial labour market performance (unemployment level and occupational status) and the subsequent labour market progress. With regards to different elements of the institutional context in these two countries, the operation of the labour market was found to be more important than the formal policies promoting economic integration of immigrants (Haberfeld et al. 2011).
NELM scholars argue further that including other household members in studying migration decisions contributes to a more nuanced understanding of the processes which shape such decisions, as not only individual considerations of the person who moves, but also those of the spouse/partner, children, and/or other family members must be taken into account (Bailey et al. 2004;Bushin 2009;Clerge et al. 2017). Family defined broadly, care obligations towards children and elderly were found to have a clear effect on household mobility decisions (De Jong 2000;Bailey et al. 2004;Bushin 2009). Couples with children, and more so couples with school-age children, were found to be less mobile than couples without children (Long 1972;Mincer 1978;Bushin 2009). Since family migration may impose costs on children, such as an interrupted schooling or terminated friendships, families with school age children were found to be less likely to migrate compared to families with children in the pre-school age (Sandell 1977).

Family reunification as a two-stage bargaining mobility game
Before proceeding, the key concept of this article, namely family reunification, must be defined. For the purposes of concept operationalisation, this article delimits family reunification to only situations wherein migrants reunite with spouses or partners in the host country. Reunification with other family members is therefore beyond the scope of this article. It also excludes cases wherein couples/families migrated together and cases of family formation after immigration, as these do not constitute family reunification.
In the framework of this paper, couples' decisions about migration and family reunification are conceptualised as a two-stage bargaining game drawing on Lundberg and Pollak's (2003) two-earner couple location problem. Reflecting the type of decision that couples need to agree upon at each stage, these are called hereafter "migration" and "reunification" stages. The migration stage represents a situation wherein one of the spouses decides to move abroad whereas the other decides to stay. In contrast to Lundberg and Pollak (2003), who conceptualise such outcome as divorce, I argue that spouses may agree on migration to be a temporary solution for, at least, two reasons. Firstly, they may see it as a temporary strategy aiming to improve family's economic situation in the short-term, wherein the migrating partner is expected to return home after some time spent abroad. Secondly, couples may plan to relocate the whole household to a new location in the long-term, but decide to start the process in a step-wise manner where the leading spouse migrates first, in line with current theories of stepwise family migration (Friberg 2012). The reunification stage represents a situation wherein the spouses must make the decision on whether to reunite or not. This can be illustrated as shown in Table 1 below. Similarly to how the location decision is made in Lundberg and Pollak (2003), here the outcome will also depend upon each spouse's preference ordering of these four possible outcomes: SG (Stay; Go), SS (Stay; Stay), RG (Return; Go), and RS (Return; Stay). Given that our data were collected among leading migrants currently living abroad, the analytical part of this article focuses on just two outcomes: the SG and the SS. The hypothetical outcomes, RG and RS (reunification in the home country), are beyond the scope of our interest.
As in the case with the couple's migration decision-making, family reunification in the country of immigration entails both material and non-material costs and benefits (Sjaastad 1962;Mincer 1978;Stark 1991). Potential material benefits of the family reunification for the migrant household can be the maximization of the household income, if the spouse/partner is likely to find a better-paid job in the immigration country. It can also free one from the burden of maintaining two homes, one at origin and another at the destination, reducing the costs associated with living separately. Similarly, family reunification is likely to bring along material costs, such as the relocation costs, having to upgrade one's residence in order to accommodate for a bigger family, costs associated with learning a new language and country-specific skills by the partner and/or children. Some of the non-material benefits of family reunification can be: improved emotional wellbeing of the partner and/or children, who no longer have to deal with the separation; better educational and professional prospects for children; upward social mobility; and an improved social life of the whole family. However, family reunification can also bring about non-material costs such as: the risk of de-skilling for the partner; and fragmentation of the social relationships with the extended family and friends left in the home country. Importantly, different members of the migrant household are likely to experience the costs and benefits associated with the family reunification differently (Borjas and Bronars 1991). As is in the case of couples' migration decision-making (Mincer 1978), the migrant household will tend to make the decision about family reunification as long as its combined material and non-material benefits will be greater than the associated material and non-material costs, and provided that spouses prefer remaining married.

Hypotheses
Migrants' labour market integration is operationalised in this paper by two variables: the incidence of unemployment and the type of contract the migrants held. From the rational choice perspective, individuals or households engage in the decision-making under the condition of imperfect information. In the context of this study, this means that migrants may run the risk of miscalculating potential costs associated with the family reunification. One may, for example, realise that one was overly optimistic about how quickly their spouse or partner would find employment or learn the new language. One strategy when making a decision under the imperfect information is to minimise potential risks. Migrants may therefore wish to first secure oneself a stable employment and save up enough money before they decide to bring their partner and/or children. From this point of view, frequent transitions into unemployment are likely to impede the migrant from securing permanent employment and, thus, postpone the decision about family reunification. Since the incidence of unemployment can serve as an Table 1 The reunification stage of the family migration-and-reunification game as an adaptation of Lundberg and Pollak's (2003)  indication of the experienced vulnerability on the jobs market and the fragmentation of migrants' career path since their immigration, migrants who have experienced more incidents of unemployment are expected to be less likely to have their partner/spouse reside with them in the immigration country (Hypothesis 1a). Furthermore, researchers agree that workers on temporary contracts generally feel less secure about their employment than those on permanent contract (Näswall and De Witte 2003;Chung 2019). In line with the logic outlined earlier, migrants with a permanent contract are expected to be more likely than migrants on temporary or no contract to have their partner/spouse live in the same country (Hypothesis 1b). With regards to the comparative perspective, one can expect elements of immigration contexts, such as the structure of immigrant-dominated job markets, to play a role in Polish migrants' different family reunification patterns in the studied countries. It is therefore reasonable to expect different employment outcomes among Polish men and women in the different labour market contexts which Norway, Sweden, and the UK provide. As a result, the tendency of the households to reunite will depend on both the leading spouse's possibility to compensate for the potential partial or full loss of the tied mover's earnings directly upon the reunification, as well as the tied mover's chance to quickly find employment. Looking separately at Norway, Sweden, and the UK with their different labour market contexts will enable us to test whether labour market integration of the leading spouse matters more in some national contexts than in others (Hypothesis 2).

Online survey
The data used in this article come from a large web survey conducted in April-May 2017 among Polish migrants in Norway, Sweden, and the UK. The survey used two strategies for participant recruitment: (1) an advertisement with a low-cost airline operating flights between Poland and the three countries of study and (2) a sponsored Facebook advertisement. Both the airline and the Facebook ad promised participants to enter a draw of valuable prizes, among which a trip worth PLN 10,000 (USD 2,800) and five iPads minis.
The airline campaign targeted all passengers flying with a low-cost carrier to Poland from Norway, Sweden, or the UK during 1-21 April 2017. The airline ad was a 738-by-339 pixels picture automatically placed in passengers' self-printed boarding passes with a short text about the survey, the logo of the funding body, and the link to the website hosting the survey. The ad was placed in 113,114 boarding passes (whenever passengers did online check-in) and, additionally, in 113,082 confirmation emails (whenever they booked a flight during the 3 weeks of the advertisement campaign).
As the airline campaign did not yield the expected response, a paid Facebook advertisement was run for 30 consecutive days between April-May 2017. The Facebook campaign consisted of a single ad which targeted all Polish-speaking Facebook users aged 20-65+ located in Norway, Sweden, or the UK. The objective of the campaign was to direct targeted users to an external website which hosted the survey. The ad consisted of an 80-s video explaining the goals of the research with a short text in Polish about the eligibility criteria, prizes, and the link to the survey. In order to ensure a broader potential reach of the ad, the campaign did not make use of Facebook's detailed targeting features (e.g., selecting potential audience based on their demographics, interests, or behaviours), contrary to Pötzschke and Braun (2017) who, in a similar study on Polish migrants in four European countries, narrowed down their selection criteria to only Polish "expats", which according its Facebook definition might have excluded certain categories of migrants (cf. Pötzschke and Braun 2017: 639). In our study, Facebook estimated the campaign's target population, or "Potential Reach" in Facebook's terms, at 80,000 users in Norway, 69,000 users in Sweden, and 800,000 users in the UK. These numbers are largely consistent with the official numbers on Polish immigrants in the three countries, mentioned earlier, with the UK figure being almost double the figure estimated in Pötzschke and Braun (2017). Concerning the performance of the Facebook strategy, the ad was on average displayed to 28.9% of its targeted users, however, this varied from 23.8% in the UK, 54.2% in Norway, and 59.4% in Sweden (see a detailed summary of the Facebook ad performance in Table 2). The unique link Click-Through-Ratio [CTR], the number of unique Facebook users per 1000 targeted users who performed a click on the link inserted in the ad, was on average 33.9 and varied between 29.8 in the UK, 31.8 in Norway, and 55.0 in Sweden. For comparison, Pötzschke and Braun (2017) reported their ad to be delivered to 18.8% of its targeted users, with the unique link CTR at 35.3, similar to that in our study.
The survey collected 5552 full and partial responses during the combined period of the two campaigns. The absolute majority of 3806 respondents who answered the question on how they learnt about the survey (i.e. full responses) were recruited via the Facebook ad (n = 3552, 93.2% of all), whereas the remainder came via the airline campaign (n = 181, 4.7%) or friends/other (n = 76, 2.0%). The survey itself comprised 12 sections with different types of questions, starting with participants' sociodemographic data and personal migration history, questions on labour market and family situation, and ending with questions on lifestyle and plans for the future. Given the responsive design of the survey, its length varied between 40 and 60 questions, and it took respondents on average 15 min to complete it, with the median time being 11 min, 50 s. The collected sample was weighted according to the respondents' country of residence (Norway, Sweden, or the UK), gender (male or female) and four age groups (20-29 years old, 30-39 years old, 40-49 years old, or 50 years and over) using the official data on marginal distributions of Polish migrant populations in Norway and Sweden and the reported data on Polish-speaking Facebook users in the UK, where official registerbased statistics were unavailable. In Facebook Ads' language, Potential Reach estimates the number of unique Facebook users who meet the targeting criteria of the ad. Reach refers to the number of unique Facebook users who saw the ad at least once and is different from Impressions which may include multiple views of the ad by the same users. Impressions specify "the number of times the ad were on screen" and, hence, is the product of Reach by Frequency, the average number of times each unique user saw the ad. Unique Link Clicks refer to the number of unique Facebook users who performed a click on the ad. CTR stands for Click-Through-Ratio and measures number of clicks per 1000 Impressions. Unique CTR can be interpreted as a response rate to the ad

Sample
The sample analysed in this article consisted of Polish-born respondents, aged 20-65, with up to 13 years of residence in Norway, Sweden, or the UK and who, at the time of the survey, were either married to or in a civil partnership with another person from Poland. Respondents with more than 13 years of residence in the immigration country were excluded from the analysis because they represent a cohort of migrants who had emigrated from Poland before 2004, thus under a different migration regime than the one that opened up after Poland's admission into the EU. Respondents who stated family reunification with their partner or spouse as the main reason for their migration were treated as "trailing" family migrants and, therefore, excluded from the sample. Further, respondents who said that they had moved abroad at the same time with their spouse/partner did not meet the definition of family reunification used in this study and, therefore, were excluded as well. Finally, in order to distinguish family reunification from cases in which family or partnership was established after immigration, we excluded respondents who, at the time of immigration, were below 20 years old, thus, unlikely to be married, and those who had reported to have met their spouse/partner in the country of immigration. Hence, the final subsample comprised 1153 respondents who had moved abroad for reasons other than family reunification, among whom were 227 women (19.7%) and 926 men (80.3%).

Variables
In this paper, the dependent binary variable 'Partner's residence', designating the place of residence of respondent's spouse or partner (0different country, 1same country), was modelled as a step-wise logistic regression on two key variables representing migrants' labour market integration: the incidents of unemployment and the type of contract held by respondents. The models were controlled for respondents' sociodemographic characteristics (age, gender, level of education, and parenthood) and contextual factors (residence length, urbanisation, and country of residence). Table 3 summarises the variables used in the modelling.
The choice of age, gender, level of education (university degree), parenthood, length of residence, and urbanisation as control variables is motivated by the insights found in earlier studies (Baizán et al. 2014;Haug 2008;De Jong 2000;Bailey et al. 2004;Bushin 2009;Mincer 1978). With regard to age, older migrants may be confronted with a number of factors that can prevent their spouse/partner from joining them abroad. For example, the partner's established career in the home country can delay family reunification or even prevent the relocation abroad. An older spouse or partner may also feel reluctant to relocate due to the fear of losing his/her social networks in the home country (Haug 2008). As a result, migrants' age is expected to be negatively associated with the odds of family reunification. Furthermore, as migration has an uneven effect on men and women (De Jong 2000), gender is likely to play an important role for family reunification. Even today, women throughout the world to a much larger degree than men are expected to cater to the social needs of the family, especially when it comes to children and the elderly. Such care expectations may exacerbate further with age, as women may face the need to take care of their own, and sometimes of their partner's, ageing parents. The odds of family reunification of male respondents are therefore expected to be lower than those of female respondents. Concerning respondents' education level, migrants with a university degree can be regarded as having more human capital, a factor which may facilitate family reunification with their partner. Further, given that a longer duration of migration increases the rate of union dissolution (Davis and Jennings 2018), couples may try to prevent such outcome. Therefore the odds of family reunification are expected to increase with the length of residence in the immigration country. The urbanisation level of the area where migrants reside can be an important factor for migrants' decision about family reunification for a number of reasons. Firstly, larger urban areas tend to offer broader and more diverse employment opportunities not only for migrants, but potentially also for their partners/ spouses, a factor which may facilitate family reunification. Secondly, they also tend to offer migrants larger co-ethnic networks and, thus, availability of co-ethnic community institutions, such as Sunday language schools, which may be important for migrant families with children. However, due to inflated housing prices living in large metropolitan areas can be more expensive than in smaller cities or towns, an additional cost factor which may discourage family reunification. Finally, having a child is likely to reduce the odds of family reunification, as children represent the partner's care obligations in the home country.

Descriptive results
Table 4 below provides descriptive summaries of the weighted overall and national subsamples according the variables outlined above. As it shows, the share of respondents who reported their spouse/partner reside in the same country varied greatly between Concerning the key variables representing migrants' labour market integration, the mean incidence of unemployment in the whole sample was reported at 0.656 times per respondent whereas the share of those holding a permanent contract comprised 88.4%. At the national level, however, respondents in Norway appear to have experienced more incidents of unemployment than those in Sweden or the UK (mean 0.89 vs. 0.60). With regards to the type of contract, respondents in the UK seemed to hold a permanent contract more often than those residing in Norway or Sweden (90.3% vs. 82.7% vs. 83%). With regards to demographic characteristics, respondents in Norway appear to be slightly older (mean age 41 years) than those in the UK (40 years) or Sweden (39 years). The share of male respondents in the subnational samples ranged from 77.3% in the UK and 87.7% in Sweden to 94.1% in Norway. The overrepresentation of men in the national subsamples reflects different gendered immigration flows to the three contexts, as well as the fact that Polish men tend to initiate the process of step-wise family migration (Friberg 2012). What concerns higher education, the share of respondents who reported holding a university degree varied between 17.6% in Sweden, 19.1% in Norway, and 23.9% in the UK. Given that the analysed samples comprised only respondents who, at the time of taking the survey, were either married or in a civil cohabitation, it comes as no surprise that the overall majority of respondents were parents  Missing values for Unemployment (n = 237, or 20.6%) and Contract (n = 397, or 34.3%) have been imputed by fitting a quasi-Poisson and a multinomial regression models, respectively, based on respondents' socio-demographic and contextual characteristics. The imputation was done independently for the national subsamples (80.9% in Sweden, 81.7% in the UK, and 87.9% in Norway). Concerning contextual characteristics, the length of residence reported by the respondents also varied across the contexts, reflecting different historical dynamics in the Polish immigration to these countries. Thus, 45.4% of all respondents in the UK (vs. 35.4% in Norway and 28.7% in Sweden) lived in there between nine and 13 years. Quite opposite, the share of respondents with up to 5 years of residence in the immigration country was largest in Sweden (48.9%) followed by Norway (35.9%) and the UK (34.0%). This gives grounds to view Polish migrants in the UK and, to some extent in Norway, as more established compared to those in Sweden. With regards to the level of urbanisation, the majority of respondents in Norway (65.3%) and Sweden (51.3%) resided in areas with up to 50,000 inhabitants, compared to only 34.7% in the UK. The share of those who reported living in areas with over 500,000 inhabitants ranged from 9.7% in Norway, 21.8% in the UK, and 22.7% in Sweden.

Multivariate results
Table 5 below summarises the results of four logistic regression models (Models 1-4) run on the overall sample and three separate models run on the national subsamples (Norway, Sweden, UK).
In Model 1, we test the role of experiencing unemployment and permanent contract for the place of residence of respondents' partner or spouse. For the reference group, which in this model consists of informants without permanent contract and who did not experience unemployment in the country of immigration, the share of those with the spouse/partner living in the same country is equal to Constant / (1 + Constant) = 0.63. Model 1 shows that there is a significant (p < 0.05) positive correlation between the permanent contract and the place of residence of respondents' spouse. Thus, holding a permanent contract is associated with a 75.3% increase in the odds of family reunification, compared to the reference group. The incidence of unemployment, however, does not seem to be strongly correlated with the predicted variable. Each incident of unemployment is associated with a 7.8% reduction in the odds of family reunification.
Model 2 adjusts the relationships identified in Model 1 for respondents' age, gender, education, parenthood, length of residence, and area's urbanisation level. Now, the reference group comprises male immigrants with a length of residence up to 5 years, residing in urban areas with population between 50,000 to 500,000 people, without children, university degree, permanent contract or prior experience of unemployment in the immigration country. Controlled for other variables, each incident of unemployment is now associated with a significant (p < 0.01) reduction in the odds of family reunification by 16.9%. However, holding a permanent contract is now only weakly positively associated with family reunification, and the association is no longer statistically significant.
In the next step, Model 3, the country of residence is added as a covariate. In this model, the respondents in the reference group are identical to those in Model 2 except that they reside in Norway. Controlled for other variables, it appears that Polish migrants with the same socio-demographic characteristics are significantly more likely to be reunited with their spouse or partner in Sweden and the UK, compared to Norway.
All other covariates maintain the direction and significance of their association with the predicted variable, except female gender which is no longer significant.
In Model 4, two interaction terms are added: between the respondents' country of residence and (1) incidence of unemployment and (2) type of contract. The reference group in this model remains the same as in Model 3. As shown in Table 5, the odds ratio associated with each incident of unemployment decrease from 0.864 in Model 3 to 0.721 in Model 4, signalling a more significant (p < 0.01) negative relationship between the incidence of unemployment and the odds of family reunification in the Norwegian context. Now, each incident of unemployment leads to a 27.9% reduction in the odds of family reunification, compared to the reference group, whereas holding a permanent contract appears to show no significant association with the outcome variable. The When the interaction between age and parenthood in the Swedish subsample is taken into account, age is significantly negatively associated with the odds of family reunification, similarly to the other two contexts. The reason why parenthood is more significantly associated with lower odds of family reunification in Sweden is likely due to the Swedish sample being on average younger and less recent in terms of the length of residence compared to the other two subsamples * p < 0.1; ** p < 0.05; *** p < 0.01 interaction terms, however, show that the strong negative relationship between the incidence of unemployment and family reunification in Norway becomes slightly mediated in two other countries, suggesting that unemployment in Sweden or the UK plays a less prominent role in migrants' propensity to reunite. Similarly, the interaction term between the permanent contract and residing in Sweden or the UK slightly moderates the relationship identified in the Norwegian case. When controlling for the two interaction terms, migrants in Sweden are less likely, though not significantly, to be reunited with their spouse or partner when compared to the reference group in Norway (35.3% lower odds). On the contrary, in the UK, compared to Norway, the odds of family reunification are significantly 168.9% higher. Finally, the three last columns in Table 5 feature Model 2 applied to the national subsamples. As we can see from the coefficients, unemployment reduces the odds of family reunification in all three contexts, significantly (p < 0.01), however, only in the Norwegian case (32% reduction in odds for each incident of unemployment). On the contrary, holding a permanent contract in the Swedish context is significantly positively associated with being reunited with one's spouse or partner (136% higher odds).

Discussion and conclusions
Looking at three European contexts with substantial numbers of Polish migrants-Norway, Sweden, and the UK-this article has investigated how migrants' integration on the labour market is related to their family reunification status. Hypothesis 1 posited that migrants' labour market integration, operationalised in this paper by the incidence of unemployment and the type of contract migrants held, was conducive to reunification with their partners in the destination country. Firstly, migrants who have experienced more incidents of unemployment were expected to be less likely to have their partner/spouse reside with them in the immigration country (Hypothesis 1a). Secondly, migrants with a permanent contract were believed to be more likely than migrants on temporary or no contract to have their partner/spouse with them (Hypothesis 1b). The results confirm both parts of the Hypothesis 1, but only in the Scandinavian contexts. In Norway, migrants who have experienced more incidents of unemployment appear to be less often reunited with their spouse/partner, supporting Hypothesis 1a. Similarly in Sweden, respondents with a permanent contract are more likely to have their spouse/ partner reside in the same country, in line with Hypothesis 1b. In the UK context, however, neither the incidence of unemployment nor the permanent contract appear to significantly predict the place of residence of migrants' spouse/partner. This confirms the Hypothesis 2 which postulated that the labour market integration of the leading spouse may matter in some national contexts more than in others. In addition, the analytical models presented in this article controlled the tested relationships for migrants' socio-demographic (age, gender, level of education, and parenthood) and contextual characteristics (length of residence, level of urbanisation, and country of residence). As expected, immigrants' age and their length of residence are clearly the most significant predictors of family reunification across all studied contexts. Urbanisation appears to be an important factor in the UK context where living in either the areas with up to 50, 000 inhabitants or the cities with over half a million residents is associated with significantly lower odds of family reunification. On the contrary, urbanisation does not appear to have a clearly strong impact on family reunification in Scandinavia.
Interestingly, the effect of parenthood, which is negatively associated with odds of family reunification across all contexts, is particularly strong in Sweden where Polish migrants seem to be on average younger and more recent when compared to Norway or the UK.
What can explain why economic integration of Polish migrants is a significant predictor of their family reunification status only in the Scandinavian contexts but not in the British? In line with the household perspective in the theory of migration decisionmaking (Sandell 1977;Mincer 1978;Stark 1991), the household's economic security upon reunification will depend on incomes of both spouses (or partners). It is therefore reasonable to expect that at the reunification stage of the family migration-andreunification game, described in this article, spouses will take into account not only the current (and future) economic situation of the leading spouse but also the trailing spouse' chances of quickly finding employment upon the reunification. Since our data did not include information about spouses' employment participation immediately upon reunification, we were unable to test this hypothesis in this study. However, according Statistics Norway, the employment rates within the first year of immigration among Polish-born women aged 20-66 years varied between 44.3% in 2006 and 65.2% in 2016, oscillating around 50% in the period (Statistics Norway 2020). It is important to bear in mind that this rate included both single and married Polish female migrants 20-66 years, independent for their reason of immigration, which may involve the issue of positive self-selection. Figure 1 below provides insight into total employment rates among Polish female immigrants in the UK and in Norway during 2007-2017. Given that the majority of our national subsamples consisted of men, I argue that Polish female migrants who joined their spouses abroad in 2007-2017 were more likely to find employment in the UK than in Norway (or Sweden, data missing). This may explain why the lead migrant's labour market integration in the UK is not significantly associated with family reunification.
Further, the implications of permanent contract for individual economic security are likely to differ across the studied contexts. In a recent comparative study of 23 European countries, Chung (2019) argued that temporary workers in Sweden are still significantly more likely to feel employment insecurity compared to temporary workers in Norway or the UK. Though this may serve as an explanation of our results in the Swedish subsample, the scope of this article does not allow us to elaborate further on legal or economic implications of permanent employment across the three countries.
The key limitation of this paper is related to the availability of collected data, namely the limited use of the partner-related variables. Respondents' own age as the proxy of their partners' age and parenthood status are the only variables that bring household characteristics into our analyses. Future quantitative models of migrants' decisions about family reunification should incorporate more partner-related characteristics. Further, features of institutional contexts, such as the availability of employment opportunities for men and women, only briefly discussed here, or migrants' entitlements to family-related welfare services can be important for migrants' decision-making vis-à-vis family reunification. The scope of this paper has not allowed us to include contextspecific variables other than migrants' length of residence, levels of urbanisation, and country of residence with its interactions with unemployment and the type of contract. Future comparative research could, for example, look into migrants' concertation in different sectors of the job market and the availability of different welfare provisions for families with children (e.g. day-care centres). In addition, given the role of language skills for migrants' integration on the labour market, further research should control for respondents' proficiency in the host country's language. Finally, it is important to emphasise that this paper has focused exclusively on migrants who arrived to Norway, Sweden, and the UK between 2004 and early 2017. Therefore, Polish migrants who arrived under a different immigration regime before 2004 or after the data were collected in spring 2017 (especially relevant for the post-Brexit context in the UK) have not been included in the analyses.
To conclude, it can be argued that in countries with less regulated labour markets and broader employment opportunities for migrants' spouses, who tend to be women, like was the case in the UK in the period of 2004-2016, the leading migrant's labour market integration may not be crucial for family reunification to occur. On the contrary, in countries with a more restricted access to permanent employment and a lesser availability of jobs for family migrants, family reunification may be delayed, and in some cases never materialise, as leading migrants face a greater expectation to secure a stable and foreseeable employment. Given that it is predominantly men who migrate first and lead the process of family migration, such gendered labour market outcomes may have a particular impact on the distribution of gender roles within the migrant households. Family immigration into less regulated labour markets-ceteris paribus-may lead to a more egalitarian distribution of gender roles as it facilitates a double-earner family model. Family immigration into more protected labour markets, on the contrary, may result in the reinforcement of the traditional male bread-winner model. This finding may be of special interest for migration policymakers in the Scandinavian countries working towards greater gender equality outcomes among its immigrant populations.