Geographical trajectories of refugees in Sweden: Uncovering patterns and drivers of inter-regional (im)mobility

Many countries actively seek to disperse refugees to counteract residential segregation or/and take measures to attract and retain international migrants in smaller communities to mitigate or reverse population decline. This study explores the regional distribution and inter-regional mobility among refugees in Sweden. It uses individual-level register data to follow two cohorts for 8 years after their arrival in Sweden, distinguishing between refugees subject to a placement policy in the 1990s and recent cohorts that either had arranged their own housing or had been assigned housing.Itusessequenceanalysisandmultinomiallogitregressiontoanalyseregion-altrajectories,andeventhistoryanalysistoexaminemobilitydeterminants.Theresultsindicatethatmostrefugeesremainedinthesametypeofregionthroughouttheperiod.Asigniﬁcantproportionofrefugeeswithassignedhousinginlargecityorsmallcity/ruralregionsstayedthereoveralongperiod,suggestingthatrefugeesettle-mentpolicieshavelong-lastingconsequences.


Introduction
In the current context of large-scale refugee migration, the issue of refugee reception has become increasingly prominent in many receiving countries. Refugees' place of settlement in their new country is the subject of particular attention from researchers and policymakers alike. One obvious reason is that large inflows of refugees in certain locations can affect local resources and constrain the capacity of local institutions to provide infrastructure and services. At the same time, there are persistent concerns that the concentration of newly arrived refugees in already immigrant-dense areas hampers their socioeconomic integration (Musterd et al. settlement and spread them geographically, with a greater or lesser degree of coerciveness. Policies of compulsory dispersal of asylum seekers have been implemented in the UK, Denmark, and Switzerland, among others. Sweden applied a refugee dispersal policy between 1985 and 1994. As will be explained, the Swedish refugee settlement policy has also recently been reformed towards increased governmental steering of asylum seekers' initial settlement location. In addition to the dispersal of refugees, many immigrant-receiving countries seek to attract and retain international migrants in smaller communities as a way to mitigate or reverse population decline in rural regions. Initiatives towards the 'regionalization of immigration' have been undertaken in Australia, Canada, and New Zealand, among others (Akbari and MacDonald 2014). In Sweden, there is also a lively political debate on the role of international migrants as a potential solution to the demographic and economic decline in rural areas (Bergstro¨m and Rehnvall 2014; Parliamentary Committee on Rural Development 2016; see also Hedberg and Haandrikman 2014 for a review of the academic literature on international migrants in rural areas).
Despite the increased policy attention on the issue, relatively little is known about the settlement patterns and relocation of recent refugee cohorts in Sweden. What type of regions does refugees settle in and relocate to? The aim of this study is thus to explore the patterns and drivers of (im)mobility for different refugee categories. In particular, we consider refugees who arrived in the early 1990s and were subject to the Sweden-wide dispersal strategy, and compare them with refugees from the late 2000s, distinguishing between those who arranged for their own accommodation upon being granted their residence permit and those who were assigned housing by the authorities. We structure our analysis in two steps. First, we apply sequence analysis to longitudinal individual-level register data, in order to identify refugees' typical geographical trajectories across different types of regions in terms of population density and labour market conditions. We also assess how individual-and neighbourhood-level characteristics are associated with refugees' likelihood of following the identified trajectories. Second, we examine the likelihood and drivers of refugees' inter-regional mobility, specifically their first move away from the region of initial settlement.
The study contributes to the literature on internal migration of refugees in Sweden in the following ways. First, it gives prominence to refugees' regional context and inter-regional mobility. The majority of Swedish studies have focused on moves between municipalities (for example Å slund 2005; Aradhya et al. 2017). Municipalities indeed play an important role in refugee reception, as they provide assistance with housing and are responsible for the organization of language courses. However, refugees' regional patterns and inter-regional mobility deserve more attention. Indeed, regions often correspond to local labour market areas (LMAs), which makes them a relevant unit of analysis with regard to refugee labour market participation, which is a core dimension of their overall integration in the host country. Individuals usually seek employment within their local LMA (or commuting zone), and not only within their municipality (or neighbourhood). There is indeed substantial empirical research showing that the regional context plays a significant role in refugees' labour market participation (Bevelander 2011;Andersson 2016;Ruist 2018;Vogiazides and Mondani 2020). The second contribution is that this study is the first, to our knowledge, to use sequence analysis to examine refugees' long-term residential trajectories across regions with different socio-demographic characteristics. Such an approach is able to identify common patterns, while highlighting aspects of timing, ordering, and duration of states (such as different residential contexts). Finally, by comparing refugees who were subject to the dispersal policy with more recent cohorts, and comparing between refugees with assigned and self-arranged housing within the more recent cohort, we will gain knowledge on the impact of different refugee reception policies in Sweden. Our findings will provide insights into the long-term effect of refugee reception policies that should be useful to policy makers in Sweden and elsewhere.

Refugee Settlement Policy in Sweden
From the mid-1980s to the mid-1990s, Sweden implemented a refugee dispersal policy-the so-called 'Sweden-wide strategy' ('Hela Sverige strategin')-which aimed at offsetting the concentration of new refugees in large metropolitan areas and achieving a balance between municipalities regarding refugee reception. Asylum seekers were assigned to a municipality where they were to reside for an 18-month introductory period (Andersson and Solid 2003). However, refugees did not face any major costs for leaving their municipality of placement, aside from a delay in enrolment into introductory activities. Since July 1994, asylum seekers have been offered the possibility to choose between arranging for their own accommodation ('Eget boende', EBO) or being accommodated by the Swedish Migration Agency ('Anla¨ggningsboende', ABO). Once they obtain a residence permit, refugees can either be assigned housing in a municipality by the authorities ('Kommunanvisad') (under the condition that they lived in accommodation provided by the Swedish Migration Agency during their asylum procedure (ABO) or arrange for their own housing ('Egenbosatt') (SFS 1994). Usually, individuals who resided in a public facility as asylum seekers tend to ask for public assistance with their settlement in a municipality (Bevelander et al. 2019). Individuals who arrange their own housing usually stay with family members or acquaintances in metropolitan regions. Those who are accommodated by the authorities tend to reside in smaller municipalities (Statistics Sweden 2016). In recent years, there has been mounting criticism against the 'Own accommodation' legislation (EBO-lagen). It is argued that the freedom of asylum seekers to settle where they want leads to residential segregation and overcrowding (National Board of Housing, Building and Planning 2015). In January 2019, the newly formed social-democratic and green government decided to reintroduce restrictions to the freedom of asylum seekers to settle wherever they want. The new policy, which entered into force in January 2020, implies that asylum seekers who settle in certain socioeconomically disadvantaged areas are no longer eligible for financial assistance (SFS 2019(SFS : 1204. This decision is part of broader efforts to reform Sweden's refugee reception policy after the high inflows Geographical trajectories of refugees in Sweden 3 of asylum seekers, which culminated in 2015 when 160,000 individuals applied for asylum in the country.

Immigrants' Settlement and Internal Migration
International migrants tend to reside in cities; this pattern has been observed in many different contexts. In the US, it is estimated that 86 per cent of the country's foreign-born population lives in a large metropolitan area (Singer 2015). The corresponding proportion among OECD countries is 64 per cent (OECD 2018a). Sweden is no exception, as nearly 82 per cent of international migrants lived in a metropolitan or larger city region in 2018, whereas that was only the case for 65 per cent of the native-born population (Statistics Sweden 2019). While migrants have traditionally settled in a few large metropolitan cities, such as New York, Los Angeles, and Chicago in the US, recent research has paid increasing attention to new destinations, or so-called 'new immigrant gateways'. These include metropolitan regions that did not previously host many international migrants, but also smaller cities and rural communities (Waters and Jime´nez 2005;Hall 2013). In Sweden, for instance, research has shown that, compared to the 1990s, non-European migrants today are increasingly present in small and larger (albeit not metropolitan) cities (Malmberg et al. 2018). Rural living is more common among asylum seekers and refugees compared to other migrant categories. Indeed, as mentioned above, many host countries have implemented policies aimed at geographically dispersing asylum seekers and refugees. Recent research shows that asylum seekers in 18 European countries are less concentrated in urban areas compared to the overall population (OECD 2018b; Proietti and Veneri 2019). Asylum seekers are especially dominant in rural areas in Ireland, Finland, and Norway. In Sweden, 37 per cent of asylum seekers resided in a rural area in 2015, while this was only the case for 20 per cent of the Swedish population (Proietti and Veneri 2019: 22).
A large body of empirical literature analyses the importance of labour market considerations for immigrants' initial and subsequent location choices (see for example Zavodny 1999 in the US; Sapeha 2017 in Australia, Lymperopoulou 2013 in the UK, and Zorlu and Mulder 2008 in the Netherlands). Research from Sweden shows that employed migrants are less likely to move internally, whereas migrants receiving social allowances are more likely to do so (Rephann and Vencatasawmy 2000;Andersson 2012).
In post-industrial society, large cities commonly offer better employment opportunities than rural areas. Employment-motivated migration tends to be directed towards more urbanized areas, and particularly metropolitan areas. In that respect, Sassen (1991) hypothesized that global cities with strong service sectors attract highly educated workers within finance, IT, and research as well as low-skilled migrant workers seeking employment in industries such as catering and cleaning. Research has shown that refugees who were placed all around Sweden as part of a dispersal policy were more likely to move away from smaller localities with high unemployment (2005; Haberfeld et al. 2019). Qualitative studies have also stressed the importance of local labour market prospects for refugees' likelihood to remain in rural/small city areas over a long period of time (Stewart and Shaffer 2015;Gilhooly and Lee 2017;Fang et al. 2018).
Employment opportunities, however, are not the only motives for residing in cities. Cities are also considered attractive for their amenities, including educational institutions, public services, and cultural and leisure facilities (Glaeser et al. 2001;Buch et al. 2014). Moreover, cities with a large established foreign-born population tend to be more 'receptive' towards new migrants (Hall 2009).
For newly arrived migrants, cities may also be appealing due to the presence of previously established migrants, as already-established migrants can provide economic assistance, information, and psychological support to more recently arrived relatives, friends, or other members of their community of origin (Massey 1990). Co-ethnic concentration allows for the maintenance of cultural practices and may entail the possibility of employment opportunities within the ethnic economy (Peach 1996;Logan et al. 2002). In this paper, we use the term ethnic preference to describe the hypothesis that migrant groups may prefer to reside in proximity to co-ethnics. This approach goes against the theory of spatial assimilation, which posits that migrants, who initially concentrate in deprived and immigrant-dense neighbourhoods, later relocate to more affluent and majority-dominated areas, as their socioeconomic situation improves (Massey 1985). In Sweden, Å slund (2005) found that refugees subject to mandatory placement in the late-eighties were more likely to relocate to municipalities with a high proportion of co-nationals and immigrants overall. Some studies have also shown that refugees were less likely to leave municipalities (Aradhya et al. 2017) or regions (Haberfeld et al. 2019) with a high proportion of co-ethnics. Finally, Bevelander et al. (2019) showed that refugees in the 1998-2010 arrival cohort who arranged their own housing were more likely to reside in cities with a higher immigrant population.
The policy context at the time of arrival also determines the geographical patterns of refugees. Refugees who were subject to a compulsory placement policy are less concentrated in urban areas and more likely to migrate internally. For example, a larger proportion of placed refugees within the Sweden-wide strategy later moved away from their assigned municipality, usually relocating into metropolitan regions and large cities (Borgega˚rd et al. 1998;Å slund 2005;Statistics Sweden 2016). Such subsequent mobility may be motivated by a desire to reside close to relatives and other co-ethnics. In that respect, several qualitative studies have discussed how dispersal policies can potentially disturb co-ethnic networks and lead to social isolation (e.g. Larsen 2011 in Denmark;Stewart andShaffer 2015 in the UK, andPovrzanovic Frykman 2009;Ro¨nnqvist 2009 in Sweden). Subsequent mobility may also be motivated by employment considerations or amenities.
Finally, a large body of literature describes how residential mobility varies over individuals' life course. In general, younger, single, and highly educated people have a higher propensity to move (Mulder and Hooimeijer 1999). In addition, the longer a person resides in a place, the less likely she is to move. This is explained by the fact that, over time, individuals develop so-called 'location-specific insider Geographical trajectories of refugees in Sweden 5 advantages', which constitute disincentives to move. Work, family ties, and attachment to the place of residence can be such drivers of immobility (Fischer and Malmberg 2001). In light of these theoretical arguments and empirical evidence, we derived the following hypotheses: Hypothesis 1: Refugees who did not choose their initial region of residence will be more likely to experience an inter-regional move, especially if their region of initial settlement is a small city or a rural region. Hypothesis 2: Refugees who lack employment are more likely to experience interregional mobility. Hypothesis 3: Refugees who initially settled in the capital region of Stockholm are more likely to remain in that region over a long period. Hypothesis 4: Refugees residing close to co-ethnics will be less likely to leave their region of initial settlement.

Data and Methods
This paper analyses the geographical trajectories of refugees in Sweden using longitudinal, annually updated individual-level register data. The data come from a compilation of administrative registers managed by Statistics Sweden and include a number of demographic, socioeconomic, and residential variables. We follow the geographical trajectories of refugees over an 8-year period, starting from their arrival in Sweden. In the administrative registers used for this study, individuals' place of residence is registered annually, on 31st December. A person is thus considered to have undergone an inter-regional move during a certain year if he or she does not reside in the same region as the year before. The refugees in our dataset were between 18 and 58 years old at the time of immigration and they did not reside in their parental home during the 8-year follow-up period (N ¼ 80,801). Our study population consists of two refugee cohorts: the 1990-93 cohort, which was subject to the Sweden-wide dispersal strategy, and the 2005-09 cohort. For the latter cohort, we distinguish between refugees with assigned housing (Kommunanvisad) and refugees with self-arranged housing (Egenbosatt). This distinction allows us to assess the impact of settlement policy on refugees' geographical trajectories. As shown in Table 1, the vast majority of refugees originate from the Middle East, North Africa (MENA) and the Horn of Africa, especially from Iraq and Somalia (88 per cent of the 2005-09 cohort with selfarranged housing; 69 per cent of the 2005-09 cohort with assigned housing and 50 per cent of the 1990-93 cohort). Over a third of the refugees in the 1990-93 cohort came from Europe, especially the former Yugoslavia. The three refugee categories also differ in terms of the type of region of initial settlement. The majority of refugees who arranged for their own accommodation (either during the asylum process or once they are granted a residence permit) settled in Stockholm (39 per cent), followed by large city regions (29 per cent). Only 8 per cent settled in a small city or rural region. In contrast, large city regions were the most common type of first region among refugees who were assigned housing by the authorities, while small city/rural regions came in second place. Only 8 per cent of refugees from the 2005-09 cohort with assigned housing initially resided in Stockholm. It should be noted that the two refugee cohorts were faced with somewhat different sociohistorical contexts on arrival. Although housing shortage was already a problem in many Swedish localities in the 1990s, it worsened over the years. In 2019, 83 per cent of Sweden's 290 municipalities reported a housing shortage (National Board of Housing, Building and Planning 2019). In the end of the 1990s, the corresponding rate was 11 per cent (Emanuelsson 2015: 55). In addition, both the 1990-93 and the 2005-09 refugee cohorts arrived in Sweden during (or just before) an economic recession. Yet while the economic downturn of the early 1990s hit refugees severely, the 2008 crisis did not affect refugees' employment rates as negatively (Ruist 2018).
In the study, we employ two complementary methods for longitudinal data analysis to explore the patterns and drivers of refugees' inter-regional (im)mobility. In the first stage of our investigation, we use sequence analysis to identify patterns of refugees' long-term trajectories across regional contexts. Sequence analysis is a family of methods to analyse longitudinal data in terms of ordered series of states and transitions between states (Billari 2001;Cornwell 2015). The main advantage of sequence analysis is that it provides a comprehensive picture of Geographical trajectories of refugees in Sweden 7 refugees' geographical trajectories across different types of regions. In a second stage, we employ event history analysis to examine the determinants of refugees' mobility away from their region of first settlement. Event history analysis is well suited to causal inquiry because it allows for time-varying covariates in regression analysis (Eerola and Helske 2016). The present study has a number of data limitations. First, refugees enter Swedish registers once they are granted a residence permit. Register data do not include any residential data from the period of determination of the asylum application (Statistics Sweden 2016). Second, in our dataset, individuals' place of residence is measured annually on 31 December. This implies that the study probably underestimates the number of mobility events experienced by refugees, as refugees are known to be particularly mobile during their first year in the country (Statistics Sweden 2016). Finally, every study on determinants of regional mobility patterns should acknowledge the possibility that selection mechanisms are in place. For instance, when choosing their region of settlement, migrants might be influenced by individual characteristics that make them sort themselves into specific types of regions (Å slund 2005). We attempted to reduce these effects by studying the drivers of inter-regional mobility at the transition from region of first settlement, using event history analysis with lagged time-varying covariates, both at the individual and at the neighbourhood level. Furthermore, individual-driven selection effects of the kind described above should be less important for the 1990-93 cohort under the Sweden-wide strategy as well as for the refugees in the 2005-09 cohort with assigned housing, since in those groups the allocation into the region of first settlement is by definition not a choice of the individual. However, this does not completely eliminate the possibility of sorting into regions and neighbourhoods upon first settlement. Among the refugees of the 2005-09 cohort, there was also a likely self-selection into choice of arranging for their own housing or asking for assistance with housing arrangements. Those choosing the former may have more economic resources or social contacts that facilitate their housing arrangements.

Region Types
A growing number of studies have applied the method of sequence analysis to explore geographical mobility. The majority of them focus on the neighbourhood unit, investigating trajectories across the neighbourhoods with different levels of deprivation (e.g. Lee et al. 2017  . Note that this sample is smaller than the study population due to the additional requirement of residence in Sweden for 8 consecutive years. Regions correspond to local LMAs, which are annually constructed by Statistics Sweden taking into account commuting patterns. Local LMAs should not be confused with Sweden's 21 County Councils, which are named 'regions'. Although LMAs are annually updated, we opted for a single classification per cohort in order to study inter-regional mobility (similar to other studies, e.g. Haberfeld et al. 2019). By allowing LMA borders to vary, we would not be able to disentangle a change in LMA of residence from a change of the LMA borders. Given that our follow-up period is relatively short and that changes in commuting patterns unfold over a long period of time, this choice should not significantly affect our results. We used the 1998 LMA classification per for the 1990-93 cohort and 2013 classification for the 2005-09 cohort, as middle points in the 8-year follow-up period. We distinguished between five types of region: 1. Stockholm, 2. Gothenburg, 3. Malmo¨, 4. Large city regions, and 5. Small city/rural regions. Large city regions are LMAs that include a large city based on the 2017 classification of municipalities by the Swedish Association of Local Authorities and Regions. When an LMA includes both a metropolitan city and a large city, it is classified as a metropolitan region. It must be acknowledged that the three metropolitan regions as well as large city regions include sparsely populated areas. Yet, as previously explained, we classified entire regions as LMAs because they are the geographical unit of interest in the study. Sweden's three metropolitan regions are separate categories because of their different economic situations and political roles. Indeed, while Stockholm is Sweden's financial and administrative centre, Gothenburg and especially Malmo¨have faced social and economic challenges related to deindustrialization. Unemployment rates have been consistently high in Malmo¨, and the region also hosts a comparatively higher proportion of refugees (Vogiazides and Mondani 2020). Figure 1 shows the mean time (in years) that a refugee spent in each type of region, by refugee category. Note that this does not have to be consecutive years, just the accumulated number of years. The patterns are relatively similar for the 1990-93 cohort and the 2005-09 cohort with assigned housing, whereas the patterns of the 2005-09 cohort with self-arranged housing differ substantially. When interpreting the results, it is important to remember that refugees in the former two cohorts had very little or no influence in the choice of the place of first settlement. On average, these two categories of refugees resided longest in a large city region, followed by small city/rural regions. In contrast, refugees from the 2005-09 cohort who arranged for their own housing were most likely to reside in Stockholm and least likely to reside in small city/rural regions. Refugees from the 2005-09 cohort who arranged for their own housing spent more than twice as much time in Stockholm compared to those who took assigned housing (about 39 per cent and 14 per cent, respectively). Residence in Malmo¨was also slightly more common among refugees with self-arranged housing. It is also worth noting certain Geographical trajectories of refugees in Sweden 9 differences between the two refugee categories that did not choose their place of first settlement. Indeed, the more recently arrived cohort spent on average a longer period in a large city region and a shorter period in Stockholm.

A Typology of Geographical Trajectories
In a next stage, we used sequence analysis to identify typical geographical trajectories across regional contexts. The five different types of region described above constitute the states in our sequence analysis, while a transition corresponds to a move between two different types of region. The sequence of a given individual may have no transitions at all or consist of one or more transitions within the 8year period; all that information is considered for the analysis. Mobility between similar types of regions (e.g. large city regions) does not count as a transition. In addition, mobility within regions (between neighbourhoods or municipalities) is outside the scope of this paper. We clustered refugees' 8-year sequences across the five region types based on their similarity. A dissimilarity score was computed by optimal matching, which implies defining costs to transform a given sequence into another one by state substitution. The substitution costs correspond to the standard implementation of constant symmetric value. Once the similarity scores were computed for all sequence pairs, sequences were grouped using a combination of two clustering techniques: the Ward method for the initial clustering, and Partition around the Medoid (Studer 2013). This procedure was carried out separately for each of the three refugee categories. For each category, we obtained 10 distinct clusters representing a typology of 10 geographical mobility trajectories. Regarding implementation, we analysed the sequence data with TraMineR, an R package for sequence analysis (Gabadinho et al. 2011). To account for the quality of the clustering in the solution, we analysed statistics on the 10-cluster solution, as implemented in the Weighted Cluster package (Studer 2013). In particular, we computed the Average Silhouette Width (ASW), which is a measure of the coherence in the assignation sequences to clusters, and of the degree to which sequence clusters are distinct from one another. It is interpreted as a measure of the plausible underlying structure in the cluster solution, and it ranges from 0 to 1. Values above 0.71 are considered as indicators of a strong underlying structure (Studer 2013). We obtained the following values of ASW for the overall cluster solution: 0.63 for the 1990-93 cohort, 0.72 for the 2005-09 cohort with assigned housing, and 0.87 for the 2005-09 cohort with self-arranged housing. Figure 2 depicts the modal sequence plots of refugees' geographical trajectories, i.e. the most frequent sequence of the cluster, for the three categories of refugees. The clusters are shown in descending order of number of individuals per cluster. The frequency of each cluster is summarized in Table 2.
A first important finding is that transitions between different types of regions were very rare. The vast majority of refugees stayed in the same type of region throughout the 8-year follow-up period. As much as 94 per cent of refugees from the 2005-09 cohort with self-arranged housing belonged to a cluster that is characterized by a lack of transition between different types of region. For the other two refugee cohorts, the corresponding proportion was 85 per cent.
There was, however, significant variation in the frequency of the different consistent trajectories among the three refugee categories in the study. In general, refugees who arranged for their own housing were more likely to follow a consistent trajectory in one of the three metropolitan regions. For instance, 38 per cent of individuals in the 2005-09 cohort with self-arranged housing followed a trajectory characterized by consistent residence in Stockholm, while this was only the case for nearly 10 per cent of the 2005-09 cohort with assigned housing. In contrast, refugees who did not choose their initial location followed mostly consistent trajectories in large city regions. Indeed, consistent residence in a large city region is the most common trajectory for the 2005-09 cohort with assigned housing and the 1990-93 cohort, involving around 40 per cent and 25 per cent of refugees, respectively. These two cohorts were also more likely to follow a trajectory of consistent residence in a small city or rural region. Nearly 20 per cent of refugees in the 2005-09 cohort with assigned housing followed that trajectory during their first 8 years in Sweden, compared to 7 per cent of refugees with self-arranged housing. It should be noted that mobility between regions of the same type is, by construction, also a consistent trajectory. The sequence analysis also revealed typical trajectories that are characterized by a transition between different types of regions. Some of them were common to the three refugee categories, whereas others differed between them. With one exception, the cluster involving a move from a large city to a small city/rural region for refugees from the 2005-09 cohort with self-arranged housing, all these trajectories included a shift towards a type of region with a higher degree of urbanization.
Among refugees who were assigned housing by the authorities, the most frequent trajectory involving a transition was an early move from small city/rural region to a large city region. About 5 per cent of refugees in the 1990-93 cohort and 2005-09 cohort with assigned housing belonged to that cluster. This move took place early on, after about a year of residence in the small city/rural region. A second cluster was characterized by the same type of mobility, yet with the transition occurring after a longer period of time (during the fourth year in Sweden). Within the 1990-93 cohort, which was subject to the 'Sweden-wide strategy', 3 per cent of the refugees belonged to a cluster dominated by this transition. Although moves towards increased urbanization were the most common type of move, they were still relatively rare when compared to the consistent trajectory clusters. Indeed, the proportion of refugees remaining in a small city/rural region exceeded that of refugees following trajectories characterized by a move away from such regions.

Determinants of Geographical Trajectories
After identifying refugees' typical geographical trajectories in Sweden with the help of sequence analysis, we estimated a multinomial logit model to examine the effects of demographic and socioeconomic characteristics on the probability of belonging to any of these trajectory clusters. The model includes the time-constant variables Gender, Region of birth, and Age at arrival. Categorical time-varying variables consist of Having a partner, Having children in the household, Having university education, and Being employed. These variables were quantified using the proportion of the 8-year sequence that an individual belongs to each category of the variable. The model also controlled for average number of intra-regional moves and the average number of inter-regional moves during the follow-up period. We used the average value of the variable over the 8-year period. We calculated average marginal effects based on the estimated multinomial logit coefficients, which are presented in Table 3. We excluded the clusters with low cluster-specific ASWs (below 0.40; see above), as well as clusters that were not common to all three refugee categories. The clusters in the regression represent 89.2 per cent of the refugees in our study population and comprise both consistent and transitiondominated clusters, thus allowing high sample coverage with a strong underlying structure.

Geographical trajectories of refugees in Sweden 15
Refugees who followed the trajectory of consistent residence in Stockholm were more likely to have arranged for their own housing, have a university degree and employment. This trajectory was also more frequent among female refugees, refugees from Africa, the Middle East, and America. Consistent residence in Malmo¨was more common among refugees who arranged for their own housing and have a university education. Yet it was negatively associated with having an employment and positively associated with European origin. Refugees who followed a consistent trajectory in small city/rural regions have generally been assigned housing by the authorities. They tended to be employed but lacked university education. Refugees in this cluster were also more likely to have a partner and children in their household. Finally, the cluster involving a transition from a large city to Stockholm was most common among refugees who did not arrange for their own housing.

Descriptive Results
In order to address the second part of our study aim, i.e. the analysis of the drivers of inter-regional (im)mobility, we used event history analysis. We focused on the mobility away from the first region of residence in Sweden, because the first region provides a natural starting point after arrival in Sweden and is not confounded by the multiplicity of transitions individuals experience later during their time in Sweden.
In a first descriptive step, we estimated hazard rates for the event of leaving the first region of settlement in Sweden. In discrete time event history analysis, the hazard is defined as the probability of an event occurring at time t, assuming that it has not yet occurred (Rabe-Hesketh and Skondal 2012: 750). Table 4 shows the hazard rates for the event of leaving the region of initial settlement, by type of region of settlement and refugee category, over the observation period, conditioned on non-emigration from Sweden. A first observation is that mobility away from the first region of residence in Sweden was relatively rare among the refugees. On average, 7 per cent of refugees in our study left their first region every year. However, there were large differences according to the type of region of initial settlement. Among the refugees who initially settled in one of Sweden's three metropolitan regions-Stockholm, Gothenburg, or Malmo¨-only about 2 per cent left their first region. In contrast, the corresponding rates for refugees who settled in a small city/rural region or a large city region reached 18 per cent and 8 per cent, respectively. There were also differences in the propensity of experiencing a first inter-regional move between the three refugee categories in the study. Indeed, refugees from the 1990-93 cohort who were subject to municipal placement within the Sweden-wide strategy were more likely to leave a large city region or a small city/rural region compared to the refugees who arrived in the period 2005-09. While on average 21 per cent of the refugees from the 1990-93 cohort placed in a small city/rural region moved away every year, the corresponding figures for refugees from the 2005-09 cohort with assigned and those with self-arranged housing were 14 per cent and 8 per cent, respectively. There was therefore a clear decrease in mobility away from non-metropolitan regions during recent years, compared to the 1990s.

Determinants of First Mobility
In a second step, we estimated a logit model for discrete time data on the probability to leave the first region of settlement. The main independent variable of the model was the Type of region. The model also controlled for socioeconomic and Geographical trajectories of refugees in Sweden 17 demographic variables known to influence inter-regional mobility, which are included in the aforementioned multinomial logit model. Finally, the variable Co-ethnic neighbours describes the proportion of a person's 1600 closest neighbours who belong to the same ethnic group. This nearest-neighbour approach bypasses the so-called Modifiable Areal Unit Problem, which is a source of statistical bias that can affect any geographical analysis based on arbitrarily defined aggregations of geographical areas (Openshaw 1984). The variable was calculated using the 'geocontext' Python script (Hennerdal 2018). As Swedish registers do not include information on individuals' ethnicity, the variable is based on country of birth data. Our dataset comprises 28 'ethnic groups', consisting of individual countries or aggregation of countries. It is therefore only a proxy of ethnicity. Indeed, many ethnic groups are geographically spread across national boundaries, notably Kurds in the Middle East.
In the event history model, time-varying variables are lagged, meaning that the risk of leaving the first region of residence during a given year was modelled as dependent on the value of the variables the preceding year. The results are reported as average marginal effects in Table 5.
A first finding is that the type of region of first settlement significantly influences refugees' likelihood of leaving the first region. Refugees who settled in a small city or rural area were on average about 12 percentage points more likely to leave their first region compared to individuals who initially settled in Stockholm. Those who resided in a large city region were six percentage points more likely to experience inter-regional migration.
In addition, a 1 per cent increase in the share of co-ethnics among the 1600 closest neighbours slightly decreased the probability of leaving the first region. Indeed, a 1 per cent increase in the proportion of co-ethnics implied an almost 0.2 percentage point decrease in the probability of relocating. Although the direction of the effect suggests the presence of an ethnic preference mechanism, we should interpret it with caution, since the magnitude of the effect is relatively small.
As expected, settlement policy also has effects on refugees' propensities to leave their region of first settlement. Compared to refugees who arranged for their own housing (and thus chose their place of settlement), Sweden-wide strategy refugees and refugees with assigned housing had on average a four and two percentage points higher probability of relocating, respectively.
Regarding the socioeconomic variables, having a university degree slightly increased the likelihood of leaving the first region, while being employed decreased it.
In line with the life course perspective, we found that refugees who moved across regions were more likely to be young, single, and without children. Men were also more mobile than women. Finally, the results revealed differences between origin groups in the propensity to leave the first region. Compared to Europeans, individuals from Africa and the Middle East had the highest propensity to relocate, while refugees from America were the least mobile group.

Discussion
A first important finding from the study is the prevalence of trajectories of consistent residence in a single type of region. Transitions between regions with different socio-demographic characteristics are generally a rare phenomenon. The vast majority of refugees in the study stayed in regions with similar population density and economic structure over the 8-year follow-up period. Only about 12 per cent followed a trajectory characterized by a transition between different types of regions. Refugees who did not choose their initial location were more likely to experience a transition between different types of regions, typically from a less to a more urbanized region. More generally, and consistent with our first hypothesis, refugees with assigned housing were more likely to leave their region of initial settlement, especially when they resided in a small city/rural region. Based on these results, it appears that some of the refugees who did not choose their region of initial settlement relocated to 'adjust' the regional context to meet their preferences.
However, despite refugees with assigned housing being more mobile, a substantial proportion of refugees who were placed outside a metropolitan region by the Swedish authorities remained in their first region over a long period. Consistent residence in a large city region was a frequent trajectory, especially for the most recently arrived refugees. As much as 40 per cent of refugees of the 2005-09 cohort with assigned housing consistently resided in a large city. In addition, a significant share of refugees who were placed in a small city or rural region remained in that region type (around 82 per cent of the 1990-93 cohort and 80 per cent of the 2005-09 cohort with assigned housing). These are rather unexpected results, given the general tendency of international migrants to reside in metropolitan regions.
An important question is whether refugees' long-term residence in less densely populated regions is the result of a choice or a constraint. The fact that a notable proportion of refugees remain in non-metropolitan regions, even when they did not decide to initially settle in such regions themselves, suggests that they are satisfied with their regional environment. Over time, refugees may develop socalled 'location-specific insiders' advantages', which act as disincentives for subsequent mobility (Fischer and Malmberg 2001). They may also feel a need for stability after the traumatic experiences leading to their seeking asylum in Sweden. Finally, many refugees may also have a preference for rural residence, particularly if they lived in a smaller community in their country of origin (Povrzanovic Frykman 2009). Yet our results suggest that immobility in non-metropolitan regions may be involuntary. First, refugees who arranged for their own housing predominantly settled and remained in metropolitan regions, which points to a general preference for urban residence. Second, transitions between different types of regions are mostly directed towards regions with higher levels of urbanization. In this regard, a recent survey study among asylum seekers showed that two-thirds of the respondents who lived in accommodation managed by the Swedish Migration Agency (ABO) hoped they would be residing elsewhere in Sweden in a year, whereas only a third of those who arranged for their own housing expressed a desire to move (Esaiasson and Sohlberg 2018). One possible explanation for the prevalence of stability in small city/rural regions among refugees with assigned housing is the increasing housing shortage in metropolitan and large Swedish cities. Indeed, the fact that mobility away from non-metropolitan regions is less common for the 2005-09 refugee cohort, who arrived at a period when Swedish metropolitan regions faced greater housing shortage, compared to the 1990s, is a further indication that immobility may be involuntary.
Our study also showed that multiple factors-economic, social, and policyrelated-influence refugees' likelihood to leave their region of first settlement. In line with our second hypothesis, being employed decreases refugees' probability of leaving the first region of residence. Inter-regional mobility, and in the case of this study its first occurrence, seems to be at least partly determined by labour market considerations. This is consistent with previous Swedish research on the inter-regional mobility of international migrants as a whole (Andersson 2012). The capital region of Stockholm is particularly favoured among refugees who arranged for their own accommodation. Moreover, consistent with our third hypothesis, refugees who initially settled in the capital region of Stockholm are more likely to remain in that region over a long period. The attractiveness of Stockholm is likely to be related to the labour market opportunities available in the region. Given its strong service-oriented economy and global competitiveness, Stockholm could be viewed as a global city, where both high-and low-skilled migrant workers are in demand (Sassen 1991). Accordingly, we found that employed and university-educated refugees had a higher probability of consistently residing in this region. This is in line with previous Swedish research showing that residence in Stockholm is associated with greater labour market entry and participation among refugees (Andersson 2016;Bevelander et al. 2019;Vogiazides and Mondani 2020). Stockholm may also be attractive due to its amenities and more receptive environment towards new migrants.
In line with our fourth hypothesis, we found that proximity to co-ethnics indeed slightly decreased refugees' likelihood of leaving their region of initial settlement, which is consistent with the ethnic preference approach. Previous studies drew similar conclusions, but measuring co-ethnic concentration at the levels of the municipality (Å slund 2005; Aradhya et al. 2017) or region of residence (Haberfeld et al. 2019). Finally, our results also highlighted differences in refugees' propensity to leave their region of initial settlement according to the settlement policy implemented. Indeed, as previously mentioned, refugees with assigned housing were more likely to leave their region of initial settlement, especially when they resided in a small city/rural region.
The coexistence of multiple motives for inter-regional mobility has been previously highlighted in some qualitative studies among refugees in Sweden (e.g. Povrzanovic Frykman 2009;Ro¨nnqvist 2009). Yet this result raises a number of challenges. First, based on register data alone, one cannot assess the relative importance of economic versus social factors. A certain motive may be more relevant to certain types of refugees. For instance, a Danish study found that preferences for residing in ethnic enclaves were strongest among the least integrated ethnic minorities, with the degree of integration being defined based on labour market participation, language skills, and social connections with natives (Skifter Andersen 2015). Moreover, economic and social factors are interrelated. A desire to live close to relatives and co-ethnics may actually be economically motivated.
Social connections with co-ethnics can indeed facilitate migrants' labour market integration, by providing employment or information about job opportunities (Ro¨nnqvist 2009;Haberfeld et al. 2019). An interesting question then is whether certain groups of refugees are more likely to find employment through ethnic networks.

Conclusion
This paper has explored the regional distribution and inter-regional mobility among refugees in Sweden. Based on rich individual-level register data, it followed two refugee cohorts for 8 years after being granted their residence permit in Sweden. It applied a comparative approach, distinguishing between refugees subject to a placement policy in the early 1990s and refugees who came in the late 2000s and that either had arranged their own housing or had been assigned housing by the authorities. By studying both comprehensive trajectories through sequence analysis and mobility events using event history analysis, our paper provides a thorough picture of refugees' long-term geographical patterns.
The main conclusion of the paper is that refugees' settlement in Sweden is not solely a metropolitan phenomenon. A substantial proportion of refugees who were placed by the authorities outside metropolitan regions continue to reside in their initial location over a long period of time. This important finding suggests that refugee placement policy has long-standing effects on refugees' regional distribution. Likewise, the drop in outmigration from non-metropolitan regions among the recent refugee cohort with assigned housing implies that government-led refugee dispersal is becoming increasingly permanent. These considerations are particularly relevant in relation to the recent reform of the 'Own accommodation' legislation that removes the right to financial assistance for asylum seekers who settle in certain socioeconomically deprived neighbourhoods, which are predominantly located in metropolitan and large city regions.
Finally, let us outline some directions for further research. First, refugees' geographical trajectories across different region types should be compared with those of migrants who were granted residence in Sweden on the grounds of employment, study, or family reunion. This would allow potential differences to be revealed in the geographical patterns based on the reason for immigration. Further comparisons could be made with the trajectories of migrants' descendants and members of the native population of similar age. Second, the inter-municipal mobility of refugees deserves further attention. Various municipal characteristics may influence refugees' mobility, including the situation of the housing market, but also 'soft factors' such as the openness and capabilities of local civil servants in charge of refugee reception. Third, refugees' subjective preferences for different types of regions of residence deserve closer attention. The process through which individuals select regions (and neighbourhoods) could be fruitfully studied by applying discrete choice modelling or by using survey data of stated preferences. Qualitative research would also enhance the understanding of refugees' subjective preferences and selection of regions of residence. Lastly, the growing presence of refugees in non-metropolitan regions invites closer academic and policy attention towards these types of regions, which could arguably be labelled 'new immigrant gateways'. Further research needs to analyse what the prospects of economic and social integration in those regions are.