Migration and Neighborhood Change in Sweden: The Interaction of Ethnic Choice and Income Constraints

Introduction

This article has a twofold aim. The first aim is to analyze segregation trends in Sweden since 1990 by the combination of ethnicity and income. This analysis is motivated by the fact that Sweden has had a rapidly expanding migrant population and that Swedish data, therefore, offers an opportunity to analyze in detail how large-scale immigration affects detailed segregation patterns. What we will see is that ethnic segregation in Sweden, measured by the dissimilarity index, despite the immigrant inflows, has remained constant or even declined, whereas income segregation has increased—both are trends that have been seen in other countries too. That ethnic segregation has declined is surprising given the expectation that increasing migrant shares would trigger white flight through a tipping-point mechanism.

The second aim of the article is to address the conundrum, that there is a simultaneous occurrence of stable or decreasing ethnic segregation, and increasing income segregation, including increased poverty concentrations. We use an elaboration of the Schelling model whereby adding resource constraints to the model provides a basis for relatively stable ethnic sorting but intensified income sorting, This approach is designed to overcome the recent observation that most articles on segregation have “concentrated on the influence of either income differences or racial preferences on residential segregation, and only a few include both income differences and racial preferences” (Li, Chang, and Wang 2020). To support our hypothesis we present evidence of income sorting in Sweden across neighborhoods with varying migration densities.

Most if not all of the previous work on urban segregation privileged the behavior of actors who selected places within the city depending on the combination of residents of similar race or ethnicity. This view of urban segregation is enshrined in the Schelling model of the process of neighborhood selection and of the notions of residential tipping, whereby the choice process of selecting similar groups leads to strong patterns of separation (Schelling 1971). A fundamental tenet of the Schelling model is that an increasing neighborhood presence of minority groups will induce members of the majority to move out—so-called white flight. This will further increase the minority share of the local population which, in turn, will strengthen majority out migration and will, eventually, lead to total segregation.

The analyses using ethnic choices did not take into account the incomes of the majority or minority groups. However, we know from recent research on income segregation that there is segregation across incomes and spatial income inequality just as there is ethnic segregation (Bischoff and Reardon 2014). In general, studies of ethnic and racial segregation and studies of income segregation still proceed in parallel and there are few studies where both ethnic outcomes and economic status outcomes are examined simultaneously. Just as the studies of ethnic segregation showed neighborhood change in response to minority flows into white neighborhoods, the studies of income segregation have suggested that a similar sorting process is occurring with respect to income. By extending the conceptualization of the segregation model we can hypothesize that the preference for similar ethnic neighborhoods can also be the process in which preference for similar status neighbors will create patterns of segregation. If higher-income individuals wish to live in neighborhoods with similar status individuals we can see this as a natural extension of the Schelling model. Thus, we can view choices arranged by a budget constraint in which income limits make some neighborhoods unattainable by lower-income individuals or any ethnicity, and the revisualization process means that those lower-income individuals will be unable to select high-status residential neighborhoods.

Currently, it has not been possible to determine whether it is ethnic preferences or household incomes that individually create the patterns of segregation. However, we can show their aggregated effect on the patterns of separation. In this article, we undertake to consider the combined effect of income and ethnicity on patterns of separation. We do not claim in this article to undertake the very difficult task of sorting out the relative contribution of ethnic preference and income-based sorting This will require more explicit modeling of the processes involved. But even though we cannot unpick the two processes, we can show how the exclusion of some groups from certain spaces, and the privileged choices of other groups, affect overall levels of segregation over time. In this way, we extend the focus from ethnic segregation alone, to the patterns created by the combination of ethnic preferences and the role of income.

Below, we will first review the existing literature and describe the changes in ethnic segregation in Sweden. The empirical overview is the setting for an experimental explanation of how white flight is compatible with stable or declining segregation. We then explore the ways in which ethnic segregation is linked to income sorting in Sweden. We show that increasing variability in neighborhood ethnic composition has, indeed, been accompanied by increasing income polarization.

Previous studies

As ethnic segregation declines, even if modestly, just as income segregation increases, it raises questions about the universality of the trend toward complete ethnic segregation across neighborhoods. The Schelling model (1971) was developed at a time of strong ethnic divisions and considerable social unrest in large American cities in the late 1960s. The model focused on agents who were dichotomous on some measures, black versus white, English speaking or French speaking, officers and enlisted men, In that sense, it was a binary model. However, contexts and attitudes have changed. Residential patterns are multi-ethnic not binary, class and status have become more salient, possibly reducing the importance of ethnicity per se, and tolerance has increased (Clark et al. 2015). Younger populations also are more open to ethnic mixing. Finally, there have been major demographic changes, from low to high immigration, and in both the US and Europe flows from a mix of global origins. It is possible to suggest that the Schelling postulate of an inescapable transition to a segregated society is no longer certain.

The extensive studies of segregation served as a basis for evaluating the role of social distance where the preference to live with people of “similar” ethnicity was central and much of the research focused on the patterns of ethnic separation. The empirical “sorting” literature in geography and economics showed how ethnic preferences were strong forces in the creation of ethnic differences across neighborhoods (Cutler, Glaeser, and Vigdor 1999; Clark and Fossett 2008; Fossett 2011; Clark 2015). In particular, Vigdor (2003) showed that when the preferences of two groups are even slightly misaligned, or incompatible with one another, the resultant housing market equilibrium accurately reflects the preferences of neither group. This can be, in part, as Benard and Willer (2007) show, that the greater the correlation between status and wealth, the more the agents (in a simulation) tend to segregate, either due to choice (for the wealthy and high status) or exclusion (for the poor and low status). Thus, the Schelling model in fact only captured part of the process, the process of choice which created exclusion.

Clearly sorting is more than just ethnic preferences and the research has focused on exploring the role of income and resources and ethnicity as forces in creating residential separation. As Bayer, McMillan, and Rueben note (2004) the sociodemographic characteristics with the greatest potential to explain black segregation relate to income and overall racial differences in sociodemographic characteristics. Together they explain a sizeable amount of the segregation of each race, especially for Hispanic households. But, as they also note, the analysis cannot get at the portion of segregation that can be explained by racial differences in sociodemographic characteristics. It is here that preferences matter. But as they concede it is difficult to specify causation or allocate proportionate explanation. Thus, for black households, for example, will separation arise because of the preferences of black households to live together, or the preferences of Asian, Hispanic, or white households to live with others of the same race, as is argued in the Schelling model. The Bayer Group (Bayer, Fang, and McMillan 2014) also shows that it is possible to have segregated neighborhoods even though socio-economic status narrows between groups. In this situation, if a sufficient number of a minority group, even though they have a similar socio-economic status to Whites, choose to live in the same neighborhood, income does not decrease segregation. Clearly, the issue of the intertwined nature of ethnic preferences and income segregation is complex.

Recent research using Swedish data provides some new findings with respect to ethnic and economic sorting. First, native Swedes tend not to move into strongly immigrant neighborhoods (Bråmå 2006). The finding is consistent with a large literature which confirms “ethnic avoidance.” This is clearly a preference finding. A second observation focuses on whether native-born populations leave neighborhoods that are becoming dominated by immigrants. Both Aldén, Hammarstedt, and Neuman (2015) and Müller, Grund, and Koskinen (2018) show some evidence of ethnic flight, leaving areas that are transitioning to ethnic concentrations. Aldén, Hammarstedt, and Neuman (2015) emphasize the role of tipping but with varying outcomes depending on the background of the immigrant population. Overall, the most recent studies tend to suggest it is more about avoidance than flight (Müller, Grund, and Koskinen 2018). Finally, consistent with Clark and Ledwith (2007) and Sampson and Sharkey (2008) and many others, is the observation that immigrants often start in “concentrated ethnic areas” which act as initial settlement areas, “ports of entry” to so speak, but then work their way up into more integrated and affluent areas. This finding is also reflected in the studies by South, Crowder, and Chavez (2005), South, Pais, and Crowder (2011), Bolt, Van Kempen, and Van Ham (2008), and Schaake, Burgers, and Mulder (2010), who use a selective migration perspective, to show that people with higher-incomes are more likely to move out of disadvantaged neighborhoods. It is a view that is supported by the Andersson, Malmberg, and Clark (2019) study which shows that sorting is by high-income second-generation immigrants, that is by Swedish-born individuals, with one or more parents who are foreign-born. These are the individuals who would be expected to be “upwardly sorting.” Again, it is about resource constraints or resource advantages—more education and higher-income increase the opportunity to move.

The conclusion that Sweden seems to have experienced sorting along socio-economic as well as ethnic lines, is at the heart of our research. If following Schelling, households have preferences regarding the income level, and or the class status of their neighbors, in particular, if higher-income neighbors are preferred to lower-income neighbors, or that individuals want to be in neighborhoods where they are “like” the incomes of their neighbors, then we will find that households with similar incomes will be more likely to be together, than are households with incomes that are different (Clark 2002). The Tiebout (1956) model also predicts residential sorting by economic status. Households that can pay for amenities such as school quality, green space, and protective services, will likely sort together into communities with other residents who also value such amenities. Households willing to pay for such amenities are likely to have higher-incomes and the preference for these amenities further structures the sorting process. In effect, the Tiebout model predicts income segregation because households with similar preferences and ability-to-pay tend to form homogeneous communities.

These observations lead to questions about the interplay of the underlying forces of ethnic preferences and the role of economic selection. It is the search for how preferences and status work together in the housing market and the patterns of ethnic concentration which are the outcome, which stimulates the empirical research in this article.

The context: Segregation trends in Sweden 1990–2017

Since 1990, the foreign-born population in Sweden has increased rapidly (See Fig. 1). Currently, 1.9 million out of a population of 10.1 million people are foreign-born. The most rapidly growing group of the migrant population are non-Europeans. In 1990, less than 200,000 persons in Sweden were born outside of Europe. In 2017, the number of non-European migrants in Sweden is close to one million.

If changes in ethnic segregation in Sweden during this period are analyzed it is possible to see two different trends. If one focuses on patterns of overrepresentation and underrepresentation of the migrant population, segregation has in fact stagnated or declined in Sweden during this period. This is evidenced by the trend in the dissimilarity index, see Fig. 2. Since 1997 dissimilarity index for non-European migrants is declining both at smaller scale levels (k = 800), medium-scale levels (k = 6,400), and at larger scale levels (k = 51,200).

But segregation can also be measured as variability in population composition. This is the approach advocated by Goldstein and Noden (2003), Jones et al. (2015), Arcaya, Schwartz, and Subramanian (2018), and others. Thus, Fig. 3 shows the standard deviation in the proportion of the neighborhood population that are non-European migrants across Swedish neighborhoods from 1990 to 2016. And as the figure clearly shows variability in population composition has increased. This is also illustrated in Table 1 which shows different percentile values for the concentration of non-European difference in concentration between neighborhoods with low migrant density and high migrant density has increased (concentration is here used to designate the proportion of the neighborhood population that belongs to a certain ethnic group).

The reason for the different trends in segregation measured using the dissimilarity index (overrepresentation and underrepresentation) and a measure of variance in the neighborhood population composition (concentration) is that the latter but not the former is influenced by the size of the migrant group. If the distribution of the minority population is uneven across the neighborhood, the variance in neighborhood population composition will increase when the migrant population is growing in size, even if the dissimilarity index is constant.

Still, both measures are needed. The dissimilarity index measures the strength of the sorting process, which is important for analyzing the forces that drive segregation. The variability in population composition, moreover, is important for analyzing the effects of segregation. For example, in neighborhood effect studies neighborhood context is often measured using concentration measures. Thus, large variances in such measures imply a potential for a large variation in outcomes. In this article, the dissimilarity index shows that, in contrast to what one could expect based on the Schelling model, an increasing migrant population has not led to stronger sorting. Variability, moreover, is important since we will argue that increased variability in neighborhood composition has increased income sorting.

Moreover, the fact that variability-in-concentration measures of segregation and segregation measures that capture overrepresentation and underrepresentation, show diametrically opposing trends during a period of strong growth in the non-European population points to a need for considering how different theoretical approaches to segregation relate to concentration vis-à-vis representation.

White flight without increasing segregation

Is it possible to have white flight without observing increased segregation? Here, we will answer this question by exploring changes in the dissimilarity index when we have a growing minority population. The analysis is reported in Tables 2 and 3. We compute the growth rate of the white population across neighborhoods in a city where the minority population is increasing but we maintain a constant dissimilarity index.

Table 1 shows a city divided into ten neighborhoods, each with a total population of 100 individuals. The population is made up of two groups: white and minority individuals, and the distribution of minorities is uneven with only 2% minority in the most-white neighborhood and 73% minority in the most minority dense neighborhood. With the distribution shown in the table, the dissimilarity index is 0.53.

Now assume that the minority population grows by 20% over a 10-year period whereas the white population remains constant. What will then be the growth of the white population in the different neighborhoods if each neighborhood’s contribution to the dissimilarity index remains the same and that the total population growth in each neighborhood is the same? The answer to this question is given in Table 3.

Table 3 shows that there has been population growth in every neighborhood. The total white population is constant, but the minority population has grown. But both the white population and the minority population has been redistributed in such a way that the absolute difference between the neighborhood share of the white population and the neighborhood share of the minority population is the same in both year 0 and year 10. This implies that the dissimilarity index is unchanged.

Table 3 also contains additional columns with measures that are used by Card et al. (2008) to demonstrate a tipping point—these are the initial share of minorities in each neighborhood and the growth of the white population relative to the size of the total neighborhood population in year 0. As can be seen in the table, this growth rate displays the same pattern of white population growth as the examples discussed by Card et al. (2008). That is, in white-dominated neighborhoods there is positive growth in the white population, whereas there is a white flight from neighborhoods with large minority shares. The “tipping point” in this case seems to be around 20%. Note, however, that this pattern, in the example presented here, is the result of an assumption of unchanging segregation, as it is measured by the dissimilarity index.

The reason why the growth of the white population is negative in areas with high minority shares is that the dissimilarity index cannot remain stable at a high level unless a relatively large share of the expanding minority population is accommodated in minority dense neighborhoods. If minority individuals would move in large numbers into white neighborhoods there would be a decline in the dissimilarity index. But with the assumption that there is no shift in the relative population size of different neighborhoods, the accommodation of new minority persons makes a reduction of the white population in neighborhoods necessary. That is, negative growth rates of the white population in minority dense neighborhoods can be seen as a form of displacement. Note, however, that the minority population is increasing also in the most-white neighborhoods.

Thus, the existence of ethnic preferences and possibly tipping effects is consistent with patterns of neighborhood sorting such as those illustrated in Table 3. What our analysis shows is, instead, that a pattern of an expanding white population in low minority neighborhoods and a declining white population in minority dense neighborhoods is not sufficient for concluding that a metropolitan area is in a process of population redistribution that will result in total segregation.

Increasing income polarization in Sweden

Thus far, we have demonstrated that an expanding migrant population can be accompanied by increasing variance in the ethnic composition while at the same time, the proportion of the non-European migrant population living in the most migrant dense-neighborhoods has not increased. We have also demonstrated that it is possible to observe white flight from migrant dense areas even if the dissimilarity index is constant. White flight, thus, does not necessarily lead to increased ethnic segregation. Instead, we will argue, it is possible that increasing differences in neighborhood concentration of migrants has triggered a process of more intense income sorting that has increased income segregation. And, as can be seen in Tables 4 and 5, there is clear evidence of increasing polarization of the income distribution in Swedish neighborhoods during the post-1990 period.

Table 4 is based on an analysis of the individualized neighborhoods of persons age 15 and above for each of the years 1990 to 2015. In a first step, the distribution of the 1,600 nearest neighbors across disposable income deciles was computed for each year using (Statistics Sweden 2018) data and a Python script developed by Hennerdal (2019). An advantage of this script is that it takes advantage of the fact that Euclidian distance relationships between grid cells are constant over time. Thus, when data has been sorted based on the distance to the focus cell, the cells that are needed to reach the required nearest neighbor population can be easily identified, and it also becomes easy to compute aggregate values for many different indicators across the selected cells. In this case, it took approximately 20 hours to compute nearest-neighbor aggregates for 10 different indicators (income deciles) for 26 different years and for more than 200,000 grid cells. It was considered that a k value smaller than 1,600 could have led to instability in the estimated decile shares, and that larger k values could have made the contextual measures less sensitive to small-scale, income-segregation patterns.

In the next step, this distribution was used to assign neighborhoods to twelve different neighborhood types using k-means clustering. This analysis was carried out on a pooled sample containing residential context data for all years with data for the same location in different years treated as separate observations. Table 4 shows how the neighborhood population is distributed across income deciles in the different neighborhood types that were identified. Thus, in the neighborhood type Rich 1, 37% of the population aged 15 years and above belong to the highest income decile. In contrast, in neighborhood type Poor 4, 46% of the population is in the lowest income decile. In neighborhood type Mixed 2, close to 10% of the population is found in each income decile which implies that these neighborhoods in terms of income distribution contain an almost perfect mix of the Swedish population.

In Table 5, the columns show the distribution of the Swedish age 15+ population changed across these neighborhood types from 1990 to 2015. The polarization is shown by the fact that the population living in the most mixed neighborhood types (Mixed 2 and Mixed 3) has declined from 34.8% to 23.1%. At the same time, the proportion of the Swedish population that lives in the three most poor neighborhood types (Poor 2, Poor 3, and Poor 4) has increased from 2.1% in 1990 to 9.6% in 2015.

What this data show is that during this period of a rapidly expanding non-European migrant population the neighborhood structure in Sweden has become increasingly polarized in the sense that fewer individuals live in mixed neighborhoods and that this polarization is characterized by the establishment of neighborhoods that have large concentrations of the poorest segments of the population. In the next section, the possibility that increasing polarization has been influenced by income sorting in response to the increased variability in the ethnic composition of neighborhoods is analyzed.

Ethnic segregation and income sorting

The theoretical link between an increasing migrant population and increasing income sorting is straight forward. It can be based on Schelling’s argument, if it is acknowledged that the effect of ethnic preferences on neighborhood selection can be mitigated by conditions in the housing market. If neighborhoods with high concentrations of migrants are less attractive than neighborhoods with high concentrations of natives it is likely that housing in the latter areas will be more expensive. Moving from a migrant dense-neighborhood to a less migrant dense-neighborhood will, therefore, be associated with higher housing cost, and hence, a reduction in consumption possibilities. Thus, staying in migrant dense-neighborhood offers low-income households a possibility to preserve their consumption standard, whereas high-income households can maintain adequate consumption levels even when they face the higher housing cost that, based on ethnic preferences, should characterize areas with high concentrations of natives. If this is the case, it is possible than an increasing migrant population can increase differences in housing cost between migrant dense and less migrant dense-neighborhoods, and in this way trigger a process of more intense income sorting.

What we will do below is to assess if such a theoretical mechanism—housing in less migrant dense areas becoming relatively more expensive and, thus, difficult to access for low-income groups—is contradicted by Swedish data on income-differentials between neighborhoods with varying migrant densities. We will do this in order to open up a discussion about whether the Schelling model might be amended in a way that recognizes income sorting as a possible result of ethnic preferences. The analysis will use Swedish register data, (Statistics Sweden 2015). This data contain information about all individuals in Sweden between 1990 and 2012, including data on income, family, education, and country of birth. The data include geocodes for the residential location of the individuals in the data set and, thus, allows a detailed analysis of residential patterns. But the analysis will be based on the individual-level data that has been aggregated to the neighborhood level. That is, we will not look at individual-level spatial sorting and hence, we will not be able to ascertain which processes that lie behind the observed patterns. This should be kept in mind since it implies that the empirical analysis will not look at causal patterns.

In the first step, we analyze the ethnic composition of egocentric neighborhoods that encompasses the 200 nearest neighbors. The focus will be on the share of migrants (foreign-born) in these bespoke neighborhoods. To identify the 200 nearest neighbors of the individuals in our data set, we use the EquiPop software. The approach used in Equipop is to expand buffers around all residential locations until the buffer contains at least 200 individuals. When this target has been reached the demographic composition of the buffer population is computed, in this case, the proportion of foreign-born. The motivation for using k = 200 level is that ethnic preferences, at least in the Schelling model framework, is about selecting neighbors with whom one potentially will have daily interactions. It is well-established that especially for face-to-face contacts, individuals living closely (within a mile) are of special importance (Wellman 1996). A higher intensity of contacts nearby is also evidenced by the fact that close neighbors (within 150 feet) are more likely to be known than more distant neighbors (Greenbaum and Greenbaum 1985). Moreover, also network relationships are strongly distance dependent (Kowald et al. 2013).

If there is income-based sorting across neighborhoods with varying concentrations of foreign-born this should be reflected in income differences between more and less migrant dense-neighborhoods. To measure income, we use individual disposable income. Disposable income is the sum of all taxable and tax-free income minus taxes and negative transfers and is computed on the household level. To obtain the individual disposable income, household disposable income is multiplied by the individuals consumption weight (=1 for adults) and divided by the household’s total consumption weight (Statistics Sweden 2011). Before computing the average income in different neighborhoods, income has been transformed into percentile values for the entire Swedish adult population. In order to simplify the statistical analysis, the population has been divided into percentile bins based on the concentration of migrants in their individualized neighborhoods.

The population is divided into three groups: Foreign-born, Swedish born with Swedish born parents (Swedish background), and Swedish born with at least one foreign-born parent (non-Swedish background). This division has been made in order to make it possible to analyze if income sorting across neighborhoods with different migrant density works in different ways depending on ethnic background.

Neighborhoods have been divided into four groups. First, into neighborhoods that are located in a densely populated region or not. The cut-off point for being in a dense region was taken to be that there should be at least 204,800 inhabitants within a radius of 30 km from the center of the individualized neighborhood. Second, into neighborhoods that are located inside or outside dense settlements, as defined by Statistics Sweden. In Sweden, dense neighborhood settlements are defined as agglomeration with at least 200 inhabitants. In dense neighborhood settlements, houses are not allowed to be more than 200 meters apart. Neighborhoods thus can be located in (1) dense settlements that are in a metropolitan area, (2) dense settlements in nonmetropolitan areas, (3) outside dense settlements but in a metropolitan area, or (4) outside dense settlements and in a nonmetropolitan area. This division of neighborhoods makes it possible to analyze if income sorting works differently in different geographical contexts. The theoretical motivation for this classification is that low-income individuals might be able to access neighborhoods with low proportions of migrants if such neighborhoods are located in areas at large distances from the center of large agglomerations. Clearly, the distribution of the housing stock underlies the outcomes across these different categorizations and influences the nature of the sorting which takes place.

Three different ethnic background classifications and four geographically different neighborhood types make it possible to analyze 12 different types of income sorting (Table 6). As we can see from the table, the foreign-born population is concentrated in dense settlements, either nonmetropolitan areas or large cities- more than 60% are in the large cities. In contrast, only 45% of Swedish born with a non-Swedish background and 37% of the Swedish native-born population are in the large cities.

We have graphed the distributions of mean percentile income across neighborhoods with different migrant densities by the three “ethnic” groups—foreign-born, Swedish born with a Swedish background, and Swedish born with non-Swedish background separately for four geographical contexts—outside dense settlements in nonmetropolitan areas, outside dense settlements in metropolitan areas in dense settlements in nonmetropolitan areas, and in dense settlements in metropolitan areas where dense settlements are defined by Statistics Sweden as settled places with more than 200 inhabitants. To assess the variability in income across neighborhoods with different concentrations of foreign-born, we have used the aggregation of individuals in different percentile bins based on the share of migrants among the nearest 200 neighbors. Thus, for each of the 12 sub-populations (defined by ethnicity and geographical neighborhood type), the mean percentile income has been computed for the different ethnic-composition percentile bins. The reason for using bins is that this allows a graphical representation of the results with graphs showing the mean percentile income on the vertical axis and the neighborhoods’ concentration of migrants on the horizontal axis. Of course, by computing the mean percentile income of the individuals in each neighborhood bin, variability is reduced, and it becomes possible to see if the increasing migrant density in the neighborhoods is correlated with an increasing concentration of low-income individuals.

The results of the analysis are presented in Figs. 4-7. In Fig. 4, there is one graph for each combination of the ethnic group and geographical context each showing the mean percentile of disposable income for neighborhoods with different migrant densities. Fig. 5, instead, shows the distribution across disposable income deciles for neighborhoods with different migrant densities, for different geographical contexts but not separately for different country-of-origin categories. Fig. 6 shows how the distribution across disposable income deciles for neighborhoods with different migrant densities separately for foreign-born individuals and Swedish born individuals.

Fig. 4 clearly shows that in dense settlements, both in metropolitan and nonmetropolitan areas, there are clear signs of an income gradient across neighborhoods with different concentrations of foreign-born individuals. The income gradient is strongest for Swedish-born individuals with a non-Swedish background (individuals born in Sweden with one or two immigrant parents) and for foreign-born individuals, but less strong for Swedish-born individuals with a Swedish background. Outside dense settlements, there is essentially no income gradient for Swedish born individuals irrespective of them having a Swedish or non-Swedish background, see also (Andersson, Wimark, and Malmberg 2020). But for foreign-born individuals, there are signs of an income gradient, even though the pattern is rather diffuse.

That there is not a clear income gradient across neighborhoods located outside dense settlements is not unexpected. Ethnic segregation is typically an urban phenomenon and there is no need for segregation processes to work the same way in urban and nonurban areas. Moreover, the lack of an income gradient in these areas could suggest that outside dense settlements low-income people can access areas with high concentrations of nonmigrants if they choose to live outside of cities.

What Fig. 5 makes clear is that this income sorting is a tendency that includes much variability. All income deciles have some representation in all neighborhood types but especially in the dense settlement in metropolitan areas, there is a strong income polarization between the migrant dense-neighborhoods and neighborhoods with lower migrant density. Lower-income groups are strongly overrepresented in the most migrant dense-neighborhoods, whereas high-income groups are strongly overrepresented in the less migrant dense-neighborhoods. Dense settlement in nonmetropolitan areas show less polarization but there is a clear sorting of low-income groups to migrant dense-neighborhoods. Outside dense settlements in nonmetropolitan areas there is no evidence of a correlation between income and migrant density, but it is also the case that there are few neighborhoods here with high migrant density.

From Fig. 6 it is clear that sorting across neighborhoods with different migrant densities is not only about income. For example, in the lowest income decile, foreign-born individuals are very strongly sorted into the most migrant dense-neighborhoods, whereas Swedish-born individuals are only very weakly sorted into the most migrant dense-neighborhood category. Moreover, also in the most migrant dense-neighborhood category, there is considerable income mixing. The lowest income decile is the largest group but a substantial number of individuals from higher-income groups are also living there. And this is, of course, especially the case among the foreign-born. It is only in the highest income decile that there is not an overrepresentation of migrants in more migrant dense-neighborhoods. From Fig. 6 it is also clear that the Swedish foreign-born population is overrepresented in the lower-income deciles. This too suggests that high migrant density in neighborhoods with many low-income earners merely reflects a sorting of low-income individuals to low-income neighborhoods. This is theoretically possible. In fact, if there was strict income sorting—with individuals living in neighborhoods where everyone belongs to the same income percentile—then the migrant density in the poorest neighborhoods would be around 70%. However, the income level of these neighborhoods would be much lower—a mean percentile below two—than is observed in the most migrant dense-neighborhood—mean percentiles around 25. This demonstrates that income sorting alone cannot explain the ethnic sorting that is found across Swedish neighborhoods.

Fig. 7, finally, shows that it cannot be ruled out that an increasing concentration of foreign-born individuals might have contributed to increasing spatial income polarization. Fig. 7 shows that, in dense metropolitan areas, the income gradients across neighborhoods with different migrant densities look very similar in 1990 and 2012. That is, patterns of income sorting appear both in 1990 and 2012. In 2012, however, there are more neighborhoods with high migrant density, and migrant density in the most migrant dense-neighborhoods has increased. And because of the gradient, this is reflected in an increasing number of neighborhoods with very low-income levels. Again, this pattern does not demonstrate a causal link from migrant density to income sorting. But it underlines the fact that the modified Schelling framework we propose in this article needs to be considered when mechanisms behind increasing income polarization are discussed.

Discussion

Theoretical models play an important role for how we interpret social and geographical trends, even to the extent that patterns suggested by theoretical models can be thought of as good descriptions even when empirical studies fail to give them support. The Schelling model is a case in point. Its suggestion of an end process in which there is total segregation has hindered the widespread acceptance of de-facto declines in ethnic segregation that have been demonstrated in a wide range of studies. This suggests that weak empirical relevance on its own is not sufficient for abandoning well-established models. There is also a need to consider the reasons why model predictions are not borne out by empirical data. In this article, we have suggested that the original Schelling model needs to be amended by the incorporation of a negative housing price-response to increasing migrant density. A price-response that can induce income-based sorting across neighborhoods. Such a price-response could explain why complete ethnic segregation is seldom observed. But it would also predict that an increasing migrant population could generate a process of more intense income segregation.

What we have accomplished empirically in this article is to consider if an amended Schelling model can or cannot account for segregation trends observed in Sweden since 1990. These trends include a larger expansion of the migrant population, constant or declining ethnic segregation as measured by dissimilarity indexes, increasing geographical income polarization and finally, a strong pattern of overrepresentation of low-income group in neighborhoods with high migrant density. These patterns, we would argue, are what one could expect based on an amended Schelling model.

We should be clear, however, that the Swedish data do not demonstrate that income sorting based on ethnic preferences is the causal factor behind the Swedish trends. Our proposition is instead that the Swedish data fit with the expectations generated by an amended Schelling model. And based on this our conclusion is that price-responses induced by increasing migrant densities leading to more intense income sorting is a process that could have a role in shaping geographical trends in countries experiencing large migration inflows.

According to the classic story, an expanding minority population could trigger a process of white flight when the local proportion of the minority population would expand beyond the threshold that is compatible with stable segregation patterns. That is, we would see a process toward increasing ethnic segregation. In this article, we have reviewed the Swedish experience from 1990 to 2017 which is, indeed, a period that has seen an expanding minority population. And to some extent, the Schelling story is still relevant. In neighborhoods with high proportions of non-European migrants, population growth for the European born population, including Swedish born has been negative. This can be seen as evidence of white flight. However, somewhat surprisingly, this white flight has not been followed by an increasing dissimilarity index. Instead, from 1997 to 2017, the dissimilarity index for non-European migrants has been declining.

At the same time, the post-1990 period has seen an increasing polarization of the income distribution in Swedish neighborhoods. The proportion of the population living in neighborhoods with a strong mixture of income groups has declined, and an increasing proportion of the population lives in neighborhoods with a concentration of low-income groups. Increasing income inequality is certainly an important factor (Andersson and Kährik 2015) but according to the amended Schelling model, the increase in the non-European migrant population can have contributed to this trend too. That is, rather than triggering a Schelling type process toward ethnic segregation, an expanding minority population could initiate a process of more intense income sorting that results in socio-economic spatial polarization.

In a Schelling model, there is a preference for living in neighborhoods with few minority individuals. If housing is a market good this would require that, in equilibrium, housing cost for comparable units should be lower in migrant-dense areas and, hence, low-income households will be concentrated there. This will modify the tendency toward complete segregation that is inherent in the classical Schelling model. Some low-income majority members will stay in minority dense neighborhoods whereas some high-income minority members will settle in majority-dense neighborhoods. Still, when the minority population is increasing, ethnic preferences will tend to redistribute the majority population toward majority-dense neighborhoods. This redistribution will be reflected in higher growth rates of the majority population in majority-dense neighborhoods. Although this can be seen as a form of white flight, we demonstrate in the article that it need not necessarily lead to an increase in the dissimilarity index. Moreover, according to the amended Schelling model, arriving migrants with low income will be sorted into migrant-dense neighborhoods where it is easier to obtain housing. This will lead to an increasing concentration of migrants in these areas and that will induce those with higher-incomes (either migrants or native-born) to move out. The result will be increasing income polarization but not necessarily more pronounced ethnic segregation.

The revised Schelling story, cannot be inferred directly from the Swedish case but, as we see it, it is an explanatory framework that is consistent with the Swedish data. Further research using microdata will be needed in order to clarify to what extent this framework is relevant, for example along the lines of (Andersson, Malmberg, and Clark 2019) and (Alm-Fjellborg 2020). From a policy perspective, this is clearly relevant if an expanding minority population does not necessarily trigger a tendency toward complete ethnic segregation but instead could initiate a process that will produce neighborhoods characterized by concentrated poverty.