Minorities moving out from minority-rich neighbourhoods: does school ethnic context matter in inter-generational residential desegregation?

Abstract

This paper aims to develop a fuller understanding of the relationship between the ethnic composition of childhood residential neighbourhoods, schools, and residential neighbourhoods later in life in producing and reproducing segregation. We apply a longitudinal research design on linked individual-level data from Estonia. Estonia is an interesting case because of the Soviet era population distribution policies and its ubiquitous state-funded educational system where minority parents can choose in which school—Russian-language or Estonian-language—their children study. We find that minority parents mostly opt for minority-dense schools and, if they do so, their children who grew up in minority-dense neighbourhoods also end up living in minority-dense neighbourhoods as adults. An inter-generational vicious circle of segregation forms. However, minority children who live in minority-dense neighbourhoods but study in majority-dense schools are more likely to end up living in majority-dense neighbourhoods later in life. Hence, intervening in school choice has the potential to contribute to inter-generational residential desegregation.

Introduction

This study focuses on segregation in schools and residential neighbourhoods within a vicious circle of segregation framework that links different life domains and generations. Previous research shows that school segregation is highly entwined with ethnic composition of neighbourhoods: if students are assigned to schools based on their place of residence, the demographic composition of neighbourhoods will largely shape the ethnic composition of schools (Denton, 1996; Mickelson, 2011). Consequently, school segregation tends to follow residential segregation (Boterman et al., 2019; Bayona‐i‐Carrasco and Domingo, 2021). Since educational outcomes tend to be lower in schools with higher shares of ethnic minorities (Agirdag et al., 2012), a vicious circle of segregation may emerge as segregated schools contribute to segregated labour markets and ethnic differences in incomes, and differences in income contribute to residential segregation among the subsequent generation of ethnic minorities (Tammaru et al., 2021). A residential neighbourhood-school link is thus crucial for the intergenerational transmission of segregation from parents to children (Bernelius et al., 2021).

However, early experiences of studying together with majority population children in schools may potentially help to break such production and reproduction of ethnic segregation (Rowley and McNeill, 2017). Ethnically diverse schools can increase first-hand interethnic contacts, promote mutual trust and acceptance, decrease prejudice and fear, and develop social ties that can also help to reduce segregation in labour and housing markets throughout the life-course (Mickelson, 2011). Therefore, studying together with majority population children may help decrease levels of residential segregation of minority population members once they start their own housing careers (Goldsmith, 2016).

Previous research on school segregation is extensive (for an overview, see Wells and Crain, 1994; Boterman et al., 2019), yet there are few empirical studies focusing on the role of school segregation on residential segregation later in life (e.g. Goldsmith, 2010, 2016; Rowley and McNeill, 2017). This study aims to develop a fuller understanding of the relationship between ethnic composition of residential neighbourhoods during childhood, at schools, and in residential neighbourhoods later in life as children start their own housing careers after leaving the parental home. More specifically, the goal of this paper is to learn what is the role of the school ethnic composition in inter-generational ethnic residential desegregation by focusing on those members of minorities who have grown up in minority-dense neighbourhoods. The study is set in Estonia, which provides interesting research setting for addressing this topic since the context of residential sorting of parents and children generations is very different. The sorting of Soviet era migrants, the parents of our research population, into the neighbourhoods of Tallinn hinged on where the new apartments were built upon arrival, and not on income, preferences, or discrimination. As people arrived from other parts of the Soviet Union, mainly from Russia, they received apartments in newly built apartment buildings through the centrally planned housing allocation system.¹ The sorting of the children’s generation into residential neighbourhoods takes place in a very different context today, characterized by the importance of income, preferences, and discrimination as in other cities in Europe and the United States.

Because of the large inflow of migrants up until the end of the 1980s and the high status of the Russian language in the former Soviet Union, migrants generally did not learn Estonian, and parallel Estonian-language and Russian-language school systems were established. The main elements of these parallel school systems still exist today. Spatially, Estonian-language and Russian-language schools are often located next to each other in the same urban neighbourhoods—most Estonian children attend Estonian-language schools (majority-dense schools), and most Russian children attend Russian-language schools (minority-dense schools). The educational system is publicly funded and all schools, irrespective of language of instruction, are treated equally in terms of financing and curricula. While most Russian parents send their kids to minority-dense schools, some do send their kids to majority-dense schools. Hence, we have an interesting research setting that allows us to longitudinally follow a full generation of young people who grew up in minority-dense residential contexts relatively randomly assigned to their parents as they arrived in Estonia. They studied either in an Estonian-language or a Russian-language school in independent Estonia, reflecting the parental expectations towards their children’s integration into Estonian society. Both types of schools are often located side-by-side and parents can choose a school for their children based on their preferences towards school language. Finally, we will document whether such selection into Estonian-language or Russian-language schools is related to the residential context in which the young people themselves live as adults. More specifically, we seek answers to the following research questions:

1. What are the school choices of minorities living in minority-dense neighbourhoods who have easy access both to minority-dense and majority-dense schools?

2. Does studying in majority-dense schools elevate the probability of living in majority-dense neighbourhoods in adulthood for minority children who grew up in minority-dense neighbourhoods?

3. How do other factors, such as parental resources, labour market success, or contacts with majority population members and host society, shape the association between schools’ ethnic composition and neighbourhoods’ ethnic composition later in life for minorities who grew up in minority-dense neighbourhoods?

We link data from the Estonian Education Information System (2005 and 2006 graduation cohorts), Estonian Population Register (2019), and Estonian Population Censuses (2000 and 2011) to follow the ethnic composition of the neighbourhood of minorities during their childhood, which school they attended (majority-dense or minority-dense) and in which neighbourhood ethnic context they settled by the age of 30. Our data also include many important correlates that may affect residential sorting later in life: family background, characteristics of the childhood neighbourhood, partner’s ethnicity, and important individual-level characteristics (socio-economic status, integration variables). In the analysis, we focus on those young people for whom Russian is their mother tongue and who grew up in minority-dense neighbourhoods, and our particular interest is on the role of schools’ ethnic composition in them living in neighbourhoods with a high share of majority population members.

Theoretical framework on links between residential and school segregation

Place stratification and spatial assimilation in a longitudinal and multi-domain perspective

In recent years, research in segregation is increasingly taking place in a longitudinal and multi-domain (neighbourhoods, workplaces, schools, etc.) framework (Piekut et al., 2019; Musterd, 2020; et al.Tammaru et al., 2021). Van Ham and Tammaru (2016) introduce the domains approach for understanding segregation by linking residential segregation, workplace segregation, and family context to understanding immigrant earnings. Boterman and Musterd (2016) use the metaphor of ‘cocooning’ to link segregation processes at places of work and residence. Park and Kwan (2017) introduce the concept ‘multi-contextual segregation’ for understanding how segregation is produced and reproduced across multiple life domains. Tammaru et al. (2021) propose a model that they call the ‘vicious circle of segregation’ for analysing how segregation is produced and reproduced across different life domains because of (i) a sequential sorting into places of residence, schools, and workplaces and (ii) the exposure effects that people experience in them.

From a longitudinal and multi-domain perspective, this sorting process and these exposure effects thus result in minorities experiencing spatial assimilation and place stratification. Spatial assimilation can be divided into the domain-specific processes of residential assimilation, school assimilation, and work assimilation. The spatial sorting relates to family characteristics (e.g. growing up in an ethnic or mixed-ethnic union), individual characteristics (e.g. preferences and resources), as well as institutional arrangements (e.g. housing and educational policies). The contextual effects arise from being exposed to and interacting with other people in different life domains. These vicious sequences of sorting and exposure occurring in different life domains often result in segregation being passed across generations, for example, children living in similar neighbourhoods to their parents (Sharkey, 2008; Hedman et al., 2015).

When placed into a longitudinal perspective, a vicious sequence of sorting and exposure would contribute to place stratification as members of the minority population may face discrimination and barriers to spatial assimilation both at places of residence and schools (Goyette et al., 2012; Pais et al., 2012). Structural constraints for minorities’ neighbourhood and school outcomes are considered an important factor in producing and reproducing segregation across domains and generations (Burdick-Will et al., 2020). Barriers to achieving residential assimilation of minorities emerge since powerful groups in cities such as real-estate agents and landlords manage and constrain access to housing, excluding minority groups they view as undesirable from certain urban locations (Charles, 2003). Even in cases where minorities themselves have not experienced discrimination, members of certain ethnic groups may still perceive that they are not welcome in certain neighbourhoods; perceptions of neighbourhoods by minorities thus start to limit their consideration set of neighbourhoods (Krysan and Crowder, 2017). The barriers minorities experience and feel are nuanced and complex, and intentional, unintentional, and perceived discrimination may reinforce each other in contributing to high levels of ethnic clustering in certain neighbourhoods.

Co-ethnic preferences, including language and culture maintenance, may further facilitate the differential residential sorting processes of ethnic groups (Clark, 2021). Ethnoburbs in American cities are an example of flourishing ethnic suburban spaces that provide homes and schools for wealthy minorities who prefer to live with co-ethnics (Li, 1998, 2009). From this perspective, schools can be seen as important institutions that help to pass ethnic identity, language, and cultural traditions between generations (Fishman, 1980). The ethnic and socio-economic composition of schools may be an important determinant of residential choice of parents as well by attracting minority families and repelling majority families (Boterman et al., 2019). For members of the ethnic majority population, mechanisms of ethnic threat may start operating as minorities start to cluster into certain parts of the city. Increases in the share of minorities tend to be related to the decrease in the perceived residential neighbourhood and school quality by members of the majority population (Goyette et al., 2012). Because of these mechanisms, members of the majority population either avoid or leave ethnic neighbourhoods (Mägi et al., 2016).

Contrary to place stratification, spatial assimilation proposes mechanisms that would help to contribute to desegregation. From a longitudinal perspective, spatial assimilation is a continuous and long-term sorting process that may take place in different life domains and across generations. When migrants enter the host society, they tend to settle in areas of minority concentration, where they can find support from their co-ethnics. However, residential segregation is often related to the feeling of disadvantage by different ethnic groups (Krysan and Crowder, 2017). Minority-dense neighbourhoods provide very few opportunities for creating first-hand interethnic contacts and building mutual respect and trust (Bolt et al., 1998). When neighbourhoods are segregated, schools can be seen as important agents of integration as minority children adapt easier to the values of the host society when learning together and socializing with members of majority population. The older the children are, the more mobile they become in urban space and the less bound they are to neighbourhood schools. Learning together with members of the majority population, consequently, would help improve their socioeconomic position and contribute to achieving residential assimilation later in life. The degree of spatial assimilation of the second and subsequent migrant generations at schools and workplaces thus becomes an important indicator of the overall integration of ethnic minorities in the host society.

Childhood socialization: perpetuation theory perspective

Schools have a double role in the processes related to place stratification and spatial assimilation. First, the residential sorting of parents may have a direct effect on the sorting of children into schools. Second, the contextual effects children experience in schools from direct contacts and interactions with teachers and peers influence their values, skills, and social networks that remain important throughout their life (cf. Allport, 1954; Braddock, 1980). According to perpetuation theory—a version of socialization theory developed for understanding the long-term effects of school life on individual lives—the social and economic circumstances in which an individual grows up have an important influence on their life outcomes, which, in turn, may be passed across generations (Braddock, 1980; Braddock and McPartland, 1989; Wells and Crain, 1994). The characteristics of the neighbourhoods in which children grow up are, therefore, likely to be the same as the neighbourhoods in which they live as adults (Sharkey, 2008; Goldsmith, 2016). Hence, minority children growing up in minority-dense neighbourhoods tend to live in minority-dense neighbourhoods as adults (de Vuijst et al., 2017).

Perpetuation theory further posits that early experiences with diversity (e.g. studying in a mixed-ethnic school) would much more likely lead minority group members to diverse settings later in life (Braddock, 1980). Thus, the perpetuation of ethnic disadvantage and residential segregation can be disrupted by studying in majority-dense schools. The perpetuation of diverse settings and experiences helps to diminish negative perceptions of interethnic contact, increase people’s abilities to cope in interethnic situations, and raise mutual trust, respect, and acceptance (Mickelson, 2011). Mickelson and Nkomo (2012) argue, for example, that mixed-ethnic schools can have a positive influence on many life outcomes later in life that are essential building blocks for socially cohesive multi-ethnic societies, helping minorities to achieve higher occupational attainment, promoting cross-ethnic friendships and contributing to residential assimilation. It has also been suggested that openness to diversity is transmitted from parents to children; those who attended ethnically diverse schools are more likely to prefer that their children have ethnically diverse schoolmates too (Braddock and Gonzalez, 2010). Thus, perpetuation of diversity can have long-term effects and intergenerational consequences on people’s lives.

In summary, perpetuation theory allows us to assume that being exposed to own-group members tends to transmit from one domain (e.g. school) to another (e.g. residential neighbourhood) in the form of vicious circles of segregation, while being exposed to diversity in one domain (school) may help to break the segregation circle in another (neighbourhood) (Figure 1). The exposure and sorting effects may transmit from parents to children, and ethnically diverse schools may act as important institutions that stop residential segregation reproducing across generations. Since the meanings of education and school changes at each educational level, we distinguish between lower secondary, upper secondary, and tertiary (university) levels. Our main interest relates to lower secondary and upper secondary since the Estonian school system has a parallel language infrastructure at these levels, but not at university level. Minorities who have obtained tertiary education in Estonia will likely have studied in a diverse environment.

Figure 1

Interactions between residential and school domains.

Empirical evidence on residence-school segregation link

Studies on the role of schools in segregation in other life domains have primarily focused on African Americans and Latino Americans in the United States. Several studies have found that African American students who attend mixed-ethnic high schools are more likely to attend desegregated colleges than those who attend segregated schools (Braddock, 1980; Dawkins, 1994). Similarly, African Americans who attend mixed-ethnic high schools are more likely to have European American co-workers than those who attend segregated ones and are more likely to have positive opinions regarding the friendliness of European American colleagues and the competence of European American supervisors (Braddock and McPartland, 1989). Past research also indicates that African American students who attend ethnically diverse schools have more diverse professional and social ties compared to segregated African American students (Crain, 1970). Even if these ties are weak, they provide ethnic minority groups with access to information and opportunities (e.g. housing or labour market opportunities) which can sort them into ethnically diverse settings later in life (Wells and Crain, 1994).

Research by Goldsmith (2016, 2010) provides evidence that studying in mixed-ethnic schools contributes to residential assimilation later in life; the ethnic composition of schools is a significant predictor of the ethnic composition of the neighbourhoods where Latino Americans and African Americans live as adults. In addition, Rowley and McNeill (2017) show that early exposure to ethnic diversity in schools is significantly related to neighbourhood ethnic diversity later in life. However, their results also suggest that schools alone are limited in their ability to perpetuate diversity in the absence of support from other institutions (e.g. fair housing policies).

Although most studies highlight the positive effect of mixed-ethnic schools on breaking the vicious circle of segregation, there are studies that have noted benefits in attending segregated schools on social mobility. For example, Gamoran et al., (2016) have found that ethnic minority students (notably, Latino Americans and Asian Americans) have better job prospects when they attend schools with co-ethnics. In addition, Carter (2010) has found that African American students who attend schools with predominantly minority students are more open to interacting with persons of other backgrounds and have higher self-esteem than African Americans who studied in mixed-ethnic schools. Therefore, while school segregation restricts opportunities for interethnic contact, the effects of attending ethnically diverse schools on outcomes in other life domains are not always clear.

In Europe, most research on school segregation has focussed on the impact of school ethnic composition on the educational attainment of students (e.g. Dumay and Dupriez, 2008; Agirdag et al., 2013). Many studies also focus on residential segregation as a driving-force behind school segregation (Boterman, 2019; Skovgaard Nielsen and Andersen, 2019; Rangvid, 2007). Additionally, research undertaken in Northern Ireland demonstrates how mixed schools can facilitate more positive opinions on other religious, socio-economic, or ethnic groups through increased intergroup contact (Hayes et al., 2007). Although these studies help to understand the reasons behind school segregation and shed some light on the relationship between school segregation and residential segregation, there is a lack of research on the long-term impact of school ethnic composition on residential ethnic composition and segregation later in life.

Furthermore, research concerning the relationship between residential segregation and school segregation is context sensitive. Most of the previous research has focused on North America and Western Europe. We extend studies on the relationship between residential segregation and school segregation to a new context, Estonia.

Case study context

The formation of ethnic minorities in Estonia contrasts significantly to North America and Western Europe. The total population of Estonia is 1.3 million, and the share of ethnic minorities stands at 31 per cent. The majority of minorities living in Estonia are of Russian ethnicity (76 per cent) or speak Russian as a mother tongue (according to the 2021 census, most Ukrainians (54 per cent) and Belarusians (85 per cent), who are the two other big ethnic minority groups in Estonia, consider Russian to be their mother tongue). This is a consequence of the fifty-year occupation of Estonia by the Soviet Union until 1991, in which Russians and people from other Soviet republics were encouraged to migrate to the other, non-Russian republics of the Union, such as Estonia. Migrants were recruited to work in the state-operated industries, in the military, and for the Soviet bureaucratic and administrative apparatus (Tammaru and Kulu, 2003). About 90 per cent of minorities settled in major cities, including in the capital city Tallinn. The total population of Tallinn is 440,000 inhabitants, and Estonians and minorities are almost equal in share. Despite otherwise egalitarian aims and low levels of socio-economic segregation, ethnic residential segregation was high since migrants settled in new housing estates built for housing migrant workers and allocated to newcomers through the central housing allocation system (Tammaru et al., 2016).

After regaining independence, the level of residential segregation between Estonian speakers and Russian speakers has increased from a dissimilarity index value of 31 in 1989, to a value of 43 in 2021 (Figure 2). The increase in residential segregation between the ethno-linguistic groups is due to different labour market trajectories with minorities becoming over-represented in lower-income jobs because of modest Estonian language skills (Saar and Helemäe, 2017). There is also a strong overlap between ethnic and social segregation in Tallinn as lower-income people have moved into the Soviet housing estates or to the most affordable housing segment, where minorities were already over-represented in 1989. The growth of ethnic segregation is almost solely due to differential residential sorting of ethnic groups already residing in Estonia since there has been little immigration into Estonia during the past 30 years.

Figure 2

Changes in residential segregation by mother tongue in Tallinn, 1989–2021.

Source: Statistics Estonia.

Schools are segregated along ethno-linguistic lines in Estonia. A parallel Russian-language education system was introduced for migrant children alongside the existing Estonian-language one during the Soviet period (Lindemann, 2011). This parallel system largely remains in place today. The Estonian education system is also relatively comprehensive in nature (Põder and Lauri, 2014): all students who enrol in public schools, which provide most school places, follow the same ‘basic education’ curricula up to ninth grade, and schools receive equal public funding irrespective of language of instruction. As a rule, students are assigned to schools based on which school’s catchment area they live in. In major cities, such as Tallinn, some schools, so-called city-wide ‘elite’ schools, are allowed to recruit students from all over the city. This choice is open for all children, irrespective of their home language. Tallinn is an interesting case because many minority-dense neighbourhoods, especially those with larger populations (e.g. large, Soviet-built housing estates) typically have a double-language educational infrastructure—Estonian-language and Russian-language schools are often located side-by-side. Families can choose whether to send their kids to the nearby Estonian-language or Russian-language school. Besides, even when both schools are not directly available in the family’s own neighbourhood, another school in a nearby neighbourhood can be chosen. In Tallinn, the accessibility of schools in adjacent neighbourhoods is good: the city is spatially compact and public transport is well-organized. As such, all minority households have the opportunity to send their child to a Russian-language or Estonian-language public school. Despite this choice, both the Estonian speakers and Russian speakers tend to send their children to a school that teaches in their mother tongue. Hence, schools are ethno-linguistically segregated, with Estonian-language schools mainly attended by Estonian speakers and Russian-language schools mainly attended by Russian speakers. However, some Russian-speaking children do study in Estonian-language schools. School choice is an important factor in shaping the labour market outcomes of young people since, on average, the learning outcomes in Estonian-language schools are higher compared to Russian-language schools (Lauri et al., 2017). Both studying in Russian-language schools and having poor Estonian language skills contribute to the Russian-speaking minority’s disadvantaged position in Estonian labour (Saar and Helemäe, 2017) and housing markets (Tammaru et al., 2016). It follows that graduating from Estonian-language schools provides many resources for social mobility and advancement in the labour market for minority children that may facilitate residential desegregation. We are therefore interested whether those Russian-speaking students who lived in minority-dense neighbourhoods but attended Estonian-language schools contribute to residential desegregation when they start their own housing careers after leaving the parental home.

Data and methods

Our empirical contribution relies on five sources of data linked together (Figure 3). First, we use Estonian Education Information System (EEIS) data for 2005 and 2006 school graduates. We selected 2005 and 2006 graduates into our study since these are the two first graduation cohorts available in the EEIS, enabling us to follow these cohorts until the age of 30. Establishing a long study period was important since young people increasingly stay longer in the parental home before starting their independent housing careers. By the age of 30, a majority of the members of the two graduation cohorts have settled in their independent homes, allowing us to contrast changes in childhood and adulthood ethnic residential contexts.

Figure 3

Overview of different datasets used for the analysis.

Most similar research in the US context is based on sample surveys. EEIS includes all students who graduated in 2005 and 2006. Following people over a long period of time poses a challenge in keeping people in the study across different survey waves, while census and register data covers all people at all points in time, avoiding the issue of attrition. Second, we derive data on the same individuals’ place of residence later in life from 2019 Population Statistical Register by linking people in the two datasets. Third, we use the 2011 Estonian census to get information about the local (Estonian) language skills of these people, as well as the educational level and mother tongue of their parents. Fourth, information about individuals’ occupational status is drawn from the 2019 Estonian EUROMOD database, which is linked to all people in the register. Finally, we used 2000 Estonian census data to characterize students’ residential environments in 2005 and 2006 since the educational register does not include information on the ethnic contexts of neighbourhoods. This choice is mainly valid since only a few neighbourhoods were reclassified between minority-dense and majority-dense between the 2000 and 2011 inter-census period.

Almost half of all Russian speakers in Estonia live in Tallinn urban region (Figure 4). The steps of selecting our research population are as follows. Our study includes Russian speakers (based on mother tongue) who finished their lower secondary (basic) or upper secondary (secondary) education in 2005 or 2006 and lived in minority-dense neighbourhoods in Tallinn urban region during their school years, and who still lived in Tallinn urban region in 2019 (6,765 students). We were able to link 89 per cent of the students in both databases. It is most likely that the 11 per cent loss between graduation in 2005 and 2006 and observing adulthood residence in 2019 is due to emigration, since many Russian speakers choose to study at universities outside Estonia (Pungas et al., 2015). We restricted our sample to students living in minority-dense neighbourhoods since Tallinn is ethnically highly segregated and our main theoretical approaches focus either on the factors that contribute to staying in ethnic neighbourhoods and on the persistence of segregation (place stratification), or that contribute to moving out from ethnic neighbourhoods and desegregation (spatial assimilation). By applying these restrictions, we have 4,877 or 72 per cent of the students remaining in our research population. As a final step, we excluded 59 people who moved out from Tallinn urban region, 32 people for whom the neighbourhood ethnic context changed without moving, and 5 students who studied in English language schools. After imposing these restrictions, our final research population comprises 4,781 Russian speakers.

Figure 4

The share of Russian speakers (2000) and the location of schools in Tallinn urban region.

Forty-two per cent of the research population lived in neighbourhoods where there are both Estonian-language and Russian-language schools (the five largest neighbourhoods), 24 per cent lived in neighbourhoods with only Russian-language schools and 14 per cent lived in neighbourhoods with only Estonian-language schools (see Figure 4). Thirty-nine per cent of the research population went to their neighbourhood schools. There were also 20 per cent of students in the research population who lived in neighbourhoods where there was neither a Russian- nor an Estonian-language school.

The aim of this study is to understand the role of school segregation and other factors behind inter-generational residential desegregation by focusing on those moving from minority-dense neighbourhoods into majority-dense neighbourhoods as adults. This is because we are interested in a qualitative inter-generational change in ethnic context from childhood to adulthood characterized as living near members of the majority population. As our research concerns residential desegregation, we exclude Russian speakers who lived in majority-dense neighbourhoods in 2005 and 2006. A simple decrease in the share of minority population (e.g. 10 percentage point decrease from 80 to 70 per cent) does not necessarily imply living in a neighbourhood with a high share of the majority population. Therefore, following previous research (e.g. Ellis et al., 2009), we use location quotient (LQ) values for defining segregated or minority-dense neighbourhoods, calculated for 2000 (origin) and 2019 (destination). LQ is a ratio that shows how concentrated a particular group is in each spatial unit as compared to the whole urban region. If the ratio is less than 1, the group is under-represented in the neighbourhood, while if it is more than 1, the group is over-represented. In this analysis, minority-dense neighbourhoods are those neighbourhoods where LQ values are greater than 1. The share of Russian speakers in these neighbourhoods is at least 40 per cent (see Figure 4). To take full advantage of the data, we include the exact share of minorities in the childhood neighbourhood as a predictor of residential desegregation. In other words, we also account for how minority-dense (ranging from 41 to 83 per cent) the childhood neighbourhood of residence was.

Given the structure of the data, with individual students nested in 92 schools, we estimate a set of multilevel (two-level) binary logistic regressions. The intra-class correlation coefficient of 8 per cent (Model 0, Table 3) also supports the use of a two-level model and indicates that, potentially, some of the variation in moving decisions can be explained by school characteristics. We fit a logistic regression model on our data, with our dependent variable indicating whether a person moved out from a minority-dense neighbourhood into a more majority-dense neighbourhood in adulthood (1) or whether they stayed in the same or moved to another minority-dense neighbourhood in adulthood (0). For a sensitivity analysis, we also fitted a multilevel mixed-effects linear regression model, with the dependent variable measuring the percentage of Russian speakers in adulthood neighbourhood (see Supplementary Appendix Table 3). Our primary independent variable is language of instruction at school (Estonian-language or Russian-language): we use it as an indicator of the schools’ ethnic context as Estonian-language and Russian-language schools have heavily Estonian and Russian ethnic compositions, respectively. We also include a dummy variable indicating whether students finished basic or secondary education in 2005 or 2006. In addition, we control for variables related to individual achievement (level of education, occupation), integration (e.g. migrant generation, Estonian language skills), partner’s ethnicity, and family background. The share of minorities in childhood neighbourhoods is included in the analysis as a continuous variable. Finally, we interacted studying in Estonian-language vs Russian-language schools with other important parental and individual characteristics, including mother’s education, highest completed level of education, and partnership status, but none of these interactions were statistically significant.

Findings of the study

Descriptive analysis of childhood residence-school-adult residence pathways

Our study focuses on minorities who grew up in minority-dense neighbourhoods and we are interested in the factors that contribute to living in majority-dense neighbourhoods as adults. Our main interest is whether the language of instruction at school (proxy for school ethnic context) is related to intergenerational residential mobility. Both Estonian-language and Russian-language schools are easily accessible for minority children living in Tallinn, and the integration policies in Estonia favour studying in Estonian to improve the social integration of minorities. However, most of the Russian speakers (93.8 per cent) who lived in minority-dense neighbourhoods studied in Russian-language schools (Table 1). Sorting into schools is related to parents’ characteristics; studying in Estonian-language schools is associated with having highly educated and intermarried parents (Table 1 and Supplementary Appendix Table 2). However, parental characteristics are not related to their children’s residential outcomes later in life (Table 1, see also Table 3).

Table 1

Descriptive statistics for the main variables of interest by language of instruction in 2005/2006 and neighbourhood type in 2019 (see Supplementary Appendix Table 1 for the whole list of the variables used in the analysis). N for the whole population is 4,781

	Language of instruction in 2005/2006		Neighbourhood type in 2019		Whole research population
	Estonian	Russian	Majority-dense	Minority-dense
	%	%	%	%	%
Language of instruction in 2005/2006			χ2(1) = 49.1, P < 0.001
Estonian			12.3	5.2	6.2
Russian			87.7	94.8	93.8
Total			100	100	100
Neighbourhood type in 2019	χ2(1) = 49.1, P < 0.001
Majority-dense	27.5	13.0			13.9
Minority-dense	72.5	87.0			86.1
Total	100	100			100
Highest completed educational level in 2019	χ2(2) = 26.9, P < 0.001		χ2(2) = 93.7, P < 0.001
Less than secondary	3.7	7.4	4.1	7.7	7.2
Secondary	38.3	49.5	34.7	51.1	48.8
High	58.1	43.1	61.2	41.2	44.0
Total	100	100	100	100	100
Citizenship 2019	χ2(3) = 60.1, P < 0.001		χ2(3) = 47.3, P < 0.001
Estonian	96.6	77.7	88.9	77.3	78.9
Russian	1.3	7.8	4.1	7.9	7.4
Undefined*	2.0	13.4	6.1	13.8	12.7
Other	0	1.1	1.0	1.0	1.0
Total	100	100	100	100	100
Partnership status 2019	χ2(4) = 107.0, P < 0.001		χ2(4) = 125.6, P < 0.001
Estonian partner	13.1	2.6	9.0	2.3	3.2
Russian partner	29.5	41.3	48.7	39.2	40.6
Separated (divorced/widowed)	8.4	8.2	7.2	8.4	8.2
Single	45.6	45.9	32.3	48.1	45.9
Partner from other ethnicity and partner mother tongue unknown	3.4	2.0	2.7	2.0	2.1
Total	100	100	100	100	100
Mother’s education	χ2(4) = 21.4, P < 0.001		χ2(3) = 13.7, P < 0.05
High	64.8	53.3	57.3	53.4	54.0
Secondary	20.8	22.7	18.2	23.3	22.6
Less than secondary	8.4	17.7	16.1	17.2	17.1
Unknown	6.0	6.4	8.4	6.0	6.3
Total	100	100	100	100	100
Father’s education	χ2(3) = 17.9, P < 0.001		χ2(3) = 11.9, P < 0.05
High	44.0	33.8	39.8	33.6	34.4
Secondary	19.5	23.6	20.6	23.8	23.4
Less than secondary	9.4	15.8	13.1	15.8	15.4
Unknown	27.2	26.7	26.5	26.8	26.8
Total	100	100	100	100	100
Parents’ mother tongue	χ2(2) = 171.0, P < 0.001		χ2 (2) = 0.84, P < 0.25
Both Russian	56.4	63.6	62.4	63.3	63.2
Estonian and Russian	13.1	1.5	2.7	2.2	2.2
Other	30.5	34.9	34.9	34.6	34.6
Total	100	100	100	100	100

Note:

*‘Undefined citizenship’ is used for those migrants from former Soviet republics who were unable and unwilling to apply any country’s citizenship after the collapse of the Soviet Union.

Regarding residential mobility, our results further show that most Russian speakers (86.1 per cent) who lived in a minority-dense neighbourhood during childhood lived in a minority-dense neighbourhood as adults too (Table 2). Although this is expected from previous research in Estonia (Mägi et al., 2016) and other contexts (cf. Braddock and McPartland, 1989; Wells and Crain, 1994), the extent of residential segregation transmitted between generations is surprising. Furthermore, 56 per cent of people lived in the very same neighbourhood as adults in 2019 as they lived during their childhood. From those who moved from their childhood neighbourhood to a different neighbourhood, 68.4 per cent moved to another minority-dense neighbourhood. It follows that Russian speakers sort into neighbourhoods where they also lived during their formative years which provide supportive social networks and Russian-language infrastructure (kindergartens, schools, clubs, etc.). However, some desegregation across generations can be documented as well; 32 per cent of people who moved and 14 per cent of our total research population lived in a majority-dense neighbourhood in 2019. For all of them, the share of Estonian-speaking neighbours increased in adulthood compared to childhood, with the average increase being 38 per cent. For those who still lived in minority-dense neighbourhoods, the average share of Estonian-speaking neighbours decreased 5 per cent in adulthood compared to childhood.

Table 2

Residentially mobile and immobile people by language of instruction at school

	N	Stayed		Moved to majority-dense neighbourhoods		Moved to minority-dense neighbourhoods
The whole research population	4,781	2,679	56.0%	665	13.9 (31.6)*	1,437	30.1 (68.4)*
Studied in Estonian	298	131	44.0%	82	27.5 (49.1)*	85	28.5 (50.9)
Studied in Russian	4,483	2,548	56.8%	583	13.0 (30.1)*	1,352	30.2 (69.9)*

Note:

*Share of those who moved.

Multilevel models of residential-school pathways

We proceed with multilevel models to obtain deeper insights into the role of school ethnic context in shaping residential outcomes later in life for minorities who grew up in minority-dense neighbourhoods. We start by presenting the results of pairwise regression or by entering and removing all our variables one by one to set their baseline values compared to the next four models where we add groups of variables in a stepwise fashion (Table 3). We find that studying in Estonian-language schools is associated with 2.52 times higher odds (predicted probability = 72 per cent) of living in majority-dense neighbourhood in adulthood compared to those who studied in Russian-language schools in the pairwise regression model. The odds ratio values of school language decrease but remain significant as we control for other variables in a stepwise fashion. Model 1 shows that students who finished secondary education have higher odds to live in majority-dense neighbourhoods later in life compared to people with lower levels of education. Model 2 controls for relevant individual level characteristics that include occupation. We find that the odds of living in majority-dense neighbourhoods are still 2.32 times higher compared to those who studied in Russian-language schools when accounting labour market success. Higher socio-economic status itself is related to adult neighbourhood outcomes as expected by the spatial assimilation theory. Both Russian speakers with a tertiary level of education as well as those employed in higher occupations have higher odds of living in majority-dense neighbourhoods in adulthood compared to other educational and occupational groups. These results support the main argument of the theory of residential assimilation, which states that higher socio-economic status favours moving out from minority-dense neighbourhoods. Interestingly, though, unemployed people are more likely to live in majority-dense neighbourhoods compared to those in lower occupations as well. This is against the expectations of the theory of residential assimilation, but we should acknowledge that there is some heterogeneity in the unemployed group, which includes both high-skilled and low-skilled people.

Table 3

Multilevel binary logistic regression analysis of residential mobility out from minority-dense neighbourhoods, odds ratios (predicted probabilities calculated for pairwise regression)

	Pairwise regression			Model 1		Model 2		Model 3		Model 4
	Exp(B)	Predicted probability	SE	Exp(B)	SE	Exp(B)	SE	Exp(B)	SE	Exp(B)	SE
Language of instruction (ref. Russian)
Estonian	2.52***	0.72	0.47	2.59***	0.49	2.32***	0.43	1.98***	0.37	2.02***	0.37
Finished educational level in 2005/2006 (ref. basic education)
Secondary	1.83***	0.64	0.16	1.83***	0.16	1.50***	0.14	1.31**	0.12	1.31**	0.12
Gender (ref. female)
Male	0.76**	0.43	0.07			0.93	0.08	1.04	0.10	1.04	0.10
Highest completed educational level in 2019 (ref. less than secondary)
Secondary	1.26	0.56	0.27			1.01	0.22	1.03	0.23	1.03	0.23
Higher	2.66***	0.73	0.57			1.71**	0.39	1.60**	0.38	1.60**	0.38
Occupational status 2019 (ref. lower occ.)
Higher occupations	2.02***	0.67	0.22			1.40**	0.17	1.38**	0.17	1.38**	0.17
Inactive	1.14	0.53	0.40			1.02	0.36	1.15	0.41	1.17	0.42
Unemployed	2.22**	0.69	0.68			1.85**	0.57	1.80*	0.56	1.82*	0.57
Unknown	1.73***	0.63	0.24			1.39**	0.19	1.42**	0.20	1.43**	0.20
Citizenship (ref. Russian)
Estonian	2.06***	0.67	0.43					1.75**	0.37	1.75***	0.37
Undefined	0.88	0.47	0.23					1.08	0.29	1.08	0.29
Other	1.67	0.63	0.81					1.67	0.82	1.68	0.83
Immigrant generation (ref. first gen.)
Second generation	0.96	0.49	0.17					0.93	0.18	0.94	0.18
Third generation	0.77	0.44	0.15					0.75	0.15	0.76	0.16
Unknown	1.31	0.57	0.38					1.38	0.43	1.41	0.44
Estonian language skills (ref. cannot speak Estonian)
Can speak Estonian	1.36**	0.58	0.16					1.02	0.14	1.01	0.14
Partnership status (ref. Russian partner)
Estonian partner	2.86***	0.74	0.52					2.51***	0.47	2.54***	0.48
Single	0.53***	0.35	0.05					0.58***	0.06	0.58***	0.06
Separated (divorced/widowed)	0.69***	0.41	0.12					0.68**	0.12	0.67**	0.16
Partner from other ethnicity and partner mother tongue unknown	1.04	0.51	0.28					0.97	0.27	0.97	0.27
Mothers’ education (ref. less than secondary)
Higher	1.03	0.53	0.12					0.86	0.11	0.87	0.11
Secondary	0.79*	0.44	0.11					0.78	0.12	0.79	0.12
Unknown	1.40*	0.58	0.26					1.31	0.27	1.30	0.27
Fathers’ education (ref. less than secondary)
Higher	1.36**	0.58	0.18					1.18	0.17	1.18	0.17
Secondary	1.03	0.51	0.15					1.01	0.15	1.01	0.15
Unknown	1.17	0.54	0.17					1.19	0.23	1.19	0.23
Parents mother tongue (ref. both parents mother tongue is Russian)
Estonian and Russian	1.12	0.53	0.31					0.92	0.26	0.91	0.26
Other	1.04	0.51	0.09					0.96	0.15	0.96	0.15
% Russian speakers in childhood neighbourhood	0.99**	0.50	0.01							0.99**	0.01
	Model 0			Model 1		Model 2		Model 3		Model 4
N	4,781			4,781		4,781		4,781		4,781
ICC (intra-class correlation coefficient)	0.08			0.05		0.04		0.03		0.03
Log likelihood	−1,904.99			−1,870.33		−1,839.78		−1,788.92		−1,786.18
Wald chi2 (df)				71.13 (2)***		128.48 (9) ***		219.18 (28) ***		226.07 (30) ***

Note: *P < 0.1; **P < 0.05; ***P < 0.001.

In Model 3, other integration and family background variables are added. We find that the odds of living in a majority-dense neighbourhood later in life are 1.98 higher for minorities studying in an Estonian-language compared to those studying in a Russian-language school. We were also interested in the association of other integration variables with settling in majority-dense neighbourhoods later in life. We find that local language skills and migrant generation are not related to living in majority-dense neighbourhoods as adults. Thus, our results do not lend support to straight-line residential assimilation where each new generation is living in less minority-dense neighbourhoods. However, other integration variables that measure first-hand contacts with members of the majority population and host society are related to inter-generational residential desegregation. Having an Estonian partner is positively related to living in majority-dense neighbourhoods in adulthood compared to having a Russian partner, although it must be acknowledged that the group of Russian speakers who have an Estonian partner is small. Another important finding is that being single is negatively related to living in majority-dense neighbourhoods. This variable partly captures those who are still living with parents which means that those minorities who leave the parental home are more likely to live in proximity to members of the majority population. The last integration variable, having Estonian citizenship, is also positively related to living in majority-dense neighbourhoods later in life.

We control for the parental background on adulthood residential context in Model 3 since many social inequalities are passed from parents to children. For example, it may be that parental cultural and economic resources are associated with the neighbourhood outcomes of their children later in life. We were able to construct two variables for parents, reflecting their level of education (proxy for economic resources) and mother tongue. Mother tongue measures whether minority children who lived in minority-dense neighbourhoods have grown up in mixed-ethnic unions, with one parent being an Estonian speaker. Interestingly, we find that while some parental background variables are weakly but statistically significantly related to the neighbourhood outcomes of their children in adulthood in a pairwise regression, but not when considering other control variables (Model 3 and Model 4). For example, in pairwise regression, minorities with highly educated fathers have 1.36 times higher odds (P < 0.05) of living in majority-dense neighbourhoods as adults, but these initial differences become insignificant in later models. It must also be admitted that the level of education is not the best proxy of parental economic resources since the education that parents received during Soviet times has lost much of its value today. Still, there is a clear positive association between level of education and income in Estonia, and people with lower levels of education suffered more in the transition from a centrally planned to a market-based economy. Since the neighbourhoods in which children grow up are also related to their parents, we include the exact share of minorities in childhood neighbourhoods in Model 4. We find that the association between the share of Russian speakers in the childhood neighbourhood and the adulthood neighbourhood is negative, indicating that the higher the share of minorities in childhood neighbourhoods, the smaller odds they have of living in majority-dense residential neighbourhoods as adults.

To summarize, the relationship between school language of instruction and residential outcome later in life changes little after controlling for the socio-economic status of people, integration variables, and parental background variables. Although the share of Russian speakers who live in minority-dense neighbourhoods in childhood and go to majority-dense schools or form mixed-ethnic unions with members of the majority population is small, our explorative results show that those variables which measure direct contacts with members of the majority population may be crucial for achieving residential assimilation and breaking the vicious circle of segregation between different life domains and generations.

Discussion and conclusions

This study has expanded existing research on segregation as a process that is produced and reproduced across multiple life domains by shedding new light on the links between the ethnic composition of schools and residential neighbourhoods during childhood and adulthood. We focused on those young members of the minority population who grew up in minority-dense neighbourhoods and the role of school ethnic composition on their ethnic residential context in adulthood. We draw our empirical evidence from Estonia, where an interesting shift in the residential sorting mechanism occurred after the country regained independence from the Soviet Union in 1991. The Soviet residential sorting of parents was shaped by central planners rather than preferences or financial resources, while the residential sorting of their children after completing their education occurs under market conditions shaped by income differences and preferences, as well as discrimination. The country also has parallel Estonian-language and Russian-language school systems where families can choose to which schools they will send their children. Both Estonian-language and Russian-language schools receive equal public funding and follow the same curricula, and the two schools are often located side-by side in urban space. Hence, as for majority parents, the school choice largely reveals the preference of minority parents: studying in Estonian-language schools alongside majority population pupils or in Russian-language schools alongside minority population pupils. The main findings of the research are as follows.

First, a vast majority of parents of Russian speakers (93.8 per cent) living in minority-dense neighbourhoods send their children to minority-dense schools. In other words, minorities living in ethnic neighbourhoods contribute to the reproduction of segregation through their children’s education. When there is a large and viable ethnic community in the city, integration with the majority population may not be of great value. A similar phenomenon can be found in the US cities where viable ethnic minority communities have emerged in the form of ethnoburbs (Li, 1998, 2009). The choice for minority-dense schools may not be so much related to perceived discrimination or disadvantage but to culture and language maintenance for minorities that help maintain the viability of the ethnic community (Fishman, 1980). While the host society considers reducing school segregation to promote assimilation and equal opportunities for social mobility, minority parents may feel that sending their kids to majority-dense schools would come with the cost of language and identity loss. Such findings are also in line with recent research in the United States showing that complex choices with long-term effects, such as where to school their children, are overwhelming for parents themselves. Parents from large ethnic communities mainly interact with other parents from the same ethnic group, leading to the decision to send their children to minority schools (Burdick-Will et al., 2020).

To facilitate ethnic mixing at schools is not an easy task when various mechanisms related to co-ethnic preferences and ethnic threat felt by majority parents are at work (Burdick-Will et al., 2020; Clark, 2021). Rising levels of residential and school segregation may start to reinforce each other, for example, through the (i) decline of the reputation of minority-dense neighbourhoods, (ii) perceived decrease in school quality, and (iii) decreased value of homes located in such neighbourhoods for members of the majority population (Goyette et al., 2012). Levels of school segregation tend to grow faster than levels of residential segregation and, as in the United States, school segregation has therefore become a major concern in most European countries with high shares of migrants and minorities (Boterman, 2019; Burdick-Will et al., 2020; Bayona‐i‐Carrasco and Domingo, 2021). When promoting spatial assimilation, it is important to see the connections between school and residential segregation. Hence, it is difficult to predict the effect of abolishing the dual-language school system in Estonia since high levels of ethnic residential segregation would still facilitate school segregation. Facilitating ethnic diversity can lead to increased ethnic mix at schools and higher levels of residential assimilation later in life. Alternatively, a greater ethnic mix in schools could increase the perceived ethnic threat among members of the majority population, resulting in them avoiding minority-dense neighbourhoods, as is happening in other European countries. The most fruitful way forward seems to be a series of careful interventions in the housing and education domains to reap the benefits of diversity and diminish ethnic threat, combined with constant monitoring of the effects of each intervention taken.

Our second main finding shows that studying in minority-dense schools is related to settling in minority-dense neighbourhoods later in life. This association changes little after controlling for labour market outcomes and integration variables. The strength of the association increases with the rising share of ethnic minorities in the childhood neighbourhood. Furthermore, we find that half of the members of the minority population who grow up in minority-dense neighbourhoods and study in minority-dense schools end up living in their childhood neighbourhood as adults. The lack of residential mobility may be due to a reduced consideration set, which is often related to minority discrimination in urban housing markets (Krysan and Crowder, 2017). In Tallinn, where minority-dense neighbourhoods are abundant, most minorities belong to the same ethnic group or speak Russian as their mother tongue and, hence, the very narrow choice set cannot be, at least entirely, explained by discrimination. Tallinn is a compact city and public transport is very efficient and free of charge, so living in a different neighbourhood may not hamper social interaction with co-ethnics living in other parts of the city. A more plausible explanation to moving to the childhood neighbourhood as adults could therefore relate to the importance of place attachment (cf. Clark, 2021); for minorities, place attachment and related ethnic networks seem to be very intimate and highly localized in urban space.

Such extensive reproduction of segregation across domains and generations provides little support to straight-line assimilation theory (cf. Alba and Nee, 2003). Likewise, local language skills per se are not related to achieving spatial assimilation; local language skills of younger minority generations are significantly better than the local language skills of older minority generations in Estonia, but they still sort into minority-dense neighbourhoods as adults. Although the reproduction of segregation across life domains and multiple generations is expected based on previous research both in the United States and Europe (Sharkey, 2008; Hedman et al., 2015), the extent of the reproduction found in this study is, however, striking. Therefore, it is especially important to learn what mechanisms would help to break away from the circle of segregation and, more specifically, the role of studying with majority population members relative to other factors that relate to achieving residential assimilation.

Our third main finding demonstrates that attending majority-dense or ethnically diverse schools is related to residential desegregation as students who studied in such schools are more likely to live in majority-dense neighbourhoods in adulthood. What is equally important is that this finding changes little after controlling for individual achievement (including labour market success), other integration variables and parental background, including parental occupational status and educational level, our proxies for social class. Our explorative research also points towards other factors that may facilitate moving into majority-dense neighbourhoods later in life. Minorities with a high level of education and higher occupational status are more likely to move out from minority-dense neighbourhoods as expected from spatial assimilation theory. Those integration variables related to establishing close contacts with majority population members and host society are also positively associated with inter-generational desegregation. Having a majority population partner is positively associated with sharing a residential neighbourhood with majority population members in adulthood. This is in line with previous research indicating that the formation of mixed-ethnic unions is strongly related to ethnic integration in other life domains (Puur et al., 2021; Rahnu et al., 2020). Our findings add that it is important to place the sequences of events that contribute to exiting the circle of segregation into a co-evolutionary perspective. The formation of cross-ethnic social contacts in one life domain may contribute to the formation of cross-ethnic social contacts in another, irrespective of the exact sequence of change: (i) sharing residential neighbourhoods with members of the majority population and studying in mixed-ethnic schools may elevate the probability of forming mixed-ethnic unions among members of the ethnic minority population (Rahnu et al., 2020), while (ii) studying in mixed-ethnic schools and having a majority partner may elevate the probability of achieving residential assimilation.

While people with an Estonian partner are more likely to live in majority-dense neighbourhoods compared to people with a minority partner, we further find that people with a minority partner also are more likely to live in majority-dense neighbourhoods compared to single people. Being single partly captures residence in parental home, implying that the formation of any type of partnership has the potential to contribute to desegregation compared to staying single. Still, close contacts with the host society are especially important as revealed by findings concerning citizenship. The Estonian context offers a unique opportunity to analyse minority attachment to the host country since minorities can choose their citizenship status. Consequently, the citizenship variable carries valuable information on the willingness of young minorities to integrate. After the dissolution of the Soviet Union, Russian speakers were not granted Estonian citizenship automatically. Instead, minorities could opt for Estonian citizenship, Russian citizenship, or another citizenship (Aasland, 2002). For older people, the choice of Russian citizenship was common since one of the requirements for obtaining Estonian citizenship relates to Estonian language proficiency. In other words, a certain level of integration is needed for receiving Estonian citizenship. Russian citizenship is also of practical value for people who often travel to Russia, while Estonian citizenship is beneficial for people travelling in Europe. We also find that minorities with Estonian citizenship have a higher probability of moving out from minority-dense neighbourhoods. Future research within a longitudinal research framework could further explore our findings that attachment to the intimate neighbourhood among minorities tends to facilitate residential segregation, while attachment to the host country tends to facilitate residential assimilation.

One aspect not examined in this study is the rationale behind the parents’ choice to send their children to Russian- or Estonian-language schools. While previously discussed factors such as integration into majority Estonian society and own-culture and language maintenance are likely to be important, other factors, such as the proximity and perceived quality of the school, may also be significant (Nieuwenhuis and Xu, 2021). Other factors may affect decision-making as well. If own-culture and language maintenance are important for parents, is their choice of children’s school based on the language of instruction, the ethnic composition of the school, the decisions of other parents in their social networks, or a combination? To further break the intergenerational transfer of segregation and reduce segregation in residential, school, and workplace domains, a better understanding of parental decision-making concerning school language of instruction is therefore needed. Finally, an important limitation in our study is that our estimates cannot be given a causal interpretation as minority students attending majority-dense schools may not be comparable with minority students in minority-dense schools.

To conclude, this longitudinally designed study reveals that, in the absence of policies that promote daily interethnic contacts, no straight-line spatial assimilation occurs over generations or with the acquisition of local language skills. If social integration and spatial assimilation are aims, then an understanding of the neighbourhood-school link is crucial for breaking the vicious circle of segregation. Housing in contemporary cities operates increasingly under market forces, while the public sector has a leading role in education. Efforts that facilitate studying with majority population members in schools would promote social contacts, including interethnic friendships, mutual trust and acceptance, and access to information in majority-population networks about different opportunities (Mickelson, 2011). Studying in majority-dense schools may contribute to residential integration later in life among minorities who lived in minority-dense neighbourhoods as children. Other forms of social contacts, most notably union formation with majority population members, and labour market success contribute further to children living in less segregated neighbourhoods in adulthood relative to their parents

Supplementary data

Supplementary data are available at ESR online.

Acknowledgements

We are sincerely grateful to Dr Liina-Mai Tooding for her advice and valuable guidance in statistics.

Funding

This work was supported by the Estonian Research Council [PRG306: ‘Understanding the Vicious Circles of Segregation. A Geographic Perspective’], the Infotechnological Mobility Observatory (www.imo.ut.ee/en), and Estonian Academy of Sciences Research Professorship of Tiit Tammaru.

Kadi Kalm is a researcher in the Department of Geography at University of Tartu with a background in human geography and regional planning. She defended her PhD thesis in 2018 and her PhD research focused on ethnic residential segregation and integration of the Russian-speaking population in Estonia. Currently, her main research focuses on spatial inequalities, domains of inter-ethnic contacts, ethnic identity and integration. She is interested in understanding the interconnections between different domains of life (e.g. residential neighbourhood, school, partner relationship) where ethnic segregation is experienced.

David Leonard Knapp is a PhD student in the Department of Geography at University of Tartu. He is interested in interdependencies between family context, residential segregation and school segregation. In addition, his research focuses on spatial and social inequalities.

Anneli Kährik is an associate professor in human geography in the Department of Geography at University of Tartu. Her main research focuses on housing geography and policy, urban geography, urban planning, urban policy, social geography, and social policy.

Kadri Leetmaa is an associate professor in human geography in the Department of Geography, University of Tartu. She is interested in urban social geography, population geography, urban planning, ethnic segregation, residential mobility, residential preferences, suburbanization, shrinking cities, and post-socialist cities.

Tiit Tammaru is a professor of urban and population geography and head of the chair of human geography at the Department of Geography, University of Tartu. His research interests include migration, residential mobility, housing, ethnic segregation across different life domains (family, neighbourhood of residence, workplace), relations between social inequalities and socioeconomic segregation, as well as comparative segregation studies. He is the first editor of the book “Socio-economic Segregation in European Capital Cities” (Routledge, 2016).

Footnotes

1.

Soviet era residential sorting of the parents of our research population is to a large degree related to their and their children’s place of residence in 2005/2006 (starting point of the analysis). It is due to the very low mobility rates in the 1990s and the first half of the 2000s, which were mainly caused by almost no new construction and severe economic hardship. Mortgages were not available and affordable housing loans only started to become common around 2004, when Estonia became a member of the European Union. Additionally, the mobility of the Russian-speaking population was even lower than Estonians and their residential patterns are similar to those developed during the Soviet period (Mägi et al., 2016). Based on these arguments, we claim that circumstances in the 2000s still reflect the inherited residential segregation.