Björn Gustafsson Department of Social Work University of Gothenburg P.O. Box 720 SE 405 30 Gothenburg Sweden and Institute for the Study of Labor (IZA), Bonn Bjorn.Gustafsson@socwork.gu.se Katarina Katz Department of Economics and Statistics Karlstad University Universitetsgatan 2 SE651 88 Karlstad Sweden Katarina.Katz@kau.se Torun Österberg Department of Social Work University of Gothenburg P.O. Box 720 SE 405 30 Gothenburg Sweden Torun.Osterberg@socwork.gu.se May 2013 Segregation from generation to generation? Intergenerational Transmission of Context among visible immigrants and the majority in Sweden Draft. Please do not cite. JEL Classification Codes: J15, J62, R23 Key words: Immigrants, Residental segregation, Neighbourhoods, Sweden 1
Abstract Since the first half of the 1980s the distribution of income in Sweden has become more unequal and many immigrants from low and middle income countries, who can be described as visibly different from the majority population were received. Residential segregation has increased and become a concern for public policy. We analyse the degree to which inequalities in neighbourhood environment persists across generations in metropolitan areas of Sweden using data on persons born 1974. The outcome variable studied is neighbourhood average household income. We report an intergenerational neighbourhood income elasticity of 0.81 and an intergenerational neighbourhood correlation of 0.38. Both numbers are considerably higher than the corresponding ones estimated for household income. We find that one half of the children of visible immigrants grew up in the poorer quartile of neighbourhoods in metropolitan Sweden and that as many as 63 per cent of these remained in the poorest quartile of neighbourhoods as adults. Those two proportions are somewhat lower than corresponding ones reported for Blacks in a US study. By contrast, of the children who grew up in the lowest quarter of neighbourhoods in metropolitan Sweden and did not have a visible immigrant background, not more than 35 per cent remained in such neighbourhoods as adults. This percentage is similar to what has been reported for Whites in the United States. Estimated regression models indicate that a number of parental and other variables contribute little to understanding the high intergenerational neighbourhood income elasticity. 2
1. Introduction Sweden has been known as comparatively egalitarian country with a largely ethnic homogenous population. However, since the first half of the 1980s the distribution of income has become more and more unequal and, during the same period waves of immigrants from low and middle income countries have reached Sweden. Many such immigrants have faced large difficulties in finding a job and are therefore found themselves at the lower end of the income distribution. A a substantial number of the immigrant children have grown up to become young adults in in less attractive neighbourhoods. Residential segregation has become an issue and raises many questions. For example: to what extent are the children of visibly different minorities growing up in low income neighbourhoods? 1 To what extent do they also live in such neighbourhoods as adults? This can also be phrased: How large is intergenerational persistence in context? Sociologists have long studied intergenerational links in occupation, class and status in single countries as well as across countries. In economics the literature on intergenerational income relations is more recent, but has grown rapidly. Many authors have used regression analysis to estimate intergenerational income elasticities or computed mobility matrixes to demonstrate the association of earnings or income across generations. The literature shows a large variation in the strength of the link between earning of sons and fathers across rich countries. 2 A number of putative mechanisms behind the association across generations have been suggested and studied. Emphasis has been placed on how differences in parent s education and other characteristics affect the education of the offspring and thereby their labour market position and earnings. 3 Much less research has been made on intergenerational links in context, however. It is not only parents that influence the life of a young person. Peers and neighbours can also play a large role by affecting values and by providing networks for the young person. Furthermore, a person s well-being is not defined only by earnings, income, occupation and class. An important aspect for many people is place of residence. Some neighbourhoods are more attractive than others for various reasons. Some such factors are a pleasant and healthy physical environment, the location of the neighbourhood in the spatial system, low perceived risk of being subject to crime, good access to commercial services, social services and transport. People often like to live in places where persons similar to themselves reside. A good position in the labour market position and a high income derived from this make it easier to live in a residence located in an attractive neighbourhood. Conversely, those with a weak labour market position, as well as minorities that are not fully accepted by the majority population, face more restricted opportunities for residential choice. 1 The term visible minority is used by the National Board of Health and Welfare for immigrants and children of immigrants from South East Europe (Greece and former Yugoslavia), Africa, Asia and Latin America (National Board of Health and Social Welfare, 2010: 184). We use this term for lack of a better. Labour market and economic outcomes for immigrants and children of immigrants from these countries differ so substantially from those of immigrants with a Nordic, Western or Eastern European background so as to call for separate analysis. 2 Se Corak (2006), Björklund and Jäntti (2009) and Blanden (2013). The last of these contrasts sociological and economic studies. 3 For a survey of the literature on mechanisms behind the intergenerational link in income see Black and Devereux (2011). Chapters in Smeeding et al (2011) and Ermisch et al (2012) are examples of recent contributions to the literature comparing countries. 3
In an article published in 2008 in the American Journal of Sociology, Patrick Sharkey examined intergenerational contextual mobility in United States defined as the degree to which inequalities in neighbourhood environment persist across generations. (Sharkey, 2008) Neighbourhood context was measured by the average income in the neighbourhood where the person resides. Sharkey found that more than 70 per cent of black children that grew up in the poorest quarter of American neighbourhoods remain in the poorest neighbourhoods as adults, compared to not more than 20 per cent of whites who started off from the same position. The intergenerational neighbourhood income elasticity was estimated to 0.64 a number higher than estimates of income or earnings elasticities computed at the individual level. Furthermore, results from estimated regression models indicated that most of the strong association cannot be attributed to measures of family background. As far as we know there has been no previous attempt to make a similar study for another country. Therefore we do not know at present, which of the results from Sharkey s study may be US specific, and to what extent not. The purpose of this paper is to use the same analytical framework as Sharkey to study, instead, the extent of intergenerational persistence in context in Sweden. Needless to say there are large social and institutional differences between the United States and Sweden which can be expected to lead to differences in results. For example, Sweden has many ambitious welfare programs but also high taxes. A robust result from empirical research is that the distribution of household income is more equal in Sweden than US. 4 All evidence tends to show a larger intergenerational income mobility between sons and fathers in Sweden than in the US. 5 The US has a disadvantaged Black population, many of whom are descendants of persons who originally were brought to the country by force. Afro-Americans have lived in US for centuries while the population of visible immigrants in Sweden is much more recent. 6 Seen from those perspectives it is surprising that we find several similarities between our results for Sweden and the results reported by Sharkey s for the US. Needless to say there are also clear differences. Why have there been so few studies of intergenerational neighbourhood relations? Most likely, this is due to problems of finding adequate data. To make an analysis, it is necessary to follow individuals over a fairly long period of time and to measure average income at the neighbourhood level. Sharkey worked with the Panel Study of Income Dynamics, a sample survey. He related the average income in the census track where the person lived when aged 26 years or older to the average income of the census track where the same person lived with his or her parents when aged 18 years or younger. The full sample for the analysis consisted of 4 464 individuals. A difference between Sharkey s study and ours is that our analysis is based on register data. This allows us to work with a much larger number of observations: 36 025 individuals. We analyse one single birth cohort, persons born in 1974 who lived in Metropolitan Sweden at age 16. In our empirical analysis, the average income in the neighbourhood where the persons lived at age 32 is related to the average income in the neighbourhood of the parental home, observed when the person was 16 years of age. 4 See for example OECD (2008) and studies that have used data from the Luxembourg Income Study, for example Brandolini and Smeeding (2009). 5 See Björklund and Jäntti (1997) as well as Jäntti et al (2006) who compared intergenerational earnings mobility between Sweden and United States using harmonised data. 6 The term visible minority is used by the National Board of Health and Welfare for immigrants and children of immigrants from South East Europe (Greece and former Yugoslavia), Africa, Asia and Latin America (Socialstyrelsen, 2010). 4
The paper is structured as follows: In the next section we discuss the importance of place for a person s well-being and development. In section 3 we discuss the Swedish context, and Section 4 describes the research design. Results on persistence in economic context across generations are presented in Section 5 while Section 6 contains an analysis of why there is a high intergenerational persistency in context. Finally, Section 7 summarises the findings of the study. 2. The importance of place and the life cycle For understand individual s behaviour as well as many types of inequalities there is often reason to look beyond the individual and the household in which he or she lives. People interact with non-household members and this can often influence their lives. An important aspect of people s lives is place of residence. People, particularly those who are not gainfully employed, spend much of their day-time in their own home and the neighbourhood where it is situated. Many factors affect how attractive a neighbourhood is considered to be. Some relate to physical characteristics of the neighbourhood, for example, the type of residential buildings and how they are placed in relation to each other; the quality of the air, and access to recreation facilitiese. Another type of characteristics of a neighbourhood that people value is its location in the spatial system: The time and resources required for traveling to work as well as to commercial and to cultural centres and places of recreation. A third type of characteristics which are often positively valued is the presence and quality of shops and commercial services within the neighbourhood. Needless to say, many parents place high value on the availability of good quality day care and schools within close distance when they consider where to live. The access to good education, social services and healthcare is thus a fourth type of characteristics of a neighbourhood valued by many. Neighbourhoods differ not only by physical characteristics. They also differ by who are living in them and what social activities take place there. Such circumstances constitute yet anothergroup of reasons why some places are more attractive than other. The risk of being exposed to crime typically varies across locations. In some neighbourhoods many people are members of civic organisations (sports-clubs, cultural clubs, choirs etc.) while in others this is not the case. The likelihood of finding neighbours matching in social traits, cultural status and interests and thereby easier to socialise with also varies across locations. Similar to this, a substantial number of persons also prefer to live close to neighbours having the same race or ethnicity. In a market economy, adults can choose where to live, within their budget restriction. Some people prefer to live close to where they grew up, for example because they have relatives and friends living there. 7 People with high income often choose to live in neighbourhoods that are attractive in the eyes of the majority population. By contrast a residence in an attractive neighbourhood is often beyond the reach of households with low income and without wealth and can be difficult to obtain for persons belonging to a racial or ethnic minority. From this follows that household income (present or the future) is a powerful predictor of a household s choice of residence. However, to change residence involves transaction costs, economic and social. Therefore there is a tendency for households to 7 For an empirical study on Swedish circumstances see Hedman (2013). 5
remain in the same home for a number of years. Due to this, as well as to other circumstances (ownership of wealth, social capital etc. and due to market regulations) the relation between present household income and neighbourhood average income is far from perfect. In addition, residential choice can have a racial and ethnic aspect. People belonging to minorities are not always welcomed as tenants or owners even if they have the financial means. Residential segregation can thus bee understood from the perspective of demand for housing which is much influenced by economic as well as demographic circumstances. There is also a supply side, however. Some neighbourhoods consist mainly run-down multi-apartment buildings and therefore have low housing costs. In other neighbourhoods most residences are new, of high quality, spacious but also expensive. In still other neighbourhoods there is more of a mix in housing quality and costs. City planning therefore has a potential of decreasing or increasing residential segregation, at least in the long run. As stated in the introduction in this paper we follow Sharkey s approach in operationalizíng context as mean household income in the neighbourhood. We focus on the relation between mean household income in the neighbourhood where the person lived when aged 32 and mean household income in the neighbourhood where the persons lived when aged 16. During the 16 years between the two points of measurement almost all young adults have left their parental home, a large majority have completed their formal education and most have also found a job. Many have also found a partner and a substantial number have become parents. Parallel to those changes most person have moved at least once from one neighbourhood to another. We now discuss different trajectories in neighbourhood average income during the first period of a person s adult life. There are good reasons to expect a larger mobility across neighbourhoods at different levels of average income among persons who grew up in high and middle high income neighbourhoods. 8 A substantial number of trajectories starts in an above average income neighbourhood continue with a move to a neighbourhood with a relatively low income after the young person has finished secondary education: The young person has left the relatively well-off parental home to study or to look for or take up their first job. During this phase of life the person usually has low purchasing power. Many are single and want a small apartment and these are often found in low income neighbourhoods. Later on, after having found a stable job and a partner, residential mobility often means a move to a neighbourhood with a higher average income. Eventually the person lives in a neighbourhood which in economic circumstances are relatively similar to those where he or she grew up. While young adulthood is characterised by very high spatial mobility, the propensity to move decreases with age. From this argument follows that one can expect the intergenerational correlation in average neighbourhood income will be low, when the second generation are in their early 20s, but will increase as the person gets older. The intergenerational transmission of education and its influence on labour market status is one reason to expect a positive intergenerational link in neighbourhood average income. However, there are also other possible channels of influence that may be important. Parents who have the resources to help their children find a home of their own or to help them buy it, tend to live in more affluent neighbourhoods. Other young adults do not have similar parental resources to fall back on. The latter is the case for many children of low-income including 8 For a statistical analysis of neighborhood residential mobility among young adults in Stockholm Metropolitan region 1990 to 2008 see van Ham et al (2012). 6
many children of immigrants. In a smaller number of cases there is even a direct intergenerational transmission of housing: Parents or some other relatives may give their house or flat to the young person or they may divide their land in a manner that allows the offspring to build a home on one part of it. A real estate owned by a parent can also make it possible for the adult child to take up a mortgage loan. 9 Race or ethnicity may be correlated with neighbourhood average income because when discrimination excludes people belonging to some identifiable group from housing in attractive areas. 10 3. The Changed Swedish scene During a long period, income inequality in Sweden was gradually decreased 11 but since the beginning of the 1980s the direction of development has been reversed. While real income at the bottom of the income distribution has changed little since the early 90s, at the top increases have been large. Thus, the distribution of household income has become more unequal (OECD, 2011, Fritzell et al 2012, Björklund and Jäntti, 2013). Several interrelated factors have contributed to this development. An important one is increased earnings and wage inequality which in turn can be traced to changes in the labour market and its institutions. (M Gustafsson, 2008, Domeij, 2008) The increased importance of capital income which is concentrated to the upper part of the income distribution is another force that has made the distribution of income less equal. Another source of rising household income inequality is elements in the extensive tax reform that was introduced in the beginning of the 90s: The progressivity in tax rates was reduced and the tax base split into two, capital income and earnings. Since then all capital income is taxed at a flat rate of 30 per cent, while previous to the reform it was added to other sources of income resulting in a higher or much higher marginal tax rate for high income earners meant. For many decades Sweden was known to have low open unemployment, partly because many workers participated in active labour market programs. This changed in the beginning of the 1990s. At that time, the downturn in the world economy hit Sweden sharper than many other rich countries. A large number of people left the work force. In particular, employment-rates among young adults fell sharply from a level which it has never fully recovered since then. The downturn also resulted in a serious debt-crisis for the public sector. Cuts in public sector programs and reforms of economic regulations and of the public sector became a top priority on the political agenda. When the economy had recovered at the end of the 90s most cuts in benefit systems were undone. To investigate the consequences of the turbulent 90s the government appointed a committee with the task of analysing the development of living standards for the population. Its final report identified three population groups as the losers of the 90s: Young adults, immigrants and single parents (predominantly single mothers). (Palme et al, 2002). Young adults, for example, had greater difficulties than previous cohorts in establishing themselves in the labour market. While most groups of the Swedish population 9 Öst (2012) show based on Swedish data that parental home ownership seems to play a primary role in determining whether the child becomes a homeowner. 10 For example Ahmed and Hammarstedt (2008) show based on a Swedish field experiment that took place in 2007 that males with Arabic/Muslim names are less likely to receive call backs when applying for a vacant rental apartment than males with a Swedish name. 11 Roine and Waldenström (2010) as well as Gustafsson and Jansson (2008) show that during the first half of the preceding century the distribution of income became more equal. Breen and Jonsson (2007) studied social fluidity during the period 1976 to 1999 due to younger anmd more fluid cohorts replacing older and less fluid ones. 7
had higher income on the eve of the 2008 world financial crisis than they had in in 1991, this was not the case for persons aged 20 to 25 (Johansson and Palmer, 2010 see also Fritzell et al, 2007). Many reforms that have taken place in Sweden since the beginning of the 90s aimed to increase the role of market allocation in the provision of public services. Much of housing policy has been dismantled. For example subsidies for housing construction have been sharply reduced and public expenditures on housing halved from 1.0 per cent of GDP in 1996 to 0.5 per cent in 2010 (Statistics Sweden). Although social services, including education, have continued be predominantly funded by the public sector, more and more of their production is undertaken by private organisations. Parallel to this users have been given more opportunities to choose providers; private or public. (Hartman ed., 2011) A debated topic is whether these reforms have been to the benefit of the public, in particular those worst off.. One of the largest changes Sweden has experienced during recent decades is a transition to a more ethnically mixed society. While the foreign born population numbered less than 2 per cent in 1950 a large number of immigrant waves have made the proportion increase to not less than 15 per cent at the end of 2011. Most immigrants who arrived during the 1950s and 60s were labour migrants who originated from countries within Europe. Finland was the largest sending country. During later decades, however, most immigrants who arrived in Sweden came due to political reasons or for family reunification. Many originate from Muslim countries and tend to be easy to distinguish from the majority population, by physical appearance and by name. The largest sender country has been Iraq. Statistics show that immigrants to Sweden who are visibly different from the majority by physical appearance and name fare considerably worse than the native population as well as western immigrants. The numbers of young adults that are visible immigrants have increased rather rapidly. The data source we describe in the next section shows that in 1990 8 per cent of 16 year old persons living in metropolitan areas of Sweden could be counted as belonging to visible minorities, according to our definition and 84 per cent had parents who were both born in Sweden. 12 In 2006 the corresponding proportions were 17 and 77 per cent, respectively. The gap in employment rates between immigrants and natives is in Sweden one of the largest among rich countries. (Dustman and Frattini, 2011) It has become difficult for many immigrants to Sweden, particularly those who belong to the visible minorities, to find a job and to keep it. There are several reasons for this. Very few immigrants master the Swedish language when they arrive in the country and most have limited knowledge about Sweden. As we will see in the description of our sample in the next section, among the parents of the cohort that we study visible immigrant parents have shorter average education than parents born in Sweden. Some immigrants have been exposed to war or repression, some have grown up under conditions of poverty. Because of this, a higher fraction of refugees than of natives of the same age, suffer from poor health. Furthermore, in Sweden as in many other countries many job offers are received through informal channels to which immigrants often have less access than natives. 12 tthe remaining 8 percent had a mixed background) 8
Other reasons why immigrants to Sweden face larger problems in finding a job, in particular a permanent job, are to do with attitudes of natives. The overwhelming proportion of all staff hiring workers is native. By now a substantial number of empirical studies have convincingly shown that employers discriminate against job applicants by name. Hypothetical job applicants with foreign-sounding names on their CVs are considerably less likely to be called in for a job interview than job applicants with native-sounding names are. 13 There is also empirical research indicating that people with foreign-sounding names enjoy a better much better income development if they change to native-sounding names. 14 In addition to this, there may be reason to do with Swedish labour market institutions. Collective barging has resulted in relatively high minimum wages in Sweden, reducing the demand for less qualified workers, for example workers with a limited knowledge of the Swedish language. It can be argued that in a hypothetical situation with a larger number of low-pay jobs, a higher proportion of immigrants would have been employed and earning wages. Certain welfare state arrangements may also contribute to lower participation rates for those who would earn a low wage. For example the combined effect of the progressive income tax, income related fees for public day care, municipal childcare allowances for parents who do not use public childcare and benefits that are negatively related to income results in low incentives for both parents to work in a household with many children. Immigrants lack of integration into the Swedish labour market has several consequences. For example, many recently arrived immigrants are not gainfully employed and therefore receive means-tested social assistance. Among immigrants who have lived in Sweden for more than a decade, the proportion receiving disability pension is higher than among natives. 15 There is now a large gap in disposable household income between natives and visible minorities and this is an important reason why visible minorities often live in other neighbourhoods than the majority population. Sweden s urban areas have become ethnically more segregated than previously. At the same time, with increased income inequality and higher levels of unemployment there has been an increase in socio-economic segregation in urban Sweden. 16 In the face of this development and in 1998, for the first time, the government formalised a metropolitan policy for Sweden. This policy consists of programs aiming to support disadvantaged areas. 17 4. Research design Sweden has three metropolitan areas with a combined population of 3.3 million. The most populous region is the Stockholm metropolitan region, including and surrounding the capital located on the east coast. In this study, 24 municipalities (city level units) are included in this region. The second largest, the Gothenburg metropolitan region is located on the west coast and includes eight municipalities. The smallest metropolitan region is located in the south of the country and consists of the city of Malmö and eight neighbouring municipalities. The 13 Ahmed and Ekberg (2009) survey a relatively large number of studies. On the hiring process se for example Åslund et al (2009). 14 Arai and Skogman Thoursie (2009) show that it pays off to change from a foreign-sound to a Swedishsounding name. 15 Gustafsson (2013) surveys studies of the disparities in social assistance receipt between immigrants and natives in Sweden. For a study of disability pension receipt se Österberg and Gustafsson (2006). 16 See for example Bitterman et al (2008). 17 On measures of the increase in segregation see Bitterman et al (2008). The metropolitan policy is described in Minisitry of Finance (1996), see also Andersson (2006). 9
population studied in this paper is persons born 1974 who lived in one of the regions in 1990, when they were 16 years old, and who also lived in one of the three metropolitan areas in 2006 when aged 32. 18 We use register information compiled by Statistics Sweden on all persons registered as residing in one of the three metropolitan regions, not a sample. In total, 40 925 persons born in 1974 were registered in 1990. We follow the 89 per cent (36 244 persons) who in 2006 lived in one of the metropolitan areas. Thus, we do not follow persons how have moved to a place outside the metropolitan areas (10 per cent of the original population). Only 1 per cent of the original population had died or emigrated. From the registers kept by Statistics Sweden we derive a number of variables at the household level for year 1990 and 2006, see Table 1. In this study we use a definition of neighbourhood which has been constructed for research purposes. The neighbourhood is defined as a population area that: - is demarcated by natural borders (large streets, green areas, etc.). - corresponds to a city district or a residential area. - possesses a number of inhabitants large enough to provide the basis for certain private or public services (most often between 4 000 and 10 000 inhabitants) - is defined as a neighbourhood by its inhabitants. Unlike definitions used in some earlier Swedish studies, it is identical for all three regions studied. It has been previously used for mapping how residential segregation has developed (National Board of Health and Social Welfare, 2010). We focus on urban neighbourhoods and have excluded rural neighbourhoods and neighbourhoods with fewer than 500 observations. From the data, we also calculate neighbourhood characteristics based on all households in the specific neighbourhood. We have calculated several variables showing the demographic composition, including education of adults in the neighbourhood in 1990. Of particular interest is the average disposable household income among all persons in the neighbourhood in the two years under study: 1990 and 2006. An important component of a household s disposable income is wages subject to income tax. In addition, there can be income from business as well as capital income - dividends and interests as well as income from capital gains from selling stocks and property. This information was derived from tax files, by Statistics Sweden. Other income components we include in the household income are receipt of social insurance benefits (such as pensions, sickness benefits and unemployment compensation) and transfers such as child allowances, housing benefits and social assistance. Statistics Sweden also obtains this information from various registers of the authorities paying the transfers. Statistics Sweden also obtains information from the tax authorities on income taxes paid by the households and after subtracting this from the sum of all income components we obtain disposable income. We derive our variable average household income in the neighbourhood by adjusting the disposable income of each household with a frequently used equivalence scale. 19 The income derived in this way is assigned to each member of the household. We arrive at average neighbourhood household income by taking the average for disposable incomes of all individuals living in the neighbourhood irrespectively of their 18 The motivation for choosing this particular age cohort is that our data-base contain information from 1990 to 2006, and therefore is the cohort born 1974 is the only we can follow from being a teenager to age 32. 19 We use an equivalence scale used by the National Board of Health and Welfare, which starts at 1.16 for the first adult person in the household, is set to 1.92 for two adults, and ads weight of 0.56, 0.66 and o.76 for each child aged 0 3 years, 4 10 years and 11 17 years respectively. Ministry of Justice (2002, 89 132) discusses the issue of choice of equivalence scale from the perspective of Swedish experiences. 10
age. 20 We also use a variable which indicates the proportion of persons of working-age who are transfer dependent in the neighbourhood.. 21 /Figure 1 about here/ Figure 1 shows Kernel density estimates of the distribution of average neighbourhood household income (as defined above) in 1990 and 2006 in the three Metropolitan areas combined. It is very clear that the distribution has shifted to the right, which means that average neighbourhood income has increased rapidly. It is also very clear that the distribution in 2006 is much more unequal than the one in 1990. Thus, residential segregation measured in this way has increased rapidly. /Table 1 about here/ Table 1 contains descriptive statistics for the total sample and the subsample of visible minorities and the majority. 22 We will make some comments on it. A first comment is that, on average, the cohort under study has experienced an increase in household income as well, although the increase was not as rapid as for average neighbourhood income. The table clearly shows that household disposable income is lower among visible minorities 1990 as well as 2006 Among visible immigrants there is a variation in years since immigration. Slightly more than half had arrived before 1980 so when they were 16 years old, most of the visible immigrant children had spent the greater part of their lives in Sweden. The visible minority and the majority parental households differ in some respects. In the former there is a larger number of children and the proportion economically excluded is almost three times larger proportion than among the majority. The education level is lower among the minority, in both generations. For example in as many as 42 per cent of minority parent household no adult had more than at most compulsory education, as compared to 15 per cent among the majority. The last part of Table 1 shows characteristics of the young adults. We find that when they are adults, in the 1974 birth cohort, the visible minorities tend to have lower levels of education, higher rates of receipt of social assistance and unemployment compensation. A larger proportion visible minority children had become a parents at an early age and a larger proportion were single parents. Following Sharkey, we investigate the relation between neighbourhood average household income by calculating matrixes for origin and destinations in the distribution of average neighbourhood income. We also estimate intergenerational neighbourhood elasticities (see Equation 1) and correlations (Equation 2). As we have found that the dispersion of neighbourhood average income in 2006 was much larger than in 1990 the correlation coefficients will be lower than the elasticities. As a comparison, we also estimate intergenerational household income elasticities. We expected these to be lower than the intergenerational neighbourhood elasticities. Note that we are relating household income across generations and thus focus on how opportunities for private consumption are related 20 Some measurement problems make our measure of average household neighborhood income somewhat noisy. For example undeclared earnings are not covered in the data. 21 Defined as persons who during a calendar year received public transfers (invalidity pensions, social assistance sickness compensaton etc) amounting to a full year worker earning a low income. 22. Majority for our purposes means not visible minority i. e. those whose parents are born in Sweden, Europe other than South Eastern Europe, Anglo-America or Oceania. 11
across generation. This differs from most of the literature on intergenerational income or earnings where the position in the labour market is in focus and individual s earnings or income are related across generations. (1) = (2) = Y 1 is average income in the neighbourhood where the person lives in 2006 and Y 0 is average income in the neighbourhood where the person lives in 1990. 0 and 1 represent the variance of Y 0 and Y 1, respectively. 5. Persistence of Neighbourhood Economic Status / Table 2 about here/ Table 2 reports our results as regards intergenerational income elasticities and correlation coefficients for neighbourhood average disposable income for the pooled samples and for each of the sub-samples. It also shows the corresponding numbers derived from the analysis of household income. We find that the regression coefficients (the elasticities) are much higher than the correlation coefficients. We also find that neighbourhood income elasticities and correlations are considerably higher than the corresponding measures of association at the household level. This is similar to what Sharkey found for the US. The elasticity for neighbourhood average income estimated for our entire sample is as high as 0.81, but the elasticity for household income is not more than 0.27. The corresponding correlation coefficients are 0.38 and 0.12. Looking at the sub-samples we find a somewhat strong association in context across generations within the visible minority population than within the majority population: While the elasticity is 0.77 among the majority it is as high as 0.88 among visible minorities. How does persistence in context in Sweden compare with what Sharkey has reported from the US? One should keep in mind that to some degree differences in results can be due to differences between the definitions of neighbourhood as well as between the samples. Therefore minor reported differences do not necessarily mean that there are differences in reality. However, the differences between our results for Sweden and those reported by Sharkey for US are not minor: Sharkey reported an elasticity of 0.64 which was equal to the correlation coefficient. Thus our intergenerational elasticity for Sweden 0.81 is clearly higher, but the correlation coefficient 0.38 is clearly lower. /Table 3 about here/ We now move to measures of intergenerational association derived from mobility matrixes. Such measures are not influenced by changes in the dispersion for neighbourhood context 12
across generations. Following Sharkey, we have divided the sample into quartiles according to neighbourhood average income in each generation. In Table 3 we report outflow values calculated for each of the two sub-populations. The numbers are within-row-percentages. We show that there are large differences between the two sub-populations, as Sharkey found between Blacks and Whites in US. Among visible minority persons, who grew up in the poorest quartile of neighbourhoods in metropolitan Sweden, as many as 63 per cent lived in the same category of neighbourhood when they were 32 years old as well. This can be compared to only 35 per cent among those in the majority population who started from the same position as teenagers. We also see that the proportion downward mobile from the highest to the lowest quartile of neighbourhoods was 23 per cent among the visible immigrants while the corresponding number in the majority population was only 12 per cent. By contrast, the proportions who lived in neighbourhoods in the top quarter of the distribution both as teenagers and as adults are almost identical for the two sub-samples; slightly less than half. The large difference between the populations in the likelihood of moving out from a bottom-position parallels Sharkey s finding of differences in mobility between Blacks and Whites in the US. However, the proportion staying in a low position across generations was still higher in the US study than for the two Swedish populations: 72 per cent among Blacks remained in the bottom quartile and 40 per cent among Whites. The proportion Whites remaining in the top quartile in the distribution for US was 63 per cent which is higher than the 44 per cent that we have report for the majority population in Sweden. (See Table 3: 44). The cross country comparison of outflow values thus leads to the same conclusion as comparing correlations in Table 2: Context is more persistent across generations in US than in Sweden. /Table 4 about here/ In Table 4 we present the intergenerational relation in context by reporting cell-percentages for the two populations in urban Sweden. Thus numbers in each panel of the table sum to 100. Again the difference between populations is clearly visible. As many as 32 per cent of the visible minority population in Sweden remained in the lowest quartile of the distribution for two generations, while the corresponding proportion in the majority population was not higher than 8 per cent, thus only one fourth as large. However, although as many as one third of all visible minorities remained in the least prosperous quartile of neighbourhoods, this is a proportion lower than the 52 per cent Sharkey reported for Blacks in United States. Consistent with this the 12 per cent of all majority persons who remained in the top quartile for two generations in Sweden is lower than the 18 per cent Sharkey reports for Whites in United States. /Table 5 about here/ Quartiles are rather broad income classes and there may be large differences within each. In order to investigate this we have split the dataset into deciles and document the results in the Appendix. Table 5 reproduces the results for the top and bottom deciles together with the corresponding results from Sharkey for United States. We see striking similarities between Sweden and US as regards the bottom decile of neighbourhoods. Slightly less than half of visible minorities who grow up in the bottom decile neighbourhood also lived in such a neighbourhood when aged 32, while the corresponding proportion among blacks in US is 13
slightly higher than half. This can be contrasted with cent of the majority in Sweden and the whites in US who started off from the same position. Among them, 19 per cent also lived in such a neighbourhood when aged 32. However, this is only one side of the reality. A larger proportion of all Blacks in US than visible minorities in Sweden grew up in the poorest decile. This explains why while 25 per cent of Blacks in US lived in the poorest deciles in two generations, the corresponding number for visible minorities in Sweden was 14 per cent. Interestingly, only very small percentages of the majority population in Sweden and of Whites in US spend two generations in the poorest decile of neighbourhoods. Table 5 also indicates that intergenerational persistence in context at the top of the distribution is higher among whites in US than among the majority in Sweden. 6. Why is neighbourhood income so persistent? Following Sharkey, we investigate to what extent the relation between neighbourhood average incomes in the two generations remain if other characteristics are considered and amend the intergenerational model in equation (1) by introducing a number of other explanatory variables, in a stepwise manner. The models are estimated on the total population, both majority and the visible minorities using Ordinary Least Squares. First, we include a number of covariates measured at the neighbourhood level in 1990 when the child was 16 years of age (Model 2). They measure the proportion minority persons in the neighbourhood (two dummy variables) and the education level of the population in the neighbourhood (three dummy variables). We also include a variable measuring how large proportion of the population who are in work active ages in the neighbourhood and depend on public transfers expecting it to have a negative association with average neighbourhood income at age 32. In a second step we add variables referring to the household in which the young adult grew up (Model 3) and a dummy for gender. In Model 4 we also add a number of characteristics of the young adults themselves as observed when aged 32. As mean neighbourhood income at age 32 and personal characteristics measured at the same age might be influenced by omitted variables estimates in this and the following specification should not be regarded as showing causality. Finally, in Model 5 we add variables measuring different amounts of positive and negative capital income. A main reason why people declare negative capital income is that they have taken a loan in order to buy a home. We therefore expect the presence of a large negative capital income to be positively associated with average neighbourhood income at age 32. /Table 6 about here/ A number of comments can be made about the results in of Table 6. A main finding, which is consistent with what Sharkey reports, from his study of the US is that even after a large number of variables have been added to the estimated model, most of the association between neighbourhood incomes remains in force. Starting with variables measured at the neighbourhood level, we find that the proportion of adults depending on public transfers in the teenagers neighbourhood appears to have a considerable negative association on neighbourhood average income when they are 32 years. It thus appears that how households in the neighbourhood acquire their income, not only the amount of income received, affects 14
intergenerational neighbourhood mobility. Another interesting observation is that in all models, the coefficient for the share of visible immigrants in the neighbourhood in 1990 is positive and estimated with high t statistics. It suggests that the high persistency of visible minorities in low income neighbours should not primarily be understood as a taste for remaining in such neighbourhoods because of their ethnic composition. Turning to variables measuring household characteristics in 1990, there are also several observations to make. We find that the effect of belonging to a visible minority, on neighbourhood average income when aged 32, is negative, and somewhat larger if more recently arrived. Still the magnitude of the dummy coefficients is not high. We also find that the disposable income of the parental household has a positive effect on neighbourhood income at age 32. The same goes for parents education. The latter coefficients are not large, however. Coefficients for several variables added to Model 4 are estimated with high precision and have the expected signs: positive for contemporary income, longer education, being married or cohabiting, and for having the first child when being over 25. There are negative signs for dummy variables indicating transfer dependency. Finally we turn to the estimates of Model 5 and find that the coefficients for capital income are interpretable. A large positve capital income as well as a large negative one are both positively associated with high average neighbourhood income. In the first case causality probably runs from receipt of income to choice of location, in the second from choice of location (housing prices) to negative capital income. /Figure 2 and Figure 3 about here/ Most variables used in the various specifications reported in Table 6 are dummy-variables and therefore it is not difficult to interpret how strong the association is between the variable and mean neighbourhood income when aged 32. Exceptions are the variables transfer dependency in the neighbourhood and log income in the parental home. In order to better understand how those variables are associated with mean neighbourhood income at age 32 we have constructed Figure 2 and Figure 3. In Figure 2 we report how log mean neighbourhood income when aged 32 is associated with log mean income in 1990 for persons who grew up in neighbourhoods with varying proportions dependent on public transfers. This is done by showing one line representing the mean dependency ratio one for the decile of neighbourhoods having the highest ratio, and one for having the highest decile. We see that in comparison to importance of the variation in log average household income is the variation in transfer dependency of lesser importance. Figure 3 constructed in a similar way shows that the variation in log household income in 1990 has very little importance for log average neighbourhood income. 7. Conclusions Since some years, inequality in Sweden is rising. The distribution of household income has become more unequal and the population of visible minorities has increased rapidly. Many visible minorities have large difficulties finding a job. Residential segregation has increased and become an issue. Following a previous study, made by Sharkey for the US, we have analysed intergenerational persistency in context for persons born in 1974, who in 1990 lived in the Metropolitan regions of Sweden, by relating average household neighbourhood income 15
where people lived when aged 32, to the average household income in their neighbourhood when they were aged 16. The study is based on register data which allows us to work with information for 36 025 individuals and we used a definition of neighbourhood which was constructed for research purposes. A main conclusion is that the relation between the average incomes of the neighbourhoods where a person lives when aged 32 and aged 16 is strong, much stronger than the corresponding relation between household incomes for our population. The intergenerational income elasticity is as high as 0.81 meaning that very much of neighbourhood income differences at age 16 are preserved at age 32. However, as the dispersion in neighbourhood average income in metropolitan Sweden doubled across the years this is compatible with an intergenerational correlation of 0.38. The latter figurer, but not the former, is lower than what Sharkey reported for the United States. Similarly to the results of Sharkey for the US, we found that most of the strength in intergenerational neighbourhood relation remains when we include a number of other variables that affect or are related to average neighbourhood income. Among results from such an exercise we choose to point out that the composition of income within the neighbourhood seems to have an independent effect on neighbourhood average income at age 32. Another main finding is large differences in intergenerational persistence in neighbourhood income between visible immigrants and the majority population. Those differences parallel differences in the United States between Blacks and Whites as reported by Sharkey. In both cases, are we talking about people who are visibly different from the majority and who are underprivileged in terms of education and income. This shows up in a particular concentration to less privileged neighbourhoods. Such a concentration is, however, not as strong in Sweden as in the United States. Keeping in mind that we have not harmonised definitions and samples with the study of Sharkey, some insights can be gained from the comparison between Sweden and US. It appears that the probability of leaving the decile of neighbourhoods having the lowest average income is similarly low if you are a visible immigrant in Sweden or a Black person in United States, while the probability of leaving the decile with lowest average household income if the person belongs to the majority in Sweden or is White in the United States is much higher. 16