Place Stratification or Spatial Assimilation? Neighbourhood Quality Changes after Residential Mobility for Migrants in Germany

Abstract

Neighbourhoods provide unequal resources and opportunities. Past research has shown that migrants are less able to move to more resourceful neighbourhoods. For Germany, cross-sectional evidence shows that migrants live in worse neighbourhoods on average, but no longitudinal analysis of changes in neighbourhood quality after residential mobility has been conducted. The present paper closes this gap and tests the place stratification model and the spatial assimilation model. Data from the German Socio-economic Panel and the MICROM dataset are used for the years 2000–09. The data are analysed using fixed-effects panel regression. The analysis shows that Turkish households are less able to improve their neighbourhood quality through moves compared with German households, while households with other ethnic backgrounds do not differ significantly from the native population.

Introduction

Local neighbourhoods are complex social entities that provide important resources and opportunities for their residents. Ideally, neighbourhoods are places of social interaction with neighbours, essential goods and services provision, and safe and healthy environments. Neighbourhoods are stratified by the opportunities and resources that they offer, where these may affect residents’ life chances (Alba and Logan, 1993). Living in underresourced neighbourhoods with fewer opportunities is associated with negative health, education and employment outcomes (for example, Brannstrom, 2004; Galster et al., 2008; Hastings, 2009). This has also recently been supported for educational attainment in Germany (Helbig, 2010). Despite this, the evidence for causal effects of neighbourhoods on residents is weak, especially in the European context (Dietz, 2002). Assuming that the ability to reside in better neighbourhoods improves life chances, it is of great sociological interest to investigate whether people are able to move to better neighbourhoods. However, previous research indicates that this path may not be open to all (for example, South and Crowder, 1998; Bolt and van Kempen, 2002). Migrants are especially less likely to improve their neighbourhood quality through residential moves than natives.¹ This may be a severe problem, because migrants often live in neighbourhoods of particularly low overall quality (Drever, 2004). Therefore, they would especially profit from moving to better areas where they could significantly improve their life chances.

The causes for the disparities in neighbourhood outcomes—i.e. the change in neighbourhood quality associated with a move—between migrants and natives are assumed to be threefold (South and Crowder, 1997; Özüekren and van Kempen, 2002; Schaake et al., 2010). First, the preferences of migrants may be different from natives. Thereby, natives and migrants choose different kinds of neighbourhood when moving and this causes disparities in neighbourhood quality. Secondly, the preferences of migrants and natives may be similar, but the realisation of these preferences may be impeded for migrants. On the one hand, realisation may be impeded because migrants on average have less financial means than natives and better neighbourhoods are more expensive to live in. Thirdly, on the other hand, realising one’s preferences may be impeded through discrimination in the housing market. In this case, gatekeepers such as landlords or real estate agents prefer natives over migrants, thus restricting the latter to neighbourhoods not preferred by the natives.

The goal of the present analysis is to show whether any disparities exist in neighbourhood outcomes associated with residential moves between migrant and native households in Germany. To explain potential disparities, I will test whether migrants move to worse neighbourhoods due to less financial means or whether they are discriminated against in the housing market. The present analysis focuses on movers, leaving aside the important problem of immobile residents who are potentially trapped in bad neighbourhoods.

There are several reasons why the effect of migrant status on changes in neighbourhood quality may be different in Germany compared with the previous analyses on the US and other European countries. First, the German housing market is characterised by relatively low rates of homeownership, a small and shrinking social housing sector, and a buoyant private rental sector which offers accommodation in good neighbourhoods (Diamond and Lea, 1992, pp. 79ff; Münch, 2009). Thus, it may be easier for migrants to access good neighbourhoods even if they cannot afford to buy a home. Secondly, the spatial disparities in Germany are smaller than in leaner welfare states due to state intervention and to planning policies (Schmidt and Buehler, 2007). Thirdly, the spatial segregation of migrants in Germany tends to be much weaker than in the US and other European countries (Musterd, 2005). This may indicate that discrimination in the German housing market is less than elsewhere, even though many migrants still experience it (Wiesemann, 2008). Migrants in Germany, moreover, exhibit a higher preference for spatial diffusion than in other countries (Hanhörster and Mölder, 2000). In addition, public policies aim explicitly at dissuading clustering of migrants and promoting social heterogeneity of neighbourhoods (Münch, 2009). Fourthly, residential mobility is relatively low in Germany (Myers, 1999). Households in Germany may be more determined to improve their neighbourhood qualities with these rare events.

Theory and Hypotheses

Neighbourhood quality is a vague concept and hence clarification is useful before analysing quality changes. To that end, the neighbourhood, its relevant characteristics and its delimitation are defined in the next section. Following this, the spatial assimilation and place stratification models are described in detail to derive the hypotheses which guide the empirical analysis.

The Neighbourhood and its Quality

As suggested by Elliott et al. (2007, p. 15), there are three core dimensions defining a neighbourhood: a neighbourhood is a relatively small spatial area constituted by a number of dwellings; inhabitants in a neighbourhood interact socially and face-to-face; and, inhabitants have some shared identity. A basic characteristic of a neighbourhood is the social composition of its residents, which may affect the life chances of the neighbourhood’s residents.² First, the social composition of the neighbourhood determines the quality of social interaction and the resources available through this interaction. For example, positive role models in a neighbourhood are a relevant resource with regard to the upbringing of children and these models are not equally distributed across all areas. It can be argued that middle and high socioeconomic status groups provide more resources than low socioeconomic status groups. In addition, social interaction in the neighbourhood is supposed to be greater and more beneficial if residential stability in the neighbourhood is high. High turnover rates will tend to reduce social interaction and increase anonymity in the neighbourhood (Atkinson and Kintrea, 2001; Elliott et al., 2007, pp. 42ff). Secondly, the social composition affects the institutional investment in the neighbourhood. Inhabitants influence the investment in their neighbourhood with their demand and by voicing their expectations. Neighbourhoods with a higher share of poor inhabitants have worse services and less shopping facilities because inhabitants are unable to voice their needs, or local businesses do not make enough profit to provide certain products. In addition, environmental stress due to noise and air pollution can be expected to be lower in neighbourhoods with higher socioeconomic status because residents are more likely to voice protest and can afford to move to less polluted areas (Atkinson and Kintrea, 2001; Crowder and Downey, 2010). The effect of the neighbourhood on its residents is moderated by the institutional environment. For example, better public transport in European cities enables residents to leave their neighbourhood temporarily more easily than in the US (Musterd, 2005).

Neighbourhood quality may be assessed via two indicators repeatedly used in past research on neighbourhood effects. First, the economic status of the inhabitants—often operationalised as a poverty measure—is widely used (for example, Quillian, 2003). The second indicator is residential stability, which is a proxy for intact social networks and familiarity with neighbours (for example, Blasius and Friedrichs, 2007). Residential instability, on the other hand, restricts the continuity of local interpersonal relationships “and the emergence of informal norms that guide and regulate behavior” (Elliott et al., 2007, p. 44).³ These indicators are two of many used in empirical research to describe neighbourhoods. Most prominently, ethnic concentration within a neighbourhood is another indicator of neighbourhood quality. However, the effects of ethnic concentration on inhabitants—especially migrants—are found to be ambiguous. Spatial concentration may increase social exclusion, or, conversely, concentration may provide access to resources for the ethnic community (Drever, 2004). Without engaging in this debate, the ethnic concentration in a neighbourhood can be considered as a distinct feature outside the general quality of the neighbourhood. However, the general quality may be correlated with ethnic concentration. In addition to objective indicators, the subjective perception of qualities of the neighbourhood may vary across individuals and consideration of subjective measures of neighbourhood quality is also desirable (Bolt and van Kempen, 2010).

Finally, the question of the spatial delimitation of neighbourhoods needs to be considered (Galster, 2001). It has been found that neighbourhood effects may vary by spatial delimitation. Hipp (2010) shows that the economic status and the residential stability in very small neighbourhoods, defined by the 10 closest neighbours, are more effective in explaining neighbourhood satisfaction than are larger areas, such as the census tract which is normally used for analysis of neighbourhood effects in the US. This is consistent with findings from Farwick (2009, p. 231) for Germany which show that very small neighbourhoods are more relevant for the analysis of individual behaviour of migrants than larger, more heterogeneous areas (also see Kearns and Parkes, 2003). Thus, small neighbourhood units are most relevant for individuals.

Neighbourhood Outcomes after Moves

The empirical analysis here is informed by the spatial assimilation and the place stratification models. The spatial assimilation model assumes that households prefer to live in neighbourhoods with residents similar to their own social status and life-course stage (Bolt and van Kempen, 2010). Thus, changes in preferences are caused by social mobility and family events. Recent research also highlights the importance of life-course events outside the family trajectory—for example, job loss—for neighbourhood outcomes (Rabe and Taylor, 2010). Preferences are assumed to be homogeneous for natives and migrants and the actualisation of the preferences is foremost a matter of financial resources (South and Crowder, 1997). In this theoretical approach, differences in neighbourhood outcomes are only due to unequal financial resources and it is assumed that discrimination does not constrain the mobility of migrants. Thus, following this model, it can be expected that

Neighbourhood outcomes are stratified by income similarly for native and migrant households (H1.1).

According to the spatial assimilation model, assimilation is also a function of the time that a migrant has lived in a country and the degree of general assimilation into the mainstream society amongst other factors (Bolt et al., 2008). Thus, it can be assumed that

Neighbourhood outcomes for first-generation migrant households deviate more strongly from the average of the natives than for second-generation migrant households (H1.2).

Higher income has been found to be positively associated with better neighbourhood outcomes (Clark and Ledwith, 2007; Schaake et al., 2010). There is no direct empirical support for the spatial assimilation model in Germany, but the relatively low ethnic concentration may indicate that dispersion is easier for migrants in Germany and thus that spatial assimilation plays a bigger role in Germany than in other countries with stronger ethnic segregation.

The place stratification model highlights the importance of structural constraints for neighbourhood outcomes (Bolt and van Kempen, 2003). Alba and Logan use the term place stratification to indicate that

places are ordered hierarchically and consequently are associated with more or less favorable life chances and quality of life for the people who reside in them (Alba and Logan, 1993, p. 1391).

To improve their life chances, households prefer to move to better-off areas. However, specific groups of the population, e.g. native Germans, manage to constrain access to these areas for members of out-groups, such as Turks, restricting the best areas for themselves. In opposition to the spatial assimilation model, the place stratification model assumes that, while native and migrant households have the same preferences, they are not equally able to actualise their preferences even with the same financial resources, because of structural constraints (South and Crowder, 1997). Thus, it can be hypothesised that

The effect of moves on neighbourhood outcomes is less positive for migrant households relative to native households on average (H2.1).

Further, it can be expected that

Financial resources have a significantly less positive effect on neighbourhood outcomes for migrant households compared with native households (H2.2).

Previous research provides evidence for both hypotheses. Univocally, the ethnic background of movers has been found to be a highly important determinant of neighbourhood outcomes (Magnusson and Özüekren, 2002; Schaake et al., 2010). For Germany there are no studies on the neighbourhood outcomes of moves, although Janssen and Schroedter (2007) analyses ethnic segregation in Germany. It can be argued that highly segregated areas in Germany have a low neighbourhood quality on average. Janssen and Schroedter find segregation to have decreased in the 2000s suggesting that migrants may improve their neighbourhood quality on average. At the same time, their analysis shows that, even when controlling for socioeconomic status, migrant households are more likely to live in highly segregated areas, which may indicate that migrant households cannot improve their neighbourhood quality to the same degree as native households with higher incomes. This finding is also supported in other outcomes of moves, such as housing tenure and room stress, for which migrants are worse off than native households in Germany after controlling for economic status (Clark and Drever, 2000).

A further model which will not be tested empirically in the present analysis, due to missing data on respondents’ preferences, is the ethnic enclave model (Schaake et al., 2010). The model assumes that ethnic concentration provides amenities and support for the migrant community. To profit from the benefits of living in an ethnic enclave, migrants may accept a lower neighbourhood quality on other dimensions.

Data, Sample and Operationalisation

For the household level, data are drawn from the Socio-economic Panel Study (SOEP), which was established in 1984 and is a nationally representative, annual panel survey of the German population including data on about 11 000 households and 20 000 individuals run by the German Institute for Economic Research. The migrant population has been oversampled to gain sufficient sample sizes to analyse this sub-group (Wagner et al., 2007); regardless, the case numbers for the migrant groups in the present analysis remain small. Inference in the multivariate analysis is only based on 581 Turkish households and 822 households with other backgrounds that move and thus may change their neighbourhood quality. This limits the generalisability of the present findings. The estimation sample consists of all households with the oldest member in the age range 18 to 79 years that have been interviewed at least twice between 2000 and 2009. The data are organised longitudinally and this leads to an analytical sample of 108 007 household-year observations from 15 524 different households including those that move and those that do not move in the observation period.

The following variables are used for the household level.⁴ The main explanatory variables are moving and ethnic background. Mover measures whether the place of residence has changed from the last interview (coded 1) or whether the household stayed put (coded 0). Regarding ethnic background, I differentiate between German, Turkish and other backgrounds. Turkish households are considered separately, because they are the largest ethnic minority in Germany and face some of the strongest discrimination in Europe (Pettigrew, 1998).⁵ The ethnic background is measured jointly at the household level for the household head and the partner in couple households and only for the head in non-couple households. The variable Turkish background measures whether the head or the head’s partner was born in Turkey, is a Turkish citizen, or whether either’s parents were born in Turkey (coded 1).⁶ If one partner has a Turkish background, while the other partner does not, the household is considered to be Turkish. If both partners are not Turkish, the variable is coded 0. The other background measures if the head or the head’s partner was born outside Germany and Turkey or one of them holds a citizenship other than German or Turkish (coded 1). If both partners are German or at least one partner is Turkish, the variable is coded 0. This category includes very heterogeneous migrant sub-groups, but cannot further be differentiated due to small case numbers. First-generation migrants, who have been born outside Germany, and second-generation migrants, who have been born in Germany by parents of whom at least one party immigrated, are differentiated. The variable is aggregated at the household level so that households with at least one partner being a first-generation migrant are considered as first-generation households; and households with at least one second-generation migrant but no first-generation migrant are considered second-generation households.

Additional independent variables are included. Income is measured as monthly, equivalised net household income. The income is averaged over a three-year time window and transformed using the log scale. I differentiate couple households from single/other households. The variable children measures the number of children below the age of 18 years in the household. The housing tenure is differentiated by owner, private tenant and social tenant. Age of the oldest person in the household is added as a control and also include as a squared term in case there are non-linear patterns, which may be caused by divergent moving behaviour over the life-course. Because the neighbourhood quality measure is strongly stratified by the degree of urbanisation of the area, a dummy coded 1 for urban areas and 0 for rural areas is included. The population density in the county is used to differentiate between urban and rural counties. Urban counties have an above average density, rural counties have not more than average density. Finally, dummy variables for the period and a dummy for East Germany are included in all models, but not shown in the result tables. I control for East Germany because of the different urban structure and average socioeconomic status compared with West Germany. In addition, the share of ethnic minorities in East Germany was only about 5 per cent in 2009, while the share in West German states was at least 13 per cent (in the state Schleswig-Holstein; Brückner and Führ, 2011). For the descriptive part of my analysis, subjective neighbourhood quality indicators from the SOEP are used that have been collected for the years 2004 and 2009.

The SOEP data are merged with information on neighbourhoods from the MICROM dataset. The MICROM dataset has been generated by a market research company for commercial purposes. The dataset can be merged to the SOEP at the household level from the year 2000 onwards. Most of the data are available at the building (on average 8 households) or street level (on average 25 households). If a building hosts less than 5 households, the building is pooled with similarly structured buildings in the direct proximity to protect privacy (Goebel et al., 2007). Due to the small size of the neighbourhood unit, higher variance in its characteristics over time is likely. In general, such small geographical areas will also result in more data ‘noise’ due to small sample size (Meen, 2009). On the other hand, these small-area data allow describing neighbourhoods at the level suggested in the literature.

The economic status and the residential stability have been identified as relevant characteristics of the neighbourhood in the literature review. The MICROM dataset offers variables on the average purchasing power as a measure for the economic status and on the average population turnover as a measure for residential stability. The purchasing power information is based on official tax data which are enhanced through estimations by the data provider. In the analysis, the average purchasing power per household in the neighbourhood is used. The variable is standardised for each observation year and East and West Germany separately, so that the annual mean equals 0 and a standard deviation is a one-unit change. The average population turnover is a categorical variable measuring the number of movers per 1 000 inhabitants. The variable is transformed so that 0 equals an average population turnover, negative values indicate an increasing turnover above average and positive values indicate a decreasing turnover below average. Both variables are combined in an additive index, in which the purchasing power is weighed more to account for the high importance of the economic situation in the neighbourhood. This neighbourhood quality index is also standardised. Thus, a one-unit increase in the index equals a standard deviation increase in the underlying unstandardised variable and a higher value indicates a better neighbourhood quality. As the goal of the present analysis is to identify the effects of residential mobility and I do not aim to model changes in the area itself over time, the mean neighbourhood quality across all years of a residential spell is assigned to a household. A residential spell is the time that a household lived at the same address. Changes in neighbourhood quality can thereby only occur if households move.

Results

Initially, I analyse whether the objective neighbourhood quality index that has been constructed actually measures relevant neighbourhood characteristics by comparing subjective evaluations of the neighbourhood with the objective measure. This step helps to contextualise findings on the change in neighbourhood quality and provides evidence for the validity of the constructed neighbourhood quality indicator. Next, the changes in quality will be analysed descriptively, before turning to the multivariate analysis to test my hypotheses. The multivariate analysis employs fixed-effects panel regression models. By separating within- and between-household variance, fixed-effects models allow controlling for unobserved time-constant heterogeneity. This alleviates the problem of self-selection in non-experimental data to some degree. Following from this, the models provide stronger support for causal claims not biased by self-selection processes, because change in the dependent variable can be related to changes in the independent variables. However, between-household variance is not considered in these models (Allison, 2009).

Validity of Neighbourhood Quality Measure

In 2004 and 2009, respondents of the SOEP were presented with some potential problems in their neighbourhood. Table 1 presents the average neighbourhood quality for those respondents that answered that this problem is relevant in their neighbourhood and for those that stated that the respective issue is not a problem in 2009. The results are similar for the year 2004. The upper panel of the table reports the results for the general sample. Across all issues, those reporting a problem also live in significantly worse neighbourhoods measured through the objective index relative to those not reporting a problem. The contrast is especially stark for crime. Those feeling unsafe in their neighbourhood because of crime live in areas that are more than half a standard deviation lower on the quality index than those respondents who feel safe (-0.61 compared with -0.09). This indicates that the quality index captures aspects of delinquency in neighbourhoods. The quality index also captures differences in the physical conditions of the neighbourhood. Respondents with lower objective neighbourhood quality are more likely to report noise, air pollution and lacking green spaces as a problem in their neighbourhood. As mentioned before, the objective index is strongly stratified by the degree of urbanisation of an area. If the results in Table 1 are only restricted to urban areas (lower panel of table), the differences between those reporting the respective problem and those that do not remain significant. This is strong evidence that the neighbourhood quality indicator also captures intraurban differences. However, it must be kept in mind that this neighbourhood quality indicator covers only a certain dimension of all neighbourhood characteristics. Other important dimensions, such as accessibility of infrastructure, may not be covered.

Table 1.

Subjective and objective neighbourhood quality indicators in comparison

Indicator	Average neighbourhood quality
	Problem	No problem
All areas
Crime	−0.61	−0.09
Noise	−0.21	−0.01
Air pollution	−0.23	−0.01
Lacking green spaces	−0.36	−0.05
Only urban areas
Crime	−0.63	−0.29
Noise	−0.39	−0.20
Air pollution	−0.39	−0.22
Lacking green spaces	−0.54	−0.23

Notes: All means are significantly different at 0.1 per cent two-tailed between problem and no problem. Respective dimension of neighbourhood constitutes a problem if respondent reports that noise, air pollution or lacking green spaces is at least a little bothersome. Crime constitutes a problem if respondent feels relatively or very unsafe in neighbourhood.

Data: SOEP v26 2009 (cross-sectionally weighted); MICROM 2000–09.

Descriptive Results

I find neighbourhood quality to be higher on average for households in rural areas compared with households in urban areas, for households with older members compared with households with younger members, for couples compared with singles, for families compared with childless households, for owners compared with tenants and for high–income households compared with low-income households (see Table 2). Movers live in worse neighbourhoods on average compared with stayers. This seems counter-intuitive, but is driven by the higher mobility of the urban population. Urban neighbourhoods have a lower quality on average and therefore movers tend to be in low-quality neighbourhoods. The average neighbourhood quality is stratified by ethnic background. On average, households with a Turkish or other background live in neighbourhoods of significantly worse quality than German households. The neighbourhood quality of Turkish households is also significantly worse than the neighbourhood quality of households with other backgrounds. Differentiating Turkish households and households with other backgrounds by first- and second-generation migrants reveals that the former are significantly worse off than the latter regarding their neighbourhood quality.

Table 2.

Average neighbourhood quality and average change

Variable	Average neighbourhood quality			Average change in neighbourhoodquality for movers
Variable	German (ref.)	Turkish	Other	German (ref.)	Turkish	Other
Total	−0.06	−0.38***	−0.20***	0.05	−0.01**	−0.00
Urban
Yes	−0.38	−0.63***	−0.44	0.01	0.00	0.00
No	0.17	−0.12***	0.03***	0.08	−0.03***	−0.01
Age
18–29	−0.48	−0.59	−0.53	−0.02	−0.04	0.05
30–44	−0.08	−0.39***	−0.25**	0.08	0.00**	0.10
45–59	−0.01	−0.33***	−0.14*	0.05	−0.00	−0.20*
60–79	0.03	−0.37***	−0.09*	0.06	−0.10* ^a	−0.07
Household type
Single/Other	−0.27	−0.60***	−0.42**	−0.02	−0.05	−0.08
Couple	0.10	−0.33***	−0.08***	0.11	−0.00***	0.06
Child(ren)
Yes	0.04	−0.37***	−0.19***	0.09	−0.04***	−0.01
No	−0.09	−0.40***	−0.21**	0.03	−0.03*	−0.07
Housing tenure
Owner	0.34	0.04***	0.19**	0.16	0.01**	0.05
Private tenant	−0.34	−0.49***	−0.35	0.03	−0.02	0.01
Social tenant	−0.75	−0.71	−0.62	0.01	0.00^a	−0.35^a
Income
1st quintile (lowest)	−0.35	−0.62***	−0.51**	0.02	−0.06*	−0.03
2nd	−0.10	−0.32***	−0.26**	0.01	0.04	0.04
3rd	−0.04	−0.29**	−0.10	0.07	−0.00	0.01
4th	0.04	−0.22**	−0.03	0.10	−0.05**^a	0.08
5th quintile (highest)	0.22	0.19	0.20	0.04	0.07^a	−0.14
Migration generation
1st	—	−0.42	−0.27	—	−0.01	−0.02
2nd	—	−0.19	−0.14	—	−0.07	0.04
Residential mobility
Stayer	−0.04	−0.37***	−0.18***	—
Mover	−0.31	−0.53***	−0.43*	—

N <100.

Notes: Differences from German households: *** significant at 0.1 per cent two-tailed; ** significant at 1 per cent; * significant at 5 per cent. Average change in neighbourhood quality from before the move to after move; characteristics measured before move.

Data: SOEP v26 2000–09 (cross-sectionally weighted); MICROM 2000–09.

The only groups for which the differences between German and Turkish households are insignificant are those in the age-group below 30 years, social tenants and households in the highest income quintile. Turkish homeowners also live in significantly worse neighbourhoods than German and other owners. In addition, Turkish households only slightly increase their neighbourhood quality in the age-groups above 30 years, on average. This is in stark contrast to German households and households with other backgrounds. For households with other backgrounds, I do not find significant differences from German households in the group of private tenants and in the third and higher income quintiles. These findings are initial evidence that the general differences between German and migrant households are not only due to divergent group characteristics, but also hold when controlling for a wide range of covariates. These findings are aligned with past literature on lower average neighbourhood quality for migrants (e.g. Drever, 2004; Bolt and van Kempen, 2002).

For the present analysis, the dynamics of neighbourhood quality are especially relevant. The average changes in neighbourhood quality for movers are significantly different for German and Turkish households, but not for German and other households (see Table 2). German households on average increase their neighbourhood quality, while Turkish households do not change their neighbourhood quality significantly through moves. The same is true for households with other backgrounds. Thus, migrants move horizontally rather than vertically with regard to neighbourhood quality (Bailey and Livingston, 2008). The finding that Turkish households on average do not improve their neighbourhood quality is alarming combined with the findings on the lower average neighbourhood quality of Turkish households. These households do not seem to be able to improve their worse neighbourhood situation through moves. Again, differences for first- and second-generation migrants can be observed, but the two groups are not significantly different in their average change in neighbourhood quality after a move.

Strong differences in neighbourhood quality change between Turkish and German households are apparent for couples and families with children. In both groups, Turkish households are significantly worse off. Turkish families with children experience a significant decline in their neighbourhood quality, the opposite being the case for German families. The findings also show that Turkish households in the lowest income quintile experience a decline in neighbourhood quality when moving, which is not the case for the German sub-sample. Further differences exist for older households. For both sub-samples, households with Turkish and other background, older households experience a decline in neighbourhood quality when moving. Again, this is not the case for German households. However, these findings need to be treated with caution because of the small sample sizes. In general, stronger differences are apparent between German and Turkish households than between German and other households without German or Turkish background. This may indicate that discrimination plays a stronger role in the neighbourhood quality outcomes of residential moves of Turkish households than other households. The difference may also be driven by divergent preferences—for example, Turkish migrants may have stronger preferences than other migrants to live in neighbourhoods with co-ethnics.

Multivariate Results

Table 3 presents results from longitudinal, fixed-effects linear regression models. The dependent variable is neighbourhood quality. A higher value indicates better neighbourhood quality. Model 1 includes all theoretically derived independent variables and has a satisfactory model fit of 8 per cent explained within-household variance. Model 2 and model 3 are extensions of this model and include interaction terms to test for the effect of migrant status on neighbourhood quality. The migrant status itself cannot be included in the fixed-effects models because it is a time-constant characteristic. Initially, the models are estimated for a pooled sample of all migrant groups. In the next step, the sample will be divided for these groups.

Table 3.

Fixed-effects regression models of neighbourhood quality

	Model 1	Model 2	Model 3
Urban	−0.598***	−0.598***	−0.593***
	(−13.58)	(−13.58)	(−13.38)
Age	0.017***	0.017***	0.017***
	(4.65)	(4.67)	(4.65)
Age squared	−0.000***	−0.000***	−0.000***
	(−4.04)	(−4.09)	(−4.07)
Couple	0.027*	0.028*	0.025*
	(2.32)	(2.29)	(2.03)
Number of children	0.014*	0.014*	0.013*
	(2.41)	(2.41)	(2.26)
Housing tenure (ref: private tenant)
Owner	0.146***	0.146***	0.147***
	(7.94)	(7.95)	(7.98)
Social tenant	−0.054***	−0.054***	−0.054***
	(−3.45)	(−3.45)	(−3.48)
Income	0.025*	0.024*	0.023*
	(2.22)	(2.08)	(2.03)
Income*Turkish		−0.005
		(−0.82)
Income*Other		0.010*
		(2.00)
Income*2nd-generation migrant			−0.001
			(-0.26)
Income*1st-generation migrant			0.010
			(1.55)
Mover	0.030***	0.035***	0.035***
	(5.56)	(5.74)	(5.77)
Mover*Turkish		−0.035*
		(−2.45)
Mover*Other		−0.011
		(−0.71)
Mover*2nd-generation migrant			−0.005
			(−0.23)
Mover*1st-generation migrant			−0.025
			(−1.92)
Constant	−0.536***	−0.534***	−0.537***
	(−3.87)	(−3.84)	(−3.85)
Observations	98 885	98 847	98 123
Households	15 245	15 230	15 119
Households with change in outcome	5 988	5 988	5 948
LL null model	−1 530.23	−1 548.64	−1 380.49
LL	2 844.59	2 851.54	2 944.55
R² within	0.08	0.08	0.08

Notes: Linear fixed-effects regression models, standardised dependent variable: neighbourhood quality; unstandardised coefficients, t statistics in parentheses. *** significant at 0.1 per cent two-tailed; ** significant at 1 per cent; * significant at 5 per cent. Models also include controls for period and East Germany not shown here.

Data: SOEP v26 2000–09 (unweighted); MICROM 2000–09.

Residential moves have a small, but clearly positive effect on neighbourhood quality. On average, a mover improves neighbourhood quality by 0.030 standard deviations. While this effect seems fairly small at first sight, it has to be noticed that this is the effect of a move controlled for a wide range of aspects that co-determine the outcome of mobility. For example, the move is controlled for the degree of urbanisation. Somebody moving from an urban to a rural neighbourhood would improve the neighbourhood quality by 0.628 (= 0.030 + (-0.598) * -1) standard deviations. Moves from rural to urban areas and the other way around have a strong effect—in fact, the strongest effect in the model—on neighbourhood quality. This variable also has the strongest explanatory power, explaining about 6 per cent of the within-variance in model 1.

According to H2.1, the effect of moves on neighbourhood quality should be less positive for migrant than for native households on average. This hypothesis is tested by including an interaction of movers with Turkish households and other households in model 2 (see Table 3). By including the interaction terms, the event of a move is allowed to have heterogeneous effects on German, Turkish and other households. A German household that moves improves neighbourhood quality by 0.034 standard deviations. Only the interaction term for Turkish households is significant and has a negative sign. On average, Turkish households do not change their neighbourhood quality through moves (0.000 = 0.035—0.035). Thus, for Turkish households, the hypothesis that moves are less beneficial for migrants in improving neighbourhood quality, even controlling for income, cannot be refuted. This finding is similar to past research in other countries (for the Netherlands: Schaake et al., 2010; Bolt and van Kempen, 2010; for the US: South and Crowder, 1998). Similar differences are not apparent for households with other backgrounds. However, households with other backgrounds are a very heterogeneous group and this may obscure differences for certain sub-groups.

In model 3, I test H1.2 which stated that neighbourhood outcomes for first-generation migrants deviate more from the average of native households than for second-generation migrants (see Table 3). I find neither first- nor second-generation migrants changing neighbourhood quality differently from German households or from each other. Thus hypothesis H1.2 is not supported in my analysis. This is in accordance with earlier findings that show only small differences in ethnic segregation between first- and second-generation migrants in Germany (Janssen and Schroedter, 2007). However, the groups of first- and second-generation migrants are very heterogeneous and, for example, different countries of origin are mixed. It was not possible to test for these differences separately for Turkish and other households because of small sample sizes.

Income has a significant, but only small effect on neighbourhood quality. This is rather unexpected because, theoretically, the effect of income on neighbourhood quality was assumed to be quite high and because previous literature suggests a strong income effect (for example, Clark and Ledwith, 2007). In fact, in cross-sectional models, which may be biased due to selection effects, the coefficient for income is estimated to be much higher.⁷ This discrepancy shows that, while income is a good predictor of between-household differences, it does little to explain within-household changes. A household with higher income is very likely to be in a good-quality neighbourhood, but income increases are associated with higher neighbourhood quality only rarely in comparison. This calls for a general reconsideration of the place stratification model and its causal claims about the role of income, support of which is mainly based on cross-sectional analyses so far.

Model 2 includes interaction effects for income and households with Turkish or other background (see Table 3). The effect of income on neighbourhood quality is not significantly different for German, Turkish and other households. H1.1, which stated that neighbourhood outcomes are similarly affected by income for native and migrant households, cannot be falsified. However, the stratification of neighbourhood quality by income is only modest. At the same time, H2.2, which assumes that financial resources have a significantly less positive effect on neighbourhood outcomes for migrant compared with native households, can be refuted. Once Turkish and other households have a higher income, they are as likely as German households to increase their neighbourhood quality through moves. This is evidence in support of the spatial assimilation model and against the place stratification model, but it should be kept in mind that income plays a small role in explaining neighbourhood changes.

The other covariates have the expected effects. The effect of age follows a curvilinear shape. Neighbourhood quality is improving faster for younger age-groups than for older age-groups. On average, couples live in better-off neighbourhoods than single/other households and the neighbourhood quality increases with the number of children in the household. Homeowners live in considerably better neighbourhoods than private tenants. Social tenants live in worse neighbourhoods than private tenants on average.

Further to investigate the differences in neighbourhood changes by ethnic background, separate models for ethnic sub-samples are estimated (see Table 4). Model 4 is estimated for the German sub-sample. The effects closely resemble the results from the pooled model, because of the high share of German households in the overall sample. Model 5 for Turkish households and model 6 for households with other backgrounds show effects mostly in the same direction as in the German sub-sample. For Turkish households, only the effects of age and mover have an opposite sign compared with the German sample. The latter difference was apparent from the interaction terms in model 3. However, only the strongest effect found in the German sample proves to be statistically significant for the other sub-samples: living in urban areas. The weak specification of the models for both migrant groups is probably due to the small sample size compared with the German sample. Nonetheless, the models tentatively point to the fact that changes in neighbourhood quality are explained by the same factors for migrants and natives. However, the overall increase in quality seems to be lower for migrant households.

Table 4.

Fixed-effects regression models of neighbourhood quality by ethnic group

	Model 4 (German)	Model 5 (Turkish)	Model 6 (Other)
Urban	−0.560***	−0.400***	−0.990***
	(−11.49)	(−3.87)	(−6.22)
Age	0.021***	−0.003	0.019
	(5.02)	(−0.33)	(1.64)
Age squared	−0.000***	0.000	−0.000
	(−4.18)	(0.47)	(−1.95)
Couple	0.021	0.034	0.031
	(1.54)	(0.75)	(0.78)
Number of children	0.015*	0.015	0.021
	(2.35)	(0.88)	(1.05)
Housing tenure (ref: private tenant)
Owner	0.143***	0.103	0.199**
	(7.33)	(1.58)	(2.78)
Social tenant	−0.069***	−0.032	−0.027
	(−3.43)	(−1.00)	(−0.82)
Income	0.016	0.041	0.045
	(1.30)	(0.93)	(1.20)
Mover	0.037***	−0.010	0.036
	(6.00)	(−0.29)	(1.25)
Mover*1st-generation migrant		0.014	−0.022
		(0.39)	(−0.65)
Constant	−0.598***	−0.450	−0.552
	(−3.61)	(−1.19)	(−1.42)
Observations	80 492	6 857	10 774
Households	12 420	1 198	1 763
Households with change in outcome	4 623	581	822
LL null model	1 956.37	824.69	−2 614.43
LL	5 628.85	985.34	−1 948.71
R² within	0.09	0.04	0.11

Data: SOEP v26 2000–09 (unweighted), MICROM 2000–09.

Conclusion

The present study has used longitudinal household data to analyse neighbourhood quality changes in Germany. The focus was on differences between the migrant and native population. The place stratification model and the spatial assimilation model were tested. The analysis shows that migrant households live in neighbourhoods with markedly lower quality than German households. Turkish households’ neighbourhood quality is especially low. Turkish households are also significantly less likely to improve their neighbourhood through moving. The strongest difference in neighbourhood quality occurs for those moving between urban and rural areas. Income plays a surprisingly small role in determining neighbourhood quality changes. Instead of income changes, transitions in the family trajectory and subsequent transitions into homeownership and rural areas seem to be more important determinants of quality changes in Germany.

None of the tested models is fully supported by the present analysis. The general disparities in neighbourhood outcomes with residential moves between German and Turkish households controlled for socioeconomic status refute the spatial assimilation theory. While German households improve their neighbourhood quality on average in the case of residential moves, Turkish households move between neighbourhoods with similar low quality on average. However, support for the place stratification model is only weak as well. Migrant and native households seem to use income to the same degree to move to better areas, but the overall stratification of neighbourhood quality changes by income is only very modest.

The differences between Turkish and German households may be driven by divergent preferences in accordance with the ethnic enclave model as Turkish households may prefer to live with co-ethnics. However, past qualitative research shows that Turkish migrants in Germany have heterogeneous preferences with regard to ethnic enclaves and only some want to live in ethnically segregated areas (Hanhörster and Mölder, 2000; Wiesemann, 2008). This explanation could not be tested explicitly in the present analysis.

Three main caveats apply to my findings because of data limitations. First, the quality indicator only captures a certain dimension of neighbourhoods. Future research should extend the quality indicator to take account of the multidimensionality of neighbourhoods. Secondly, the preferences of households were not considered. The consideration of preferences of Turkish and other households is mandatory to understand fully the dynamics of neighbourhood change. Thirdly, the small sample size for Turkish and other households prevented a further differentiation into first- and second-generation migrants as well as sub-groups of other ethnic groups and calls for caution in interpreting the results.

Footnotes

Appendix

Table A1.

Descriptive statistics

Variable	Mean	S.D.	Minimim	Maximim
East	0.24	0.43	0.00	1.00
Urban	0.40	0.49	0.00	1.00
Age	50.77	14.83	18.00	79.00
Household type
Single/other	0.32	0.47	0.00	1.00
Couple	0.68	0.47	0.00	1.00
Children	0.50	0.88	0.00	9.00
Housing tenure
Owner	0.48	0.50	0.00	1.00
Private tenant	0.49	0.50	0.00	1.00
Social tenant	0.03	0.18	0.00	1.00
Income	7.33	0.52	0.00	12.76
Mover	0.09	0.29	0.00	1.00
Ethnic background
German	0.82	0.39	0.00	1.00
Turkish	0.07	0.26	0.00	1.00
Other	0.11	0.32	0.00	1.00
Migrant generation
1st	0.04	0.21	0.00	1.00
2nd	0.13	0.34	0.00	1.00
Neighbourhood quality	0.00	0.94	−2.90	7.82

Data: SOEP v26 2000–09 (unweighted); MICROM 2000–09.

Acknowledgements

The data used in this publication have been made available by the German Socio-economic Panel Study (SOEP) at the German Institute for Economic Research (DIW), Berlin, which also provided a work place for analysing the data. The author is especially grateful to Jan Goebel from the DIW for his support and advice. The author also wishes to thank Rory Coulter, Nate Breznau, Daniel Horn and two anonymous referees for comments that substantially improved the paper. All remaining errors are the author’s alone.

Notes

Funding

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

References

Alba

R. D.

Logan

J. R.

(1993) Minority proximity to whites in suburbs: an individual-level analysis of segregation, American Journal of Sociology, 98, pp. 1388–1427.

Allison

P. D.

(2009) Fixed Effects Regression Models. Los Angeles, CA: Sage.

Atkinson

Kintrea

(2001) Disentangling area effects: evidence from deprived and non-deprived neighbourhoods, Urban Studies, 38, pp. 2277–2298.

Bailey

Livingston

(2007) Population Turnover and Area Deprivation. Bristol: Policy Press.

Bailey

Livingston

(2008) Selective migration and neighbourhood deprivation: evidence from 2001 census migration data for England and Scotland, Urban Studies, 45, pp. 943–961.

Blasius

Friedrichs

(2007) Internal heterogeneity of a deprived urban area and its impact on residents’ perception of deviance, Housing Studies, 22, pp. 753–780.

Bolt

Kempen

R. van

(2002) Moving up or moving down? Housing careers of Turks and Moroccans in Utrecht, the Netherlands, Housing Studies, 17, pp. 401–422.

Bolt

Kempen

R. van

(2003) Escaping poverty neighbourhoods in the Netherlands, Housing, Theory and Society, 20, pp. 209–222.

Bolt

Kempen

R. van

(2010) Ethnic segregation and residential mobility: relocations of minority ethnic groups in the Netherlands, Journal of Ethnic and Migration Studies, 36, pp. 333–354.

10.

Bolt

Kempen

R. van

Ham

M. van

(2008) Minority ethnic groups in the Dutch housing market: spatial segregation, relocation dynamics and housing policy, Urban Studies, 45, pp. 1359–1384.

11.

Brannstrom

(2004) Poor places, poor prospects? Counterfactual models of neighbourhood effects on social exclusion in Stockholm, Sweden, Urban Studies, 41, pp. 2515–2537.

12.

Brückner

Fuhr

(2011) Bevölkerung mit Migrationshintergrund [Population with migration background], in: Statistisches Bundesamt, Wissenschaftszentrum Berlin für Sozialforschung (Eds), Datenreport 2011 [Data report 2011]. Bonn: Bundeszentrale für politische Bildung.

13.

Castles

Miller

M. J.

(2003) The Age of Migration: International Population Movements in the Modern World, 3rd edn. New York: Guilford Press.

14.

Clark

W. A. V.

Drever

A. I.

(2000) Residential mobility in a constrained housing market: implications for ethnic populations in Germany, Environment and Planning A, 32, pp. 833–846.

15.

Clark

W. A. V.

Ledwith

(2007) How much does income matter in neighborhood choice?, Population Research and Policy Review, 26, pp. 145–161.

16.

Crowder

K. D.

Downey

(2010) Interneighborhood migration, race, and environmental hazards: modeling microlevel processes of environmental inequality, American Journal of Sociology, 115, pp. 1110–1149.

17.

Diamond

D. B.

Lea

M. J.

(1992) Housing finance in developed countries: an international comparison of efficiency, Journal of Housing Research, 3(1), pp. 1–271.

18.

Dietz

R. D.

(2002) The estimation of neighborhood effects in the social sciences: an interdisciplinary approach, Social Science Research, 31, pp. 539–575.

19.

Drever

A. I.

(2004) Separate spaces, separate outcomes? Neighbourhood impacts on minorities in Germany, Urban Studies, 41, pp. 1423–1439.

20.

Elliott

D. S.

Menard

S. W.

Rankin

Elliot

. (2007) Good Kids from Bad Neighborhoods: Successful Development in Social Context. Cambridge: Cambridge University Press.

21.

Farwick

(2009) Segregation und Eingliederung: Zum Einfluss der räumlichen Konzentration von Zuwanderern auf den Eingliederungsprozess [Segregation and integration: on the effects of spatial concentration of immigrants on integration]. Wiesbaden: VS Verlag für Sozialwissenschaften.

22.

Galster

. (2001) On the nature of neighbourhood, Urban Studies, 38, pp. 2111–2124.

23.

Galster

Andersson

Musterd

Kauppinen

T. M.

(2008) Does neighborhood income mix affect earnings of adults? New evidence from Sweden, Journal of Urban Economics, 63, pp. 858–870.

24.

Goebel

Spiess

C. K.

Witte

N. R. J.

Gerstenberg

(2007) Die Verknüpfung des SOEP mit MICROM-Indikatoren: Der MICROM-SOEP Datensatz [Linking SOEP and MICROM data: The MICROM-SOEP dataset]. Data Documentation 26, Deutsches Institut für Wirtschaftsforschung, Berlin(http://www.diw.de/documents/publikationen/73/diw_01.c.78103.de/diw_datadoc_2007-026.pdf; last modified 17 January 2008; accessed 22 November 2009).

25.

Hanhörster

Mölder

(2000) Konflikt- und Integrationsräume im Wohnbereich [Spaces of conflict and integration in residential areas], in: Heitmeyer

Anhut

(Eds) Bedrohte Stadtgesellschaft [The threatened urban society], pp. 347–400. Weinheim: Juventa.

26.

Hastings

(2009) Poor neighbourhoods and poor services: evidence on the ‘rationing’ of environmental service provision to deprived neighbourhoods, Urban Studies, 46, pp. 2907–2927.

27.

Helbig

(2010) Neighborhood does matter! Soziostrukturelle Nachbarschaftscharakteristika und Bildungserfolg [Socio-structural characteristics of neighbourhood and educational attainment], Kölner Zeitschrift für Soziologie und Sozialpsychologie, 62, pp. 655–679.

28.

Hipp

(2010) What is the ‘neighbourhood’ in neighbourhood satisfaction? Comparing the effects of structural characteristics measured at the micro neighbourhood and tract levels, Urban Studies, 47, pp. 2517–2536.

29.

Janssen

Schroedter

J. H.

(2007) Kleinräumliche Segregation der ausländischen Bevölkerung in Deutschland, Eine Analyse auf der Basis des Mikrozensus [Residential segregation of the foreign population in Germany: an analysis based on German microcensus data], Zeitschrift für Soziologie, 36, pp. 453–472.

30.

Kearns

Parkes

(2003) Living in and leaving poor neighbourhood conditions in England, Housing Studies, 18, pp. 827–851.

31.

Magnusson

Özüekren

A. S.

(2002) The housing careers of Turkish households in middle-sized Swedish municipalities, Housing Studies, 17, pp. 465–486.

32.

Meen

(2009) Modelling local spatial poverty traps in England, Housing Studies, 24, pp. 127–147.

33.

Münch

(2009) ‘It’s all in the mix’: Constructing ethnic segregation as a social problem in Germany, Journal of Housing and the Built Environment, 24, pp. 441–455.

34.

Musterd

(2005) Social and ethnic segregation in Europe: levels, causes, and effects, Journal of Urban Affairs, 27, pp. 331–348.

35.

Myers

S. M.

(1999) Residential mobility as a way of life: evidence of intergenerational similarities, Journal of Marriage and the Family, 61, pp. 871–880.

36.

Özüekren

A. S.

Kempen

R. van

(2002) Housing careers of minority ethnic groups, experiences, explanations and prospects, Housing Studies, 17, pp. 365–379.

37.

Pettigrew

T. F.

(1998) Reactions toward the new minorities of western Europe, Annual Review of Sociology, 24, pp. 77–103.

38.

Quillian

. (2003) How long are exposures to poor neighborhoods? The long-term dynamics of entry and exit from poor neighborhoods, Population Research and Policy Review, 22, pp. 221–249.

39.

Rabe

Taylor

M. P.

(2010) Residential mobility, quality of neighbourhood and life course events, Journal of the Royal Statistical Society: Series A, 173, pp. 531–555.

40.

Schaake

Burgers

Mulder

C. H.

(2010) Ethnicity at the individual and neighborhood level as an explanation for moving out of the neighborhood, Population Research and Policy Review, 29, pp. 593–608.

41.

Schmidt

Buehler

(2007) The planning process in the US and Germany: a comparative analysis, International Planning Studies, 12, pp. 55–75.

42.

South

S. J.

Crowder

K. D.

(1997) Escaping distressed neighborhoods: individual, community and metropolitan influences, American Journal of Sociology, 102, pp. 1040–1084.

43.

South

S. J.

Crowder

K. D.

(1998) Leaving the ‘hood’, residential mobility between black, white, and integrated neighborhoods, American Sociological Review, 63, pp. 17–26.

44.

Wagner

G. G.

Frick

Schupp

(2007) The German socio-economic panel study (SOEP): scope, evolution and enhancements, Schmollers Jahrbuch, 127, pp. 139–170.

45.

Wiesemann

(2008) Wohnstandortentscheidungen türkischer Migranten im Vergleich [Residential choice of Turkish migrants in comparison], in: Hillmann

Windzio

(Eds) Migration und städtischer Raum: Chancen und Risiken der Segregation und Integration [Migration and urban space: chances and risks of segregation and integration], pp. 193–211. Opladen: Budrich.