Long-term Neighbourhood Effects on Education,Income and Employment among Adolescents in Oslo

Abstract

The study of neighbourhood effects has spread within Europe over the past decade. This article extends previous European research by focusing on Oslo, Norway. The main question relates to individual development among adolescents: does the social composition of the neighbourhood affect the socioeconomic status later in life? The study applies a multilevel approach and utilises longitudinal register-based data. The results reveal small but significant effects of neighbourhood deprivation on educational achievement and, even less pronounced, on income. Some effects on unemployment are also observed, but only in the short run. The strongest associations obtain for concentration of welfare recipients in the neighbourhood, which emphasises the importance of social value and social participation. A crude comparison suggests that neighbourhood effects in Oslo are slightly smaller/larger than similar effects in Swedish/UK cities.

Introduction

Studies of neighbourhood effects often focus on child and adolescent development, and some on long-term outcomes (i.e. poverty, lack of skill, unemployment, social isolation, etc.). The results have been analysed in numerous review articles and can be summarised as follows. American research has documented an independent but modest effect of area deprivation on individual development and welfare (see reviews by Leventhal and Brooks-Gunn, 2000; Sampson et al., 2002; Pebley and Sastry, 2004). European research is much more scarce and flies in different directions. Andersson (2004) investigated three medium-sized Swedish cities (Västerås, Jönköping and Gävle), and found that neighbourhood context during adolescence affects employment status and educational achievements at age 25. A later study (Andersson and Subramanian, 2006) corroborated these results in relation to educational outcomes, now for the whole of Sweden. In Germany, Obertwittler (2007) reported significant neighbourhood effects on adolescent delinquency in Cologne and Freiburg—however, only among natives, and based on cross-sectional data. A third type of pattern was identified in a UK study of examination results and a Finnish study of secondary school completion. In these cases, a neighbourhood effect appeared primarily in areas of concentrated advantage (Gordon and Monastiriotis, 2006; Kauppinen, 2007).

It appears, therefore, that European research exposes greater variation than American research. This impression also applies to studies of adult populations. Swedish research has found clear impacts of residential context on labour market participation (Musterd and Andersson, 2006) and earnings (Andersson et al., 2007; Galster et al., 2008; Musterd et al., 2008). Dutch research (Musterd et al., 2003; van der Laan Bouma-Doff, 2007), German research (Drever, 2004; Friedrichs and Blasius, 2005) and British research (Buck, 2001; McCulloch, 2001; Bolster et al., 2007; van Ham and Manley, 2010) provides a less consistent picture: some analyses support a ‘neighbourhood effects story’, others do not. There may be some pattern to this diversity, but it is hard to detect.¹ Sweden seems to be the only European country with a fairly uniform evidence base.

The fact that Sweden stands out is highly relevant for the present paper. Drawing on Norwegian data, we expect to find a certain affinity with Swedish research. Norway and Sweden share a history of universalism and comprehensive redistribution. The social structure in the two countries is very similar and the cultural discourses tend to revolve around the same issues. Some relevant evidence has already been collected by a group of economists. Raaum et al. (2006) analyse the effects of neighbourhood characteristics (census tracts) on educational achievements and earnings in Norway as a whole. Their results point towards a sharply decreasing but still significant effect of residential location: “The impact of neighbourhood is reduced by half from 1960 to 1970” (Raaum et al., 2006, p. 200). They do not rigorously test explanations for the decrease, but link it to policy changes aimed at increasing equality of opportunity, like expanding municipal services and education reforms.

The lack of previous research has directed us towards a wide rather than a narrow thematic scope. The main question is phrased as follows: does the extent of deprivation in the neighbourhood during adolescence influence socioeconomic status later in life? One sub-question looks at consistency across three different outcomes: do we observe the same pattern for educational achievement, income and unemployment? A second sub-question traces short-term versus long-term impacts: does the effect of neighbourhood influence attenuate over time? This question can only be analysed for unemployment, given the character of the educational variable (educational level) and the design of the study. Then, finally, we also address the extent to which different neighbourhood indicators yield a similar outcome: are there important differences between indicators such as unemployment, low income, low education and receipt of welfare benefits?

Our paper restricts the geographical scope to Oslo, the capital of Norway. This city has a deep-rooted geography of deprivation and disadvantage. The level of poverty lies far above the national average and so does the level of income inequality. Factors such as ethnicity, family structure, place stigma and housing market pressure compound the poverty challenge and set Oslo apart from the rest of Norway. Oslo, in short, exhibits many characteristics of a ‘large city’ and is a suitable context for a study of neighbourhood effects.

We have deliberately chosen a recent time frame. The Norwegian debate on neighbourhood effects, often called ‘segregation effects’, can be traced at least to the 1970s. At this stage, however, it was not regarded as a serious issue. An emerging urban policy was more focused on housing shortage, sub-standard housing, municipal economy and loss of manufacturing employment. Neighbourhood composition and socialisation became an important concern in the early 1990s, partly in response to events and politics abroad. A White paper (Kommunaldepartementet, 1990–91) pointed at the spatial concentration of poverty and affluence as the main disadvantage of large cities. Later on, a number of new policies emerged, particularly area-based initiatives.

Our target population consists of the 1976 and 1977 birth cohorts. We follow these cohorts through two transitional stages, from age 14 to 18 (exposure to neighbourhoods) and from age 20 to 29 (outcomes). A major aim behind the design is to reduce the problem of ‘reverse causality’—i.e. the effect of unmeasured characteristics of individuals who select particular neighbourhoods.

The paper is organised as follows. The next section presents the context, the social landscape of Oslo. Then we briefly discuss relevant literature followed by a detailed description of data and methods. Finally, we give the results, before concluding in the last section.

Neighbourhood Characteristics and Policies in Oslo

The social landscape of Oslo is marked by two divides, one between east and west and one between the inner and the outer city. Oslo inner east was built as a working-class area and went through a classic phase of degradation and filtering in the period of suburban expansion, crudely from the 1950s to the 1980s. The outer city evolved as an extension of the existing social space, with high-income suburbs in the west and low- to middle-income suburbs (‘satellite towns’) in the east. This basic structure has been reproduced to the present day, although two pronounced changes appear at a lower scale. First, the inner east has gained a more varied character through gentrification. Secondly, the outer east has become increasingly deprived, partly through the influx of poor immigrant families. Both of these changes commenced during the first stage of observation, when the target population was 14–18 years. The main point, however, is the split pattern of deprivation. Deprived neighbourhoods in Oslo are found both in the inner and the outer city. Going back 15–20 years, the extent and intensity of deprivation were particularly pronounced in the inner city (Wessel, 2000).

We believe, intuitively, that increasing social and ethnic variation affects social life in the neighbourhoods. There exists some evidence in this direction, particularly for Oslo inner east. Both Pløger (1997) and Haslum (2005) emphasise that people ‘rub shoulders’ but do not interact across social and ethnic lines. Some reports from the industrial era in the 1950s, 1960s and 1970s describe an entirely different situation, where the street constitutes a rich communicative environment.

Neighbourhood deprivation has been targeted through a number of policies and schemes. Perhaps the most significant measure is a progressive redistribution system which secures additional budget resources in poor townships. The pro-rata budget for children and youths may vary by a factor of 2.5 from the richest to the poorest township (Oslo kommune, 2009). Area-based programmes have existed for many years, mostly on a provisional and limited basis. A break-through for this type of policy came, as noted, in the 1990s. A multisectoral programme for Oslo inner east ran for 10 years and was immediately succeeded by a similar programme for a part of the outer east (Groruddalen). Crucially, none of these initiatives rested on documented neighbourhood effects.

Theoretical Perspectives

Social theory emphasises that children and adolescents are strongly affected by social environments. The neighbourhood is one such environment, but it is not the main focus of attention. It is, rather, an elusive and poorly understood part of the subject-matter (Pebley and Sastry, 2004). The root of the problem, to our mind, lies in a combination of a flexible core concept, ‘the neighbourhood’, and causal complexity. A large number of mechanisms may produce neighbourhood disadvantage, some of which reflect population characteristics and some connections with the world outside the neighbourhood. These causal properties have been discussed in numerous reviews, so we will confine ourselves to a brief presentation. We make use of two main categories, internal social relations and external conditions.²

Internal Social Relations

Internal social relations point towards the issue of community and territory. This is a classic theme in urban research, which involves numerous mechanisms (Galster and Santiago, 2006; Andersson et al., 2007; Galster, 2011). The most important ones for our study relate to learning and socialisation. Peer group theory suggests that attitudes to school efforts, higher education and drugs, etc. among youths, or a smaller group of youths, affect other youths at the same age. A simple association is often assumed: the greater the concentration of like-minded individuals, the stronger the ‘normative climate’. Attitudes and norms may have a positive or a negative character, and may alleviate or enhance poverty. Role models emphasise the ‘power of the example’: individuals (adults and youths) may learn from local people with a successful professional career or with a strong and positive engagement for the local community. Learning can even be traced to social networks that stretch beyond the single neighbourhood. Social networks have, in addition, a resource-producing potential. A number of studies have shown that external ties (both weak and strong ties) can facilitate access to employment information.

Internal social relations provide, as a whole, a relevant theoretical perspective. The share of disadvantaged neighbours may influence youths’ norms, attitudes and behaviours. We cannot, however, separate between different types of socialisation, nor can we estimate the power of social networks (i.e. access to information and resources).

External Conditions

External conditions are various relationships between deprived neighbourhoods and the surrounding environments (Pebley and Sastry, 2004; Galster and Santiago, 2006; Andersson et al., 2007). Place stigma is a negative label that sustains or reinforces invisible borders between people according to their place of residence. It is an unmanageable mechanism that tends to rest on widespread representations, often related to population characteristics. Its potential effects are twofold: it may influence the self-esteem of the residents and their place attachment, and/or it may shape reactions from the outside world. Trickle-down results are decreasing investments, decreasing house prices and increasing social filtering. ‘Spatial mismatch’ obtains when neighbourhoods have restricted access to appropriate jobs. This situation typically arises when work disappears through closure or relocation. Institutional resources are a third type: schools, childcare providers, public libraries and recreational programmes/organisations may differ in quality between different urban districts and the actual location within each district may influence access for the residents.

Place stigma has been documented for Oslo (Hansen and Brattbakk, 2005) and might be relevant for our study. Spatial mismatch, on the other hand, is both unresearched and less relevant. Oslo is a well-connected city, particularly along the major transport corridors. It is also a city with a high motor vehicle density, which shapes people’s activity spaces. Given the mentioned profile of budget resources, it seems to be a weak intuitive connection between municipal services and neighbourhood effects. However, we have to omit schools in this context, since schools have a separate allocation system, and since we do not combine school and neighbourhood contexts.³

The most important neighbourhoods effects are, as we perceive it, produced through internal social relations. We expect these effects to be quite small.

Data and Empirical Approach

We focus on adolescent development because a number of earlier studies suggest that neighbourhood effects will have the strongest effect on children and young people. The study focuses on the entire urban space and includes information about the whole population of Oslo. The dataset is compiled from a range of statistical registers (demography/population, education, income and social benefits) and contains a large number of demographic and socioeconomic variables. The city of Oslo is sub-divided into 92 neighbourhoods, which is a fairly low geographical scale. The selected cohorts (1976 and 1977) lived in the defined 92 neighbourhoods in Oslo for a five-year period, from 14 through to 18 years of age. We use three indicators of socioeconomic status: educational level, market income and employment/unemployment. Individual control variables relate to gender, ethnicity, family status and family background (parents’ socioeconomic status) during childhood and adolescence. The analysis is based on multilevel modelling.

Geographical Levels

The sub-division into 92 neighbourhoods has been constructed by expert-bureaucrats in the municipal administration of Oslo, supplemented by the local urban district administrations. The construction reflects a compromise between different single criteria: coherent area coverage, common physical characteristics, uniform population size (threshold of 3000 persons), distinctive place names and collective images. Important also, the neighbourhoods are based on census enumeration districts (EDs), which is the lowest geographical level for population statistics. The 15 administrative urban districts of Oslo consist of 4–8 neighbourhoods and 23–44 EDs. The neighbourhoods, in turn, have 3–13 EDs.

A potential problem concerns school commuting. The 1976/77 cohorts attended lower secondary school at a time (1989–92) when municipal schools received approximately 92 per cent of all pupils (Oslo kommune, 2004). Equally important, the default option throughout the 1990s was to attend the nearest local school. Many of the defined neighbourhoods are more or less concurrent with catchment areas for the lower secondary schools. This suggests a huge and complex overlap between neighbourhood effects and school effects. The upper secondary schools in Oslo were divided into six regions, and two out of three pupils went to a school in their own region (Hagen, 1995). These regions are much bigger than the neighbourhoods, but this still shows that most of the pupils went to upper secondary schools in their own part of the city.

Population Characteristics

The youth cohorts born in 1976 and 1977 include 7272 persons as registered living in Oslo in one or more of the years from 1990 until 2006. This number is reduced to 5516 when we focus on the youths who lived in the same neighbourhood over a five-year period, from when they were 14 to 18 years old (in the years 1990/91, 1992/93 and 1994/95). Broken down, the number varies from seven youths in an inner-city neighbourhood to 159 in a western suburban neighbourhood, with 60 as the mean value. This variation is taken into account by the statistical programme we use, MLwiN.⁴

The total workforce (18–66 years) in Oslo amounted to 311 247 persons in 1993, varying from 1177 to 5845 at the neighbourhood level. The mean size was 3383 persons. This population is the working-age neighbours of the youth cohorts born in 1976 and 1977, and it is information about them and their composition that is the basis of the area variables (level 2 variables) in our study. For some variables, the whole group is used and, for others, more limited sections, based on standard age-spans.

Dependent Variables

Our three dependent variables are important indicators of socioeconomic status and may have been influenced by internal social relations (socialisation, social networks) in the neighbourhood during childhood and adolescence. They have a straightforward construction.

Educational attainment measures completion of a college or university degree (at least bachelor) at age 29. The variable is constructed as a binary and we use a logistic multilevel regression model. A total of 5493 youths born in 1976 and 1977 are included in this analysis.

Market income includes wage from full-time work, income from self-employment and capital income (property and investments), and is measured as an average for 2004 and 2005 (age 28 and 29 years). No kinds of transfers (positive or negative) or tax reductions are included, in order to capture pure benefits of education and labour market participation. The dependent variable is continuous and we have removed a few outliers at the top end of the distribution. A total of 3064 youths born in 1976 and 1977 are included in this analysis. The technique is multilevel OLS regression.

Labour market position relates to employment/unemployment. We use, as noted, three measurement points: 21, 25 and 29 years old. Each extension therefore measures an increasing distance from the period when the neighbourhood may have affected the youths. The analyses are based on 5516 youths born in 1976 and 1977. The variable is dichotomous and we use a logistic multilevel regression model.

Independent Variables

The independent variables of special interest in this study are measured at the neighbourhood level (level 2). The individual-level (level 1) variables are only interesting as control variables.

Explanatory variables: level 1—individual characteristics

It is well established by social science that parents have a major influence on many aspects of the future of their children. Social background is to a great extent inherited from one generation to the next. We use a variety of 20 available, recognised indicators of socioeconomic status and demographic characteristics of the parents to control for effects stemming from family background. Socioeconomic status (SES) of the parents includes education (father and mother), income, employment status and receipt of welfare benefits (social assistance, disability pension, rehabilitation benefits and transition benefits for single mothers). Parents’ age and civil status are also included. Demographic characteristics of the youths include sex, national background and number of siblings.

Explanatory variables: level 2—neighbourhood characteristics

We use seven different socioeconomic indicators at the neighbourhood level from 1993: share of individuals with a low level of education, defined as primary or lower secondary school (40–49 years); low income, defined as annual market income below 100 000 Norwegian kroner in 1993 (40–59 years, male); receipt of unemployment benefit (18–66 years); receipt of disability pension (18–49 years); receipt of transitional single-parent benefit (18–39 years); receipt of social assistance (i.e. economic aid for adjusting to difficult living conditions (18–66 years)); receipt of rehabilitation benefit (18–66 years).

Multiple Membership Structures and Selection

We further apply models with multiple membership structures that allow youths to be ‘members’ of several subsequent neighbourhoods, where the time spent in each neighbourhood is weighted and recorded. There are several reasons for this choice. First, we increase the coverage of the chosen cohorts. Secondly, and importantly, we may analyse whether life chances are affected when people live in multiple neighbourhoods. Thirdly, as we include the moving families we get a better grasp of selective migration in and out of the neighbourhoods.

The latter problem emerges through complicated processes based on resources, restrictions, preferences and so on. A mix of observed and unobserved characteristics may cause selective in- and out migration between neighbourhoods. In other words, the measured effects may not be ‘real’ neighbourhood effects, but rather selection effects of some unobserved characteristics linked to the residents in each area. Another relevant point here is that youths are ‘dependent migrants’; they do not decide where to live. This reduces the selection problem substantially, although it does not eliminate it: some unobserved characteristics may be transferred across generations, or linked to the youths themselves.

Descriptive Statistics

Table 1 reports descriptive statistics at level 1. We see, above all, that youths from moving families have a lower score on the dependent variables than the stable majority. This also holds for their parents’ income, educational attainment, unemployment and receipt of public transfers (not shown). Moving families further diverge in terms of national background (high share of non-Westerners) and family composition (high share of single mothers). These differences compel us to include the moving families and thus to see whether the estimated effects are attenuated or reinforced.

Table 1.

Descriptive statistics for the dependent variables high level of education, market income and unemployment among youths born in 1976 and 1977

Variable	Mean	S.D.
High level of education, age 29 (2005/06), mean (percentage)
Stationary (n = 5493)	46.8	0.499
Mobile (n = 886)	31.8	0.467
Market income, average at age 28 and 29 (2004 and 2005), median (NOK)
Stationary (n = 3064)	334 087	145 525
Mobile (n = 400)	314 938	124 774
Unemployment rate, at age 21 (1997/98), mean (percentage)
Stationary (n = 5516)	2.2	0.146
Mobile (n = 894)	4.3	0.202
Unemployment rate, at age 25 (2001/02), mean (percentage)
Stationary (n = 5516)	4.9	0.217
Mobile (n = 894)	8.2	0.274
Unemployment rate, at age 29 (2005/06), mean (percentage)
Stationary (n = 5516)	3.2	0.175
Mobile (n = 894)	6.4	0.254

Compared with most of Europe, the level of unemployment has been relatively low in Norway, both among adolescents and adults. The economic recession in the late 1980s and early 1990s led to increasing unemployment, peaking at 6.1 per cent in 1992. The rate then dropped until 1999, followed by a more fractured pattern (stability, growth, decline, stability, growth).

Table 1 partly illustrates this fluctuation, but also shows that our 1976 and 1977 cohorts diverge somewhat from the average pattern. The unemployment rate for these cohorts dropped to a low in 1997 and then moved to a high in 2001, followed by a decline during 2001–06.

Table 2 shows the statistical variation, including relative gaps, at level 2 (neighbourhoods). Tentatively, the greatest variation obtains for low income and low level of education.

Table 2.

Descriptive statistics for neighbourhood variables for the 1976 and 1977 cohorts in 1993

Share of people in the neighbourhood with	Mean (percentage)	S.D.	Minimum (percentage)	Maximum (percentage)
Disability pension	2.7	1.27	0.8	7.0
Unemployed	5.7	2.23	2.4	11.5
Social assistance	8.2	5.16	1.4	22.5
Transitional benefits for single parents	6.2	3.03	1.2	15.4
Low level of education	20.7	11.30	2.1	45.8
Low level of income	26.7	12.19	10.3	57.0
Rehabilitation benefits	4.1	1.34	1.7	7.2

Results

We present the results in two major steps. First, we report the plain variation across neighbourhoods in terms of education, income and labour market position (employment/unemployment). The next step introduces statistical control for individual characteristics, individual/family background and neighbourhood characteristics.

Educational Level

Neighbourhood effects on education or educational level tend to be greater than effects on earnings (Andersson, 2004; Raaum et al., 2006; Galster, 2007). A similar pattern emerges in this study. Our educational attainment measure (the proportion of people with a completed university or college education) yields a differential between neighbourhoods (Figure 1) which is quite steep and which amounts to 15.3 per cent of the total variance. Prior to control for individual characteristics and social background, it seems that the neighbourhood variation explains a relatively large part of the variation in the youths’ educational attainment. After such control, the neighbourhood importance is strongly reduced, from 15.3 per cent to 2.0 per cent (statistically significant).

Figure 1.

Neighbourhood-level residuals for high levels of education: variance components model.

Table 3 reports the variance in educational attainment at two levels, individuals and neighbourhoods, allowing for individuals to be nested within neighbourhoods. Model A includes seven neighbourhood measures, introduced one at a time, with no further control for individual background. All reported effects are significant at the 1 per cent level, although the size of the estimate varies a lot. Model B introduces a full control for 20 individual variables. Now, the putative neighbourhood effects are reduced to half, but all of them are significant at the 1 per cent level. Adding the moving families (model C) produces essentially similar results. Some coefficients are slightly attenuated and some others slightly reinforced.

Table 3.

Effects of neighbourhood characteristics on high level of education

Share of people in the neighbourhood with	A models (n = 5493)		B models (n = 5493)		C models (n = 6379)
	b	S.E.	b	S.E.	b	S.E.
Disability pension	−0.485	0.047	−0.203	0.032	−0.185	0.032
Unemployed	−0.344	0.024	−0.165	0.022	−0.141	0.022
Social assistance	−0.143	0.011	−0.069	0.010	−0.060	0.009
Transitional benefits for single parents	−0.205	0.018	−0.085	0.013	−0.079	0.012
Low level of education	−0.068	0.003	−0.032	0.004	−0.030	0.003
Low level of income	−0.056	0.005	−0.026	0.005	−0.022	0.004
Rehabilitation benefits	−0.481	0.038	−0.197	0.029	−0.183	0.026

Notes: Results in bold are significant at the 1 per cent level. Estimation procedures: penalised quasi-likelihood (PQL) and 2nd order for A and B models, and Markov chain Monte Carlo methods (MCMC) for C models.

We are therefore left with a significant outcome: increasing share of disadvantaged neighbours leads to lower future educational attainment for the youths. The strongest effects appear for receipt of disability pensions and rehabilitation benefits, and for unemployed.

The inclusion of the moving families gives us a better grasp of the selection issue. Children who move frequently may have different long-run outcomes from those of stationary children, partly due to family disadvantage and partly due to the experience of moving. In this particular case, the selection bias appears to be modest and we choose to interpret the effect as a causal connection between neighbourhood context, particularly internal social relations, and educational achievement. The magnitude of the effect is quite small and may partly reflect schoolmate characteristics rather than neighbour characteristics. A rough comparison suggests an effect size for youths’ educational attainment closer to the Swedish than to the UK evidence (Andersson and Subramanian, 2006; Bergsten, 2010; Gordon and Monastiriotis, 2006).

Market Income

The variation in market income between neighbourhoods relative to variation within neighbourhoods is only 2.7 per cent. This crude effect is statistically significant and so are 5 out of 92 differences at the neighbourhood level (not shown).

Table 4 presents logistic regression results for the neighbourhood variables in 21 separate models, extending from level two (model a) to both levels (model b) and further to a multiple membership structure (model c). All effects are significant at the 1 per cent level in all three models. Again, the effects are sharply reduced when we introduce individual attributes (model B) and include the mover stratum (model C). The second step roughly divides the effects in half, whereas the third step leads to trivial changes.

Table 4.

Effects of neighbourhood characteristics on market income

Share of people in the neighbourhood with	A Models (n = 3064)		B Models (n = 3064)		C Models (n = 3465)
	b	S.E.	b	S.E.	b	S.E.
Disability pension	−1.375	0.251	−0.659	0.248	−0.698	0.224
Unemployed	−1.028	0.155	−0.617	0.177	−0.560	0.160
Social assistance	−0.410	0.070	−0.220	0.076	−0.205	0.069
Transitional benefits for single parents	−0.665	0.089	−0.347	0.098	−0.316	0.089
Low level of education	−0.200	0.027	−0.116	0.031	−0.111	0.028
Low level of income	−0.163	0.032	−0.097	0.036	−0.089	0.031
Rehabilitation benefits	−1.453	0.219	−0.758	0.226	−0.703	0.207

Notes: Results in bold are significant at the 1 per cent level. Estimation procedures: marginal quasi-likelihood (MQL) and 1st order.

Further, all of the models reflect a pattern where the share of disadvantaged neighbours during adolescence exerts a negative effect on market income at age 28/29. We observe a fairly similar picture in terms of single indicators as in Table 3: the strongest effects appear for the share of disability pensioners, jobless people and people receiving rehabilitation benefits. One potential explanation for this, which we will return to, is that benefit rates are strong indicators of social exclusion.

We thus catch a glimpse of a multiplier effect at the neighbourhood level. The size of the effect is smaller than we expected and appears to lie somewhere between previous results for Sweden (Andersson et al., 2007; Galster et al., 2008) and the UK (Bolster et al., 2007). It is important to note, however, that these studies focus on effects for adults in a shorter run, whilst our study looks at long-term effects for adolescents. The results also square with Norwegian research at the national level, relating to earnings in 1990–95 (Raaum et al., (2006). Whether the observed effect is causal or not is difficult to assess, given the weak association and the potential for interrelated school effects. If it is causal, it might emerge from neighbourhood socialisation, or at least from the broader category ‘internal social relations’.

Unemployment

Our final dependent variable gauges unemployment in 1997, 2001 and 2005, when the cohort persons were 21, 25 and 29 years old. The gradient for unconditional differences between neighbourhoods (not shown) changes a lot over time. It was steepest in 1997, when it accounted for 10.7 per cent of the variation, via 4.2 per cent in 2001, and dropped to a flat level in 2005.⁵ This pattern of declining disadvantage seems plausible and in accordance with the proposed mechanisms: neighbourhood effects from childhood or adolescence will diminish as people grow older.

Table 5 shows logistic regression results for the neighbourhood variables in 63 models, using the established chronology, and covering three points in time. Model A is significant at the 1 per cent level for all years and for all indicators. As before, the results change considerably when we move to a full multilevel analysis (model B). The few neighbourhood effects that remain significant in model B are reduced by more than 50 per cent and obtain only at age 25 in 2001/02. We further note some differences between model B (non-movers) and model C (movers and non-movers). Most coefficients (16 out of 21) are larger in model C and several effects (5) change from insignificant to significant status. Cautiously, one might speculate whether social disadvantage correlates with involuntary mobility (Buck, 2000). Then again, we cannot rule out some disturbing period effects, given the oil-driven and fluctuating economy. A second limitation concerns life-course choices and engagements. Youths at age 21 are often students and are less likely to register as unemployed. This may explain why more coefficients are significant at age 25 than at age 21 (model B).

Table 5.

Effects of neighbourhood characteristics on unemployment at the age of 21, 25 and 29

Share of people in the neighbourhood with	Age 21 (1997/98)						Age 25 (2001/02)						Age 29 (2005/06)
	A Models(n = 5516)		B Models(n = 5516)		C Models(n = 6410)		A Models(n = 5516)		B Models(n = 5516)		C Models(n = 6410)		A Models(n = 5516)		B Models(n = 5516)		C Models(n = 6410)
	b	S.E.	b	S.E.	b	S.E.	b	S.E.	b	S.E.	b	S.E.	b	S.E.	b	S.E.	b	S.E.
Disability pension	0.284	0.086	0.117	0.128	0.148	0.093	0.302	0.047	0.099	0.065	0.090	0.055	0.208	0.058	0.141	0.080	0.156	0.069
Unemployed	0.213	0.049	0.078	0.083	0.115	0.058	0.189	0.029	0.075	0.044	0.066	0.037	0.113	0.037	0.022	0.056	0.033	0.048
Social assistance	0.084	0.021	0.020	0.036	0.044	0.025	0.084	0.012	0.037	0.018	0.038	0.015	0.052	0.015	0.015	0.023	0.029	0.020
Transitional benefits for single parents	0.102	0.037	0.004	0.054	0.019	0.039	0.125	0.018	0.035	0.026	0.048	0.022	0.072	0.023	0.036	0.032	0.048	0.028
Low level of education	0.045	0.010	0.018	0.016	0.022	0.011	0.037	0.006	0.007	0.008	0.009	0.007	0.026	0.007	0.014	0.010	0.011	0.009
Low level of income	0.038	0.009	0.010	0.015	0.021	0.010	0.033	0.006	0.016	0.008	0.012	0.007	0.024	0.007	0.007	0.010	0.009	0.009
Rehabilitation benefits	0.280	0.086	0.032	0.126	0.093	0.092	0.333	0.047	0.124	0.061	0.132	0.054	0.194	0.056	0.126	0.076	0.099	0.070

Notes: Results in bold are significant at the 1 per cent level, and underlined results are significant at the 5 per cent level. Estimation procedures: penalised quasi-likelihood (PQL) and 2nd order for A and B models, and Markov chain Monte Carlo methods (MCMC) for C models.

The unstable effects in Table 5 complicate an assessment of different indicators. We note, nevertheless, that three welfare indicators (rehabilitation benefits, transitional benefits for single parents and social assistance) have significant effect at age 25 (models B and C).

It is difficult to place these results in the European context of neighbourhood studies. Musterd and Andersson (2006) report a consistent negative effect of neighbourhood deprivation on labour market outcomes, but this study relates to adults. Van Ham and Manley (2010), by contrast, also studying adults, find a conditional and dubious relation between neighbourhood deprivation and labour market performance in Scotland. The effect only applies to homeowners and not to social renters, and the authors interpret it as a result of self-selection and not of causation.

Conclusions and Comments

Our main aim in this paper has been to determine the existence of neighbourhood effects—i.e. a causal association between the extent of deprivation in the neighbourhood during adolescence and socioeconomic status later in life. The presented evidence is supportive, but slightly mixed. We have established that differences between neighbourhoods, before the introduction of control variables, account for a moderate proportion of the variation in educational attainment (15.3 per cent). The influence is somewhat less for unemployment at age 21 (10.7 per cent) and much less for unemployment at age 25 (4.2 per cent) and market income at age 29 (2.7 per cent). These effects are sharply reduced, but do not vanish, after control for socioeconomic background. All neighbourhood indicators show small but significant long-term effects on youths’ educational achievement and income. The corresponding effect on unemployment remains significant for three neighbourhood indicators at ages 21 and 25.

In essence: the main question raised at the beginning of this paper is answered positively. Neighbourhood deprivation during adolescence appears to have a lagged effect on socioeconomic status. The first sub-question, by contrast, is answered negatively. We do not observe the same pattern for education, income and employment/unemployment. The differences are striking and require a complex explanation. Given space limitations, suffice it here to mention one methodological point: it is problematic to measure income at an early stage in the lifespan. The income level of young Norwegian adults is much less differentiated than it is for older ones. Youths with a low level of education start their working career earlier and many blue-collar jobs are well paid. Young academics, by comparison, have studied for several years and are not fully compensated for this in the first years of their career. Educational attainment is, accordingly, a better indicator when we examine people at age 29.

It remains to see whether different outcomes are equally stable or unstable as people grow older (see the second sub-question). The current analysis documents a rapidly declining/disappearing effect of neighbourhood deprivation on unemployment. This pattern is probably influenced by a changing level of unemployment and, further, by exposure to new social contexts (neighbourhoods, schools/universities, work places, etc). The results make sense in theoretical terms and resonate with a British study of neighbourhood deprivation and income growth (Bolster et al., 2007).

The third sub-question aimed at neighbourhood characteristics and their relative effect on education, income and employment/unemployment. We were not able to measure social mechanisms directly, but had to rely on compositional characteristics as proxies for socialisation, social control, place stigmatisation, etc. Notwithstanding this limitation, some characteristics appear to matter more than others. We have found surprisingly weak effects of low income and low education, and surprisingly robust effects of ‘welfare dependency’ (public transfers, particularly disability pension and rehabilitation benefit). What is at issue here is the compounded nature of socioeconomic status, but also the local mix of politics and economy. Labour market participation in Oslo fluctuates around 79–80 per cent, slightly below the national average. A life outside the labour market is, in this context, a life on the fringes of society. It may influence people’s identity, motivation and well-being. It may further influence peer group norms, attitudes and behaviours, as suggested by Wilson (1987) in his study of the “truly disadvantaged”. For the moment, though, we have to leave this theme to further research.

Our conclusions are obviously tentative. We have not inspected the importance of geographical scale, the functional form of neigbourhood effects (i.e. the distinction between linear and non-linear relationships) or the overlap between school contexts and neighbourhood contexts. Meanwhile, awaiting further results, some questions arise.

First, why are the estimated effects so small? A frequent line of argument suggests that values, norms, activities and social networks have been stretched over an increasingly larger area (Fischer, 1984; Friedrichs, 1998; Pebley and Sastry, 2004). Raaum et al. (2006, p. 200) capture this story succinctly: “The neighbourhood is not what it used to be”. Still, while ‘rescaling’ is an obvious fact, we would certainly warn against theoretical closure. There is little knowledge about subtle connections between neighbourhoods, labour markets, housing markets and extended ‘territorial bases’ (i.e. the district, the city, the regions, etc.). Nor is there sufficient knowledge about the ramifications of rising inequality. We believe, for instance, that growing affluence at the top intensifies the competition for desirable neighbourhoods. Such a chain of impacts is highly relevant in the Oslo context.

Secondly, how do our results compare with existing evidence in other countries? Some of our results indicate a marginal difference between Swedish and Norwegian neighbourhood effects. Both countries have high-quality data, but the Swedish data are richer in terms of individual and family background. This may explain the difference, or part of the difference. However, we would also like to investigate the importance of spatial stigmatisation. Poor neighbourhoods in Sweden are to a large extent located in suburban and peri-urban hinterlands. Such places exist in Oslo as well, but they have a larger proportion of owner-occupied housing, a lower rate of unemployment and, quite often, a smaller size (Brattbakk and Hansen, 2004). We should not forget, either, that Oslo contains clusters of poverty in the inner city. These differences are the results of historical events and they may well influence the continuous production of myths and narratives.

Thirdly, what are the consequences in terms of local and national policies? Our response is, again, hesitating and ambiguous. We lack basic information about causal connections between neighbourhood deprivation and individual outcomes. Neighbourhood intervention is an equally murky subject; we simply do not know whether targeted programmes are effective or not. At present, we can only advise a supplementary and small-area approach in the use of area-based programmes. The budget criteria system is, by and large, a far more important instrument.

Footnotes

Acknowledgements

The authors would like to thank a number of persons for comments on an earlier draft of this paper. These persons are: George Galster, Kelvin Jones, Torkild Hovde Lyngstad, Arne Mastekaasa and Oddbjørn Raaum. Thanks finally to the Editor and referees for useful comments.

Notes

Funding Statement

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

References

Andersson

(2004) From valley of sadness to hill of happiness: the significance of surroundings for socioeconomic career, Urban Studies, 41(3), pp. 641–659.

Andersson

Subramanian

S. V.

(2006) Explorations of neighbourhood and educational outcomes for young Swedes, Urban Studies, 43(11), pp. 2013–2025.

Andersson

Musterd

Galster

Kaupinnen

T. M.

(2007) What mix matters? Exploring the relationships between individuals’ incomes and different measures of their neighbourhood context, Housing Studies, 22(5), pp. 637–660.

Bergsten

(2010) Bättre framtidsutsikter? Blandade bostadsområden och grannskapseffekter: en analys av visioner och effekter av blandat boende. PhD thesis, Uppsala University.

Bolster

Burgess

Johnston

R. J.

Jones

. (2007) Neighbourhoods, households and income dynamics: a semi-parametric investigation of neighbourhood effects, Journal of Economic Geography, 7(1), pp. 1–38.

Brattbakk

Hansen

(2004) Post-war large housing estates in Norway: well-kept residential areas still stigmatised?, Journal of Housing and the Built Environment, 19(3), pp. 311–332.

Buck

(2000) Housing, location and residential mobility, in: Berthoud

Gershuny

(Eds) Evidence on the Dynamics of Social Change from the British Household Panel Survey, pp. 133–160. Bristol: The Policy Press.

Buck

(2001) Identifying neighbourhood effects on social exclusion, Urban Studies, 38(12), pp. 2251–2275.

Drever

A. I.

(2004) Separate spaces, separate outcomes? Neighbourhood impacts on minorities in Germany, Urban Studies, 41(8), pp. 1423–2489.

10.

Fischer

C. S.

(1984) The Urban Experience. New York: Harcourt, Brace, Jovanovich.

11.

Friedrichs

(1998) Do poor neighbourhoods make their residents poorer? Context effects of poverty neighbourhoods on residents, in: Andress

H.-J.

(Ed.) Empirical Poverty Research in Comparative Perspective, pp. 77–99. Aldershot: Ashgate.

12.

Friedrichs

Blasius

(2005) Social norms in distressed neighbourhoods: testing the Wilson hypothesis, in: Friedrichs

Galster

Musterd

(Eds) Life in Poverty Neighbourhoods, pp. 11–30. Abingdon: Routledge.

13.

Galster

(2007) Should policy makers strive for neighbourhood social mix? An analysis of the western European evidence base, Housing Studies, 22(4), pp. 523–545.

14.

Galster

(2011) The mechanism(s) of neighbourhood effects theory, evidence, and policy implications, in: van Ham

Manley

Bailey

Simpson

Maclennan

(Eds) Neighbourhood Effects Research: New Perspectives, pp. 23–56. Dordrecht: Springer.

15.

Galster

Santiago

A. M.

(2006) What’s the ’hood got to do with it? Parental perceptions about how neighbourhood mechanisms affect their children, Journal of Urban Affairs, 28(4), pp. 201–226.

16.

Galster

Andersson

Musterd

Kauppinen

(2008) Does neighbourhood income mix affect earnings of adults? New evidence from Sweden, Journal of Urban Economics, 63(3), pp. 858–870.

17.

Gordon

Monastiriotis

(2006) Urban size, spatial segregation and inequality in educational outcomes, Urban Studies, 43(1), pp. 213–236.

18.

Hagen

(1995) Adresse eller karakterer? En evaluering av inntaksordningen benyttet ved opptak til grunnkurs i allmenne og økonomisk/administrative fag i Oslo videregående skoler, høsten 1994. Oslo: Fafo.

19.

van Ham

Manley

(2010) The effect of neighbourhood housing tenure mix on labour market outcomes: a longitudinal investigation of neighbourhood effects, Journal of Economic Geography, 10(2), pp. 257–282.

20.

Hansen

Brattbakk

(2005) Endringsprosesser i norske drabantbyer. Byggforsk skriftserie 6, Oslo.

21.

Haslum

(2005) Lag, landskaper og tre områder i endring, in: Aspen

(Ed.) By og byliv i endring: studier av byrom og handlingsrom, pp. 177–222. Oslo: Scandinavian Academic Press.

22.

Kauppinen

T. M.

(2007) Neighbourhood effects in a European city: secondary education of young people in Helsinki, Social Science Research, 36(1), pp. 421–444.

23.

Kauppinen

T. M.

(2008) Schools as mediators of neighbourhood effects on choice between vocational and academic tracks of secondary education in Helsinki, European Sociological Review, 24(3), pp. 379–391.

24.

Kommunaldepartementet (1990–91) Om storbyenes økonomiske situasjon. Stortingsmelding No. 11, Kommunaldepartementet [The Ministry of Municipal Affairs], Oslo.

25.

Laan Bouma-Doff

W. van der

(2007) Confined contact: residential segregation and ethnic bridges in the Netherlands, Urban Studies, 44(5/6), pp. 997–1017.

26.

Leventhal

Brooks-Gunn

(2000) The neighbourhoods they live in, Psychological Bulletin, 126(2), pp. 309–337.

27.

McCulloch

(2001) Ward-level deprivation and individual social and economic outcomes in the British Household Panel Study waves 1 to 8, Environment and Planning A, 33(4), pp. 667–684.

28.

Musterd

Andersson

(2006) Employment, social mobility and neighbourhood effects: the case of Sweden, International Journal of Urban and Regional Research, 30(1), pp. 120–140.

29.

Musterd

Ostendorf

de Vos

(2003) Neighbourhood effects and social mobility: a longitudinal analysis, Housing Studies, 18(6), pp. 877–892.

30.

Musterd

Andersson

Galster

Kauppinen

T. M.

(2008) Are immigrants’ earnings influenced by the characteristics of their neighbours?, Environment and Planning A, 40(4), pp. 785–805.

31.

Obertwittler

(2007) The effects of neighbourhood poverty on adolescent problem behaviours: a multi-level analysis differentiated by gender and ethnicity, Housing Studies, 22(5), pp. 781–803.

32.

Oslo kommune (2004) Statistisk årbok for Oslo. Utviklings- og kompetanseetaten, Oslo.

33.

Oslo kommune (2009) Byrådets budsjettforslag for 2009: dokumentasjon av budsjettgrunnlaget for bydelene: grønt hefte. Byrådsavdeling for finans og utvikling, Oslo.

34.

Pebley

A. R.

Sastry

(2004) Neighborhoods, poverty, and children’s well-being, in: Neckerman

K. M.

(Ed.) Social Inequality, pp. 119–145. New York: Russell Sage Foundation.

35.

Pløger

(1997) Byliv og modernitet. Oslo: NIBRs plusserie.

36.

Raaum

Salvanes

K. G.

Sørensen

E. Ø

. (2006) The neighbourhood is not what it used to be, The Economic Journal, 116(January), pp. 200–222.

37.

Sampson

Morenoff

Gannon-Rowley

(2002) Assessing ‘neighbourhood effects: social processes and new directions in research, Annual Review of Sociology, 28, pp. 443–478.

38.

Wessel

(2000) Social polarisation and socioeconomic segregation in a welfare state: the case of Oslo, Urban Studies, 37(11), pp. 1947–1967.

39.

Wilson

W. J.

(1987) The Truly Disadvantaged: The Inner City, the Underclass, and Public Policy. Chicago, IL: University of Chicago Press.