The Conditionality of Neighbourhood Effects upon Social Neighbourhood Embeddedness: A Critical Examination of the Resources and Socialisation Mechanisms

Abstract

An immense body of literature has been published on the effects of the residential neighbourhood on individual socio-economic outcomes. Numerous studies have designated these neighbourhood effects to the socialisation and resources mechanisms. This study argues that social contacts and interactions in the neighbourhood are the minimal condition for these mechanisms to operate. Following this argument, this study examines whether these particular mechanisms will operate more strongly, and thus whether the magnitude of neighbourhood effects will be higher, for individuals who are socially more embedded in their neighbourhood. These conditional neighbourhood effects upon social embeddedness in the neighbourhood are examined for 3272 individuals within 246 neighbourhoods in the Netherlands. Surprisingly, it is found that the association between neighbourhood's socio-economic conditions and resident's income is not different for individuals with a different degree of neighbourhood-specific social contacts and interactions. Consequently, this study challenges the core of the neighbourhood effects argument on socio-economic outcomes by questioning the often applied socialisation and resources mechanisms.

Keywords::

• Neighbourhood effects
• Mechanisms
• Socialisation
• Social networks
• Effect heterogeneity

Acknowledgements

I thank the three anonymous referees and the editor of Housing Studies for their very helpful comments. My thanks are extended to Tom van der Meer for his support and efforts to advance this study and Herman van de Werfhorst for his patient reading and useful comments.

Notes

¹ Based on the theoretical substantiation of the social network and socialisation mechanisms by Galster (2012), I prefer to refer to these mechanisms from now on as network resources. The socialisation mechanism is also renamed network socialisation, to emphasise the fact that also this mechanism essentially transmits through the social contacts in the network.

² The neighbourhood effects literature sets out from Wilson's perspective on social isolation and concentration effects on social and occupational mobility, while taking a broader scope on conditions and outcomes. Neighbourhood effects have been identified on educational achievement, sexual activity and teenage pregnancy, deviant behaviour, school dropout rates, crime rates and health outcomes (for an overview see Ellen & Turner, 1997).

³ Besides these mechanisms, environmental, geographical and institutional mechanisms are also considered in the theoretical review literature (Galster, 2012). The latter mechanisms are, however, less considered in empirical studies on neighbourhood effects on socio-economic outcomes such as employment and income.

⁴ These social-interactive mechanisms are endogenous (Manski, 2000): the resident's socio-economic status is affected by the aggregation of the socio-economic statuses of the residents in the neighbourhood. It is challenging to assess endogenous effects from data, due to the so-called ‘reflection problem’. I elaborate on this in the Methods section.

⁵ The body of academic work on neighbourhood effects on socio-economic outcomes are very restricted to residents in poor, disadvantaged neighbourhoods. This provides only a one-sided view of the neighbourhood effects argument, as residents in this specific type of neighbourhood are known for being generally more locally oriented in their contacts and therefore more ‘exposed’ to the neighbourhood (Campbell & Lee, 1990; Ellen & Turner, 1997; Small & Feldman, 2012; Young, 2003).

⁶ Pinkster notes that socialisation in the neighbourhood can also operate through indirect interaction, where just residing in the same space makes residents susceptible to the behaviour of their co-residents (Pinkster, 2007). This study focuses, however, on direct interaction through local contacts of the resident and consider indirect socialisation as an subordinate effect that is already covered by the main effect of the neighbourhood.

⁷ This network socialisation process in the neighbourhood is also often referred to as the ‘contagion model’ or ‘epidemic theory’, which implies that (non-)normative behaviour is ‘contagious’: residents are influenced by the behaviour and beliefs of their co-residents through contact with them (Crane, 1991; Friedrichs & Blasius, 2003; Pinkster, 2007).

⁸ Examples of studies that account for social neighbourhood embeddedness focus their research on outcomes such as the acceptance of deviant norms (Friedrichs & Blasius, 2003), juvenile delinquency (Oberwittler, 2004) and immigrants having German friends (Farwick, 2007).

⁹ Galster et al. (2010) hypothesised that neighbourhood effects are less strong for older residents, residents who work more hours and have a higher income, and more strong for residents with children. The authors could not confirm all hypotheses with their analyses: they find that regardless of gender, only residents with children and who do not work full time experience larger neighbourhood effects. It is, however, empirically unclear whether social embeddedness in the neighbourhood is the conditioning factor here. Appendix A1 shows in the upper two rows the expectations that Galster et al. (2010) postulated on the associations between magnitude of neighbourhood effect and individual characteristics for the socialisation and network neighbourhood effects mechanisms (see Galster et al., 2010, p. 2921). The six rows below test whether neighbourhood embeddedness is indeed lower for older residents, residents who work more hours and have a higher income, and higher for residents with children (no clear hypothesis was posed for females). My analysis shows that these proxies are not sound: both bivariate correlations and multilevel regression analyses (individual neighbourhood embeddedness nested within neighbourhoods) show that proxies do not always predict the contact with neighbours (measured by (1) the absolute number, (2) the share of members of a resident's core network residing in the same neighbourhood and (3) more general, personal contact in the neighbourhood) in the right direction. Furthermore, the explanatory power of all these proxies together is low: the individual-level variance of our three neighbourhood embeddedness measures is hardly reduced by including the proxies.

¹⁰ Because the number of Moroccans and Turks living in more rural areas is very small, including Moroccans and Turks in these areas would lead to clustering effects, as interviewers would have to interview basically all Moroccans and Turks in those areas to obtain a sufficient number. It could also possibly hamper the sampling process; in order to reach out and interview Moroccans and Turks in these rural municipalities, the number of sampled municipalities should increase (de Graaf et al., 2010b). For this reason, the oversampling of Moroccans and Turks was restricted to municipalities with the highest urbanisation degrees (ranging for very strong urbanisation [> = 2500 addresses per km²] to moderate urbanisation [1000–1500 addresses per km²]).

¹¹ Local authorities have drawn random samples from the population registry based on age and country of birth of the respondent and the parents. The local authority then provided the name, date of birth, sex, ethnicity and address of the individual. The overall response rate of the survey was 52 per cent.

¹² The minimum number individuals per neighbourhood is 1, the maximum 92. On average, 13.3 individuals per neighbourhood are included.

¹³ It could also be the case that residents have no important contacts at all. This type of resident would then automatically receive a score of 0 on both measures, while a resident who has at least one important contact but has none of those contacts living in the same neighbourhood also scores 0. Because these two situations are conceptually very different, only residents with at least one important contact are included in the analysis and can score a 0 on these measures (thereby excluding 3.7 per cent of individuals in our sample that report no important contacts). Additional analyses with a slightly larger sample also, including respondents without any contacts (with a dummy indicating having no important contacts), showed very similar outcomes. Furthermore, it could be the case that the most important contacts residing in the same neighbourhood are family members of the resident. An additional model with a smaller sample which excludes those individuals of which all of their core contacts are both reported as family members and neighbours (excluding 14.2 per cent of the individuals in our sample) led to very similar results.

¹⁴ Question ‘What is the net monthly income of you and your partner (if applicable) together? (partner with whom you live together or are married)’. Unfortunately, I could not estimate the individual income as almost half of the respondents did not answer the follow-up question on the individual contribution of respondent to the household income.

¹⁵ This is most effectively shown in studies by Buck (2001) and Bolster et al. (2007), who present evidence that including a range of individual and household characteristics attenuate the neighbourhood effects, showing the importance of including these control variables.

¹⁶ Table A2 in the Appendix shows the factor loadings and uniqueness of this factor. Source of neighbourhood characteristics is ‘Key Figures Districts and Neighbourhoods 2009’. The principal component factor (PCF) analysis was based on the total of 258 neighbourhood districts in the sample. The factorability of the following nine items was examined: (1) percentage of (self-)employed people in the 15–64 age group; (2) the average income (wages, transfers, other) after tax (per income recipient); (3) the average income (wages, transfers, other) per person; (4) percentage of income recipients with income lower than 40 per cent of national income distribution; (5) percentage of income recipients with income higher than 80 per cent of national income distribution; (6) percentage of transfer recipients (employment disability insurance, unemployment benefits, welfare); (7) percentage of welfare benefits per 1000 households; (8) employment disability recipients per 1000 individuals aged 15–64; and (9) percentage of unemployment benefit recipients per 1000 individuals aged 15–64.

¹⁷ The index was based on all neighbourhoods in the data-set, the final sample was confined to 246 neighbourhoods. From the standardisation, it logically follows that the standardised deprivation index in the final sample has a mean close to 0 and standard deviation close to 1. The skewness of the neighbourhood deprivation index for the 246 neighbourhoods is 0.564 and the kursosis is 3.117, so the distribution is approximately symmetric.

¹⁸ Random slopes on the measures on the most important contacts and the more general, personal contacts were not included because no significant random slope variance was found. As argued by Snijders & Bosker (1999), however, despite the fact that there is no significant random slope, a specific cross-level interaction can still be tested.

¹⁹ Neighbourhood-level variance 0.564 and individual-level variance 7.991.

²⁰ The main models only show residents nested in neighbourhoods, but respondents are also nested within 35 municipalities. A three-level intercept-only model shows an intraclass correlation on the municipality level of 2.4 per cent, and of 4.4 per cent on the neighbourhood level. I conducted robustness checks for the main models where the clustering at the municipality level is included. These three-level models produced very similar results.

²¹ Model not shown due to space limitations.

²² I calculated a modified z-score both for the dependent variable income and the independent variable level neighbourhood deprivation. This modified z-score is determined based on the outlier resistant median of absolute deviation about the median. An individual case is an outlier when this modified z-score is greater than 3.5 (Iglewicz & Hoaglin, 1993). I conducted a stricter test with a modified z-score of 2. The findings withstand these strict checks for outliers on both the dependent and independent variable.

²³ Including each item from the index of neighbourhood deprivation separately leads to very similar results, only the percentage of welfare benefits and employment disability recipients on the neighbourhood level had no independent significant association with income. I also created threshold dummies for residents living in neighbourhoods scoring below the 20th percentile of deprivation (the least deprived areas) and a dummy for living in neighbourhoods scoring above the 80th percentile of the index of deprivation (the most deprived areas) because the mechanisms of socialisation and resources could also be non-linear, threshold-like (Galster, 2008). The neighbourhood effects might only occur after a minimum threshold of respective role models and resources has been reached before it can either enhance or limit residents' socio-economic opportunities (Andersson et al., 2007; Galster, 2008). Additional analyses which included the threshold dummies showed comparable outcomes. For reasons of parsimony, these analyses are not included in this study.

²⁴ The dependent variable income is grouped in income categories. The outcomes thus have interval censoring, as the exact income of each individual is not known. I converted the outcome variable to an interval variable using two approaches: the mid-point strategy and interval regression. For the mid-point strategy the assumption is made that individuals within one category are evenly spread across the category and the mid-point of each category is taken as the new value of the income variable. For the interval regression the assumption is made that the ordinal variable is derived from a continuous unobserved variable; for this approach I created two variables, indicating the lower and upper bound from the categories. As the outcome variables are now measured in euros and not on the 1–16 scale, the coefficients are rather different, but still show the same results: there is a small, but significant negative relationship between the level of neighbourhood deprivation and income.

Additional information

Funding

This research was supported by a Research Talent grant from The Netherlands Organization for Scientific Research [NWO: project no. 406-11-038].