Neighbourhood effects and endogeneity issues

A recent body of research suggests that the spatial structure of cities might influence the socioeconomic characteristics and outcomes of their residents. In particular, the literature on neighbourhood effects emphasizes the potential influence of the socioeconomic composition of neighbourhoods in shaping individual’s behaviours and outcomes, through social networks, peer influences or socialization effects. However, empirical work still has not reached a consensus regarding the existence and magnitude of such effects. This is mainly because the study of neighbourhood effects raises important methodological concerns that have not often been taken into account. Notably, as individuals with similar socio-economic characteristics tend to sort themselves into certain parts of the city, the estimation of neighbourhood effects raises the issue of location choice endogeneity. Indeed, it is difficult to distinguish between neighbourhood effects and correlated effects, i.e. similarities in behaviours and outcomes arising from individuals having similar characteristics. This problem, if not dequately corrected for, may yield biased results. In the first part of this paper, neighbourhood effects are defined and some methodological problems involved in measuring such effects are identified. Particular attention is paid to the endogeneity issue, giving a formal definition of the problem and reviewing the main methods that have been used in the literature to try to solve it. The second part is devoted to an empirical illustration of the study of neighbourhood effects, in the case of labour-market outcomes of young adults in Brussels. The effect of living in a deprived neighbourhood on the unemployment probability of young adults residing in Brussels is estimated using logistic regressions. The endogeneity of neighbourhood is addressed by restricting the sample to young adults residing with their parents. Then, a ensitivity analysis is used to assess the robustness of the results to the presence of both observed and unobserved parental covariates.