borhood. Neighborhood types were defined through a data analysis step based on their social
composition. We estimated simultaneously three probit equations relating respectively to unem-
ployment, neighborhood type, and accommodation in the public housing sector, thus allowing to
deal with endogeneity of the two residential variables with respect to unemployment. Potential
dependencies within neighborhoods were accounted for by the estimation of a robust variance
matrix. Demographic characteristics were used as exclusion restrictions. Estimation of this
system by simulated maximum likelihood used the GHK simulator.
We observed that the endogeneity bias on coefficients of residential variables is relatively
high. Our study also shows that the particular situation of public housing renters provides
a valuable opportunity to estimate the impact of neighborhood on socioeconomic outcomes in
France. Contrary to Oreopoulos (2003), this is not because the location of public housing tenants
is exogenous, but because the tenure helps us to explain neighborhood choice by an exogenous
characteristic.
Our results do not provide any support to the hypothesis according to which public housing
accommodation would affect job search behavior and, in particular, would reduce residential
mobility sufficiently so as to increase unemployment probability. As to residential location, we
clearly observe a neighborhood effect on unemployment affecting, in particular, public housing
renters. According to our results, living in one of the deprived neighborhoods (which represent
35% of Lyon’s population) would increase the unemployment probability by 1.2 points. These
results both add to the literature on neighborhood effects and give insight into a much debated
policy issue in France and in other countries, that is, the effect of the location of public housing
in cities on individual socioeconomic outcomes.
Of course, due to the chosen framework, this study does not allow us to estimate sep-
arately endogenous and contextual effects, because mean unemployment rate and neighbors’
characteristics supposed to influence unemployment are used simultaneously in the classifica-
tion of neighborhoods. Therefore, we are not able to test for the existence of a social multiplier,
nor for specific mechanisms such as the role of social networks, stigma, or role models, but we
keep these issues for future work. Further, we only estimate the change in unemployment occur-
ring with a change of neighborhood type, and not a continuous effect. However, this strategy is
relevant with respect to the fact that several correlated variables generate neighborhood effects
and that some of them may have a non-continuous impact.
21