2 Model specification
2.1 Identification strategy
In his widely cited article, Manski (1993) considers two effects by which the social group may
impact an individual’s behavior. Individual behavior can be influenced either by the average
behavior of his/her reference group, or by average characteristics of the members of this group.
The first effect is referred to as an endogenous effect, while the latter is called a contextual effect.
Moreover, similar behaviors in a group can be the consequence of exposure to common unob-
served factors giving rise to correlated effects. Correlated effects may be caused by simultaneity
in behaviors, common shocks or non random group selection. The goal of contemporaneous
work on neighborhood effects is to disentangle these different kinds of mechanisms, in particular
because endogenous and contextual effects, if shown to exist, have different policy implications
(Moffit, 2001; Glaeser and Scheinkman, 2001). Recent empirical studies highlight the reduc-
tion of estimated neighborhood effects that stems from correcting for several biases (Ginther et
al., 2000; Krauth, 2005). The endogeneity of group membership in particular is likely to gen-
erate large biases, because individuals sort themselves into neighborhoods depending on their
observable and unobservable characteristics.
The goal of this work is to estimate the intensity of neighborhood effects, focusing on the
correction for selection into neighborhoods. Indeed, we do not try to disentangle endogenous and
contextual effects, but we aim at providing an estimate of their global effect on unemployment
probability. Our identification strategy consists in dealing with the endogeneity of neighborhood
choice by estimating simultaneously two probits for unemployment and the choice of neighbor-
hood.2 Estimating simultaneously the two probits is a simple way to correct for endogeneity
(Greene, 1998) that, to our knowledge, has not been used in the context of neighborhood effects.
We treat neighborhood choice as a dummy variable indicating whether each neighborhood3 of
Lyon may be considered, on the basis of the social characteristics of its residents, as likely to
generate negative spillovers in terms of unemployment. Specifically, the neighborhood type is
2Various strategies have been developed to correct for the endogeneity of neighborhood choice. Instrumen-
tal variables methods were often used, but Rivkin (2001) shows that using aggregate variables as instruments
may actually increase the endogeneity bias. Quasi-experimental situations such as the Gautreaux Program and
the Moving To Opportunity program provided more reliable estimates of neighborhood effects on labor-market
outcomes (see Oreopoulos, 2003 for a review). However, we are not aware of any such possibility in the French
case. A third strand of literature uses aggregate statistics and their variation in space to assess the importance
of neighborhood effects (Glaeser et al., 1996; Topa, 2001).
3 See Section 3 for the definition of neighborhoods.