a variable for public housing accommodation is included in the neighborhood equation. The po-
tential endogeneity of public housing accommodation is dealt with by estimating a third probit
model for housing tenure with the two former probits.
Finally, this strategy also permits us to test for potential detrimental effects of public
housing accommodation on unemployment. Indeed, residing in a public housing unit may affect
labor-market opportunities of individuals by constraining their residential location choices and
subsequent residential mobility. In France, public housing renters are at risk of not obtaining
another public housing unit if they move home. This may explain that annual mobility rates of
public renters are at 10 percent against 16 percent in the private sector (Debrand and Taffin,
2005). Higher mobility costs of public renters may raise their reservation wage, thus increasing
their unemployment probability. In order to test for such an effect, the public housing variable
is also included in the unemployment probit equation.
2.2 Empirical model and econometric method
This study only deals with couple households, because the case of single adults suffers from a se-
lection bias, young adults being less likely to form a separate household if they are unemployed.5
Moreover, because dealing with women would imply to explain not only unemployment, but also
labor-market participation, our study only concerns the household head.
Although the classical theory of job search ends up in the estimation of unemployment
duration models, our dataset only allows us to estimate the probability of unemployment. This
reduced form is assumed to gather both how neighborhood characteristics affect the arrival
rate of job offers and how they impact reservation wages. Unemployment is then explained,
in a classical manner, by individual characteristics relative to experience (that will be proxied
by age and its square to allow for a non-linear effect), education and previous occupation. The
individual’s nationality is included in order to account for potential discrimination by employers.
The spouse nationality is used as a proxy for the access to information on job opportunities
through the network of relatives, as opposed to the social network provided by the neighborhood.
Lastly, the two residential variables of neighborhood type and public housing accommodation
are included as explanatory variables of unemployment in order to test our hypotheses.
As our model includes two endogenous observed discrete variables on the right hand side
5 Note that because we deal explicitly with residential sorting, we did not find useful to work on young adults
still living with their parents.