defined through a data analysis step, in which neighborhoods are classified as deprived or not
according to characteristics likely to influence information on job opportunities, role models,
peer effects in human capital acquisition or to generate statistical discrimination. This method-
ology is also motivated by the idea that individual outcomes are influenced by a wide variety of
neighborhood characteristics. Introducing separately all of them is not desirable because of the
high degree of correlation observed between such variables, which may cause instability in the
parameters and significance levels (O’Regan and Quigley, 1998). The neighborhood type is then
used to estimate simultaneously a probit model of unemployment and a probit model of location
in a disadvantaged neighborhood. The simultaneous probit accounts for the correlation between
unobservables by explicitly estimating the correlation matrix of residuals. Including neighbor-
hood type in variables affecting unemployment allows to test for the presence of neighborhood
effects.
Our identification strategy also takes advantage of the French process of assignment of
households to public housing units. Indeed, although the identification of a simultaneous probit
model does not formally require exclusion restrictions (Wilde, 2000), we have exclusion restric-
tions that are grounded on demographic criteria used by French public housing offices for giving
access to public housing. In order to be eligible for public housing, French households must
have an income below a certain threshold. Moreover, because demand largely exceeds supply,
applications are ranked on a waiting list, subject to several criteria (for instance, households
with a disabled person, or single-parent families are considered as having priority) and available
housing units are proposed to households following their rank on the waiting list. They may then
accept or refuse the proposal, and in the latter case may receive new proposals later. In 2002,
one quarter of households housed in the public housing sector had rejected at least one offer
before accepting one; half of these refusals were justified by the fact that “the housing unit was
in a neighborhood that did not fit household’s preferences” (Insee4, 2002 French Housing Survey).
Thus, the French public housing application process allows households to choose their neighbor-
hood and forbids us to consider a priori the location of public housing renters as exogenous, as
done by Oreopoulos (2003). Yet, as will be clear in our results, public housing accommodation
is a strong determinant of the location in deprived neighborhoods and helps us identifying the
effect of neighborhood on unemployment. Indeed, it is first worth noting that public housing
units represent almost one half of the French renting sector (17% of the housing stock in 2002;
Insee, 2003) and that a large part of those housing units belong to large projects located in the
periphery of urban cores, thus providing a powerful source of income segregation. Consequently,
4 French National Institute for Statistics.