Flatau et al. (2003) find similar results for Australia, where public renters do not have higher
unemployment probability once the endogeneity of tenure is accounted for. In our case, the
apparent effect of public housing on unemployment is entirely due to its indirect influence
through its positive effect on living in a deprived quarter, which itself raises unemployment.18
Indeed, the deprived neighborhood variable exerts a positive effect on unemployment
probabilities. This variable is endogenous in the unemployment equation, that is, unobserved
variables influencing unemployment are negatively correlated with unobserved characteristics
affecting neighborhood choices. The negative sign of the correlation indicates that individuals
having a higher propensity for unemployment than explained by their observed characteristics
are less likely to live in a deprived neighborhood. While surprising at first sight, this result is in
line with the observation of mixed neighborhoods in Lyon’s urban core, that are not classified as
deprived but that simultaneously have unemployment rates above the average and host younger
individuals, with potentially less predictable paths.
As explained in the first part of the paper, estimating neighborhood effects requires to
deal with correlated effects. Our identification strategy allows us to deal with the endogeneity of
neighborhood choice. The estimated neighborhood effects could still be suspected to suffer from
other biases due to random shocks common to all individuals in deprived neighborhoods. This
is why our estimation method corrects the variance matrix of coefficients in order to account for
dependencies within neighborhoods. This correction slightly changes the coefficient standard
errors, but does not change the significativity with respect to conventional thresholds.19 This is
not surprising, since neighborhoods which are classified as deprived are spread in different parts
of the city. There is no reason why each of the deprived neighborhoods would be concerned
by a shock that would not affect the other types of neighborhoods in the same area. Note
also that the estimations were performed for different initial values of correlation coefficients
and all of them converged to the same correlation matrix and produced very similar coefficients.
Other specifications differing with respect to exogenous explanatory variables were also estimated
without changing the baseline results.
Because the public housing variable is not endogenous in the unemployment equation,
nor in the neighborhood equation, and because it does not affect unemployment probability
as soon as the endogeneity of neighborhood is properly dealt with, we base the assessment
of neighborhood effects on the simultaneous estimation of two probits of unemployment and
18 This result is confirmed by the estimation of a simultaneous model of two probits for unemployment and
public housing accommodation, showing that public housing has no effect on unemployment probability.
19 Detailed results available from the authors upon request.
16