Using Matching Estiamtors to Evaluate Youth Employment Programs: Evidence from France, 1986-1988

In this paper we apply the statistical framework developed by Imbens (1998) and Lechner (1999) to identify and to estimate the causal effects of multiple treatments under the conditional independence assumption. The example concerns youth employment programs which were set up in France during the eighties to improve the labor market prospects of the most disadvantaged and unskilled young workers. The empirical analysis makes use of non- experimental longitudinal micro data collected by INSEE (Institut National de la Statistique et des Etudes Economiques, Paris) from 1986 to 1988. In the first section, we show that under the conditional independence assumption, matching only on the ratio of the conditional propensity scores is sufficient to remove the selectivity bias and it is therefore possible to use directly a one dimensional kernel function for implementing the matching techniques which were recently studied by Heckman, Ichimura and Todd (1998) and Heckman, Ichimura, Smith and Todd (1998). The specification and the estimation of the propensity scores are key issues. Due to the fact that our sample is extracted from the stock of unemployed people at a given date (August 1986), a natural specification of the treatment probabilities may be derived from a competing-risks duration model that takes into account the stock sampling bias correction. Our results show that different programs have differentiated effects, but also that there are variations in the post-program effects for participants in the same program : in general the relative effectiveness tends to increase with respect to the ratio of the scores.