47

This paper compares the results of an experimental evaluation with those that would have been obtained using non-experimental methods. We study the example of a job search assistance programme targeting young graduates. The programme was initially evaluated using a randomised controlled trial. Absent such a design, the only possible procedure would have consisted in using a non-experimental method and comparing the outcome of participants with those of non-participants. We remove the experimental control group from the sample and use propensity-score matching methods to best control for the observed heterogeneity between individuals who enter the programme and those who do not. The results obtained with matching methods are radically different from the experimental estimates.