Estimating equilibrium effects of job search assistance

Identifying policy-relevant treatment effects from randomized experiments requires the absence of spillovers between participants and nonparticipants (SUTVA) or variation in observed treatment levels. We find that SUTVA is violated for a Danish activation program for unemployed workers. Using a difference-in-differences model, we show that nonparticipants in the experiment regions find jobs more slowly after the introduction of the program than workers.