An important analytical obstacle is the sample selection problem. Since non-employment levels are high and workers are unlikely to represent a random sample from the population of former recipients, estimates that fail to account for sample selection could be seriously biased