Random Parameters and Self-Selection Models.

Most empirical work employs models with fixed parameters. In contrast, I discuss the specification of self-selection models with random parameters. I demonstrate that in a self-selection model, misspecification of the parameter structure as constant causes biased estimates, especially of the selectivity effect. I estimate a self-selection model of moonlighting with random parameters and find that the selectivity effect, which was almost absent in the constant parameter version of the model, is indeed significant.