In most models in which firms and workers learn about worker productivity through repeated observations of on-the-job performance, the amount of information revealed about workers is exogenously given and constant across jobs. In this paper, we examine what happens when the amount of information gathered about workers can be altered by assigning workers to different jobs. We show that informational differences across jobs naturally give rise to experimentation. That is, there is a trade off between current period output in order assign workers to jobs that reveal a substantial amount of information workers’ skills. We find that while experimentation is the most valuable when workers are young and inexperienced, the optimal level of experimentation is initially very small, rises as workers gain experience and then eventually declines. As a result, our model suggests that wage growth may be driven partly by a decline in experimentation as workers age. We also show that random productivity shocks can have long-lasting and, in some cases, permanent effects on both wages and wage growth.
