Abstract:

We analyze a model of organizational learning where agents’ performance reflects time-invariant unobservable ability, privately-chosen effort, and noise. Our main result is that, even when performance is almost entirely random, maximizing the probability of identifying the best agent (“selective efficiency”) requires biasing final selection in favor of early winners. Making luck persistent, e.g. through fast-tracks, is thus rationalized by the pursuit of selective efficiency. Agents’ strategic efforts amplify the persistence of luck. Organizational learning also affects the persistence of initial advantages stemming from identity. Identity-dependent biases, e.g. gender-specific mentoring, create incentives that make selection both more efficient and more equitable.

Joint with Mikhail Drugov and Marc Möller.

Paper available here.