Statistical discrimination from composition effects in the market for low-skilled workers

In a random search environment with two racial groups each composed of identical numbers of high and low productivity workers, firms use an imperfect screening device (interviews) to control hiring. If inconclusive interviews lead firms to hire majority workers but not minority workers, then the unemployment pool for majority workers is of higher average quality. This can justify the initial hiring choices. Color-blind hiring always eliminates racial disparities but is not necessarily beneficial; in the USA it would improve welfare with only a brief small increase in white unemployment.