14.662 S15 Labor Economics II Lecture 21

The statistical models of discrimination we covered previously were static models. In contrast, Altonji and Pierret (2001) use a dynamic model of employer learning to develop a test of statistical discrimination by firms. The theoretical underpinnings of their paper build closely on the paper by Farber and Gibbons (1996). Before discussing Altonji and Pierret (2001), we are first going to walk through the Farber and Gibbons (1996) model in detail, and from there we will be able to discuss the Altonji and Pierret (2001) in more depth. The starting point of the Farber and Gibbons (1996) paper is the following. When a worker enters the labor market, her education and some other characteristics are observable by employers, but it is likely that these observable characteristics convey only partial information to the employer about the worker’s productive abilities. However, over time as the worker accumulates experience in the labor market further information is likely to be revealed. One of the key insights of the Farber-Gibbons paper was to realize that, at least in some datasets, the econometrician may observe variables measuring productivity that are not observed by employers such as AFQT scores. It is then possible to ask how employers learn about these (unobserved to employers) productivity measures as they gather information about the worker’s productivity over time. Farber and Gibbons are interested in the question of what implications this type of employer learning has for wage dynamics. This paper was very influential, in part because it developed a tractable framework with empirically testable implications that were supported by the data.