Scribd
Upload
27 views1 page

Employer Learning and Statistical Discrimination

The paper by Altonji & Pierret (2001) investigates whether employers statistically discriminate among younger workers based on observable characteristics like education and race, and how this reliance changes as they gain experience. The authors develop a model of Employer Learning with Statistical Discrimination (EL-SD) and find evidence supporting statistical discrimination, particularly related to education rather than race. However, they acknowledge limitations in addressing racial discrimination in employment and suggest further research on the impact of additional observable productivity correlates on statistical discrimination.

Uploaded by

rgong987
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
27 views1 page

Employer Learning and Statistical Discrimination

The paper by Altonji & Pierret (2001) investigates whether employers statistically discriminate among younger workers based on observable characteristics like education and race, and how this reliance changes as they gain experience. The authors develop a model of Employer Learning with Statistical Discrimination (EL-SD) and find evidence supporting statistical discrimination, particularly related to education rather than race. However, they acknowledge limitations in addressing racial discrimination in employment and suggest further research on the impact of additional observable productivity correlates on statistical discrimination.

Uploaded by

rgong987
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Altonji & Pierret (2001)

(i) what the question is about

This paper asks if employers “statistically discriminate” among younger workers on the basis of easily observ-

able variables such as education, race, and other clues to a worker’s labor force preparation? As they learn over

time, do they rely less on such variables? These questions arise as employers must make new hires on the basis of

limited information from resumes, recommendations, and interviews. As firms learn more about the productivity

of a hire over time, they may revise their beliefs and pay the worker differently. The answers to these questions

have implications for education signaling models, statistical discrimination models, earnings dynamics, and the

design of mechanisms for hiring and firing workers.

The authors establish and test a model of Employer Learning with Statistical Discrimination (EL-SD). Firms

observe variables s such as schooling that affect productivity, but cannot observe variables z such as AFQT scores

that also affect productivity. Under EL-SD, if s and z are positively correlated then the wage coefficient on z will

rise with experience and the wage coefficient on s will fall. Using NLSY79 data for young males, which includes

AFQT score, father’s education, and wages for older siblings as z variables, the authors find evidence for statistical

discrimination. They also find statistical discrimination on race has a minor role in determining racial wage gaps.

(ii) what was well done in your opinion

The authors clearly motivate their empirical test with an extension of Farber and Gibbons (1996). The model

generates two propositions. First, the regression coefficient on z is nondecreasing in years t and the regression

coefficient on s is nonincreasing in t. Second, the change in the s coefficient with respect to time is negatively

related to the product of the change in the z coefficient with respect to time and the coefficient from regressing z

on s. Their second proposition offers a valuable solution to a difficult identification problem in studying statistical

discrimination: whether the correlation between wage and an easily observable variable arises because firms

discriminate on the basis of that variable or because the variable is correlated with information on productivity

firms use but the econometrician does not observe. The authors also do well in accounting for on-the-job training

that can drive wage growth separately from s and z.

(iii) what was not well-done in your opinion

The authors do not fully study the issue of statistical racial discrimination and racial wage gaps. As they point

out, they do not address the issue of discrimination in employment. The composition workers they study is taken

as given, yet discrimination on the extensive margin is a central issue in debates on racial inequality. Blacks and

whites don’t have equal opportunity.

(iv) a possible extension

The signal value of schooling tends to fall as workers accumulate more experience. A potential extension to the

paper could consider whether expanding the number of productivity correlates a firm observes reduces statistical

discrimination on schooling. Under a policy counterfactual where firms get access to government collected z

characteristics during the hiring process, do average labor market returns to schooling fall?

You might also like