Employer learning and statistical discrimination in the Canadian labour market

Abstract

Statistical discrimination is frequently applied to illustrate different economic opportunities among equally able individuals. We use statistics from 1994, the second wave of the Survey of Labour and Income Dynamics, to analyze the income received from paid work jobs as the measure of an individual’s economic opportunity. At the same time, Heckman’s two-stage procedure is performed to account for possible bias that arises from estimating with only a pool of paid workers. We are interested in testing the following hypotheses: whether employers statistically discriminate among potential workers on the basis of education and immigration status if they have limited information about those workers and whether they learn to revise their judgments as new information is obtained. The results confirm the employer learning and statistical discrimination based on years of schooling hypotheses for the Canadian labour market. The labour market returns to initially unobservable characteristic increases with time spend in the labour market. In addition, wage becomes less related to education that employers initially use to infer an individual’s productivity. On the other hand, immigration status is not very informative about the productivity of a worker and the results do not support the hypothesis of statistical discrimination on the basis of immigration status. This paper points out the challenges faced by traditional labour market policies in a world of statistical discrimination and employer learning.