Thread by @mogens_fosgerau

1/ Here is a paper “Costly Screening and Categorical Inequality” by @mogens_fosgerau, @rajivatbarnard and @JorgenWeibull. #EconTwitter @ERC_Research

papers.ssrn.com/sol3/papers.cf…

2/ We unify and extend several strands in the literature on categorical inequality, including statistical discrimination, prejudice, and social capital.

3/ The model can be illustrated in a single figure.

4/ p (horizontal axis) is the probability that a candidate is qualified, r (vertical axis) is the screening intensity. The blue curve is the screener’s response to p, the red curve is the population of candidates’ response to r. There are three equilibria, two of them are stable.

5/ In equilibrium, we assume that the screener’s belief regarding p, the share of qualified candidates in a group, is correct. Candidates are able to anticipate how strictly they will be screened.

6/ The screener is active when faced with candidates with p in a screening interval. All candidates from groups with lower p are just rejected, all candidates from groups with higher p are just accepted. Candidates with high p will face easier test.

7/ One stable equilibrium occurs at (0,0): Candidates invest no effort in becoming qualified. The screener is inactive and just rejects everybody.

8/ There are two equilibria where the screener is active. The equilibrium to the right is stable. We can compare stable active equilibria under different circumstances.

9/ One group may be costlier to screen, e.g. if it has ethnicity different from the screener. Then the screening interval becomes more narrow. In (stable active) equilibrium, the disadvantaged group then makes less effort, they are accepted less often, and receive lower wages.

10/ Prejudice (in the sense of Becker, 1957), may mean that the screener places lower value on accepting qualified candidates from a disadvantaged group. Screening will be tougher for them, they respond by making less effort and end up being paid less on average.

11/ Disadvantage may also arise if opportunities differ and the qualification is harder to obtain. The disadvantaged group will be screened harder and be paid less on average.

12/ So we have a quite general stylized model that can be summarized in a single figure. It can be used to think about different ways inequality may arise and persist due to discrimination. The predictions are testable.

13/ PS. We use the very cool and very general Bregman information cost. It has the Shannon mutual information as a special case.