Bayesian v. Frequenist analysis of employment discrimination

In the article 'Hierarchical Models for Employment Decisions', Joseph Kadane and George Woodworth (Journal of Business and Economic Statistics, April 2004, Vol 22. No.2) advocate the use of Bayesian analysis to assist the trier of facts in the determining if the employer had discriminated against the protected groups of workers.

Conceptually, the Bayesian statistical analysis approach turns the traditional approach to analyzing statistical evidence of discrimination upside down. The traditional approach, or what statisticians refer to as the frequentist approach, begins by assuming that no discrimination has occurred and the employment decisions were made in a age, race, or gender neutral manner. From that point, the data is used to determine the likelihood that a age, race, or gender neutral employment process would have generated the employment outcomes that are at issue in the lawsuit.

In short, the frequenist approach ask:

'Assuming that the defendant is utilizing a age, gender, or race neutral employment process, what is the probability that the unbiased employment process could have generated the observed employer's employment decisions?'


If there is a very small probability that a neutral employment process would have generated the outcome then it may be inferred that discrimination has occurred.

The Bayesian approach turns the approach around and upside down The Bayesian approach ask the question:

'Given that we observe the defendant's employment decision outcomes, what is the likelihood that the employment decision was age, race or gender neutral?'

If there is a small probability that the data is consistent with a neutral employment process, then it may be inferred that some type of discrimination has occurred.

What are the pros and cons of each approach? To be discussed....!