Fair screening with respect to age using a resume correspondence experiment
Algorithms are used in almost every step of the hiring process, assisting hiring decisions, if not directly making them. While the use of algorithms can help us eliminate the effects of the discriminatory tendencies of humans on hiring decisions, it can also amplify the effect of discriminatory hiring practices. Previous studies on algorithmic hiring focused on miti-gating discrimination against race and gender, leaving a gap in the literature regarding the methods to address age discrimination in hiring algorithms. Combining the data generation techniques used by resume correspondence studies from the economics of discrimination literature and a particular definition of fairness, counterfactual fairness presented by Kusneret al.(2018), from the computer science literature, I propose an algorithm that is fair withrespect to age: I show that counterfactuals in resume correspondence experiments can be used to decompose the effects of age and experience in hiring decisions; moreover, a relaxation of counterfactual fairness can be satisfied by a simple adjustment to the parameters of the machine learning model. The correction in model parameters is designed to address age discrimination, yet this method is of interest to researchers who study other settings where the sensitive attributes, such as age, gender, or race, of the individuals causally affect non-sensitive characteristics relevant to decision-making.