Disparate treatment and adverse impact in applied attrition modeling

Abstract

Speer and colleagues (2019) provide an excellent overview of key practices on applied attrition modeling. With our commentary, we wish to elaborate on a decision point Speer and colleagues left open in the development of attrition models, namely the decision to examine protected classification information. Our contribution seems particularly relevant given popular press discussions regarding discriminatory employment activities enacted via artificial intelligence. For instance, Amazon developed an algorithm with the purpose of recruiting job candidates with the highest potential. Unfortunately, the process resulted in a bias against female candidates that could not be remedied in a timely fashion resulting in leaders terminating the project (Dastin, 2018). Our concern is that a similar outcome might result here if biases against protected classes are not examined thoroughly. Unfortunately, while there is value in studying protected classes and their relation to turnover, we have observed that legal teams might resist people analytics teams' efforts to examine protected classes in projects such as the development of attrition models, and so we hope to speak to those practitioners who are facing such an obstacle. Therefore, with our commentary, we call attention to what has in our observation been a problem in practice: gaining permission to analyze protected class information on employees in building attrition models. Drawing upon the adverse impact and disparate treatment literature, we highlight how both including and failing to acknowledge the role of protected class information can introduce legal

Empirical attrition modelling and discrimination: Balancing validity and group differences

“…Given the nature of this paper, the results will not be rehashed in detail. The approach initially suggested by Castille and Castille (2019) and more thoroughly articulated in this paper was successfully applied within this study and resulted in removal of adverse impact for this particular dataset. This paper, along with Speer et al (2019), therefore serves as guides for practitioners in conducting attrition modelling.…”

“…The current paper expands upon recent commentaries on attrition modelling (e.g. Castille & Castille, 2019;Speer et al, 2019) by providing deeper discussion and an applied demonstration of the complexities involved when dealing with group differences in attrition scores, including the need to test for alternative models with less adverse impact but similar validity. This paper is applicable to both researchers and practitioners, as it examines the conditions and strategies for discrimination testing and for establishing validity evidence for attrition models.…”

“…high potential identification). Although Castille and Castille (2019) broadly discuss group differences with attrition models, the current paper provides a much more detailed look at these issues at a level of specificity that is necessary for practitioners to responsibly perform attrition modelling. Such a focus is important both in terms of maintaining legal protection when using attrition models and other data-driven HR models, but also to ensuring that companies are achieving their diversity goals.…”

“…In this section, I outline steps to account for group differences in attrition scores as recommended by Castille and Castille (2019). However, the framework can easily be applied to any data-driven HR model.…”

“…In the (3) third step suggested by Castille and Castille (2019), researchers act upon the findings from step 2. If adverse impact is found, researchers have several options, though these are not thoroughly explicated in Castille and Castille's paper.…”