Empirical attrition modelling and discrimination: Balancing validity and group differences

Abstract

Abstract Attrition models combine variables into statistical algorithms to understand and predict employee turnover. People analytics teams and external vendors use attrition models to offer insights and to develop organisational interventions. However, if attrition models or other data‐driven models inform employment decisions, model scores may then be subjected to civil rights laws and diversity concerns resulting from group differences in scores. This paper discusses adverse impact when building attrition models, outlining how researchers test for adverse impact in this context, strategies to reduce group differences and how attrition modelling and other human resources ‘big data’ predictions fit within larger validity frameworks. Procedures were applied to field data in an applied demonstration of an attrition model with disparate impact. Model revisions resulted in adverse impact reductions while simultaneously maintaining model validity. Collectively, this paper provides timely attention to important aspects of the people analytics, turnover and legal domains.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 26 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.