Association for Public Policy Analysis & Management

Job applicants' work history, especially as reflected in resumes and job applications, is commonly used to screen job applicants; however, there is little consensus in the employee selection literature as to how to systematically model work history and translate information about one’s past into predictions about future work outcomes. Drawing on and expanding the extant literature and applying machine learning (ML) techniques, we develop novel, indirect, and objective measures of demands–abilities fit, affective tone, passion for job, and propensity for job dissatisfaction, job hopping, and involuntary turnover to predict, prior to hire, the employees’ subsequent subjective and objective performance evaluations, as well as voluntary and involuntary turnover. We empirically examine our theoretical model on a longitudinal sample of 16,071 applicants for teaching positions in a public-school district. About 15% of these applicants were hired for whom we observe their objective and subjective measures of performance as well as turnover. ML enables us to enrich work history data by connecting applicants’ previous jobs to their occupational characteristics through the U.S. Department of Labor’s O*NET occupational information system. Consistent with our theoretical model, we find that unobservable determinants of hiring—expected to be highly correlated with recruiters' decision-making biases—, demographic variables, and spelling accuracy only predict subjective evaluations of teacher performance. Additionally, we show that demands–abilities fit and pre-hire propensity for job dissatisfaction are linked to subjective and objective performance evaluations, as well as turnover. We also demonstrate that our model can improve the quality of the selection process, while lowering the risk of adverse impact. The theoretical, methodological, and practical implications of these findings are discussed.