Development and validation of a predictive score for personnel turnover: a data-driven analysis of employee survey responses

Employee turnover is a challenge for public sector employers. In this study, we used machine learning to develop and validate models to predict actualized turnover of Finnish public sector workers. The development cohort data (N=52 291) included 158 variables from 2018. We defined overall turnover (regardless of reason) and net turnover (excluding workers in retirement age) through eligibility to a follow-up survey in 2020. The validation cohort included 9030 hospital workers who responded to survey in 2017, with turnover assessed in 2019. Area under the curve (AUC) value was 0.75 (95% CI: 0.74-0.76) for overall turnover and 0.75 (95% CI 0.73-0.76) for net turnover. The validation yielded similar AUC values. Key predictors of turnover were younger age, shorter job tenure, and turnover intentions totaling over 70% of the net gain. Work-related exposures, of which low threat of lay-off and satisfaction with challenges at work were most important, had considerably lower predictive power (about 1% each). These results may offer insights for public sector employers in their efforts to reduce employee turnover.