One longstanding concern in employee selection is that some of the most valid predictors of employee success also result in adverse impact against minority groups. Adverse impact occurs when members of a protected group are hired at a substantially lower rate than others. To remedy this, organizations may try to use alternative tests or predictors. However, in the age of big data, the number of options seems endless. New research (Rottman et al., 2023) offers a way to utilize machine learning to select a workforce that is both diverse and successful.
COMPARING TWO MACHINE LEARNING TECHNIQUES
The researchers used data from the video interviews of over 32,000 job applicants applying for teaching positions. They used the audio from these interviews to extract words that were used to predict employee success. Performance data was then obtained from a smaller subset of the teachers.
Two machine learning techniques were compared. The first technique, called iterative selective removal, statistically removed predictors that led to unfair differences between groups of employees. The second technique, called multi-penalty optimization, trained a model to optimize both diversity and validity, and created a penalty for any group differences. The researchers found that the second method was more effective in predicting employee success, while the first method decreased group differences, but was moderately less able to predict employee success.
PRACTICAL APPLICATIONS FOR ORGANIZATIONS
The researchers suggest that organizations use the multi-penalty optimization method when they are dealing with larger applicant pools with many data points. This method is promising given its efficiency, robustness, and ability to reduce the need for subjective judgment. Another benefit of this method is that multiple protected group categories (e.g., race and gender) can be taken into consideration, something previous methods could not do as easily. Thus, as big data and machine learning become more popular, organizations could benefit from using these tools to help with employee selection.
Rottman, C., Gardner, C., Liff, J., Mondragon, N., & Zuloaga, L. (2023). New strategies for addressing the diversity–validity dilemma with big data. Journal of Applied Psychology, 108(9), 1425–1444.
Image credit: istockphoto/fotogestoeber