Can Machine Learning Be Used to Make Better Hiring Decisions?

Topic(s): fairness, performance, selection, turnover
Publication: Journal of Applied Psychology (2019)
Article: Using Machine Learning to Translate Applicant Work History into Predictors of Performance and Turnover
Authors: S. Sajjadiani, A.J. Sojourner, J.D. Kammeyer-Mueller, E. Mykerezi
Reviewed by: Marissa Post

Virtually every company requires prospective job candidates to share their work history through resumes and job applications as a basis for screening. However, the practices used by hiring managers to assess candidate work history varies between—or even within—companies. As a result, these evaluations may have varying degrees of success when trying to predict job performance or turnover. How can organizations better select high potential candidates and filter out poor candidates without anyone slipping through the cracks?

New research (Sajjadiani et al., 2019) suggests that using machine learning can work as an alternative method of work history screening. The findings suggest that a shift to machine learning techniques may allow companies to improve their selection decisions while reducing the risk of adverse impact, which occurs when a legally-protected group is underrepresented in the workplace.


Machine learning occurs when a computer uses trial and error to generate and improve an algorithm until it achieves maximum utility. In the area of personnel selection, this refined algorithm could potentially be used to help an organization select the best employees. The researchers explain that their approach to machine learning deviates from typical methods in which data scientists fail to evaluate why various inputs into an algorithm serve as predictors. This typical approach poses a challenge for practical usage because it can yield results that are difficult to explain to stakeholders and may even pose legal challenges when results are not clearly linked to job requirements.

Instead, the authors draw upon previous research to select factors that are linked to performance and turnover. Factors of interest include relevant work experience, tenure history, and other turnover history, such as involuntary turnover, leaving to seek jobs that are a better personal fit, and leaving to avoid “bad” jobs (e.g., seeking jobs requiring minimal effort, or attributing negative work outcomes to the workplace itself). By establishing this theoretical basis, the algorithm can be systematically trained and refined based on the relevant research.


The researchers used 15,000 applicants to public school teaching positions to test the hypothesis that their algorithm could be taught to use the aforementioned theoretical underpinnings to predict turnover and performance. Based on the results, they created a list of the applicants they would have hired had the algorithm actually been used in the selection process.

To holistically evaluate the success of their algorithm, the researchers used a teaching performance evaluation, student evaluations, expert observation ratings, students’ standardized test scores, and turnover. In addition to these measures, adverse impact was also included as a way to measure success of the algorithm.

When comparing their recommended candidate pool to the actual group that was hired by the school district, the researchers found that their modeling outperformed the school district’s selection with respect to overall performance. Risk of adverse impact by gender was roughly equal for both, though the machine learning technique resulted in slightly lower risk of adverse impact with respect to race.


The techniques employed in this study leverage something that is readily accessible to employers: text-based data. While cumbersome and time-consuming for selection personnel to navigate, text-based data from job applications can be handled by machine learning with greater ease and fairness, leading to improved business outcomes compared to conventional methods.

Organizations interested in pursuing machine learning methods for selection should be advised to pay special attention to transparency and consistency. Clearly communicating reasoning behind the techniques will ensure that stakeholders appreciate the fairness and effectiveness of machine learning.

Finally, while this study presents compelling evidence in favor of machine learning, it has not to date been validated in other job contexts outside of teaching. Therefore, these methods should be used cautiously and only with proper validation.


Sajjadiani, S., Sojourner, A.J., Kammeyer-Muller, J.D., & Mykerezi, E. (2019). Using machine learning to translate applicant work history into predictors of performance and turnover. Journal of Applied Psychology, 104(10), 1027-1225.