Insights at UBC Sauder

Study finds using machine learning can hire better teachers

Study finds using machine learning
Posted 2019-09-05
scroll_arrow

A new study led by UBC Sauder School of Business researcher, Sima Sajjadiani, reveals that using machine learning to screen teachers’ job applications could improve the quality of hired teachers and lower the risks of adverse impacts. 

Machine learning, a branch of artificial intelligence, is a method of data analysis that automates analytical model building. It is based on the idea that systems can learn from data, identify patterns and make decisions with minimal human intervention.

“In the study, we applied machine learning techniques to the applicants’ work histories in their teaching job applications and extracted from the unstructured text they had provided in their applications some measures, such as applicants’ work experience relevance, tenure history, and attributions for previous turnover,” said Sajjadiani. “We found that applicants whose work experience is more relevant to the job they are applying for tend to be more effective as teachers, and stay longer with the school district, but applicants whose tenure in previous jobs is shorter are less effective and leave more quickly. Also, interestingly, if the reason for which applicants left previous jobs was seeking a better position, they tend to be better teachers if hired. On the other hand, those who attribute their previous turnovers to avoiding their undesirable positions, tend to perform worse on all the measures of performance we used in this study.” 

The researchers’ machine learning model also decreases discrimination against minority applicants by reducing the likelihood of recruiters’ biases in the selection process. 

“Recruiters tend to bias towards a certain race and gender. One of the criticisms against applications of machine learning is that it perpetuates this type of biases, because the machine learns from existing practices which are biased. In developing our model, we were mindful of this issue and to mitigate it, we started by informing our application of machine learning with the findings of previous research and theories on this topic and then combining machine learning techniques with other statistical tools,” said Sajjadiani. “For example, instead of using all the variables we could get out of the text analysis in our model, we excluded variables that obviously have racial or gender bearings. For instance, applicants who stated caretaking as the reason for why they left previous jobs are usually women. We did not include this measure in our model, because it could indirectly result in discriminating against women.”

The researchers examined data from 16,071 external applicants for teaching positions at the Minneapolis Public School District (MPS) between 2007 and 2013. MPS continually faced a high turnover rate due to its location in an underprivileged district. As a consequence, the school district was eager to find a solution to the problem and retained Sajjadiani and her colleagues to explore ways to improve employee selection. 

Sajjadiani adds that improving the quality of teacher selection has a substantial impact on any nation’s economy, welfare and human capital at relatively minimal cost.

“Teachers represent a significant number in the workforce and also contribute to the quality of human capital by educating the future workforce and impacting their potential successes,” said Sajjadiani. 

While this study was conducted specially for teacher selections, the researchers note that the findings are relevant for all industries. 

“Using machine learning techniques to screen applications, if used appropriately, not only has the potential to increase the quality of the selection, it also can speed up the process and lower costs,” said Sajjadiani. “But employers should not blindly use these techniques. They should be very careful what type of data they feed the algorithms to ensure the outcomes are accurate and fair.” 

Sima Sajjadiani of the UBC Sauder School of Business co-authored “Using machine learning to translate applicant work history into predictors of performance and turnover” along with Aaron J. Sojourner, John D. Kammeyer-Mueller and Elton Mykerezi of the University of Minnesota. The study is forthcoming in the Journal of Applied Psychology.