STEM employment in Spain: An explainable machine learning analysis from a national survey

Abstract

.

This study examines the variables associated with STEM (science, technology, engineering, and mathematics) employment using the 2019 survey on the labour market insertion of Spanish university graduates. The outcome distinguishes graduates employed in a STEM occupation about five years after graduation from all other graduates, whether employed in non-STEM occupations or not employed. Using explainable machine learning methods (feature selection, a Bayesian-optimized decision tree and association rule mining), we find that, among characteristics fixed at graduation, holding a STEM degree is the strongest predictor of STEM employment, followed by gender: a tree based only on these characteristics reduces to a single split on the STEM degree and correctly classifies 77% of the graduates employed in STEM occupations and 84% of the others. Advanced ICT skills and graduates’ retrospective assessment that their degree helped them to find a job, both reported in 2019, are also associated with STEM employment. Gender differences are consistent with the lower participation of women in STEM degrees and their lower rate of STEM employment among STEM-degree holders, and the main results hold when the analysis is restricted to employed graduates.