Developing a Continuous, Rather Than Binary, Classification for Measuring STEM Jobs

Abstract

This paper presents our review and synthesis of the literature on STEM classification, and our results for a novel approach towards understanding, categorizing, and tracking STEM attributes in the workplace. We found two deficiencies in the way STEM is traditionally discussed, which we attempt to address in this work. The first is that the key components of STEM tend to be discussed holistically in the literature, rather than discretely as Science, Technology, Education, and Mathematics. The second is that our ability to track changes in S.T.E.M. concentrations in the workplace, both geographically and temporally, is underdeveloped. Further, we have found that this second deficiency is due, in part, to how STEM occupations are categorized; i.e., "STEM" tends to be a binary designation, rather than measured on a continuum for each job, and each component of S.T.E.M. It is also due to the lack of a "gold standard" measurement of the quantity of S.T.E.M. for all occupations. Here, we present a novel approach for machine learning algorithms using a "bag of words" method. These algorithms are trained on a small selection of Standard Occupational Classification (SOC) occupations, using ratings for each component of S.T.E.M. as the exemplars on which to train (SOC 2019). Recognizing that such a classification scheme is new, and that one of the goals of this project is to solicit Subject Matter Expert (SME) feedback, the resultant model of S.T.E.M. measurements across these occupations is designed to easily incorporate multiple distinct models and alternative approaches