Possible mechanisms behind the choice gap

A submitted ranking may faithfully report preferences formed under incomplete or inaccurate information. Women and men may hold different beliefs about studying and working in STEM, or may search and learn about different sets of programs before applying.

The administrative records begin with the submitted list. They do not observe the information applicants encountered, the programs they considered and rejected before submission, or their beliefs about admission and careers.

Women and men may expect different returns from the same degree or place different weight on non-pecuniary job attributes such as stability and flexibility. A difference in rankings can therefore be consistent with applicants making choices over expected work as well as over courses of study.

The paper cites evidence on expected returns, program tastes, and preferences over non-pecuniary job attributes, but the data used here contain no direct measures of those expectations or valuations.

Identity, belonging, and self-efficacy may affect whether an applicant regards STEM as an available or suitable field. These processes can shape a truthful ranking without appearing separately in the submitted list.

The paper lists these mechanisms as possible explanations; it does not estimate their contribution and does not use the application records to compare them.

Applicants may anticipate different treatment in STEM occupations. Such expectations can affect field rankings even when applicants face similar admission prospects at the point of application.

The cross-country wage-parity correlations are descriptive and do not identify anticipated discrimination or any other mechanism. The paper includes expectations of discrimination among the explanations that remain open.

Distinguishing these explanations requires information observed before submission. The paper identifies beliefs and search as priorities: elicited expectations about study and work, together with records of which programs applicants investigated before constructing their lists. Linking those measures to administrative rankings would address a different question from the descriptive decomposition reported here.