Discrimination in STEM May Result From Inaccurate Aptitude Beliefs

Persistent discrimination against women in STEM fields has been well-documented. However, much of the research has focused on static settings, in which individuals are evaluated based on a single interaction. Aislinn Bohren, Assistant Professor of Economics — alongside co-authors Alex Imas, William S. Dietrich II Assistant Professor in Behavioral Economics in the Department of Social and Decision Sciences, and Michael Rosenberg (BS 2017), a Wayfair data scientist and former CMU research assistant — investigated a more dynamic context, in which highly visible data on prior reputation affected subsequent evaluations.

In “The Dynamics of Discrimination: Theory and Evidence,” the researchers present a theoretical framework regarding the source of discrimination. If discrimination is based on preferences — e.g., if an evaluator holds a preference against rewarding women — then the existence of data on prior performance will not change the discrimination across time. However, if discrimination is based on beliefs — e.g., if an evaluator believes men have a higher ability than women — then a visible reputation score can mitigate the effects.

Bohren and her colleagues evaluated interactions within an online forum for students and researchers in STEM fields. They posted mathematical questions from male and female accounts that either had no existing reputation or had a high reputation derived from positive past evaluations. Forum users evaluated questions based on their subjective perception of quality, such as how interesting or useful the questions were, and evaluated answers based on whether they were correct — a more objective assessment.

“The data showed no significant difference in how answers posted from accounts with male and female usernames were evaluated, which suggests that the source of discrimination is not preference-based,” Bohren said. “But for accounts with no prior evaluations, questions from accounts with female usernames received lower average evaluations. Taken together, these results are consistent with a belief-based source of discrimination.”

When the accounts were associated with a high reputation, the pattern reversed: Questions from female accounts received more positive evaluations on average than male accounts with similar reputations. “A high-reputation female account may come to eventually receive more positive evaluations due to the fact that, in the presence of initial discrimination, a woman needed to generate higher quality content to achieve the same high reputation as a male peer on the forum,” Bohren said. “It is important to note that in a world without discrimination, these women would have received even more positive evaluations to reflect their higher performance.”  ―