Francesca Miserocchi (Harvard University)

Discrimination through Biased Memory

Abstract. This paper shows that decision-makers make more stereotypical decisions when they struggle to recall individual-level information, penalizing women in male-dominated fields. Analyzing administrative data from Italian public schools, I find that when teachers need to assess a larger number of students, girls are less likely to be recommended for top-tier scientific high school tracks compared to boys with the same math ability. Notably, this bias vanishes for teachers who report checking student data in class registers, relying less on memory alone. To directly assess the extent to which limitations and biases in recall generate more stereotypical decisions, I conducted two experiments. In the first, teachers provided track recommendations for a series of student profiles. When teachers cannot check individual data and must rely on memory, they recall a limited set of individual signals and disproportionately retrieve stereotype-consistent information. As a consequence, large gender-based disparities in track recommendations emerge, with girls 39% less likely to be recommended to STEM tracks than identical boys. Eliminating memory constraints by allowing teachers to check individual-level information reduces the gender gap by 80%, mitigating the misallocation of talent. A second large-scale online experiment generalizes this mechanism. Taken together, the results highlight how memory limitations and biases amplify discriminatory behaviors and suggest that simple, cost-effective interventions facilitating access to individual-level information can mitigate such biases.