STEM and Teens: An Algorithm Bias on a Social Media

Abstract

We study whether online platforms might reproduce offline stereotypes of girls in the STEM disciplines, and if this bias can be reduced. For this purpose, we estimate ad distribution on a popular social media platform via a field experiment by setting up a randomized online ad campaign on behalf of a French computer science school. The ad campaign targeted students in high schools in France. The treatment aims to estimate whether a message aimed at prompting girls is displayed to girls more than to boys. The article contributes to work that aims to shed light on the possible biases generated by algorithms. Our results show that on average, girls are less likely to see the ad than boys; this difference in the number of impressions is not attributable to difference in ad costs between girls and boys. The treatment ad which was aimed to be shown to more girls had a crowding-out effect, since overall, it was displayed less to both boys and girls.

Keywords: Gender-gap, discrimination, algorithm bias, STEM education

JEL Classification: J16, I24