STEM and Teens: An Algorithmic Bias on a Social Media

Abstract

We study whether online platforms might reproduce offline stereotypes of girls in the STEM disciplines. The article contributes to work that aims to shed light on the possi- ble bias generated by algorithms. We run a field experiment based on launching online ad campaigns on a popular social media platform on behalf of a French computer sci- ence engineering school. We randomize the ad campaign in order to estimate whether a message aimed at prompting girls is more displayed to girls than to boys. The ad campaign targets students in high schools in the Paris area. Our results show that on average girls received 25 fewer impressions than boys, but were more likely to click on the ad if they come across it. This bias is moderated for science oriented high schools with a large majority of girls enrolled in science track. This group of girls receive more impressions compared to other girls. The treatment ad aimed at targeting more girls has a crowding-out effect, with an ad which was, overall, less shown to all.