STEM and Teens: An Algorithmic Bias on a Social Media

Abstract

We propose a field experiment on a social media platform to investigate the sensitivity of an algorithm to different ad texts in the case of distribution of an ad related to science, technology, engineering and mathematics (STEM) education. We ran advertising campaigns on behalf of an engineering school targeting high schools in France, one with gender-unspecified ad text (control) and one with women-specific ad text (treatment). The treatment ad with the women-specific text was displayed less to all users. Although it might seem that the ad algorithm was being discriminatory, the distribution pattern is explained by the sensitivity of the algorithm to the ad text. The outcome was the result of both text length and the meaning of the word in the treatment ad especially in the case of displays to individuals aged over 18 years.

Keywords: Advertising, Algorithm bias, Field experiment

JEL Classification: M31, M37