Algorithmic Bias? An Empirical Study into Apparent Gender-Based Discrimination in the Display of STEM Career Ads

With Amalia Miller

The delivery of online ads has changed, so that rather than choosing to deliver advertising via a certain medium, instead within the same medium advertisers can choose which users their ads are shown to or allow an algorithm to pick the ‘right’ users for their campaign. In this paper we show initial data that suggests this shift in optimizing delivery based on cost-effectiveness can lead to outcomes consistent with apparent data-based discrimination. We show data from a field test of a social media ad for STEM jobs that was explicitly intended to be gender-neutral in its delivery. We show that women were far less likely to be shown the ad. This is not because they were less likely to click on it – if women ever saw the ad, they were more likely than men to click. We present evidence of the mechanism by which this apparent data-based discrimination occurs. The likelihood of showing ads to men rather than women does not reflect underlying factors which might bias the algorithm away from gender equity such as labor participation rates or female education within the country. Instead, it reflects the fact that younger women are a prized demographic and as a consequence are more expensive to show ads to. This means that an ad algorithm which simply optimizes ad delivery to be cost-effective, will deliver ads that were intended to be gender-neutral in what appears to be a discriminatory way, due to crowding out.