[Paper Review] Discrimination through Optimization: How Facebook's Ad Delivery Can Lead to Biased Outcomes

Muhammad Ali, Piotr Sapieżyński+4 more|arXiv (Cornell University)|Nov 7, 2019

Consumer Market Behavior and Pricing33 references116 citations

TL;DR

This paper demonstrates that Facebook's ad delivery algorithm can produce discriminatory outcomes along gender and racial lines, even when advertisers use neutral targeting, due to platform-level optimization for relevance and ad performance. The study reveals that budget allocation and ad content significantly skew delivery, leading to unequal visibility for housing and employment ads despite inclusive targeting settings.

ABSTRACT

The enormous financial success of online advertising platforms is partially due to the precise targeting features they offer. Although researchers and journalists have found many ways that advertisers can target---or exclude---particular groups of users seeing their ads, comparatively little attention has been paid to the implications of the platform's ad delivery process, comprised of the platform's choices about which users see which ads. It has been hypothesized that this process can ad delivery in ways that the advertisers do not intend, making some users less likely than others to see particular ads based on their demographic characteristics. In this paper, we demonstrate that such skewed delivery occurs on Facebook, due to market and financial optimization effects as well as the platform's own predictions about the relevance of ads to different groups of users. We find that both the advertiser's budget and the content of the ad each significantly contribute to the skew of Facebook's ad delivery. Critically, we observe significant skew in delivery along gender and racial lines for real ads for employment and housing opportunities despite neutral targeting parameters. Our results demonstrate previously unknown mechanisms that can lead to potentially discriminatory ad delivery, even when advertisers set their targeting parameters to be highly inclusive. This underscores the need for policymakers and platforms to carefully consider the role of the ad delivery optimization run by ad platforms themselves---and not just the targeting choices of advertisers---in preventing discrimination in digital advertising.

Motivation & Objective

• To investigate whether Facebook's ad delivery system can produce discriminatory outcomes independent of advertiser targeting choices.
• To examine how platform-level optimization—driven by relevance predictions and financial incentives—affects ad visibility across demographic groups.
• To assess the impact of ad content and budget on skewed delivery of employment and housing ads.
• To highlight the role of the platform's delivery algorithm as a previously underappreciated source of systemic bias in digital advertising.

Proposed method

• Conducted controlled experiments using real Facebook ads for housing and employment opportunities with neutral targeting parameters.
• Varied ad content and budget levels across different demographic groups to measure delivery disparities.
• Analyzed delivery outcomes using platform-reported metrics on ad visibility and engagement.
• Compared delivery rates across gender and racial groups to detect systematic skew.
• Evaluated the influence of Facebook's relevance scoring and optimization algorithms on ad distribution.

Experimental results

Research questions

• RQ1 Can Facebook's ad delivery system produce discriminatory outcomes even when advertisers use neutral targeting?
• RQ2 To what extent do ad content and budget influence the skew in ad delivery across demographic groups?
• RQ3 How do platform-level optimization mechanisms, such as relevance prediction and financial incentives, contribute to unequal ad visibility?
• RQ4 Are there measurable disparities in ad delivery for employment and housing ads along gender and racial lines?

Key findings

• Facebook's ad delivery system produced significant skew in ad visibility along gender and racial lines for employment and housing ads, despite neutral targeting parameters.
• Ad content and budget allocation were found to significantly contribute to the skew in delivery outcomes.
• The platform's optimization for relevance and performance led to reduced visibility for certain demographic groups, even when advertisers intended inclusivity.
• Skewed delivery occurred even when advertisers did not explicitly exclude any groups, indicating that the platform's algorithmic processes are a key source of discrimination.
• The study reveals that platform-driven delivery mechanisms can perpetuate bias independently of advertiser intent, highlighting a systemic issue in digital advertising ecosystems.