Algorithmic Microtargeting? Testing the Influence of the Meta Ad Delivery Algorithm

Political campaigns are increasingly resorting to targeted advertising on social media platforms to reach potential voters. Social media companies like Meta provide advertisers with a wide variety of targeting criteria based on user demographics, behaviors, and inferred interests. However, there is a different entity ultimately responsible for deciding who is delivered which ad: the ad delivery algorithm. The algorithm is set up to learn which ads are more “relevant” for which audience and may deliver ads to certain audiences without the explicit intention or knowledge of the advertiser themselves. Recent research finds that the delivery of ads that run at the same time, with the same target audience, and the same budget can be heavily skewed along gendered and racial stereotypes. In the realm of political advertising, researchers have demonstrated that ads are more likely to be delivered to ideologically similar audiences, and presenting a liberal ad to liberal users costs almost half as much as showing the same ad to a conservative audience. Studies on the role of ad algorithms in political advertising have focused on the United States, and to the best of our knowledge, no similar effort has been made to discover how ad algorithms influence political ad delivery in an European multi-party context. To fill this gap in the literature, we collaborated with three Dutch political parties to place 135 identical ads on Facebook and Instagram targeting 9 different audiences in the run-up to the country's municipal elections on March 16th, 2022. To make sure any difference in pricing and delivery occurs only due to the party placing the ad and the target audience, the ads run with the exact same settings, at the same time, using the same daily budgets, texts, and images. We hypothesize that audiences who are more aligned with the party platform and the ad content are cheaper in cost and delivered to more people. We find clear differences in pricing and delivery between parties and audiences, however not always in line with our pre-registered expectations. We find evidence that certain parties are charged more than other parties, with one party, in particular, paying 7.5 to 10% less on average to reach 1000 users, despite using the exact same settings. We also consistently find that it is harder for political parties to reach certain audiences, for example, lower-educated citizens. Our findings have several implications for political parties and democracy. First, unbeknownst to them, certain parties spend more money to woo prospective voters, resulting in an unequal playing field. Furthermore, some groups of people are systematically less likely to receive political advertisements and more expensive to reach, potentially isolating these groups from receiving election-related information. Finally, our findings imply that banning or severely limiting advertiser targeting capabilities, as some legislators are proposing, may be counterproductive since it will only give more power to the black box of the ad delivery algorithm to target specific groups without oversight or transparency.