Systematic discrepancies in the delivery of political ads on Facebook and Instagram

Abstract

Political advertising on social media has become a central element in election campaigns. However, granular information about political advertising on social media was previously unavailable, thus raising concerns regarding fairness, accountability, and transparency in the electoral process. In this paper, we analyze targeted political advertising on social media via a unique, large-scale dataset of over 80000 political ads from Meta during the 2021 German federal election, with more than 1.1 billion impressions. For each political ad, our dataset records granular information about targeting strategies, spending, and actual impressions. We then study (i) the prevalence of targeted ads across the political spectrum; (ii) the discrepancies between targeted and actual audiences due to algorithmic ad delivery; and (iii) which targeting strategies on social media attain a wide reach at low cost. We find that targeted ads are prevalent across the entire political spectrum. Moreover, there are considerable discrepancies between targeted and actual audiences, and systematic differences in the reach of political ads (in impressions-per-EUR) among parties, where the algorithm favors ads from populists over others.

Figures (10)

• Figure 1: Top-10 targeting criteria by total spending.
• Figure 2: a; Distributions of impressions-per-EUR across the ads of each party. An orange cross indicates the mean of the distribution. b; Difference between average impressions-per-EUR of a party and average impressions-per-EUR in the overall sample. c; Distribution of impressions-per-EUR across the ads of each party. An orange cross indicates the mean of the distribution.
• Figure 3: a, Discrepancy in the age distribution between the actual and target audience (in %). We find that ads by most parties (except AfD) are seen by more users between 25--34 than originally intended. b, Comparison of actual and target audience by age for political ads published by AfD. A red color indicates areas where the difference between the actual and targeted audience is negative (green for positive). Younger and older users see the ads less often than originally intended by the party. c, Discrepancy in the gender distribution between the actual and target audience (in %). We find large differences between male and female audiences for right-wing parties (e.g., Union, AfD), implying that ads are seen by considerably fewer females than originally intended due to the algorithmic ad delivery.
• Figure 4: Coefficient estimates and 95 % confidence intervals for a targeting strategies, b demographics, and c ad characteristics. Statistically significant coefficients ($p<0.05$) are indicated by black circles , all others by gray circles . Sentiment, weekday, platform dummy, and party dummy are categorical variables, and the reference categories are "neutral" sentiment, "Monday", "both platforms", and "AfD".
• Figure 5: Average difference between actual vs. predicted impressions-per-EUR based on our machine learning model over 10 runs.