The effects of political advertising on Facebook and Instagram before the 2020 US election

Summary

This large-scale field experiment, conducted in collaboration with Meta, estimates the causal effect of political advertising on Facebook and Instagram in the six weeks before the 2020 US presidential election. Roughly 63,000 users were randomly assigned to have political ads removed from their feeds. The authors find no detectable effects on political knowledge, polarization, perceptions, or related attitudinal outcomes, though they caution that the study was underpowered to rule out small effects on turnout and vote choice. Descriptively, they document that presidential ads were aimed mostly at co-partisans and that fundraising — not persuasion — was the dominant ad category.

Key Contributions

• Large-scale randomized causal evidence on social media political advertising during a high-salience US election.
• Descriptive documentation of presidential ad targeting (largely co-partisan) and content mix (fundraising-heavy).
• An empirical contribution to platform regulation debates, suggesting attitudinal effects of political ads may be smaller than often presumed.
• A methodological template — platform-collaborative randomized ad removal — for studying advertising effects in real feeds.

Methods

A six-week pre-election randomized controlled experiment with 36,906 Facebook and 25,925 Instagram users, in which treated users had political ads suppressed from their feeds. Outcomes spanned political knowledge, affective and issue polarization, perceptions, turnout, and vote choice, complemented by descriptive analyses of ad targeting and content categories.

Findings

• Presidential ads were predominantly targeted at co-partisan supporters rather than persuadable voters.
• Fundraising appeals were the single most common type of political ad.
• No detectable effects of ad removal on political knowledge.
• No detectable effects on affective or issue polarization.
• No detectable effects on political perceptions.
• Effects on turnout and vote choice could not be ruled out due to limited statistical power.

Connections

This paper belongs to the Meta–academic 2020 US Election Research collaboration and sits alongside Voelkel2026-lc, another experiment from that program testing platform-level interventions on attitudinal outcomes; both report striking nulls on polarization despite substantively different treatments. It also speaks to broader work on persuasion and messaging effects in political communication, where similar small-or-null findings have been recurrent.