Did Political Advertising on Facebook and Instagram Affect the 2020 US Election?

Our research team partnered with Meta, the owner of Facebook and Instagram, to invite a sample group of users representative of the platforms’ overall user base to participate in a study about political ads during the six weeks before Election Day. We documented several notable observations. First, the number of ad impressions—or instances in which an ad appeared in a user’s feed—varied widely between users. For example, participants in swing states and those who voted in 2016 saw more impressions than others. Furthermore, the same ads were often shown multiple times to the same user. In those six weeks, the average Facebook participant saw 132 impressions of 72 political ads.

Second, users primarily saw ads for their own political party. Although most presidential ad impressions on Facebook were Democratic—67 percent according to one measure—Democratic ads were shown mainly to Democrats and Republican ads to Republicans. Both types of ads were most likely to be shown to their strongest supporters. Independents were shown the fewest political ads, including the fewest presidential ads. Relatedly, participants who reported being undecided voters saw fewer political ads than those who reported that their minds were made up. On Instagram, almost all ads that we could associate with a party were Democratic, which may partly reflect Instagram’s younger user base and the tendency of young voters to favor the Democratic Party.

Third, most political ads were targeted using advertiser-provided user data (i.e., list targeting). Overall, about 62 percent of political ad impressions shown to Facebook participants were list-targeted, compared with 71 percent on Instagram. The share was even higher for presidential ads on Facebook, ranging from 73 to 90 percent. This suggests that campaigns may intentionally target ads to their own supporters, which makes sense because most ads were about fundraising. Specifically, 46 percent of presidential ad impressions to Facebook participants sought donations, 26 percent were persuasive, 17 percent collected user information, and 5.5 percent advertised merchandise. Donation ads were targeted toward users of the same party, while ads shown to supporters of the opposite party mostly involved persuasion.

Our study also involved an experiment that randomly replaced all political ads from some users’ feeds with nonpolitical ads during the six weeks before Election Day. Participants were aware that their feeds might be modified, but they did not know that this would concern political ads, and they were not notified whether a modification occurred. Participants completed a survey that measured political knowledge, polarization, perceived legitimacy of the election, political participation, Trump’s favorability, and voter turnout. We did not detect significant effects of removing ads on any of these outcomes, and our estimates are precise enough to rule out even fairly small effects.

Given the large share of ads focused on donations, we were especially interested in one aspect of political participation: campaign contributions. However, our research finds no significant effect of removing ads on total donations, based on both self-reported data and campaign contribution data, and we can rule out that viewing political ads during the six weeks prior to Election Day increased contributions by more than $2.50 per person. That amount is small, but given the low price of ads, a return of $2.50 per person could have been enough to make those ads profitable. These findings raise three possibilities: few participants contributed via donation ads, those who did would have contributed through other means if they had not viewed the ads, or removing political ads affected contributions but by an amount too small for our research to detect.

Our research finds some evidence that removing political ads affected two outcomes. First, removing all political ads for Facebook participants may have slightly increased their engagement with political content other than advertisements. The largest increase in engagement occurred via more likes, comments, and reshares on content from politicians running for office. Additional evidence suggests that this may have occurred because participants began spending more time on the platform and shifted their behavior while on the platform. Second, removing list-targeted ads on Facebook may have slightly increased the probability that participants reported voting for Trump. However, several other findings suggest that this effect may have also occurred by chance. First, removing all political ads on Facebook did not affect participants’ probability of voting for Trump, and most presidential ads were list-targeted. Second, this effect seems to have been driven by users who would have seen fewer political ads than the typical user, but we would have expected this group to experience the smallest effects. Third, removing political ads did not affect other outcomes that one would expect to change if participants became more likely to vote for Trump: Trump’s favorability, favorability toward Republicans, agreement with Republican beliefs, and Republican vote share in state-level races.

Our research also examines whether removing political ads from Facebook and Instagram affected specific groups—namely, Republicans, Democrats, independents, undecided voters, voters who identified as black or Hispanic, and users who likely would have seen more political ads than the typical user. We found no evidence of any effects, even among users who would have seen many political ads, a group we had expected to experience the largest effects.

We also examined whether the effects of removing ads differed depending on which type of ads users were most likely to see. We compared participants who were predicted to see more Democratic ads with those who were predicted to see more Republican ads. If Democratic ads were effective at persuading voters, then removing them should have made users who would have seen more of them relatively more supportive of Trump. But we did not find evidence of this effect.

Finally, we calculated the implied returns to advertising spending by presidential advertisers. The average ad in our study cost $25.10 per thousand impressions, and each dollar spent on ads gained no more than 0.03 votes for the advertiser’s preferred candidate. Furthermore, our findings reveal that $1,000 of additional pro-party spending yielded no more than 8.7 votes among people who otherwise would not have voted, 30 net votes (which accounts for new voters and voters who switched candidates), and $2,208 of additional contributions. For comparison, prior research estimates that television ads yield 6 votes per $1,000. Thus, advertising may have generated reasonably positive returns for political parties.

Taken together, our findings suggest that Facebook and Instagram advertising had, at most, a minimal impact on political outcomes in 2020; however, its low cost may have made it worthwhile for campaigns. Our findings rebut claims in the popular press and prior studies that digital ads are likely to substantially influence political attitudes and election outcomes.

Note
This research brief is based on Hunt Allcott et al., “The Effects of Political Advertising on Facebook and Instagram Before the 2020 US Election,” National Bureau of Economic Research Working Paper no. 33818, May 2025.