October 1, 2020
President Trump and other conservatives have long cited biases and inequities in treatment from mass media and social media sites. Because of concerns with the First Amendment’s protection of free expression, it is important to study whether Facebook’s ad removal policy is unfavorably biased against a certain party. Using the Political Ads Library, we analyzed political campaign ads on both sides of the aisle.
Results Summary
We found that, in looking at the campaigns at an individual level, the highest proportion of ads removed belonged to the Trump campaign. Using a difference of proportions test, we determined that more Republican ads were taken down than those of Democratic campaigns.
Taking advantage of the Political Ads Library, we decided to examine the ads that Facebook has taken down, particularly with regard to their party affiliations. As social media sites are increasingly used as news sources, their ability to restrict content and thus free expression are of increasing concern. As such, this study becomes markedly important because of its First Amendment implications. The Supreme Court has consistently protected the First Amendment— such as through New York Times Co. v. Sullivan, which established a high standard for libel— with regard to politics in the media to “foster an informed public” through active debate [1]. If establishing this informed public is a primary goal, Facebook’s 2020 advertising policy should not target a specific political party’s ads. To that end, we tested the following null hypothesis: Facebook takes down the same percentage of ads for both Republicans and Democrats.
In 2019, Facebook tasked Senator Jon Kyl and Covington & Burling LLP with producing a report on this perceived anti-conservative bias. Although the report did not find definitive evidence of an anti-conservative bias, it highlighted Facebook’s potential to restrict free expression due to a lack of transparency in the removal of ads [2]. Their methods consisted primarily of interviews with “key conservative organizations, individuals, and lawmakers who either use, study, or have the potential to regulate Facebook” [3].
We decided to focus our research on the campaigns for President in the 2020 elections in order to see the distribution of ads taken down broken down by party. Going through each campaign, we tracked how many ads the campaign had on Facebook, how many Facebook took down, and then calculated a percentage. The percentages were calculated at an individual campaign level as well as a party level. From there we ran a difference of proportions test in order to see if Facebook was taking down a higher proportion of either Democratic or Republicans ads at the 95% confidence level.
| Candidate | Party | Total Ads | Ads Taken Down | Percent of Ads Taken down |
|---|---|---|---|---|
| Trump | R | 108,410 | 1,641 | 1.51% |
| Harris | D | 85,043 | 1,287 | 1.51% |
| O'Rourke | D | 10,255 | 51 | 0.50% |
| Sanders | D | 5,991 | 1 | 0.02% |
| Warren | D | 7,252 | 26 | 0.36% |
| Steyer | D | 8,350 | 2 | 0.02% |
| Booker | D | 4,908 | 1 | 0.02% |
| Gillibrand | D | 2,344 | 8 | 0.34% |
| Yang | D | 2,006 | 4 | 0.20% |
The table above illustrates which campaigns were included in the study and the data of how many ads they ran on Facebook, how many Facebook took down, and then the proportion of ads that Facebook took down. The highest proportion of ads that Facebook took down belonged to the Trump campaign.
When we added up the results for each party we got the following table:
| Party | Total Ads | Ads Taken Down | Percent of Ads Taken down |
|---|---|---|---|
|
R |
108,410 |
1,641 |
1.51% |
|
D |
126,149 |
1,380 |
1.09% |
When running our difference of proportions test, we got a z-score of 8.88. This means that we can reject our null hypothesis at the 95% confidence level in favor of our alternative hypothesis that Facebook takes down a higher proportion of republican ads then democratic ads.
While we were able to show that Facebook does indeed take down more republican ads then democratic ads on a percentage basis, this does not prove causation. This means that we cannot decisively say that Facebook is taking down more Republican ads because they do not like the party. There could be a confounding variable in that the Trump campaign intentionally ran ads that they knew would be taken down on a higher percentage basis than democratic campaigns did. Another limitation of our study is that Trump was the only Republican candidate that was included. This was not intentional but rather a side effect of the fact that he was the only Republican candidate who ran a large campaign during this time. This certainly could have skewed our results and in order to truly prove that Facebook is biased, a study would need to be done that included a larger number of candidates from different years.
Ryan Chung and Paul Marino are undergraduates at Harvard University in GOV 1430 Tech Science to Save the World in the Spring of 2020.