Can Differences in 'Misinformation' Sharing Explain Political Disparities in Social Media Suspensions?
A new study finds that conservatives are especially likely to share information from sources that a "politically balanced" sample of Republicans and Democrats deemed untrustworthy.
During last week's vice presidential debate, the Republican candidate, Sen. J.D. Vance (R–Ohio), decried "the threat of censorship." One example he cited was "big technology companies" that are "silencing their fellow citizens."
Vance not only glided over the legally crucial distinction between government-directed censorship and content rules enforced by private social media companies on their own platforms. He took it for granted that left-leaning social media moderators apply a double standard in deciding which posts violate those rules—a longstanding conservative complaint that has inspired constitutionally controversial state laws aimed at promoting balance by restricting moderation practices. A recent study of this issue, reported last week in Nature, casts doubt on Vance's assumption by suggesting an alternative explanation for disparities he attributes to political bias.
Based on an analysis of posting behavior and subsequent suspensions on Twitter, Oxford Internet Institute professor Mohsen Mosleh and four other researchers confirmed that Republicans and conservatives were much more likely to run afoul of moderators than Democrats and progressives were. But they also found that right-leaning social media users were much more likely to share information from "low-quality news sites." Those findings, the authors say, suggest that "differences in misinformation sharing can lead to politically asymmetric sanctions."
I know what you're thinking: Since "misinformation" is a vague, subjective, and highly contested category, it can easily serve as a cover for bias against particular opinions or ideologies. But Mosleh et al. took that possibility into account by judging the quality of news sites based on "trustworthiness ratings" by a nationally representative and "politically balanced" sample of 970 Republicans and Democrats. They also considered how sites ranked when they were rated only by the Republicans.
Just to give you a sense of the rankings, sources with relatively high trustworthiness ratings, based on the average judgments of the bipartisan sample, included the New York Post, the Daily Mail, and AOL News as well as ABC News, The New York Times and The Washington Post. Sources with the lowest ratings included Before It's News, Real News Right Now, and Not Allowed To.
After the 2020 presidential election, Mosleh et al. found, "pro-Trump" or "conservative" Twitter users "were indeed substantially more likely to be suspended" than "pro-Biden" or "liberal" Twitter users. But the former group "also shared far more links to various sets of low-quality news sites," even as judged by "politically balanced groups of laypeople" or "groups of only Republican laypeople." Mosleh et al. found "similar associations between stated or inferred conservatism and low-quality news sharing (on the basis of both expert and politically balanced layperson ratings) in 7 other datasets of sharing from Twitter, Facebook and survey experiments, spanning 2016 to 2023 and including data from 16 different countries."
The main investigation involved random samples of 4,500 Twitter users who "shared at least one #VoteBidenHarris2020 hashtag" in October 2020 and 4,500 users who "shared at least one #Trump2020 hashtag" during the same period. Mosleh et al. found that "people who used Trump hashtags shared news from domains that were on average rated as significantly less trustworthy than people who used Biden hashtags." Based on the bipartisan trustworthiness ratings, for example, "the median Trump hashtag poster shared four times more links to low-quality websites compared with the median Biden hashtag poster."
Nine months later, the researchers found that "accounts that had shared #Trump2020 during the election were 4.4 times more likely to have been subsequently suspended than those that shared #VoteBidenHarris2020." But in light of the previously measured difference in sharing behavior, Mosleh et al. write, that disparity "does not necessarily indicate a causal effect of a user's politics on suspension," given "the potential for political orientation to be confounded with the tendency to share misinformation."
Mosleh et al. found further evidence that "the tendency to share misinformation" is politically skewed when they analyzed data from seven other sources, including information about "YouGov respondents' on-platform Facebook sharing in 2016," "prolific respondents' on-platform Twitter sharing in 2018," and "the on-platform sharing of Twitter users sampled in various ways in 2021." And again, that association was apparent based on the "politically balanced" trustworthiness assessments as well as "fact-checker ratings."
These results are consistent with previous research, Mosleh et al. say. They cite studies finding that "links to websites that journalists and fact-checkers deemed to be low-quality 'fake news' sites were shared much more by conservatives than liberals on Facebook" during the 2016 election and the 2020 election and on Twitter during the 2016 election and during Donald Trump's first impeachment.
Other studies have found that "conservatives on Twitter were much more likely to follow elites [who] made claims fact-checkers rated as false compared with Democrats" and that "Republican-oriented images on Facebook were much more likely to be rated as misleading than Democratic-oriented images." Mosleh et al. also note evidence from surveys that "present participants with politically balanced sets of headlines," which "typically find that conservatives indicate higher sharing intentions for articles deemed to be false by professional fact-checkers than liberals."
Such associations can be seen in other countries as well as the United States. "A survey experiment conducted in 16 countries found widespread cross-cultural evidence of conservatives sharing more unambiguously false claims about COVID-19 than liberals," Mosleh et al. note. "An examination of Twitter data found that conservative political elites shared links to lower-quality news sites than liberal political elites in the USA, Germany and the UK."
Conservatives might dismiss much of that earlier research, given the prominent role played by presumably left-leaning "journalists" and "professional fact-checkers." The advantage of this study is that it relies on news quality ratings by "laypeople" with a variety of political orientations. And although we still don't know for sure exactly why those 2020 Twitter users were suspended, the association between partisan allegiance and sharing behavior that Mosleh et al. report is at least one plausible explanation for the disparity in consequences that Republicans like Vance blame on politically biased moderators.
"It is important to keep in mind that people who share content on social media are not representative of the general public, and therefore the consistent pattern we observe here does not necessarily generalize to comparisons of the average liberal versus conservative or Democrat versus Republican," Mosleh et al. write. They add that "the pattern that we observe in these data may be different at other points in time."
It is hardly surprising, for example, that Trump supporters would be more likely than Biden supporters to share claims about a supposedly stolen presidential election in 2020. After the 2016 election, by contrast, Democrats surely were more likely than Republicans to share dubious excuses for Hillary Clinton's defeat.
Beyond circumstances like those, this study does not tell us why partisan allegiance was correlated with reliance on news sources that survey respondents considered untrustworthy. That association, Mosleh et al. note, does not necessarily mean conservatives are "psychologically inclined" to "share more misinformation"; perhaps they were simply "exposed to more misinformation," possibly "because, at least during the study period, conservative elites share[d] more misinformation than liberal elites."
In any case, this study complicates the story told by Republicans such as Vance, Trump, Florida Gov. Ron DeSantis, and Texas Gov. Greg Abbott. "There is clear evidence of a political asymmetry in misinformation sharing among social media users in the USA," Mosleh et al. write. And since that asymmetry is apparent even when "misinformation" is defined based on its origin in sources that Republicans as well as Democrats consider untrustworthy, it "cannot be easily attributed to partisan bias on the part of those determining what counts as misinformation." Mosleh et al. therefore conclude that "differential treatment of those on one versus the other side of the aisle does not on its own constitute evidence of political bias on the part of social media companies."
