Skip to main content
ABC News
Political Twitter Is No Place For Moderates

Twitter has never been a bastion of the Queen’s English, but recent reporting has unearthed a variety of polarizing posts about American politics that are presumed to be of Russian origin, in part because some of them use awkward English. The revelations have led to recriminations against the tech platform and an invitation for the company to testify before Congress this week.

The Russians might have occasionally gotten their words mangled, but they were right when it came to mimicking their targets, attacking from both the right and the left. It turns out that American Twitter users who tweet about politics overwhelmingly come from the extremes of the political spectrum as well.

A study I recently undertook with University of Pennsylvania colleagues Daniel Preotiuc-Pietro, Ye Liu and Lyle Ungar sheds light on who discusses politics on Twitter. We found that tweeting about politics is the exception, not the rule, and that the use of political words is concentrated among the small group of respondents who term themselves either “very conservative” or “very liberal.”

Our research surveyed 3,938 Twitter usersnationally representative panel I helped oversee in January 2016, which found 26 percent of respondents to be “conservative” or “extremely conservative” while 17 percent were “liberal” or “extremely liberal.”

">1 who together generated 4.8 million tweets that we downloaded in August 2016. We then coded the most common 12,000 words based on whether they were political, either because they used an explicitly political term (say, “president”) or a proper noun that was likely to be political (“Romney,” “Maddow” or “CNN”). In all, we found just 293 of the 12,000 words to be political.

Even though we might expect a sample of Twitter users to be more political than the U.S. population overall, we were surprised to see that the use of political words was concentrated among the small group of respondents who term themselves “very conservative” or “very liberal.” Take a look at at the chart below, which illustrates who is using these political terms. The C-shaped pattern makes the core result clear. It’s primarily those people on the ideological extremes who use political terms with any frequency. In other words, if someone tweets about politics, they are likely to label themselves as on the edges of the American ideological distribution. People who deem themselves “very conservative” or “very liberal” seem to be telling us as much about their level of political engagement as about their policy positions.

We also studied a related question: What words are most predictive of tweeters who are very conservative or very liberal? It turns out that religious words and words describing family relationships are among the most distinguishing marks of strong conservatives’ Twitter vocabulary, as well as the use of hashtags like “#pjnet” — referring to the conservative Patriot Journalist Network — and handles such as @foxnews or @realdonaldtrump. By contrast, liberals stand out for their use of terms like “gay,” “racism” and “Sanders.”

Strong liberals’ language also tends to include more adjectives, adverbs, conjunctions and comparisons, which may indicate an effort to convey more nuance. Also, some moderate liberals employ sexual terms so heavily in their tweets that they rendered our paper one of the rare computer science papers that would merit an R rating. But what distinguishes even people who are very liberal and very conservative is principally their use of non-political terms, whether those terms relate to religion or sexual orientation.

We asked a more representative sample of American adults2 about their Twitter usage and found that as of November/December 2016, just 12 percent of our respondents reported using the social media platform. So Twitter posters are already a distinct minority, and even among those who do tweet, users who routinely tweet about politics are an even smaller and more atypical minority. Political tweeters are even more polarized than the nation as a whole.


  1. We identified the users through an opt-in panel run through Qualtrics, meaning that these respondents come from various parts of the U.S. but are decidedly not a random sample of the population. As is commonly the case with online convenience samples, the respondents skew left relative to the national distribution: Among those who placed themselves on the traditional left-right spectrum, 17 percent self-identify as conservative or very conservative, while 37 percent say they are liberal or very liberal. Compare that to a nationally representative panel I helped oversee in January 2016, which found 26 percent of respondents to be “conservative” or “extremely conservative” while 17 percent were “liberal” or “extremely liberal.”

  2. The long-running panel survey I mentioned earlier.

Dan Hopkins is a professor of political science at the University of Pennsylvania, and his research focuses on American elections and public opinion.