We're the world's oldest democracy, and the only way that can come unraveled is if we have disrespect for institutions that lead to Americans turning on Americans. A lot of that starts with the words we're using.
—Representative Brian FitzpatrickThe peaceful transfer of power is an essential component of a healthy electoral democracy (Dahl, Reference Dahl1989), and over two centuries of presidential elections, losing parties have observed this norm, offering losers’ consent because of deeply held values or long-term electoral incentives. When Trump supporters attacked the US Capitol on January 6, 2021, their attempt to overturn the results of a free and fair election represented a breakdown of this core democratic tradition. Yet, Americans fundamentally disagree over what happened, who was responsible, and what consequences are appropriate, if any (Rakich, Reference Rakich2022). While a majority of Americans agree that January 6th was “an attack on democracy,” most Republicans think “it is time to move on” (Weiner et al., Reference Weiner, Clement and Guskin2024). These disagreements are not trivial, as a stable democracy depends upon public support for democratic norms among both elites and the mass public (Levitsky and Ziblatt, Reference Levitsky and Ziblatt2018) and partisan loyalty undermines people's willingness to prioritize democratic ideals (Simonovits et al., Reference Simonovits, McCoy and Littvay2022).
To better understand how views of January 6th became polarized, we examine how members of Congress initially framed events. We collect every tweet posted by members of the House and Senate on January 6th and study the language used to describe the attack while it unfolded. Nearly every member condemned the attack, posting tweets that expressed outrage, called for action, and communicated their safety.Footnote 1 Media accounts reflected this widespread disapproval, reporting bipartisan alarm with headlines such as “Democrats, Republicans blame Trump for inciting ‘coup’ as mob storms Capitol”Footnote 2 and “The Day Trump Broke the GOP.”Footnote 3 We find that while unified in condemning the attack, members portrayed two fundamentally different threats. Democratic members were more likely to describe events as a threat to democracy, recognizing the attack's political implications. Republicans were more likely to portray the attack as a legitimate protest that got out of hand, recognizing the grievances of those involved while condemning violence. However, Republican members representing districts where Trump did poorly were more willing to describe the events as an attack on democracy.
We examine how members of Congress framed the January 6th attack on Twitter for two reasons. First, members’ tweets provided a first-hand account that informed media coverage of the day. Members use Twitter to share information about their activities, cultivate a political brand, and influence the policy agenda (Russell, Reference Russell2021). While most people do not follow elected representatives on Twitter, journalists comprise a significant portion of their audience (Kreiss et al., Reference Kreiss, Lawrence and McGregor2018). Politicians’ tweets are routinely quoted in media coverage (Heim, Reference Heim2021), expanding their reach and influencing public perceptions of the events and people involved (Dumitrescu and Ross, Reference Dumitrescu and Ross2021). On January 6, this dynamic was amplified as members had a closer view of the attack than any media. As Trump supporters overwhelmed the Capitol police, disrupted congressional proceedings, and sent elected officials and their staffs scrambling, members were “on the ground” reporters. They took to Twitter to describe their experiences, and their tweets were reprinted by news outlets, from The Washington Post to the Montgomery Advertiser, shaping public understanding of events.
Second, political elites play a unique role in upholding democratic norms (Rosenblum and Muirhead, Reference Rosenblum and Muirhead2020). The “carriers of the creed” argument claims that the United States survives as a liberal democracy because of elite consensus on democratic values (McClosky, Reference McClosky1964), although that commitment is waning (Layman et al., Reference Layman, Lee and Wolbrecht2023). Americans take their cues from political elites (Zaller, Reference Zaller1992) and put trust in leaders in times of crisis (Albertson and Gadarian, Reference Albertson and Gadarian2015). When elite rhetoric is divided, polarization is exacerbated (Druckman et al., Reference Druckman, Peterson and Slothuus2013), and public trust in democratic institutions is undermined (Clayton et al., Reference Clayton, Davis, Nyhan, Porter, Ryan and Wood2021). Understanding areas of consensus and division in elite communication has important implications for public support of democratic norms and how contemporary polarization influences our ability to respond effectively to national crises (Boussalis et al., Reference Boussalis, Coan and Holman2024).
Partisan divergence in congressional messaging is common, as members compete by promoting their party brand while distinguishing themselves from the opposition (Lee, Reference Lee2016). There are two main reasons to expect this dynamic would not characterize communication on January 6th. First, common threats have unifying effects, reducing identification with subgroups and fostering alliances between adversaries (Flade et al., Reference Flade, Klar and Imhoff2019). Second, supporting democratic norms—such as accepting legitimate election results—should be in the best interest of both parties, either because they value democratic traditions or anticipate long-term gains (Weingast, Reference Weingast1997). Further, after losses in the 2018 and 2020 elections, Republicans in Congress might have been eager to distance themselves from an unpopular president.
In the immediate aftermath of the attack, there were multiple signs of wavering Trump support. Members of both parties expressed widespread disapproval of those involved (Anderson and Coduto, Reference Anderson and Coduto2024), corporations paused campaign contributions (Li and Disalvo, Reference Li and Disalvo2023), and Republican Party identification decreased (Eady et al., Reference Eady, Hjorth and Dinesen2023; Loving and Smith, Reference Loving and Smith2024). Pundits suggested the attack would be a breaking point in the Republican Party. Yet within the next year, the people who attacked the Capitol were recast by some Republican members of Congress as “peaceful patriots,” and the public view of January 6th was split along partisan lines, with a majority of Republicans viewing the attack as a “legitimate protest,” and most Democrats continuing to describe it as an “insurrection” (Monmouth University Polling Institute, 2022).
Our research demonstrates that the seeds of partisan division were present from the moment the Capitol barriers were breached. Beneath the near-universal condemnation, members offered two fundamentally different narratives of events. In one, there is a clear breakdown of democratic norms in which armed Trump supporters sought to overturn the election. In the other, a legitimate protest turned into a riot, representing a breakdown of law and order. This starkly partisan divide hinders Congress's ability to address political violence and other undemocratic behavior (Adler and Wilkerson, Reference Adler and Wilkerson2013), undermines public support for basic democratic principles (Kingzette et al., Reference Kingzette, Druckman, Klar, Krupnikov, Levendusky and Ryan2021), and ultimately threatens American democracy.
1. Data and methods
We identified the official, personal, and campaign handles of every House and Senate member in office on January 6th and downloaded all tweets posted by these accounts between 1 p.m. and 9 p.m. on January 6, 2021.Footnote 4 We removed retweets and links, yielding 1861 tweets from 604 accounts or 497 members. We then pre-processed the text to identify common word stems, remove uninformative words, and preserve commonly occurring word pairs and negations, such as “rule of law” and “not protest.”Footnote 5
We use a semi-supervised machine-learning approach developed by King et al. (Reference King, Lam and Roberts2017) to create two dictionaries corresponding to frames we label as “attack on democracy” and “lawless protest.” This approach uses an iterative process to identify all relevant keywords in the data based on an initial seed of high-discernment terms and is intended to account for limitations in human ability to create comprehensive dictionaries.Footnote 6
Our attack on democracy frame is built around words and phrases directly implicating political structures and processes. Trump supporters forcibly breached the police perimeters around the Capitol to prevent Congress from counting electoral votes and nullify the 2020 presidential election. Many of those people also sought to intimidate or punish political enemies, as evidenced by the vandalization of the building and chants of “Hang Mike Pence.” We construct the attack on democracy dictionary to capture tweets portraying the attack as an attempt to subvert the Constitution by including language such as attack_democracy, coup, insurrection, and sedition.
In contrast, our lawless protest frame is built around the idea of a protest that got out of hand. Importantly, no members took to Twitter on January 6th to applaud the attack. Lawless protest tweets emphasize free speech and do not recognize the constitutional threat, instead condemning acts of lawlessness and disorderly behavior. We construct the lawless protest dictionary to capture tweets emphasizing protesters’ First Amendment rights while objecting to violence or focusing on law and order. Figure 1 displays the word stems for both dictionaries and their frequency of use.Footnote 7
Frequency of usage for “Attack on Democracy” and “Lawless Protest” dictionary terms.
After creating the dictionaries, we identified all instances of these keywords in members’ tweets. Each tweet is categorized into one of three frames:
• Attack on Democracy—any usage of words in the attack on democracy dictionary
• Lawless Protest—usage of words in the lawless protest dictionary and no usage of words in the attack on democracy dictionary
• Neither—no usage of words from either dictionaryFootnote 8
We classify all 116 tweets that use both protest and democracy language as “attack on democracy” tweets because they capture the political implications of the attack. A tweet that describes participants as protesters and condemns the attack as an attempted coup recognizes the violation of democratic norms.Footnote 9 In contrast, a stand-alone lawless protest frame provides an alternate version of events—a group voiced legitimate political speech, and things got out of hand.
2. Results
Figure 2 displays the number of tweets posted by each member and the number of members using each frame. Nearly every member tweets during the Capitol attack, with the median member tweeting twice and 40 members not tweeting.Footnote 10 Frame usage by member shows a clear partisan distinction; 58.4 percent of Democrats and 12.5 percent of Republicans posted only tweets classified as “attack on democracy,” whereas 6.2 percent of Democrats and 52.9 percent of Republicans posted only lawless protest tweets.
Tweet frequency and frame utilization at member level.
Figure 3 disaggregates the data to the tweet level and displays the number of tweets using each frame and the proportion of tweets using each frame across party and chamber. Among the 1861 tweets in our dataset, 32.7 percent used attack on democracy language, and 17.6 percent used only lawless protest language. Roughly one-half of the tweets do not use either frame. We observe consistent patterns of frame usage within-party and across chambers, with tweets from Democrats in both chambers using the attack on democracy frame more frequently and tweets from Republicans in both chambers using the lawless protest frame more frequently.
Frame utilization at tweet level as count and proportion.
We also examine how the tweets unfolded over time (Figure 4). While frames vary over the 8 hours, the two parties offer differing characterizations from the start.Footnote 11 During every hour, Republicans were more likely to tweet with a lawless protest frame, and Democrats were more likely to tweet with an attack on democracy frame.
Tweet usage over time.
Next, we estimate models to understand the dynamics of how members framed the January 6th attack. First, we examine which members were active on Twitter. We use a negative binomial model where the unit of observation is the member, and the dependent variable is the number of tweets they posted between 1 p.m. and 9 p.m. Some members tweet more than others, so we control for the average number of tweets (logged) they post each year. To account for the possibility that members from Trump-supporting districts remained quiet rather than risk Trump's ire, we control for Trump Margin, which is the difference between the share of the vote received by Donald Trump and the share of the vote received by Joe Biden for that state or district in the 2020 election (scaled by x10). We also control for each member's chamber, party, number of terms served, if they are in party leadership, and their margin of victory in the 2020 election.
Model results, presented in Figure 5, illustrate that overall Twitter activity is the strongest predictor of how much members tweeted on January 6th.Footnote 12 After controlling for tweets per year, we find that Senators tweeted less, as did members representing districts and states where Trump performed better in the 2020 election. Figure 5 also displays the model predictions and shows a statistically significant decrease in the predicted number of tweets as Trump support increases, consistent with the possibility that members from these districts were strategically quiet on January 6th.Footnote 13
Model results estimating number of tweets posted by member.
For our study of frame utilization, tweets are the unit of analysis, reflecting our interest in the overall narrative presented on Twitter. To test our hypotheses, we estimate multinomial logistic regression models in which the dependent variable is the choice between the three frame options: attack on democracy, lawless protest, or neither. Our key independent variables are whether the tweet came from a member of the Democratic Party (Democrat), and electoral support for Trump in the tweeting member's constituency (Trump Margin). Our partisan hypothesis predicts that Democrats are more likely to use the attack on democracy frame and Republicans are more likely to use the lawless protest frame. Our Trump hypothesis predicts that for Republican members, increased district-level support for Trump is associated with a decreased likelihood of using the attack on democracy frame and an increased likelihood of using the lawless protest frame.
We also include controls for the number of congresses a member served (Seniority); whether they are in party leadership (Leadership); whether they faced a primary challenger in their last election (Primary); and the difference between their vote share in the previous election and their general election opponent (Electoral Safety). We also control for Tweet Type to account for whether a tweet was a quote, reply, or part of a thread, and Time Trend, which is an indicator for each hour.Footnote 14
Our first model predicts the likelihood that a tweet used either frame, with “neither” as the reference category. Across all tweets, the clearest distinction in frame usage is partisan, with Democrats 3.38 times more likely than Republicans to choose the attack on democracy frame over neither frame. Similarly, Democrats are 0.26 times less likely than Republicans to use the lawless protest frame. In substantive terms, the probability of an average Democratic tweet using the attack on democracy frame is about the same as that of an average Republican tweet using the lawless protest frame, around 0.40 (Figure 6a). At the same time, Democrats are significantly less likely to use the lawless protest frame, and Republicans are significantly less likely to use the attack on democracy frame, illustrating the clear partisan divide in how members portrayed the attack on the US Capitol.
Model results estimating frame utilization by tweet.
As the strength of partisanship obscures other relationships, we re-estimate the models by party. For Democratic tweets, none of our predictors are statistically significant, consistent with widespread usage of the attack on democracy frame by Democratic members (Figure 6c). For Republican members, several factors are associated with frame usage (Figure 6d). Tweets from senior Republicans were more likely to use the attack on democracy frame, while tweets from Republicans with a 2020 primary challenge were more likely to use the lawless protest frame. Consistent with our hypothesis, support for Trump in a Republican member's constituency is a significant predictor of a tweet invoking the attack on democracy frame. After controlling for primary threat and electoral security, a one-point increase in Trump support is associated with a decrease in the likelihood a tweet uses the attack on democracy frame by a factor of 0.96. While this may seem small, Figure 6b illustrates that increased support for Trump has a substantive relationship with frame usage. For tweets from generic Republicans representing a constituency that supported Trump and Biden equally, the predicted probability of the attack on democracy frame is 0.137. For tweets from similar members representing constituencies where Trump won by 50 points, the predicted probability of the attack on democracy frame is 0.023. These findings further support our argument that both partisanship and support for Trump shaped how members framed the January 6th attack.
3. Discussion
As Trump supporters broke into the US Capitol building, members of Congress offered diverging perspectives of the attack. Most Democrats depicted an attack on democracy, whereas Republicans portrayed a lawless protest. We argue that how January 6th was characterized influences whether the attack is viewed as a threat to democracy in which armed Trump supporters attempted to overturn the results of an election or a legitimate display of free speech that turned violent. Understanding the language members used on January 6th helps us understand why Republican members were “outraged” by the attack on the Capitol, but that anger did not translate into action. Their initial framing of the attack as a lawless protest informs subsequent objections to the January 6th Committee, described by the Republican National Committee (RNC) as “a Democrat-led persecution of ordinary citizens engaged in legitimate political discourse” (Wise, Reference Wise2022).
Members’ tweets from January 6th also provide insights into intra-party divisions within the Republican Party. The two Republicans who served on the January 6th committee, Liz Cheney (R-WY) and Adam Kinzinger (R-IL), were censured by the RNC and no longer serve in the House, but they are not the only minority voices who left Congress. Republicans who framed January 6th as an attack on democracy were more likely to draw a primary challenger and less likely to be reelected in 2022.Footnote 15 In contrast, Republicans who posted only lawless protest tweets were more likely to reject the Arizona and Pennsylvania electors and less likely to support impeaching Trump and establishing a January 6th commission. They have also been more electorally successful than Republicans who portrayed the attack as an attack on democracy.
By highlighting the partisan divisions present in the first hours of the attack, this research lays the groundwork for future work on how partisanship shapes interpretations of January 6th and the role of elite messaging in maintaining democracy's guardrails.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2024.51.
To obtain replication material for this article, https://doi.org/10.7910/DVN/N64C76
Open access
