The Big Lie among state legislators
Measuring the prevalence and correlates of 2020 election denial

Introduction

How committed are political elites to the basic norms of democracy? Classic accounts of demo­c­ra­t­ic resilience in the United States suggest that political elites almost uniformly subscribe to the ideals of liberal democracy and that they play a key role in safe­guard­ing it. However, recent events raise concerns that some elected officials may not defend democracy and might even undermine it.

The 2020 pres­i­den­tial election and its aftermath provided com­pelling evidence for both propo­si­tions. On the one hand, incumbent president Donald J. Trump refused to acknowl­edge his loss and fomented a violent attack on the Capitol in an attempt to stop con­gres­sion­al cer­ti­fi­ca­tion of the election results. On the other hand, key political actors from Trump’s own party resisted pressure to manip­u­late the election results and proved instru­men­tal in ensuring a peaceful transfer of power. The lessons to be drawn from the 2020 election are decidedly con­tra­dic­to­ry: the election was both “a miracle and a tragedy,” simul­ta­ne­ous­ly improving and wounding faith in the com­mit­ment of elites to demo­c­ra­t­ic norms. 

Understanding the behavior of political elites is thus crucial as we approach the 2024 election. Media attention to election denial was greatest for Trump and his allies in Congress, of course. Similarly, numerous Republican can­di­dates for statewide elected office in 2022 did not recognize the results of the 2020 pres­i­den­tial election. However, we know much less about the behavior of lower-level officials – a major blind spot given the key role of states in con­duct­ing elections, America’s history of sub­na­tion­al author­i­tar­i­an­ism, and the potential for state-level demo­c­ra­t­ic backsliding.

In this project, we measured the preva­lence of election fraud claims in tweets from state leg­is­la­tors drawing on data from December 2020-February 2023. Our main findings were as follows:

• Republican state legislators made or endorsed claims of election fraud over 2,000 times on Twitter.

• However, this activity was heavily concentrated among a handful of legislators and states (most notably in Arizona).

• Finally, state legislator ideological extremity is a strong predictor of Twitter endorsement of election fraud claims, whereas the ideology of constituents was not.

Democratic erosion from the top

A recent lit­er­a­ture in political science has high­light­ed the con­di­tion­al nature of citizens’ com­mit­ment to democracy: the vast majority express support for democracy in principle but many fail to sanction norm vio­la­tions from parties or can­di­dates they support. Citizens’ con­cep­tions of those norms them­selves are also malleable – people may ratio­nal­ize trans­gres­sive actions taken on behalf of policies one favors. The public cannot be counted on to act as a bulwark against demo­c­ra­t­ic erosion. 

Given these vul­ner­a­bil­i­ties, prominent schol­ar­ship has cast political elites as crucial defenders of demo­c­ra­t­ic norms against populist author­i­tar­i­an impulses. Yet the com­mit­ment of American political elites to the basic norms of democracy, long assumed to be nearly universal, appeared to soften in the last few years as officials across the country pursued partisan objec­tives at the expense of core demo­c­ra­t­ic values. 

The 2020 pres­i­den­tial election exem­pli­fied this trend: despite the absence of evidence that fraud altered the outcome of the election, many Republicans at all levels of gov­ern­ment refused to accept Joe Biden as the rightful winner. News orga­ni­za­tions including The New York Times, The Washington Post, and Bloomberg published sys­tem­at­ic tallies of election-denying can­di­dates for various high-level offices. The results under­scored the gravity of the situation: nearly a third of Republican can­di­dates for Attorney General and approx­i­mate­ly half of Republican can­di­dates for U.S. House did not accept the results of the 2020 election.

Whether indi­vid­u­als in positions of power accept the results of elections and abide by demo­c­ra­t­ic norms while in office matters for democracy. First, elected officials shape demo­c­ra­t­ic per­for­mance through their exercise of their formal powers. In the United States, which has an unusually decen­tral­ized election admin­is­tra­tion system, state officials play a large role in deter­min­ing who is allowed to vote and how con­gres­sion­al districts are drawn. In addition, elected officials exert an important indirect influence on demo­c­ra­t­ic per­for­mance. Through their words and actions, indi­vid­u­als who hold formal power can influence the political envi­ron­ment. Recent research shows that norms of behavior are pow­er­ful­ly shaped by elite rhetoric: what politi­cians say and do sets the tone for con­stituent conduct and attitudes. Candidates and office­hold­ers who reject the results of an election, regard­less of the strength of evidence under­ly­ing their claims, can weaken the public’s con­fi­dence in elections.

The behavior of higher-level elected officials, such as governors and members of Congress, is con­cern­ing and worthy of attention. However, other office­hold­ers with sub­stan­tial influence on election admin­is­tra­tion – both formal and informal – have received less scrutiny. In par­tic­u­lar, the role of state leg­is­la­tors in prop­a­gat­ing false claims about election security deserves more attention. This project examines the behavior of these important but largely over­looked political actors. Scholarship on rep­re­sen­ta­tion in state leg­is­la­tures shows that mech­a­nisms of account­abil­i­ty are weak, which makes detailed and sys­tem­at­ic inves­ti­ga­tions of their members’ conduct even more important.

Social media data collection 

In the past, both Republicans and Democrats have expressed doubts about the legit­i­ma­cy of election results. However, questions about the legit­i­ma­cy of the 2020 election have been raised almost exclu­sive­ly by Republicans. We therefore focus on the behavior of GOP state legislators. 

We con­struct­ed a list of over 7,000 state Republican leg­is­la­tors in all 99 leg­isla­tive chambers across all 50 states using data from Ballotpedia, an online political ency­clo­pe­dia. We then scraped leg­is­la­tors’ Twitter and Facebook posts from December 2020-February 2023 and used keyword searches to identify tweets about election fraud. Each post was then read by at least two human coders who coded whether it endorsed claims that the 2020 pres­i­den­tial election was marred by fraud. In total, we iden­ti­fied a total of 2,268 tweets by U.S. state leg­is­la­tors that promote election fraud claims during the study period. (The appendix includes a detailed descrip­tion of our methods.)

Results 

We first analyze the preva­lence of election fraud tweets over time in the figure below, which reveals sub­stan­tial variation. Unsurprisingly, we first see a flurry of activity in the aftermath of the 2020 election. Between December 15, 2020 and January 5, 2021, state leg­is­la­tors posted 284 tweets that endorsed claims of election fraud, an average of nearly 13 such tweets per day. A sharp break occurs after the storming of the Capitol on January 6th. Between January 7 and January 20, the day when Joe Biden was inau­gu­rat­ed as president, the rate drops to approx­i­mate­ly 2 per day (a total of 26 tweets). 

We observe little fraud claim activity during the rest of winter and spring 2021, but there are two notice­able peaks in July 2021 and September 2021. Manual inspec­tion of the data makes clear that these are not random events. In mid-July, many of the tweets we cat­e­go­rize as endorsing election fraud claims refer to a segment presented on Tucker Carlson’s show on Fox News on July 14. On July 15, the Arizona State Senate held hearings on election fraud that also attracted com­men­tary from state leg­is­la­tors. We can similarly account for the September 2021 spike: on September 24, the long-awaited forensic audit performed by the “Cyber Ninjas” in Maricopa County was made public. 
Since then, there has been very little activity. Most notably, there was no observ­able spike in election fraud claims following the 2022 midterm elections in which Republicans took control of the House of Representatives but failed to gain a majority in the U.S. Senate (cor­re­spond­ing with public opinion trends after the election).

However, it is important to note that just a few state leg­is­la­tors are respon­si­ble for most of the prob­lem­at­ic tweets that we identify. This pattern can be seen in the figure below, which shows the cumu­la­tive pro­por­tion of tweets as the x‑axis moves from the most prolific state leg­is­la­tor to the least prolific (sep­a­rate­ly for all tweets and for tweets endorsing claims of election fraud). As the figure shows, the supply of tweets is highly skewed in general: the most prolific Twitter users produce a dis­pro­por­tion­ate share of all tweets. However, the pro­duc­tion of tweets that we classify as endorsing claims of election fraud is far more con­cen­trat­ed at the top. 

Many of the most prolific fraud tweeters may be rec­og­niz­able to avid news consumers. Most notably, Arizona state senator Wendy Rogers, a well-known defender of election fraud claims, appears 889 times in our data, which rep­re­sents nearly 40% of all instances we identify. The next most prolific state lawmaker is Mark Finchem (210 tweets), a former member of Arizona House of Representatives who was the Republican nominee for Secretary of State in 2022. Doug Mastriano, the Republican nominee for Pennsylvania Governor in the same election cycle, is third with 110 tweets. Taken together, this trio accounts for more than half of all tweets that we identify as endorsing claims of election fraud. In total, we find qual­i­fy­ing tweets from 175 unique Twitter accounts belonging to 171 state leg­is­la­tors – approx­i­mate­ly 4% of all Republican state lawmakers and 11% of those with an active Twitter presence. 

We find at least one qual­i­fy­ing fraud tweet from state leg­is­la­tors in 41 different states, but we also see huge variation across states driven by a handful of vocal election deniers. Most notably, Arizona stands out as the epicenter of election fraud discourse. Four of the five most prolific tweeters, and nine of the top sixteen, are from Arizona. Taken together, they account for approx­i­mate­ly 66% of all fraud tweets we observe. Following Arizona, a handful of states are respon­si­ble for the majority of the remaining activity: we record 13 leg­is­la­tors producing 145 qual­i­fy­ing tweets in Pennsylvania, 11 leg­is­la­tors producing 101 tweets in New Hampshire, 18 leg­is­la­tors producing 82 tweets in Texas, and 17 leg­is­la­tors producing 39 tweets in Wisconsin.

Are the leg­is­la­tors who endorse claims of election fraud ide­o­log­i­cal­ly extreme? We combined our data with estimates of state leg­is­la­tor ideology created by Boris Shor and Nolan McCarty These estimates, which were produced using a com­pre­hen­sive record of roll call voting data over time in the 50 states combined with a recurring survey of state leg­isla­tive can­di­dates, are designed to be com­pa­ra­ble across time, chambers, and states. 

In panel A of the figure below, each dot rep­re­sents a Republican state leg­is­la­tor. The dots in red at the top of the figure represent lawmakers who tweeted at least once in support of claims of election fraud, while the dots in black at the bottom show lawmakers who did not tweet in support of these claims. The mean Shor-McCarty ideology score of leg­is­la­tors who endorsed claims of election fraud is 1.18, compared to 0.88 for other leg­is­la­tors (higher values are more con­ser­v­a­tive). Panel B shows predicted prob­a­bil­i­ties of a Republican leg­is­la­tor endorsing claims of election fraud on Twitter as a function of Shor-McCarty scores (estimated using a logistic regres­sion model). Among the least con­ser­v­a­tive Republican lawmakers, the estimated prob­a­bil­i­ty of endorsing claims of election fraud is essen­tial­ly 0. At the 25th per­centile of con­ser­vatism among Republican state leg­is­la­tors (an ideology score of approx­i­mate­ly 0.65 in the Shor-McCarty data), the estimated prob­a­bil­i­ty remains low at about 0.02. The same is true for median ideology (the 50th per­centile), where the estimated prob­a­bil­i­ty is 0.03. The estimated prob­a­bil­i­ty of endorse­ment only rises sub­stan­tial­ly at the most con­ser­v­a­tive end of the ide­o­log­i­cal spectrum, reaching 0.05 at an ideology score of 1.30 (87th per­centile among Republicans) and 0.10 at an ideology score of 1.84 (98th per­centile among Republicans). 

However, it is not the case that leg­is­la­tors who endorse claims of election fraud are uniformly con­ser­v­a­tive. In fact, nearly 23% of the lawmakers that we identify as endorsing claims of election fraud have an ideal point that is less con­ser­v­a­tive than the GOP median. The least con­ser­v­a­tive lawmaker who posted at least one tweet endorsing claims of election fraud – Rhode Island state rep­re­sen­ta­tive Michael Chippendale – is ranked at the 1st per­centile of con­ser­vatism among all Republican leg­is­la­tors across the country.

Strikingly, district ideology has even less pre­dic­tive power. We repeated the procedure described above, but this time using a measure of public ideology in each state leg­isla­tive district that was developed by Chris Tausanovitch and Christopher Warshaw¹. Once again, each dot rep­re­sents a Republican state leg­is­la­tor in panel A of the figure below. The dots in red at the top of the figure represent lawmakers who tweeted at least once in support of claims of election fraud, while the dots in black at the bottom show lawmakers who did not tweet in support of these claims. This time, the x‑axis shows the measure of public ideology for each state leg­isla­tive district; higher values indicate policy pref­er­ences that are right-leaning, while lower values indicate policy pref­er­ences that are left-leaning. 

In contrast to the findings related to leg­is­la­tor ideology, we find no clear dis­crep­an­cy between the ideology of districts whose members did or did not endorse claims of election fraud. The mean district ideology in districts rep­re­sent­ed by leg­is­la­tors who endorsed claims of election fraud is approx­i­mate­ly 0.154, while it is 0.148 for districts rep­re­sent­ed by leg­is­la­tors who did not (this dif­fer­ence is not sta­tis­ti­cal­ly sig­nif­i­cant). Panel B of the figure rein­forces this point: the predicted prob­a­bil­i­ty of a Republican state leg­is­la­tor endorsing election fraud claims on Twitter is essen­tial­ly the same across the range of ideology in districts rep­re­sent­ed by Republicans. 

This result should be inter­pret­ed carefully because our sample is restrict­ed to Republican leg­is­la­tors, which nec­es­sar­i­ly restricts variation in district ideology. If we observed more Republican leg­is­la­tors in left-leaning districts, it is plausible that the rela­tion­ship between district ideology and endorse­ment of election fraud claims would be stronger. In other words, we cannot claim that district ideology writ large has no observ­able link to leg­is­la­tors’ state­ments on election fraud. We can only say that, among Republican leg­is­la­tors, rep­re­sent­ing a strongly con­ser­v­a­tive district is not asso­ci­at­ed with a higher prob­a­bil­i­ty of endorsing claims of election fraud than rep­re­sent­ing a less con­ser­v­a­tive, or even a fairly liberal, district — a striking finding given the strong asso­ci­a­tion between fraud endorse­ment and leg­is­la­tor ideology. 

Appendix

1. We accessed the Ballotpedia page that lists the 99 state legislative chambers in the United States. Each chamber contains an embedded URL leading to its full Ballotpedia page. We scraped all 99 embedded URLs. 

2. We used the Wayback Machine’s availability API to identify the latest archived version of a given state legislative chamber’s Ballotpedia page that preceded November 7th, 2022. On that date, most states held state legislative elections that substantially altered the composition of their legislative bodies. Since our project seeks to track behavior since December 2020, accessing archived versions of the Ballotpedia pages simplified the process. 

3. We scraped all 99 archived versions of the Ballotpedia state legislative pages using the URLs provided by the Wayback Machine’s availability API. Each page contained a table of current members, along with embedded URLs to each member’s Ballotpedia page. We extracted this information from each page, which resulted in a dataset of 7,383 state legislators. 

4. We scraped the Ballotpedia pages of all 7,383 legislators. In the vast majority of cases, a legislator’s page contains an information box with basic biographical information (name, office, party, etc.) as well as a list of social media accounts.

5. A minority of state legislative seats had not been held by the same individual since December 2020. There were two broad categories of such cases: (i) a seat that was listed as empty, and (ii) a seat whose current holder assumed office after December 2020.

   1. In the latter case, we accessed the current officeholder’s Ballotpedia page and extracted information on their predecessor (which was usually contained in the infobox at the top-right of the page or in a box at the bottom of the page). 

   2. In the former case (empty seat) or whenever the above method did not yield information on a predecessor, we queried the Wayback Machine’s accessibility API to find an archived version of the Ballotpedia page of their state legislative chamber, with a target date 180 days before the current officeholder took office. 

   3. We then scraped this archived version of the Ballotpedia state legislative chamber page, extracted the table of then-current members, located the relevant legislative seat, and extracted information on the then-officeholder (if any). Using the Twitter API, we scraped all tweets from the Twitter accounts of Republican state legislators identified above between December 2020 and January 2023. 

6. For each legislator identified through the process described above, we scrape their Ballotpedia page and store information on their social media accounts. 

7. We used keyword searches to flag tweets that could endorse claims of election fraud. Keywords such as “dead people vot,” “illegal vot,” “rigged elect,” and “steal elect” were among those used to identify tweets related to election fraud narratives. Additionally, co-occurring keyword pairs like “vot” and “fraud” or “elect” and “integrity” were detected to enhance precision. The criterion used to flag tweets can be found here. 

 

Undergraduate research assis­tants consulted each tweet iden­ti­fied through the above process and deter­mined whether or not each tweet should be clas­si­fied as endorsing claims of election fraud. The full codebook that research assis­tants referred to can be found here. Tweets that met any of the following criteria were clas­si­fied as endorsing claims of election fraud: 

 

1. Claims that there was fraud in the 2020 or 2022 elections, unless the tweet specifies that such fraud was extremely rare. 

2. Calls for, endorsements of, or updates on non-routine election audits. 

3. Statements that question the veracity of claims made by election authorities regarding the legitimacy of elections. 

4. Claims that the outcome of the 2020 election is unknown. 

5. Usage of dubious evidence to support claims of electoral fraud. 

 

Tweets that met the following criteria were not clas­si­fied as endorsing claims of election fraud (unless they also met one of the above criteria):

 

1. Statements about “election integrity” that do not include specific claims regarding the 2020 or 2022 elections. 

2. References to routine audits (e.g., audits triggered automatically because of narrow margin). 

3. Calls for future audits. 

4. Statements decrying a federal takeover of elections. 

 

Each tweet was coded by at least two research assis­tants. Tweets for which coders made different decisions were then assigned to a third coder as a tiebreak­er. For the final data, Cohen’s kappa is 0.70, indi­cat­ing fairly good con­sis­ten­cy in coding decisions across research assistants. 

1.  The Tausanovitch ideology dataset covers 93% of districts (7% are missing).