INTRODUCTION
In 2022, denying the 2020 election’s outcome became an explicit campaign strategy for many Republican candidates. In the end, a number of high-profile candidates who denied the 2020 election lost their races in 2022, leading some to argue that the American electorate had rejected this movement. National Public Radio (NPR), to choose one of many examples, ran a headline declaring “Midterm results show voters reject election denialism.”Footnote 1 Others disagree. For example, pointing to many election-denying candidates who won their races, a 538 article wrote that “election denial is alive and well.”Footnote 2 What can data tell us about the extent to which voters did or did not reject these candidates in 2022?
To provide hard data on this important question, we gather new 2022 primary and general election data on key statewide and federal offices—U.S. Senate, U.S. House, governor, secretary of state, and attorney general—and combine them with systematic data on which candidates explicitly denied the 2020 election outcome publicly. Using several statistical approaches to account for partisan differences across states and offices, we estimate that election-denying candidates for statewide offices underperformed their fellow Republicans who did not deny the 2020 election by roughly 3.2 percentage points, on average. While this difference is small by some standards, it constitutes a large enough vote-share swing to have changed important, close elections in recent cycles.Footnote 3 For example, in 2022, there were 20 statewide races in our data in which the winning candidate received less than 53% of the two-party vote—with these races clustered in the most-contested battleground states including Arizona, Georgia, Pennsylvania, and Wisconsin. As such, the estimated underperformance seems relevant for the debate over the electability of Republican candidates in the 2024 primary cycle.
Our study adds to two very recent studies of the 2022 election cycle. Jacobson (Reference Jacobson2023) analyzes the midterm congressional elections as a whole, and uses one of the measures of election denialism we also use in regressions that suggest a modest or null underperformance to election-denying candidates.Footnote 4 Bartels and Carnes (Reference Bartels and Carnes2023) studied the U.S. House and found that, among incumbents, election-denying candidates received higher average vote shares, using a measure of election denying based on votes in Congress. We expand on these studies in three key ways. First, because measuring election-denying candidates is a subjective exercise, we make sure that results are not driven by a particular data source by incorporating four different measures of election denialism that apply to both incumbents and nonincumbents. Second, we analyze all relevant statewide and federal offices, including both incumbents and nonincumbents.Footnote 5 Third, we develop analyses that are explicitly focused on estimating the gap in performance between election-denying and non-denying candidates, holding electoral factors fixed, without including any potentially mediating variables like campaign finance measures that may be posttreatment and could bias estimates.Footnote 6 To do so, we employ state fixed effects so that we can compare the gap in performance for election-denying and non-denying candidates who ran on the same ballot, and we do not control for variables like spending that could be mediators. Together, these three factors allow us to provide the only comprehensive analysis of the precise underperformance of election-denying candidates across offices and incumbency status in the 2022 election cycle.
Our results also help to advance a long-running literature on candidate positioning and electoral outcomes in American elections. Empirical studies consistently estimate an advantage to more moderate candidates (Ansolabehere, Snyder, and Stewart Reference Ansolabehere, Snyder and Stewart, III.2001; Broockman and Kalla Reference Broockman and Kalla2020; Canes-Wrone, Brady, and Cogan Reference Canes-Wrone, Brady and Cogan2002; Caughey and Warshaw Reference Caughey and Warshaw2023; Hall Reference Hall2015; Hall and Thompson Reference Hall and Thompson2018), and behavioral data suggest that swing voters remain important in determining election outcomes (Fowler et al. Reference Fowler, Hill, Lewis, Tausanovitch, Vavreck and Warshaw2022; Hill, Hopkins, and Huber Reference Hill, Hopkins and Huber2021). However, an important vein of behavioral research argues that Americans are too uninformed and/or too partisan to care much about other considerations like candidate positions (e.g., Achen and Bartels Reference Achen and Bartels2016; Campbell et al. Reference Campbell, Converse, Miller and Stokes1960). As partisanship has increased in the American electorate, the relationship between challenger moderation and electoral performance has weakened in congressional elections (Canes-Wrone and Kistner Reference Canes-Wrone and Kistner2022), and declined for both incumbents and nonincumbents in state legislative races (Handan-Nader, Myers, and Hall Reference Handan-Nader, Myers and Hall2022). How much candidate positions matter in the current political climate is therefore very much in question.
Assessing how much the overall relationship between candidate positions and election outcomes has changed in congressional elections is challenging, for two related reasons. First, our ability to measure candidate positions has become threatened by the increasing polarization of Congress itself, which is causing problems with roll-call-based measures of incumbent ideology used in many studies of elections (Duck-Mayr and Montgomery Reference Duck-Mayr and Montgomery2023; Tausanovitch and Warshaw Reference Tausanovitch and Warshaw2017). Second, to the extent that symbolic, partisan “culture war” issues have become more salient than the traditional left–right divide over economic policies, standard approaches to scaling candidates may be less useful than before.
Beyond the general election, it is also important to understand the extent to which Republican primaries preferred election-denying candidates. Perhaps surprisingly, we estimate that election-denying Republicans outperformed primary opponents by only roughly 2 percentage points.Footnote 7 Because the estimate does seem positive, however, it is at least roughly consistent with research exploring the perceived tradeoff primary voters face between voting for more-extreme candidates they prefer and voting for less-extreme candidates more likely to win the general election (e.g., Aranson and Ordeshook Reference Aranson, Ordeshook, Niemi and Weisberg1972; Brady, Han, and Pope Reference Brady, Han and Pope2007; Hill Reference Hill2015; Owen and Grofman Reference Owen and Grofman2006).
There are two important limitations to our analysis. First, our analysis only measures the gap in electoral performance of statewide Republican candidates who denied or did not deny the 2020 election. Because it compares Republican candidates to one another, it differences out any penalty that may have accrued to the Republican party as a whole because of its association with the election-denying position of former President Trump and other candidates. Second, our estimate does not reflect the causal effect of a candidate switching her position on election denialism; rather, it summarizes how much worse election-denying candidates did than non-denying candidates. If election-denying candidates differ from non-denying candidates in their same states—for example, if they are less experienced, more ideologically extreme, or otherwise less popular candidates—these other differences would contribute to our estimate as well.
EMPIRICAL APPROACHES TO ACCOUNT FOR STATE PARTISANSHIP
We collect certified statewide election returns for 2022 directly from official state websites (Malzahn and Hall Reference Malzahn and Hall2024).Footnote 8 We combine these data with information on 2020 Republican presidential vote share collected from Dave Leip’s Election Atlas (Leip Reference Leip2023). Ultimately, we analyze data for 42 states, excluding eight states that either did not have any two-party contested elections for Senate, governor, secretary of state, and attorney general (six states) or which use nonstandard election rules that lead to more than one general election candidate in one or both parties (two states).Footnote 9
Classifying Republican candidates as denying the 2020 election is partially a subjective exercise. To avoid making our own judgment calls, we rely on three external datasets of election-denying candidates. The first, from States United Democracy Center (SUDC), is a list of Republican candidates for governor, secretary of state, and state attorney general who made public statements expressing skepticism about the 2020 election. SUDC is a nonpartisan organization founded by three former government officials that focuses on issues around American democracy.Footnote 10 SUDC identifies Republican as having denied the election if, in the evaluation of the organization’s experts, they claimed former President Trump was the rightful winner of the 2022 election instead of President Biden, spread lies regarding the election to the press or on social media, called for an audit of election results after they were certified, attended “Stop the Steal” rallies, or filed litigation to overturn election results (States United Democracy Center 2022). As this dataset is the only one to evaluate candidates in primary elections, it provides us unique leverage to examine the performance of election-denying candidates in primary elections. However, it does not include U.S. Senate or House candidates.
The second is a dataset from FiveThirtyEight (538) that the news organization created by contacting every Republican nominee and asking them about the 2020 election (FiveThirtyEight 2022). A key advantage to this dataset is that they include all statewide and congressional general election candidates; however, these data do not cover primary elections. Rather than classifying candidates as denying the 2020 election, 538 lists six possible kinds of stances each candidate could take. These stances are: “fully accepted,” “accepted with reservations,” “avoided answering,” “no comment,” “raised questions,” and “fully denied.” We classify candidates in these data as denying the 2020 election only if they “fully denied” the 2020 election, though in the Supplementary Material we show that results are robust to expanding this definition.
The third is a dataset from the Washington Post (WaPo) that uses a method nearly identical to SUDC to inductively classify election deniers. Unlike SUDC, WaPo includes all statewide and congressional candidates, giving us better coverage. In addition to using these three classifications separately, we also combine them by generating a variable that classifies any Republican candidate as denying the 2020 election if they are classified as doing so by SUDC, 538, or WaPo.
A concern with these measures could be that some election-denying candidates are not detected because they do not show up saliently in news sources as an election-denying candidate. The 538 data are reassuring in this regard because they are exhaustive—they contacted every candidate under study. Our consolidated measure is also helpful because it casts as wide a net as possible across all three sources. Nevertheless, there could be candidates who at different points denied the election but were not detected by these sources. To the extent these candidates exist, it could attenuate our estimate of the underperformance to election-denying candidates by leading us to accidentally include some denying candidates in the control group of our regressions. However, given the strength of our data sources, we think this bias is likely to be small if it exists.
Focusing only on Republican statewide candidates, we run regressions of the form
$$ \begin{array}{rl}{\mathrm{Repub}\hskip0.3em \mathrm{Vote}\hskip0.3em \mathrm{Share}}_{is}=\beta {\mathrm{Deny}\hskip0.3em \mathrm{2020}}_i+\delta {X}_{is}+{\epsilon}_{is},& \end{array} $$
where the outcome is the vote share for Republican statewide candidate i in state s in the 2022 election. The variable
$ \mathrm{Deny}\hskip0.3em \mathrm{202}{0}_i $
is a binary variable indicating whether candidate i officially denied the results of the 2020 election. The variable
$ {X}_{is} $
represents a control for either state-level presidential vote share or state fixed effects to account for possible confounding where states with more election-denying candidates in 2022 are states where Republican vote shares are generally higher. Finally,
$ {\epsilon}_{is} $
is the error term which we expect to feature autocorrelation within states.
UNDERPERFORMANCE OF ELECTION-DENYING CANDIDATES IN 2022
Before presenting our estimates, it is valuable to understand the distribution of election-denying candidates across races and offices. Figure SI.2 in the Supplementary Material plots the number of denying and non-denying candidates across office and incumbency status. In all four offices, non-denying candidates outnumber denying ones by a wide margin. Using our combined measure of election denialism, we find 10 election-denying candidates in AG elections, 19 in gubernatorial, 13 in senatorial, and 10 in secretary of state races. While we find roughly equal numbers of incumbent and nonincumbent election-denying candidates for governor, we find fewer incumbent and more nonincumbent election-denying candidates for AG, senator, and especially secretary of state, where there are actually no election-denying incumbent candidates. Finally, there are a total of 20 states in which we see Republican election-denying and non-denying candidates on the same ballot.
Figure 1 explains which states featured the most underperformance by election-denying candidates, focusing on states that featured Republican candidates who did and did not deny the 2020 election—as mentioned above, this focuses on 20 states in our data. The plot shows the election-denying candidates’ average vote share in the state on the horizontal axis, and the non-election-denying candidates’ average vote share in the state on the vertical axis. Points above the 45-degree line are states where the non-denying candidates outperformed the denying candidates. As the figure shows, while many states are quite close to the 45-degree line, there are more states above and to the left of the line, indicating places where election-denying candidates underperformed.
Figure 1. Comparing 2022 Vote Shares of Statewide Republican Candidates Who Denied vs. Did Not Deny the Results of the 2020 Election and Ran in the Same State
Table 1 presents the formal estimates. The first four columns present our more-precise statistical estimates where we pool the data while controlling for state-level 2020 presidential vote share, which allows us to include more states. The second set of four columns present estimates where we instead use state fixed effects, focusing on comparisons between Republican candidates running in the same state who vary in their election denialism. This latter specification can address potential changes in state partisanship between 2020 and 2022, but at the cost of using less of the data and therefore lowering our statistical precision. Specifically, state fixed effects only allow us to estimate a penalty for states with variation in the candidate denier classification, leaving us with as few as 11 states for the SUDC measure and as many as 21 states for the WaPo measure.
Table 1. Underperformance of Election-Denying Candidates in 2022 Races for Governor, Senator, Secretary of State, and Attorney General
Note: Sample is Republican candidates for attorney general, secretary of state, governor, and U.S. Senate. Robust standard errors clustered by state in parentheses. Columns labeled States United use election denier classifications from SUDC, which do not include Senate or House races. Columns labeled 538 use election denier classifications from FiveThirtyEight. Rows labeled WaPo use election denier classifications from the Washington Post. Columns 1–4 include controls for 2020 presidential vote share.
Columns 1 and 5 use the SUDC classification of election-denying candidates (which do not include Senate races), columns 2 and 6 use the 538 classification, columns 3 and 7 use the WaPo classification, and columns 4 and 8 use the combined measure we created, described above.
As the results show, we find a relatively consistent underperformance for election-denying Republicans compared to other Republicans, ranging from as large as 4.5 percentage points (column 5) to 2.0 percentage points (column 6). Our most precise estimates (in terms of lowest standard error) are in columns 2–4, across which we estimate at least a 2.6 percentage point underperformance using the 538 classification. Column 8 remains our ex ante preferred specification, where we use the combined measure and state fixed effects and estimate an underperformance of 3.2 percentage points relative to other Republicans.
LACK OF UNDERPERFORMANCE FOR ELECTION-DENYING CANDIDATES IN THE U.S. HOUSE
We can also perform a similar analysis for the U.S. House. Because we do not have access to statewide offices’ vote shares by congressional district, we cannot pursue the fixed-effects strategy from above, where we compute the difference between election-denying and non-denying candidates within the same constituency. But we can control for presidential vote share at the House district level, paralleling the other strategy we took at the state level previously. Table 2 presents the results. As the table shows, we find no evidence for underperformance by election-denying candidates at the House level. In our preferred specification in column 3, we estimate a 0 percentage point difference, with a 95% confidence interval lower-bound of −0.6 percentage points. Though we use a different measure of denialism, this is broadly consistent with the results in Bartels and Carnes (Reference Bartels and Carnes2023).
Table 2. Neutral Performance of Election-Denying Candidates in 2022 Races for House
Note: Sample is Republican candidates for U.S. House. Robust standard errors clustered by state in parentheses. Columns labeled 538 use election denier classifications from FiveThirtyEight. Columns labeled WaPo use election denier classifications from the Washington Post. Columns labeled combined classify candidates as denying the election if either WaPo or 538 classify them as an election denier. Columns 1–3 include controls for 2020 presidential vote share.
Statewide races differ from House races in ways relevant to an election denial penalty. To start, statewide offices have more direct and concentrated power over election administration and certification, which makes their position on this issue more consequential. Secretaries of State directly administer state elections. Attorney generals play key roles in any state-level election litigation. Governors can block or sign state legislation to fund elections or change election laws. Additionally, with smaller districts that are often much more partisan, House candidates who deny the election may have been able to do so while alienating a smaller proportion of their voters. Moreover, in safe Republican districts where primary elections matter more than general elections, incumbents who accepted the results of the 2020 election may have had more to fear. Indeed, Representatives Anthony Gonzalez (R-OH), Adam Kinzinger (R-IL), and John Katko (R-NY) all opted to retire rather than face a potential election-denying opponent in their next primary, indicating the expectation of a penalty. As another example, Rep. Tom Rice (R-SC) lost a primary challenge to Rep. Russell Fry who fully denied the 2020 election.
In the Supplementary Material, we present estimates for the House, Senate, and statewide races that consider incumbency status. Interestingly, we find variation across offices. In governor and AG races, the underperformance of election-denying candidates appears to be larger for nonincumbents than for incumbents, while the opposite appears to be the case for the House and Senate.
POSSIBLE ADVANTAGE OF ELECTION-DENYING CANDIDATES IN PRIMARY ELECTIONS
In the Supplementary Material, we also explore whether election-denying candidates for secretary of state, governor, and attorney general are advantaged in primary elections. Perhaps surprisingly, we imprecisely estimate only a small advantage of 0.019 with a standard error of 0.031. This advantage is consistent with the notion that there is a tradeoff between primary elections that favor more-extreme candidates and general elections that favor more-moderate candidates (Hall Reference Hall2015), but the imprecision in our estimates makes it difficult to draw strong conclusions.
CONCLUSION
Understanding the degree to which voters in 2022 rejected election-denying candidates is important for understanding how American democracy might function in future election cycles and for assessing how the potential links between candidate positions and electoral outcomes are changing in congressional elections. Although we know that many election-denying candidates lost key state-level races in 2022, we lack a precise empirical sense of how strongly the American electorate punished candidates who espoused these views. The purpose of this study is to put together the data necessary to quantify this underperformance systematically. We find that election-denying candidates underperformed in 2022 by a margin substantial enough to suggest that it could tip close elections in the future. On the other hand, the underperformance of election-denying candidates that we document is also small enough to suggest that many voters were willing to continue supporting Republican candidates even if they denied the results of the 2020 election.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055424001084.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the American Political Science Review Dataverse: https://doi.org/10.7910/DVN/JPKJSJ.
ACKNOWLEDGMENTS
For comments and helpful suggestions, the authors thank Justin Grimmer and Kelly Rader as well as participants of the EOLDN-Hoover Conference.
FUNDING STATEMENT
This research was supported in part by funding from the States United Democracy Center.
CONFLICT OF INTEREST
The authors declare no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The authors affirm this research did not involve human participants.
Open access
Comments
No Comments have been published for this article.