The paper
Tadeas Cely, Marc S. Jacob, and Sean J. Westwood (2026). Levels of Office and Voter Accountability for Democratic Norm Violations. Public Opinion Quarterly 90(2), 310–335.
doi:10.1093/poq/nfag004 (opens in a new tab) · Replication data: doi:10.7910/DVN/87SZYD (opens in a new tab)
Main text
p. 310, introduction
Published
At the state level, 25 election denialists currently hold office, and in 2024, an additional 15 denialist candidates ran for statewide office (Grimmer et al. 2023).
Correction
At the state level, 25 election denialists currently hold office, and in 2024, an additional 15 denialist candidates ran for statewide office.
Basis The cited source, Grimmer et al. (2023), “Who Are the Election Skeptics? Evidence from the 2022 Midterm Elections” (doi:10.31219/osf.io/pe2zg), is a study of voters’ election skepticism. It contains no counts of officeholders or candidates, and a 2023 preprint cannot report 2024 candidacies. The citation is withdrawn; the article gives no other source for the two counts.
p. 319, note to Table 1
Published
The violations of circumventing the Congress and gerrymandering were selected based on Graham and Svolik (2020).
Correction
The violations of circumventing the Congress and reducing the number of voting places in areas dominated by the out-party were selected based on Graham and Svolik (2020).
Basis Table 1 lists the Electoral fairness violation as “There should be fewer voting places in areas dominated by out-party”; no gerrymandering violation appears in either study.
p. 320, Benchmarking Effect Sizes
Published
The highest defection rate in our data, 51 percent in Study 1, is well within the range of 36 percent to 60 percent reported in previous experiments by Graham and Svolik (2020), Mummolo, Peterson, and Westwood (2021), Peterson (2017), and Krishnarajan (2023).
Correction
The highest defection rate in our data, 62.5 percent in Study 1 (two norm violations), is above the range of 36 percent to 60 percent reported in previous experiments by Graham and Svolik (2020), Mummolo, Peterson, and Westwood (2021), Peterson (2017), and Krishnarajan (2023). Among Democrats facing a local candidate with two violations, the rate is 68.1 percent.
Basis The printed 51 percent is the two-violation coefficient in the Figure 2 model (0.513), the increase in defection relative to a norm-compliant local candidate, not a defection rate. The article itself reports a 63 percent defection rate for two violations (p. 322).
p. 321, Study 1 Results
Published
In the initial study, conducted between February 19 and March 3, 2024, we sampled 2,000 respondents (before exclusions)
Correction
In the initial study, conducted between February 16 and February 28, 2024, we sampled 2,000 respondents (before exclusions)
Basis Earliest and latest interview start dates in the Study 1 data.
p. 323, Study 1 levels of office
Published
Overall, we found no significant differences based on different electoral races.
Correction
Overall, defection from norm-violating copartisans was slightly higher in local races than in state and national races, the opposite of our hypothesis. With one violation, defection from local candidates was 2.6 percentage points higher than from state candidates (p = 0.040) and 1.9 points higher than from national candidates (p = 0.13); with two violations, the differences were 2.5 points (p = 0.016) and 2.1 points (p = 0.048). The preregistered test of H1, which regresses defection on the level of office across all tasks, gives the same pattern: defection was 2.4 points lower in state races (p = 0.004) and 2.1 points lower in national races (p = 0.012) than in local races.
Basis Contrasts from the article’s Figure 2 model (analysis.R, model M3: defection on level of office, number of violations and their interaction, standard errors clustered by respondent). The H1 regression follows the Study 1 preregistration (osf.io/ekzhp, section 5) with the same clustering. The differences run against H1, so the rejection of H1 stands; Study 2 shows no level differences.
p. 324, Study 1 by party
Published
The same findings hold when we break down the effects for Democrats and Republicans. The only exception is that Republicans defect less from norm-violating local candidates than Democrats. The difference in defection rates is about 6 percentage points, and it is statistically significant (p < 0.05).
Correction
The same findings hold when we break down the effects for Democrats and Republicans, except that Republicans defect less than Democrats from norm-violating candidates at every level of office. With one violation, Republicans’ defection rate is 8.9 percentage points lower than Democrats’ for local candidates, 6.5 points lower for state candidates and 6.9 points lower for national candidates; with two violations, the gaps are 10.4, 8.9 and 7.7 points (all p ≤ 0.009). The partisan gaps do not differ significantly across levels of office (p ≥ 0.21).
Basis The printed 6 points is the One Violation × Republican term in the Supplementary Material section 9 table (p = 0.097), which uses local races as the reference category and therefore compares the parties only in local races. Computed for each level from the same model, the partisan gap in the effect of one violation is 6.2 points (local), 4.0 (state) and 11.4 (national, p = 0.002); none of the three-way terms is significant (p ≥ 0.29).
p. 325, Study 2
Published
The second study was conducted between April 29 and May 13, 2024
Correction
The second study was conducted between April 26 and May 23, 2024
Basis Earliest and latest interview start dates among respondents in the three survey weeks the analysis keeps (transformation2.R drops weeks 86 and 89).
p. 325, Table 2
Published
Table 2 (“Factual manipulation check”), percent correct by treatment: “National” 63.7 (SE = 0.96), “State” 63.2 (SE = 1.36), “Local” 72.4 (SE = 1.34); “N (respondents)” 2,511.
Correction
Corrected Table 2. Each respondent answered the check once; the standard errors are 1.70 (national), 1.67 (state) and 1.60 (local), from the 2,424 respondents who answered it. The local share remains significantly higher than the national share (8.7 points, p < 0.001).
Corrected Table 2. Factual manipulation check, one observation per respondent; standard errors of each percentage. Sample Percent correct All treatments 66.4 By treatment National 63.7 (SE = 1.70) State 63.2 (SE = 1.67) Local 72.4 (SE = 1.60) N (respondents) 2,424 Basis The printed standard errors come from a regression on task-level data, in which each respondent appears three times: 0.96 is the intercept (national) standard error, and 1.36 and 1.34 are the standard errors of the local and state differences, printed against the state and local rows. The archived analysis2.R stops before producing the table, because it uses checkB before creating it and calls ggpredict on an undefined object. The overall share in the data is 66.4 percent; the article prints 66.3.
p. 327, Study 2 by party
Published
Whereas 33.7 percent of Democrats (CI95 = 31.3 percent–36 percent) defect from candidates who violate norms of electoral fairness (i.e., removing polling places in areas dominated by the out-party), only 19 percent of Republicans (CI95 = 17 percent–21 percent) do so. Similarly, 39.4 percent of Democrats (CI95 = 37 percent–41.8 percent) defect from candidates who challenge electoral integrity (i.e., claiming that elections are fraudulent), while only 22.3 percent of Republicans (CI95 = 19.8 percent–24.7 percent) respond in the same way.
Correction
Whereas 33.7 percent of Democrats (CI95 = 29.6–37.8 percent) defect from candidates who violate norms of electoral fairness (i.e., removing polling places in areas dominated by the out-party), only 19.0 percent of Republicans (CI95 = 15.1–23.0 percent) do so. Similarly, 39.4 percent of Democrats (CI95 = 35.2–43.6 percent) defect from candidates who challenge electoral integrity (i.e., claiming that elections are fraudulent), while only 22.3 percent of Republicans (CI95 = 18.2–26.3 percent) respond in the same way.
Basis Confidence intervals clustered by respondent, from the Figure 3 model with party interactions. The four estimates rest on 573, 410, 564 and 395 violation choices. The printed intervals are narrower than the pooled interval for electoral fairness (24.6–30.5 percent, p. 326), although each subgroup holds only part of the pooled data. The partisan differences remain significant.
p. 328, Defections at different levels of government
Published
In addition to this model, we condition these effects on a set of preregistered covariates (with and without adjustments for multiple hypothesis testing), showing no significant differences either.
Correction
In addition to this model, Supplementary Material sections 6–9 report how preregistered covariates relate to defections in Study 2. Only the party breakdown in section 9 conditions the level-of-office effects, and none of these analyses is adjusted for multiple hypothesis testing.
Basis The article and Supplementary Material report no Study 2 analysis that conditions the level-of-office effects on the preregistered covariates with an adjustment for multiple testing, and the replication archive contains no code for one.
p. 331, Appendix A (Dates of Data Collection; Sample Sizes)
Published
Data collection for Study 1 took place between February 19 and March 3, 2024, and for Study 2 between April 29 and May 19, 2024. [...] Study 1 included 2,000 respondents (after exclusions), and Study 2 included 3,000 respondents.
Correction
Data collection for Study 1 took place between February 16 and February 28, 2024, and for Study 2 between April 26 and May 23, 2024. [...] Study 1 included 2,000 respondents before exclusions (1,699 partisans after exclusions), and Study 2 included 3,000 respondents.
Basis See the corrections to pp. 321 and 325. The main text (p. 321) describes the 2,000 Study 1 respondents as the sample before exclusions.
Supplementary information
Supplementary Material section 3 (Priming Importance of Democracy Intervention)
Published
Therefore, we do not further condition on this treatment as preregistered as exploratory hypothesis.
Correction
The Study 2 preregistration lists the prime as a primary hypothesis (H3): defection from norm-violating copartisans increases when respondents are primed about the threat to democracy, and more so at higher levels of office. Neither part is supported. The prime changes the effect of a norm violation on defection by 4.9 percentage points in federal races (p = 0.10), 2.3 points in state races and −1.7 points in local races. Relative to federal races, the violation × prime × level terms are −0.067 for local races (p = 0.11) and −0.026 for state races (p = 0.55). The estimates rise with the level of office, as H3 predicts, but the differences are not significant.
Basis Preregistration osf.io/6r3js, section 2.1 (primary hypotheses) and section 5 (the interaction of the prime and the level of government). Estimated as in the table in this section: defection on violation, prime and level of office, with standard errors clustered by respondent. The main text (p. 320) refers to this section for the priming results across electoral scenarios.
Supplementary Material section 8, table Specific Norm Support and Party Defections (Study 2)
Published
Norm Violation × Support for Norm 0.07 [...] Observations 7267
Correction
Corrected table. With each norm matched to its own support item, the interaction is 0.077 (published 0.070), and the conclusion holds.
Corrected table, Specific Norm Support and Party Defections (Study 2): Electoral Integrity matched to the election-denial item and Division of Powers to the executive-order item. 95% confidence intervals clustered by respondent. Predictors Estimates CI p (Intercept) 0.25 0.21 – 0.29 <0.001 Norm Violation −0.12 −0.17 – −0.06 <0.001 Support for Norm at Stake −0.04 −0.05 – −0.03 <0.001 Norm Violation × Support for Norm 0.08 0.06 – 0.09 <0.001 Observations 7,267 R2 / R2 adjusted 0.047 / 0.046 Basis The archived transformation2.R pairs the Electoral Integrity violation with the polling-station item and the Division of Powers violation with the election-denial item, and never uses the executive-order item, although this section defines the election-denial and executive-order items separately. The mismatch affects 2,971 of the 7,267 choices. The section’s statements about supporters and opponents of election denial rest on the mismatched items.
Supplementary Material section 9 (Democrats and Republicans)
Published
The only statistically significant difference could be found among norm-violating local candidates, where Republicans defect less than Democrats. [...] Focusing on the second study, we found the same divergence between Democrats and Republicans in defections from norm-violating local candidates. Whereas Democrats defect in 29% of cases, Republicans only in 22% of cases.
Correction
Republicans defect less than Democrats from norm-violating candidates at every level of government. Because the tables use local races as the reference category, their Republican interaction terms compare the parties only in local races (see the correction to p. 324). [...] The second study shows the same partisan gap at every level: Democrats defect from norm-violating candidates in 29% of local, 27% of state and 29% of federal cases, and Republicans in 21%, 21% and 21%. The partisan gap in the effect of a violation is 7.1 points (local), 6.2 (state) and 7.1 (federal), all with p ≤ 0.039.
Basis Study 2 defection rates are shares of violation choices by party and level of office; the data give 21% for Republicans in local races, where the section prints 22%. The partisan gaps in the effect of a violation come from the section’s Study 2 model with standard errors clustered by respondent.
Supplementary Material section 11 (Benchmarking Treatment Effects)
Published
The 51% maximum defection rate from our Study 1 falls squarely within this range, indicating that the magnitude of our core result is consistent with those reported in the existing literature.
Correction
The maximum defection rate in our Study 1, 62.5% (two norm violations), lies above this range; among Democrats facing a local candidate with two violations, it is 68.1%.
Basis See the correction to p. 320.