U.S. democracy before the 2026 midterms

Bright Line Watch October 2026 survey

After an unprecedented decline in expert evaluations of U.S. democracy in the first months of Donald Trump’s second administration, ratings stabilized and then partially rebounded in late 2025 and 2026. Democrats are expected to make substantial gains in the upcoming midterm elections, which would likely cause Congress to resist Trump more forcefully. However, the prospect of Republican losses has heightened concerns that the federal government will interfere with the election or that Trump will again claim it was stolen. It is therefore crucial to measure how experts and the public assess the state of American democracy and the integrity of our election system in this fraught pre-election moment.

The results reported below were collected in Bright Line Watch’s 30th survey of political scientists (hereafter referred to as experts; 492 responses collected October 3–17, 2026), which was paired with a representative public sample of Americans (2,750 responses collected October 6–14, 2026). We collected expert survey data in partnership with Luth Research, which managed survey invitations to preserve participant anonymity. Panel management and data collection for the public surveys were conducted by YouGov. All public survey results reported below incorporate YouGov’s post-stratification weights.

Executive summary

Expert ratings of U.S. democracy are up slightly from July 2026. Though the difference is not statistically significant, the pattern is consistent with an ongoing rebound from the observed low in April 2025. However, experts still rate the U.S. closer to a hypothetical illiberal democracy than to a strong democracy. Public ratings of democracy are also statistically unchanged, though ratings among Democrats fell slightly.
Experts project no change in their ratings of U.S. democracy by 2027, but expect improvement by 2032. The public forecasts a slight improvement from today by 2027 but no further change in 2032.
Public confidence in the 2026 election is greatest among Democrats, but it has increased since July among supporters of both parties. Experts are most confident overall.
Republicans are more likely to be confident that eligible voters can cast a ballot than are Democrats; but less likely to be confident that election administrators will count fairly and accurately. Experts are more confident in election administrators than are supporters of either party.
Republicans are much more likely than Democrats to think that voter and election fraud, of various kinds, will happen a few thousand times or more; these perceptions are mostly similar to those in October 2024. Estimates of the frequency of all types of election fraud are far lower among experts than among Republicans.
Between 42% and 51% of experts think we will see a few thousand cases or more of each of the following: eligible voters being prevented from voting by election officials, eligible voters being deterred from voting by poll watchers, and eligible voters being deterred from voting by law enforcement, immigration, or military personnel.
Experts and leading AI forecasters with a record of matching or exceeding the accuracy of top human forecasters are extraordinarily confident that the midterm elections will happen on time and give low probabilities to many potentially disruptive election-related events. The unadjusted median forecast probability of a free and fair election is 75% for experts; we do not report AI forecasts for this question.
Among potentially disruptive election-related events, expert and AI forecasters think the most likely event is that five or more of the best-known Republicans will publicly allege that fraud or other illegal conduct affected the election results. Forecasters also give a meaningful chance to the Trump administration or a GOP governor deploying the National Guard, military personnel, or federal or state law enforcement officers at or near polling places, and to Trump or an appointee asking the Guard, military personnel, or other federal agents to seize voting equipment or ballots. When asked to assume either of the last two scenarios, experts give lower assessments of the likelihood of free and fair elections; both are among the six scenarios with the lowest median assessments.
The gap between Democrats and Republicans on the likelihood of free and fair elections is quite small, but partisans divide widely on whether prominent officials from each party will publicly allege that fraud or other illegal conduct affected the election results, whether the Trump administration will request voting equipment or ballot seizures, whether at least one candidate will win only because of voter fraud, and on whether the federal government will restrict in-person voting hours or locations.
When asked to rate U.S. democracy retrospectively, experts rate democratic performance in previous years five to ten points more favorably than they did at the time.
When asked whether political scientists should take an active role in debates about democracy, experts are most favorable, followed by Democrats; Republicans are much more skeptical. The public is also polarized over whether political scientists should communicate with the public about threats to democracy.
Exposure to news coverage stating that experts rate U.S. democracy as diminished decreased ratings of democracy among Democrats and the public as a whole. It also increased support for political scientists playing an active role in debates about democracy among Democrats. We found no measurable effect among Republicans in either case.

Ratings of U.S. democracy

A core goal of Bright Line Watch is to track broad assessments of U.S. democracy from both experts and the public. Since our first survey in 2017, we’ve asked experts (and starting from our third survey, the public) to rate U.S. democracy on a 0–100 scale. The figure below plots mean ratings from 2017 to today (October 2026) along with projected ratings of the U.S. in 2027 and 2032.

Expert and public ratings of U.S. democracy: 2017–2032

Current ratings and expectations for 2027 and 2032

Dotted lines represent forecasts; bars show 95% confidence intervals. Public estimates are weighted; expert estimates are unweighted. Gray bands denote Trump administrations; the intervening period denotes Biden. Public includes Democrats, Republicans, and independents; party groups shown separately are subsets of Public. Source: Bright Line Watch, 2017–2026.

We find that experts rate U.S. democracy at 61 on average, up slightly from the average of 60 observed in our July survey, though the difference is not statistically significant. The trend is consistent with the pattern of an ongoing rebound from the observed low of 53 in April 2025. The expert rating of 61 remains below the lowest rating observed under Biden (66, in November 2021) and is statistically indistinguishable from Trump’s first-term low (61, in October 2020). Compared to April 2025 expert ratings of other countries, the U.S. is closest to Poland (63) and Mexico (60) and remains well below Western democracies such as the United Kingdom (83) and Canada (88). Despite the persistent signs of rebound, expert ratings place the U.S. closer to a hypothetical illiberal democracy (which experts previously rated at 47) than to a strong democracy (93).

Public ratings are also statistically unchanged at 51, which is just narrowly above the lowest point of Trump II (49 in April 2025) and slightly below the previous rating (52). However, this pattern of aggregate stability masks differences among partisans. As with experts, ratings among Democrats dropped to their lowest point during the first year of Trump’s second term (42 in January 2026) and have since then shown signs of tempered recovery. Unlike experts, however, Democrats’ ratings fell slightly since our last survey, declining from 50 to 47. Republican ratings were statistically unchanged and are statistically indistinguishable from their first ratings in Trump’s second term (59 in February 2025).

We also ask experts and the public to project the state of democracy in 2027 and 2032. Experts project no change from today to 2027. Their expectations for 2027 are rosier now, however, compared to our prior measure in February–March 2026 (56 to 61). Looking to 2032, experts forecast a further rebound to 65. This projected improved rating is statistically indistinguishable from the last ratings given during the Biden presidency (November 2024, 67) and does not differ measurably since we last asked in February–March 2026.

The public overall projects slight improvements from today to 2027 (51 to 53) but no further change in 2032. The forecasted rise in 2027 is driven by Democrats who expect an increase from 47 to 52; Republicans expect no change. Despite this disagreement, partisans end up converging on their 2032 ratings. Republicans project no change from now to 2032, while Democrats project a sharp turnaround by 10 points, matching Republicans at 57.

Confidence in U.S. elections

With the midterm election approaching, we again surveyed the American public about their confidence that votes nationwide will be counted as voters intended.

Confidence that votes nationwide are counted as intended

Before and after each election; labels mark the last pre-election survey

Public and partisan estimates are weighted; expert estimates are unweighted. Party groups include leaners. Bars show 95% confidence intervals. Dashed lines mark federal elections. Public includes Democrats, Republicans, and independents; party groups shown separately are subsets of Public. Source: Bright Line Watch, 2020–2026.

Democrats are more confident overall, but as the figure above shows, confidence in the 2026 election has increased since July not only among Democrats, whose rate of confidence increased from 66% to 73%, but among Republicans, for whom confidence rose from 58% to 67%. Experts are most likely to be confident (85%), though not measurably different from July (82%). In recent history, the pattern of public pre-election confidence in 2026 is closest to 2020 — the gap in confidence between the parties is small and fewer than two in three express confidence in the vote overall.

We also asked the public and experts whether eligible voters would be able to cast a ballot and whether election administrators would count the vote fairly and accurately. The figure below presents results from these questions and compares them to data from surveys conducted of the public before past elections (when available).

Voter access and vote counting before each election

Share somewhat or very confident; last pre-election survey in each cycle

Public and partisan estimates are weighted; expert estimates are unweighted. Party groups include leaners. Bars show 95% confidence intervals. Shares combine somewhat confident and very confident responses. Public includes Democrats, Republicans, and independents; party groups shown separately are subsets of Public. Source: Bright Line Watch, 2020, 2022, 2024, and 2026.

Republicans are more likely to be confident that eligible voters can cast a ballot than are Democrats (85% versus 70%); the reverse is true for election administrators counting fairly and accurately (80% of Democrats versus 69% for Republicans). Experts are more likely to be confident in election administrators than either group (90%) but statistically indistinguishable from Democrats in their confidence in ballot access for eligible voters (68%). The voter access question again most resembles 2020, the most recent federal election conducted with Trump as president. For election administrators, the pattern of convergence between parties from 2024 to 2026 resembles what we see for election confidence above.

In recent years, public confidence in elections has been shaken by false or unsupported claims about widespread voter and election fraud, which were the core of President Trump’s effort to overturn his defeat in the 2020 election. We therefore surveyed experts and the public to see how many people endorse false or unsupported claims about widespread fraud as well as how many have specific fears about this election. The graph specifically plots the share who say each type of fraud will occur a few thousand times or more.

Expert and public beliefs about election fraud

Combines respondents asked about ‘a midterm election’ and about ‘the midterm election this November.’

Public estimates are weighted; expert estimates are unweighted. Bars show 95% confidence intervals. Combines the typical-midterm and this-November question versions. Party groups include leaners. Response options: “Never,” “Fewer than a dozen,” “A few dozen,” “A few hundred,” “A few thousand,” and “A million or more.” The plotted share combines “A few thousand” and “A million or more.” These are reported beliefs, not measured fraud. Gray connectors show Republican-Democratic differences. Public includes Democrats, Republicans, and independents; party groups shown separately are subsets of Public. Public survey: October 6–14, 2026; expert survey: October 3–17, 2026. Source: Bright Line Watch, October 2026.

As the figure highlights, Republicans are far more likely to think that a few thousand cases or more of each type of fraud will take place (52–67%) compared to Democrats (10–19%). Experts are also more skeptical than Republicans for each type (5–10%).

To determine how perceptions of this election might vary, we randomized whether the fraud questions reported above asked about the prevalence of each type of fraud in “a midterm election” (a general reference to elections) versus “the midterm elections this November” (a specific election; this phrasing is most consistent with prior wording). Across experts, the public, and partisan groups, we found only a handful of significant wording differences after adjusting for multiple comparisons (three of 24 fraud comparisons: experts more often expected ballot stealing or tampering, 17% versus 3%, and voting-machine manipulation, 8% versus 2%; Democrats less often expected impersonation, 7% versus 16%, with this November listed first). We report pooled results above to summarize differences between audiences. Separate results for each wording appear in the Appendix.

The figure below compares the expected pre-election prevalence of each type of fraud in our survey conducted before the 2024 presidential election with what we found in this survey before the 2026 midterm election (using only data from the “this November” phrasing to align with the 2024 survey).

Change in expected election fraud, 2024–2026

Open circles: September/October 2024. Arrowheads: October 2026. Arrows show change; confidence intervals are not displayed. Change labels appear only when BH-adjusted p < .05; nonsignificant arrows are faded. BH adjustment covers six year-to-year fraud contrasts within each audience; tests are unpaired. Public and party estimates are survey-weighted; party groups include leaners. Public includes Democrats, Republicans, and independents; the party groups in the right panel are subsets of Public. Both surveys used the same response scale. The 2024 questions concerned the upcoming presidential election; the 2026 questions concerned the midterm election this November, excluding the typical-midterm version. Source: Bright Line Watch, September/October 2024 and October 2026.

As the figure shows, expectations about these election fraud cases are mostly similar to those in the October 2024 pre-election survey except for two cases among Democrats, who were less likely to expect at least a few thousand cases of people voting who are not U.S. citizens (15% to 10%) and individuals pretending to be someone else when voting (13% to 7%).

Using the same response scale of expected prevalence, we also asked experts and the public about how often potential barriers to voting prevent eligible voters and ineligible voters from participating.

Expert and public expectations about voting barriers

Measured separately for “a midterm election” vs. “the midterm election this November”

Public estimates are weighted; expert estimates are unweighted. Bars show 95% confidence intervals. Circles: a typical midterm; triangles: this November. Typical-midterm marks fade when the unpaired difference is not significant after BH adjustment (adjusted p >= .05). BH adjustment covers 12 fraud/barrier wording contrasts per audience. Confidence intervals are pointwise. Public includes Democrats, Republicans, and independents. Public survey: October 6–14, 2026; expert survey: October 3–17, 2026. Source: Bright Line Watch, October 2026.

The distinction between asking about “a midterm election” and “the midterm elections this November” was greater for barriers to voting. Most notably, between 42% and 51% of experts think we will see a few thousand cases or more of eligible voters being prevented or deterred from voting in this election by each of the following: election officials, poll watchers near polling places, or the presence of law enforcement, immigration, or military personnel near the polls. Fewer expect these incidents to happen so often in a typical midterm. In addition, 15% expect law enforcement, immigration or military personnel to deter a few thousand or more ineligible voters in this election, compared to 4% in a typical midterm. The public was also more likely to expect these barriers to have an impact in the midterm election this November than in a typical midterm, with increases of six to nine percentage points. (See Appendix for results disaggregated by party. As expected, Republicans were more likely to expect ineligible voters to be deterred a few thousand times or more for each barrier and Democrats were more likely to expect the same for eligible voters.)

Finally, we measured public and expert perceptions of whether the U.S. is mostly or fully meeting various democratic standards. The graph below presents how those perceptions have changed over time in pre-election polls dating back to 2018.

Election standards before each federal election

Share saying the United States mostly or fully meets each standard

Public and partisan estimates are weighted; expert estimates are unweighted. Party groups include leaners. Bars show 95% confidence intervals. Shares include unsure respondents in the denominator. Source: Bright Line Watch, 2018–2026.

Expert evaluations and those of Republicans in the public sample are statistically unchanged from 2024 after adjusting for multiple comparisons, except for an increase among Republicans in the perception that votes have equal impact (from 42% to 56%). By comparison, Democrats are less likely to say American democracy is meeting the standard in question for six of the seven standards that we measured.

Forecasting election threats

Many concerns have been raised about interference in the 2026 elections or other kinds of events that might interfere with the process. We asked the public, experts, and leading AI forecasters (FutureSearch, Laertes, and Preseen; each with a record of matching or exceeding the accuracy of top human forecasters) to estimate the likelihood of 14 possible objectively defined events, as well as the overall likelihood that the midterms will be free and fair and that a candidate would win only because of voter fraud (we do not report AI forecasts for these last two questions).

Following past practice, we adjust expert event forecast medians for excess pessimism using a model calibrated on 76 resolved questions. We do not adjust the forecast of free and fair elections or the conditional forecasts shown in the main report. For the probability of midterm elections being held on time, we calibrate the complementary negative event and then convert back. Public and AI forecasts are not adjusted. We also show raw expert event forecasts for transparency. Conditional forecasts and the free-and-fair baseline remain unadjusted throughout the report.

Forecasts of 2026 election-related events: Expert and AI forecasters

Raw and pessimism-adjusted expert medians alongside AI forecasts

Expert medians are unweighted. Open green circles: raw expert medians; green diamonds: pessimism-adjusted expert medians. Free and fair elections are shown only as an unadjusted expert median; no adjustment is applied to this question. AI forecasts are medians across available systems: FutureSearch October 12; Preseen October 5; Laertes October 20, 2026. No AI forecast is reported for free and fair elections or a candidate winning only because of fraud. Source: Bright Line Watch, October 3–17, 2026.

Experts and AI forecasters are highly confident that the midterm elections will happen on time and give very low probabilities to a series of disruptive events, including a federal limit or ban on counting ballots using machines. The unadjusted expert median forecast of free and fair elections is 75% (we only apply the pessimism adjustment to events).

Among the set of 14 potentially disruptive events considered, forecasters think the chances are greatest that five or more of the 25 best-known Republicans will publicly allege that fraud or other illegal conduct affected the election results (expert median: 75%; calibrated experts: 53%; AIs: 60%); that the Trump administration or a Republican governor will deploy National Guard members, military personnel, or federal or state law enforcement officers at or near polling places (expert median: 50%; calibrated experts: 14%; AIs: 38%); and that Trump or an appointee will ask the National Guard, military personnel, or other federal agents to seize voting equipment or ballots (expert median: 50%; calibrated experts: 14%; AIs: 40%).

When we disaggregate the AI forecasts and compare them with the calibrated expert forecasts on the 14 events with objective outcomes, we find that, after adjustment for past pessimism, experts are less pessimistic than all three AI systems on 10 of the 14 events and less pessimistic than the median AI forecast on all 14.

Expert, FutureSearch, Preseen, and Laertes forecasts

Pessimism-adjusted expert medians and AI forecasts; 14 common events

Expert medians are unweighted and pessimism-adjusted. AI forecasts: FutureSearch October 12; Preseen October 5; Laertes October 20, 2026. Only events forecast by all four sources are shown. Gray segments span the displayed forecasts. Source: Bright Line Watch, October 3–17, 2026.

Divergence among AI forecasters is greatest on three forecasts. The greatest by far is on whether Trump or an appointee will ask the National Guard, military personnel, or other federal agents to seize voting equipment or ballots before March 31, 2027, which FutureSearch estimated as having a 26% chance of taking place (citing “robust institutional resistance and the likelihood that post-election rhetoric will consist of general fraud claims rather than explicit orders to take possession of equipment”) versus 40% for Preseen and 63% for Laertes (which emphasizes the chance of such an action taking place later in the resolution period). FutureSearch is also more skeptical on whether five or more well-known Republicans or Democrats will publicly allege that fraud or other illegal conduct affected the election results (41% and 10%, respectively) than is Preseen (60%, 27%) or Laertes (66%, 30%). The rationales provided by the AI forecasters for these predictions are available here.

Forecasts by the public tend to be closer to the middle on these questions than those from experts or AI forecasters.

Forecasts of 2026 election-related events: Partisans and the public

Points show weighted median probabilities. Party groups include leaners. Public includes Democrats, Republicans, and independents; party groups shown separately are subsets of Public. Source: Bright Line Watch, October 6–14, 2026.

Interestingly, the partisan gap on the likelihood of free and fair elections is quite small (median forecasts of 68% for Democrats versus 63% for Republicans). The partisan gaps are twenty percentage points or more on whether prominent Democrats or Republicans will publicly allege that the election results were affected by fraud or other illegal conduct, whether the Trump administration will request voting equipment or ballot seizures, whether the federal government will restrict the hours or locations of in-person voting, and whether at least one candidate will win only because of voter fraud.

We also asked experts to judge the likelihood of free and fair elections under hypothetical scenarios. These responses describe experts’ subjective judgments, not estimates of the causal effects of those events. The figure shows differences between group medians using all available answers at each endpoint, not paired individual changes. The Appendix shows unadjusted public and partisan conditional median differences.

Experts’ judgments of election fairness under hypothetical scenarios

Difference between conditional and baseline median assessments

Expert medians are unweighted. These are unadjusted, subjective probability judgments under hypothetical scenarios; the baseline median assessment is 75%. Points show differences between group medians, using all available answers at each endpoint, not paired individual changes. Differences describe judgments, not estimated causal effects. No confidence intervals or significance indicators are shown. Source: Bright Line Watch, October 3–17, 2026.

Their unadjusted median assessment was 75% without an assumed event. Three events have conditional medians of 50%, 25 percentage points below baseline: a federal in-person voter ID requirement, an AP-declared winner having the result overturned, and a proof-of-citizenship registration law. Seven events have medians of 30–40%, 35–45 points below baseline, including deployment near polling places (35%, or ‑40 points). The lowest conditional medians are for a request to seize voting equipment or ballots (20%, or ‑55 points), a candidate winning only because of fraud (20%, or ‑55 points), and a ban on state mail-in ballots (21%, or ‑54 points). These are descriptive differences between median subjective judgments, not estimated causal effects or tests of statistical significance.

Retrospective evaluations of democracy

In our July survey, we asked our public sample to rate U.S. democracy at several points in the 2017–2024 period. We found that Democratic ratings of the 2017–2021 period were consistently more favorable than the ones that Democrats provided at the time (unlike among Republicans). We therefore measured evaluations of democracy for that period among experts in this survey.

Expert democracy ratings: recalled versus rated at the time

Expert means are unweighted. Retrospective ratings: October 3–17, 2026. Historical benchmarks average ratings collected during each year. Shading shows the difference between recalled and contemporaneous mean ratings. Source: Bright Line Watch, 2017–2026.

In each case, experts’ mean ratings of democracy in past years are 5–10 points higher on average than the ratings we reported at the time. (The difference within the set of experts whose responses we can link over time is 4–8 points.) We interpret this pattern as consistent with a contrast effect in which negative evaluations of the current status quo make the past seem relatively more favorable than it did at the time. Alternatively, they may rate it more favorably because the fears they held at the time were not realized. Notably, experts rate 2024 more favorably today than they did at the time. By contrast, Democrats in the public rated 2024 less favorably in our July survey than they did at the time (-10), which may be a reaction to the result of the election rather than the state of the political system in the year in question.

Political scientists and the headline experiment

Given the mission of Bright Line Watch, we asked both experts and the public whether political scientists should take an active role in public debates about issues related to democracy or focus on establishing sound scientific facts and stay out of public debates. For obvious reasons, we were especially interested in the effect of exposure to political science reports on public assessments of the state of U.S. democracy. To do so, we randomized public exposure to an adapted excerpt from a news article about a Bright Line Watch report with the headline “U.S. democracy has settled into diminished state, experts find” and then asked them to rate U.S. democracy a second time on a 0–100 scale.

Among those who were not exposed to the article, experts were most likely to say political scientists should take an active role (81%) followed by Democrats (61%) and the public as a whole (41%). Republicans were most skeptical (19%). We observe similar polarization on the question of whether political scientists should communicate about threats to democracy with the general public. Among experts, 93% agreed that political scientists should communicate about democratic threats. Among the public, 61% overall shared this opinion, with eight in ten Democrats but only 50% of Republicans agreeing.

Effect of report on threats to U.S. democracy

Survey-weighted treatment effects, adjusted for earlier democracy ratings

Separate survey-weighted linear models control for the earlier rating in the same survey. Points show article-minus-no-article coefficients; bars show HC2-robust 95% confidence intervals. Public includes Democrats, Republicans, and independents. Party groups (including leaners) are subsets. Source: Bright Line Watch, October 6–14, 2026.

Exposure to the article significantly reduced democracy ratings by two points on average among the public as a whole and three points among Democrats (conditional on their ratings earlier in the survey). Among Republicans, however, the two-point effect we observed was not statistically significant, though the difference in effects versus Democrats was not significant either.

Effect of democracy threat report on confidence in political scientists

Percent with a great deal or a fair amount of confidence in their findings

Public estimates are weighted. Bars show 95% confidence intervals. Article and no-article groups are randomly assigned. Party groups include leaners. Confidence combines a great deal and a fair amount; all five response categories remain in the denominator. Within each article/no-article group, Public includes Democrats, Republicans, and independents; party estimates are subsets. Source: Bright Line Watch, October 6–14, 2026.

Exposure to the article made Democrats and the public as a whole more likely to express confidence in political scientists “when they make claims about democracy in America”, while confidence among Republicans did not change measurably. Similarly, support for political scientists playing an active role in debates about democracy increased from 61% to 76% among Democrats and from 19% to 23% among Republicans (although the latter shift falls short of statistical significance). As a result, overall public support for playing an active role increased from 41% to 48%. The article had no measurable effect on agreement that political scientists should communicate about democratic threats.

Political scientists in public debate

Left: percent choosing an active role. Right: percent agreeing or strongly agreeing with public communication.

Public estimates are weighted; expert estimates are unweighted. Bars show 95% confidence intervals. Article and no-article public groups are randomized; experts are a separate comparison group. Within each article/no-article group, Public includes Democrats, Republicans, and independents; party estimates are subsets. Public survey: October 6–14, 2026; expert survey: October 3–17, 2026. Source: Bright Line Watch, October 2026.

Survey data and replication materials: data page.

Appendix

Statistical methods

For binary-threshold outcomes, we converted responses to binary indicators using pre-defined thresholds and estimated sample and subgroup differences in proportions using two-sided, unpaired weighted Welch t‑tests. For continuous outcomes, we used the same approach to compare means. In both cases, we reported proportions or means, pairwise differences, 95% confidence intervals, and p‑values.

Public descriptive estimates and Welch comparisons used survey weights. Expert comparisons used equal weights across expert respondents and unpaired Welch t‑tests. We applied the Benjamini–Hochberg false discovery rate adjustment across the full tested democratic-performance battery within each comparison and audience, across the 12 election-fraud and voting-barrier wording comparisons within each audience, and across the six historical election-fraud comparisons within each audience. Other comparisons used unadjusted, two-sided p‑values, with statistical significance assessed at .05. Confidence intervals were pointwise rather than multiplicity-adjusted.

For categorical items with substantively distinct response groupings, we defined separate binary indicators for each grouping and estimated subgroup differences for each indicator.

For the article experiment’s democracy-rating outcome, we estimated separate survey-weighted linear regressions for the public, Democrats, and Republicans, controlling for each respondent’s earlier democracy rating in the same survey. We used HC2 heteroskedasticity-robust standard errors and tested partisan differences using a treatment-by-Republican interaction in a model that also included treatment, a Republican indicator, and the earlier rating.

Forecast probability summaries are presented as medians. The main conditional comparison shows descriptive differences between group medians, not estimated causal effects or tests of statistical significance.