Monetary Incentive Effects on Partisan Disbelief

Participants are randomly assigned to either an incentive or control condition before they report beliefs about group accuracy rates.

Control: Participants report their best guesses of the accuracy rates of (i) Republican supporters, (ii) Democratic supporters, and (iii) non-partisans on each judgment task (and, for two tasks, the accuracy rates of college vs non-college groups), with no performance-based payment tied to these belief reports.

Incentive treatment: Participants receive information that they may earn a $1 bonus based on how close their reported beliefs are to the realized accuracy rates of the corresponding target group(s). The bonus is implemented as a probabilistic payment that decreases with the absolute prediction error (in percentage points) between the participant’s stated belief and the realized accuracy rate.

2026-02-14

2026-02-28

Perceived partisan accuracy gap (partisan disbelief), respondent-level.
Among partisan respondents (Republican or Democratic supporters), the main endpoint is the perceived difference in accuracy between (i) supporters of the respondent’s own party and (ii) supporters of the opposing party, averaged across the six partisan fact tasks.

Effect of incentives on perceived partisan accuracy gap.
The main experimental endpoint is the difference in the respondent-level perceived partisan accuracy gap between the incentive condition and the control condition among partisan respondents.

Outcome 1: Perceived partisan accuracy gap (partisan disbelief)

Raw inputs: For each of the six partisan fact tasks, respondents report their perceived accuracy rates for Republican supporters and for Democratic supporters.

Construction: For each partisan respondent, compute a task-level perceived gap as “perceived accuracy of own-party supporters minus perceived accuracy of opposing-party supporters.” Then average this perceived gap across the six partisan tasks to form one respondent-level measure of partisan disbelief.

Outcome 2: Effect of incentives on partisan disbelief

Raw inputs: The respondent-level partisan disbelief measure above and the respondent’s randomized assignment (incentive vs control).

Construction: Compare the average respondent-level partisan disbelief between the incentive group and the control group, restricting the analysis to partisan respondents, following the specification in the Pre-Analysis Plan.

Perceived education-group accuracy gap (education disbelief), respondent-level.
The perceived difference in accuracy between college graduates and non-college graduates, averaged across the two education tasks.

Non-partisans’ perceived partisan accuracy gap.
Among non-partisan respondents, the perceived difference in accuracy between Republican supporters and Democratic supporters, averaged across the partisan fact tasks.

Overall perceived accuracy levels by target group.
Average perceived accuracy rates assigned to each target group (Republican supporters, Democratic supporters, non-partisans; and, in the education tasks, college graduates and non-college graduates).

Belief–reality calibration (absolute prediction error).
The absolute difference between a respondent’s perceived accuracy rate for each target group and the corresponding realized accuracy rate, summarized across tasks and/or target groups.

Confidence and own performance (descriptive).
Respondents’ confidence in their own true/false answers and their own factual accuracy on the judgment tasks.

Heterogeneity in partisan disbelief and calibration.
Differences in the primary outcome and in calibration measures by pre-specified respondent characteristics (e.g., affective polarization, demographics, education, income, partisan composition of friends, and beliefs about college completion across parties).

Education-group accuracy gap (education disbelief)

Raw inputs: For each of the two education tasks, respondents report perceived accuracy rates for college graduates and for non-college graduates.

Construction: Compute “perceived accuracy of college graduates minus perceived accuracy of non-college graduates” for each of the two tasks, then average across the two tasks to create a respondent-level measure.

Non-partisans’ perceived partisan accuracy gap

Raw inputs: For each partisan fact task, non-partisan respondents report perceived accuracy rates for Republican supporters and Democratic supporters.

Construction: Compute “perceived accuracy of Republican supporters minus perceived accuracy of Democratic supporters” for each task, then average across tasks to create a respondent-level measure for non-partisans.

Overall perceived accuracy levels by target group

Raw inputs: Respondents’ perceived accuracy rates for each target group in each task.

Construction: Average perceived accuracy rates by target group, optionally also averaging across tasks.

Belief–reality calibration (absolute prediction error)

Raw inputs: Respondents’ perceived accuracy rates for each target group and the realized accuracy rates for each target group (computed from respondents’ actual answers).

Construction: For each respondent–task–target combination, take the absolute difference between the perceived accuracy rate and the realized accuracy rate; then summarize by averaging across tasks and/or targets.

Confidence and own performance

Raw inputs: Respondents’ confidence ratings and their true/false answers to the factual questions.

Construction: Create summary measures such as average confidence across tasks and overall factual accuracy (share of correct answers) across tasks.

Heterogeneity analyses

Raw inputs: The primary and secondary outcomes above and pre-specified covariates collected in the survey.

Construction: Stratify outcomes or estimate differential patterns across subgroups defined by the pre-specified characteristics, following the Pre-Analysis Plan.

We field an online survey experiment in the United States with a target sample size of 1,500 respondents. The study measures respondents’ factual judgments on simple true/false statements and their beliefs about how accurate different social groups are in answering those statements. The key experimental feature is a randomized incentive condition: before respondents report beliefs about group accuracy, half are informed that they may earn an additional $1 bonus based on the accuracy of their belief reports, while the other half receive no accuracy-linked bonus. The primary outcome is the perceived accuracy gap between one’s own party and the opposing party, averaged across the partisan fact tasks, and the main comparison is between the incentive and control conditions. Full details are provided in the Pre-Analysis Plan.

Not available

Randomization is implemented by the survey platform using a computer-based random number generator to assign respondents to the incentive or control condition with equal probability.

Individual respondent.

No

1,500 individuals

1,500 individuals

1,500 individuals

Let x be the share of partisans among all respondents. which we do not know before the survey. Then, assuming a usual two-sided test, a significance level of 5%, and a power of 80%, the MDE for each hypothesis in standard deviation units is Hypotheses 1 and 2: 2.80*sqrt (750*x) Hypothesis 3: 3.96*sqrt (750*x)