Crowd-Wisdom Polling

Definition

Crowd-wisdom polling asks respondents about expected winners or the voting behavior of friends, relatives, and coworkers, seeking information beyond each respondent’s stated personal vote choice.

Current Synthesis

The method can sample perceptions of a respondent’s social environment, but those networks are neither random nor independent. In the episode’s 2022 test, incorporating crowd-wisdom questions moved a highly accurate conventional national House estimate far away from the final margin. The questions may still be informative as measures of expectations, perceived local norms, or turnout motivation, but that is a different claim from improving vote-share forecasts.

Key Claims

• One respondent’s social-circle estimate can contain information about people a conventional poll did not directly reach.
• Homophily, selective observation, media exposure, and shared error can make social-network reports systematically unrepresentative.
• The 2022 Marist test is negative evidence against using these questions as an automatic direct forecast correction.
• Expected-winner and social-circle questions may have research value for turnout, bandwagon, or underdog effects even when they harm winner prediction.

Evidence

Forecast test

• We try to fix election polling (update) reports a conventional Marist estimate of a three-point Republican national House lead against a final 2.8-point margin, while the experimental questions would have shifted the estimate to roughly eight points.

Alternative measurement value

• We try to fix election polling (update) says Marist retained the questions ahead of the 2026 midterms while studying whether expectations help explain turnout or bandwagon and underdog effects.

Counterevidence & Qualifications

• The episode does not provide the experiment’s full model, sample, uncertainty intervals, preregistration, or performance across multiple elections.
• A single poor national result does not show that every formulation or use of crowd-wisdom questions fails.
• Respondents’ expectations may predict behavior without accurately measuring vote share, so the target outcome must be specified before judging performance.
• Social-circle reports can duplicate correlated opinions and should not be treated as independent additional respondents.

What Changed

• Established a distinction between crowd-wisdom questions as forecast adjustments and as measures of expectations or motivation.
• Added the 2022 Marist result as direct negative evidence for the forecast-adjustment use.

Related Concepts

• Election Polling Representativeness - defines the direct-sampling problem crowd-wisdom questions try to supplement.
• Survey Mode Effects - offers a parallel warning that measurement design can change observed results.
• Scientific Sampling Discipline - explains why additional observations do not help when their provenance and dependence are misunderstood.
• Marist Poll - organization that tested and retained the questions for further study.