Rumors and Health Care Reform: Experiments in Political Misinformation

Contributions

• Identifies "fluency" — ease of information processing — as a key mechanism underlying rumor power; shows that merely repeating a false rumor increases its perceived truth, even without additional corroboration
• Demonstrates that corrections are most effective when paired with source credibility effects: a Republican debunking Democratic health-care rumors is more persuasive than either non-partisan or Democratic debunking of the same content
• Tests the persistence of rumor effects and correction efficacy over time, showing that corrections fade in relative effectiveness but rumors persist
• Shows that correction effects vary by individual attention: people who closely read correction materials show stronger effects than those who skim
• Reveals backfire effects in specific contexts: Democratic corrections for Republican audiences sometimes reduce receptiveness to health-care policy support

Method

Two large-scale experimental studies conducted during 2010, immediately following the passage of the Affordable Care Act (ACA). Study 1 (N ≈ 1,600) tested rumors about "death panels" — false claims that the ACA would require elderly and infirm patients to meet with government officials to discuss "end of life" options. Berinsky varied the presentation of rumors (alone vs. paired with corrections) and the source of correction information (non-partisan authority, Republican source, or Democratic source).

Treatments presented short news stories about the ACA with different factual content: (1) control (no rumor), (2) rumor only (death panel claims), (3) rumor + non-partisan correction (factual rebuttal), (4) rumor + Republican correction (using statements from Republican sources debunking the rumor), and (5) rumor + Democratic correction (using statements from Democratic sources debunking the rumor).

Respondents were asked whether they believed the death panel rumor and whether they would support the ACA. A second wave of follow-up interviews measured persistence of effects one week later.

Study 2 (N ≈ 1,000) fielded in October 2010 using a different panel, testing similar conditions with slightly different treatments and follow-up intervals to measure decay of correction effects.

Results

Full sample (Study 1):

• Control: 50% reject the rumor
• Rumor only: 45% reject the rumor (5 percentage-point decrease in rejection)
• Rumor + non-partisan correction: 57% reject the rumor (7 percentage-point increase)
• Rumor + Republican correction: 58% reject the rumor (8 percentage-point increase, most effective)
• Rumor + Democratic correction: 54% reject the rumor (4 percentage-point increase, least effective)

Attentive subsample (those who passed screener questions indicating close attention):

• Control: 57% reject
• Rumor only: 46% reject (11 percentage-point decrease)
• Rumor + non-partisan correction: 60% reject
• Rumor + Republican correction: 69% reject (strongest effect, 15 percentage-point increase relative to rumor-only)
• Rumor + Democratic correction: 60% reject

Effect persistence (Study 1, week 2): Effects fade over time. Roughly one week later, the difference between conditions shrinks, with rumor rejection rates converging across groups. The rumor-only condition remains below control (57% vs. 58% control), suggesting some persistence of the rumor effect even after a week.

Policy support (Study 2): Presenting the rumor alone decreases support for ACA policies (51% support in control vs. 42% support in rumor-only condition). Corrections improve policy support, but the effect depends on source: Republican corrections increase support most (48% relative to 42% in rumor-only). Democratic corrections paradoxically reduce support to the lowest level (37%), suggesting a boomerang effect when the in-party (Democrats) attempts correction.

Notes

Strengths:

• Rigorous experimental design with multiple waves and attention-based subgroup analysis; field-tested during an active political controversy rather than artificial laboratory scenarios
• Clear identification of fluency as a mechanism distinct from logical argument evaluation
• Demonstrates that partisanship of the correction source matters more than the factual content alone, a finding often overlooked in naive correction efforts
• Tests persistence and shows effects are not one-shot but decay over time

Limitations:

• Focuses on a specific historical event (2010 ACA debate); generalizability to other rumors and time periods unclear
• The "fluency" explanation, while plausible, is inferred from observed patterns; more direct measures of subjective information ease would strengthen the claim
• Sample and effect sizes in attentive subsample are modest; backfire effects in some conditions approach or do not reach statistical significance
• Does not test interactive correction strategies (e.g., providing both non-partisan facts and partisan source endorsement)

Follow-ups:

• Later work by others (Nyhan, Reifler, Pennycook, Rand) extends these findings to COVID-19 and political misinformation in the 2016–2020 period, confirming persistence of source credibility effects and partisan asymmetries
• Questions remain about optimal source selection when both parties actively deny inconvenient truths; Berinsky's work shows that out-party correction works when out-party has better credibility on that specific claim, but many modern rumors have no credible partisan out-group willing to debunk them