Findings, in five figures.
Certainty difference (majority − minority) vs group controversy.
Controversy and certainty difference across 135 moral dilemmas.Each point represents one moral dilemma post in MORALMOMENTS dataset. The x-axis shows group controversy, defined as the proportion of the minority judgment, ranging from 0 (full consensus) to 0.5 (max- imum division). The y-axis shows the certainty difference, the proportion of judgments that are certain in the majority group minus that in the minority group. The marginal histogram (top, pink) shows the distribution of posts across group controversy levels, and the marginal histogram (right, blue) shows the distribution of certainty difference values across all posts.
Overall weakening and strengthening rates by treatment.
Left: weakening rate (%) — the share of participants whose stance weakened after the intervention, stacked into judgment changed (dense hatch) and judgment unchanged with lower confidence (light hatch). Right: strengthening rate (%) — same verdict held with higher confidence. Bars run Control · Controversy · Certainty · Controversy + Certainty; headline labels carry Barnard-exact significance vs Control (BH-corrected: *** .001 · ** .01 · * .05). Hover any bar for the exact breakdown and sample size.
Quadrants — weakening.
Left: forest plot of the weakening logistic regression (log-odds, 95% CI) — Initial confidence, the disagreement-rate and in/out-group-certainty signals within each treatment arm, and the decision-style / intellectual-humility traits; a hollow point marks a non-significant coefficient. Right: proportion of participants whose stance was weakened — a flipped verdict OR an unchanged verdict with reduced confidence — split across treatments and post-level dimensions. Control: overall rate. Controversy: majority vs minority side. Certainty: in-group more vs less certain than the out-group. Controversy + Certainty: full 2×2 cross. Hover any bar or point for exact values.
Quadrants — strengthening.
Left: forest plot of the strengthening logistic regression (log-odds, 95% CI) — same predictors as Fig 3, fit on participants below ceiling initial confidence; a hollow point marks a non-significant coefficient. Note the signs largely mirror Fig 3: signals that push toward weakening push away from strengthening. Right: proportion of participants whose stance was strengthened — same verdict, higher confidence after intervention — split the same way as Fig 3. Hover any bar or point for exact values.
Engagement and acceptance of each treatment variable.
Top: engagement rate — the share of participants in each condition who referenced a given treatment variable (disagreement rate, in/out-group certainty) in their written deliberation; n is the condition's sample. Bottom: among those engagers, the accept vs reject split, with each bar broken into the participant's outcome (weakened / maintained / strengthened) as a share of engagers — so accept + reject spans 100%. Brackets mark significant pairwise differences (Barnard one-sided: *** .001 · ** .01 · * .05). Hover any bar for exact counts; click an accept/reject bar in the bottom panel for its outcome breakdown and example deliberations.