When deliberation backfires: reflection and overconfidence in moral judgment

Abstract

Across six experiments, we show that instructing participants to think carefully before rendering a moral judgment reliably increases their confidence in that judgment without improving its accuracy, consistency, or resistance to framing effects. The effect is robust to incentives, expertise primes, and time pressure manipulations. We argue that deliberation activates a post-hoc rationalization process that produces fluent, coherent narratives — and that people mistake narrative coherence for epistemic warrant. These findings have direct implications for deliberative democratic theory and for legal contexts where careful deliberation is assumed to improve judgment quality.

Introduction

A cornerstone assumption of deliberative models of judgment — in law, democratic theory, and folk epistemology — is that careful reflection improves decision quality. Jurors are instructed to deliberate. Students are told to check their work. The assumption that reflection is epistemically virtuous is so deeply embedded in Western intellectual culture that questioning it can seem perverse.

And yet there is reason for suspicion. Research on verbal overshadowing (Schooler & Engstler-Schooler, 1990), choice blindness (Johansson et al., 2005), and confabulation (Nisbett & Wilson, 1977) all document dissociations between the fluency of a post-hoc account and its reliability as a report on the underlying mental process. If something similar operates in moral judgment — if deliberate reflection generates confident but epistemically unwarranted conclusions — the normative case for deliberation weakens considerably.

The present work tests this possibility across six experiments. We focus on a specific prediction: prompting deliberation before a moral judgment should increase confidence in that judgment without producing a corresponding improvement in accuracy, consistency, or resistance to framing effects.

Method

Participants and Design

Across six experiments we recruited 3,416 adults through Prolific Academic (Experiments 1–5; M age = 34.1, SD = 11.8; 52% women) and in-person laboratory testing at three Canadian universities (Experiment 6; N = 304). All experiments used a between-participants design with random assignment to a deliberation-prompt or control condition.

Materials and Procedure

Moral judgment scenarios were adapted from the trolley-problem literature (Thomson, 1985; Foot, 1967) and supplemented with novel vignettes spanning personal/impersonal, direct/indirect, and omission/commission dimensions. Participants in the deliberation condition read: "Before making your decision, please take at least 60 seconds to think carefully and consider all relevant factors." Control participants read: "Please make your decision." Confidence was measured on a 0–100 slider immediately after each judgment.

Accuracy was operationalised as (a) correspondence with the modal response of a hold-out panel of trained ethics students, and (b) within-participant consistency across structurally equivalent scenarios separated by a two-week delay. Framing susceptibility was measured as the absolute judgment difference between positively versus negatively framed versions of the same scenario.

Results

Deliberation reliably increased expressed confidence across all six experiments (pooled Cohen's d = 0.71, 95% CI [0.61, 0.81]). This effect held regardless of scenario type, participant age, incentive condition, and self-reported ethics expertise.

In contrast, deliberation had no reliable effect on accuracy (d = 0.03, 95% CI [–0.08, 0.14]) or two-week consistency (d = 0.06, 95% CI [–0.05, 0.17]). In Experiments 3 and 4, deliberation modestly increased susceptibility to framing effects (d = 0.28, 95% CI [0.09, 0.47]). Exploratory think-aloud data (Experiment 5) showed that deliberation produced longer and more syntactically coherent justifications, and justification length strongly predicted reported confidence (r = 0.54) independent of judgment quality.

Discussion

The results are consistent with the hypothesis that deliberation activates post-hoc narrative construction and that people use narrative fluency as a proxy for epistemic quality. When deliberation is prompted, participants generate more elaborate justifications; the elaborateness of those justifications drives confidence — regardless of whether the underlying judgment has improved.

These findings stand in tension with dual-process accounts that treat deliberative processing as generally more reliable than intuitive responding. Our data suggest instead that what deliberation adds, at least in the moral domain, is not better reasoning but better rationalisation. The implication for legal contexts is direct: if deliberation produces confidence without accuracy, the deliberative norms embedded in adversarial legal procedure may warrant critical re-examination.

References

Cushman, F. (2013). Action, outcome, and value. Personality and Social Psychology Review, 17(3), 273–292.
Foot, P. (1967). The problem of abortion and the doctrine of double effect. Oxford Review, 5, 5–15.
Haidt, J. (2001). The emotional dog and its rational tail. Psychological Review, 108(4), 814–834.
Johansson, P., Hall, L., Sikström, S., & Olsson, A. (2005). Failure to detect mismatches between intention and outcome in a simple decision task. Science, 310, 116–119.
Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know: Verbal reports on mental processes. Psychological Review, 84(3), 231–259.
Schooler, J. W., & Engstler-Schooler, T. Y. (1990). Verbal overshadowing of visual memories. Cognitive Psychology, 22(1), 36–71.
Thomson, J. J. (1985). The trolley problem. The Yale Law Journal, 94(6), 1395–1415.