We analyzed the effects of the agent’s reason for the belief (good, unspecified, bad), the truth of the belief (true, false), and the outcome of the action (neutral, bad) on participants’ judgments of moral blameworthiness, judgments of belief justification, and neural responses in each region of interest (ROI). Because the conditions were not completely crossed (i.e. there was no condition in which the belief was true, but the outcome negative), the effects of truth and outcome were analyzed separately in all subsequent analyses. The effect of truth was measured by comparing the conditions with neutral outcomes (lucky agents’ true beliefs and extra lucky agents’ false beliefs). The effect of outcome was measured by comparing the conditions with false beliefs (extra lucky neutral outcomes and unlucky negative outcomes).
Moral Blameworthiness Judgments (fMRI Experiment)
Subjects’ judgments of moral blameworthiness (Fig. 2) were affected by the agent’s reason for the belief (good, unspecified, bad), the truth of the belief (true, false), and the outcome of the action (neutral, bad).
A 3 [Reason: bad vs. unspecified vs. good] ×2 [Truth: true vs. false] repeated measures ANOVA of the neutral outcome conditions, revealed main effects of reason (F(1, 17) = 8.9 p = 0.002, partial h
2 = 0.51) and truth (F(1, 18) = 86.1 p = 2.8 × 10−8, partial h
2 = 0.83) on judgments of moral blameworthiness. The interaction between reason and truth was not significant (F(1, 18) = 1.0 p = 0.37, partial h
2 = 0.11). Even when all the outcomes were neutral, agents with bad reasons were judged as more blameworthy than agents with unspecified reasons (t(18) = 2.7 p = 0.01), and agents with unspecified reasons were judged as more blameworthy than agents with good reasons (t(18) = 3.5 p = 0.002). Agents with false beliefs were judged as more blameworthy than agents with true beliefs across the reason conditions: when agents’ reasons for their beliefs were good (t(18) = 5.7 p = 2.2 × 10−5), unspecified (t(18) = 3.8 p = 0.001), and bad (t(18) = −6.1 p = 1.0 × 10−5).
A 3 [Reason: bad vs. unspecified vs. good] ×2 [Outcome: neutral vs. bad] repeated measures ANOVA of the false belief conditions, revealed main effects of reason (F(1, 17) = 11.4 p = 0.001, partial h2 = 0.57) and outcome (F(1, 18) = 5.4 p = 0.03, partial h
2 = 0.23) on judgments of moral blameworthiness. The interaction between reason and outcome was not significant (F(1, 18) = 0.90 p = 0.43, partial h
2 = 0.10). When their beliefs were false, agents with good reasons were still judged to be less morally blameworthy than agents with unspecified reasons (t(18) = 3.4 p = 0.003) or bad reasons (t(18) = 4.8 p = 1.5 × 10−4). The difference in moral judgments of agents with unspecified versus bad reasons did not reach significance (t(18) = 1.6 p = 0.12). Although there was a significant main effect of outcome in the overall analysis, in pairwise comparisons agents causing bad outcomes were judged significantly more morally blameworthy than agents causing neutral outcomes only when agents had bad reasons for their beliefs (t(18) = 2.1 p = 0.046). The effect of bad outcomes did not reach significance when agents had unspecified reasons (t(18) = 1.5 p = 0.15) or good reasons (t(18) = 1.5 p = 0.15) for their beliefs.
Participants’ reaction times to make these judgments were not affected by the agent’s reason, the truth of their beliefs, or the outcome of their actions.
Belief Justification Judgments (Behavioral Experiment 1)
As predicted, subjects’ judgments of belief justification (Fig. 3) were influenced by the agent’s reason for the belief (good, unspecified, bad) and the truth of the belief (true, false). A 3 [Reason: bad vs. unspecified vs. good] ×2 [Truth: true vs. false] repeated measures ANOVA of the neutral outcome conditions, revealed significant main effects of reason (F(1, 22) = 118.5 p = 1.6 × 10−12, partial h
2 = 0.92) and truth (F(1, 23) = 31,9 p = 9.4 × 10−6, partial h
2 = 0.58). Whether the belief was true or false mattered more for belief justification judgments, however, if the agent had a good or unspecified reason for his or her belief; if the agent had a bad reason for his or her belief, participants judged the belief to be unjustified even if it turned out to be true, producing a significant interaction between reason and truth (F(1, 22) = 6.4 p = 0.006, partial h
2 = 0.37). In pairwise comparisons, when the outcomes were all neutral, agents with good reasons were judged as having more justified beliefs than agents with unspecified reasons (t(23) = 5.7 p = 7.7 × 10−6), and agents with unspecified reasons were judged as having more justified beliefs than agents with bad reasons (t(23) = 4.7 p = 1.1 × 10−4). False beliefs were judged to be less justified than true beliefs when the reasons for the beliefs were good (t(23) = 5.7 p = 7.8 × 10−6) or unspecified (t(23) = 4.0 p = 0.001). However, false beliefs were only marginally less justified than true beliefs when the agent had a bad reason for the beliefs (t(23) = 1.9 p = 0.07).
The outcome of the action (neutral vs. bad) had a small effect on judgments of belief justification (Fig. 3). A 3 [Reason: bad vs. unspecified vs. good] ×2 [Outcome: neutral vs. bad] repeated measures ANOVA of the false belief conditions, revealed a significant main effect of reason (F(1, 22) = 60.8 p = 1.1 × 10−9, partial h
2 = 0.85) and a marginal effect of outcome (F(1, 23) = 3.8 p = 0.06, partial h
2 = 0.14), with no interaction (F(1, 22) = 0.99 p = 0.39, partial h
2 = 0.08). For the false belief conditions, beliefs based on good reasons were judged more justified than beliefs based on unspecified reason (t(23) = 5.8 p = 6.1 × 10−6), and beliefs based on unspecified reasons were judged more justified than beliefs with bad reasons (t(23) = 4.6 p = 1.2 × 10−4). Similar to the pattern for moral blame judgments, when agents had bad reasons for their false beliefs, those beliefs were judged significantly less justified when they led to bad versus neutral outcomes (t(23) = 2.3 p = 0.03). In other words, the same false beliefs based on the same bad reasons were judged to be less justified when they led to bad outcomes (as opposed to neutral outcomes). This effect of bad outcomes on belief justification judgments was limited to bad reasons, though; there was no effect of bad outcomes on belief justification judgments when agents had unspecified reasons (t(23) = 1.3 p = 0.22) or good reasons (t(23) = 0.83 p = 0.41) for their beliefs.
Behavioral Experiment 2
Forty-two new participants read the same set of fifty-four moral scenarios but made both judgments of moral blameworthiness and judgments of belief justification for each scenario. This design allowed us to accomplish two goals. First, this experiment allowed us to examine the relationship among the different variables by mediation analyses, specifically, to determine (1) whether the influence of reason on moral judgments was mediated by the influence of reason on belief justification judgments, (2) whether the influence of truth on moral judgments was mediated by the influence of truth on belief justification judgments, and (3) whether the influence of outcome on moral judgments was mediated by the influence of outcome on belief justification judgments, or, alternatively, whether the direct influence of outcome on moral judgments mediated the influence of outcome on belief justification judgments. Second, this behavioral experiment, together with the moral judgment data collected in the fMRI experiment, allowed us to test whether false beliefs contribute more to moral luck than bad outcomes.
We first replicated the general pattern of effects reported in the initial behavioral and fMRI experiments. For moral blameworthiness judgments, a 3 [Reason: bad vs. unspecified vs. good] ×2 [Truth: true vs. false] repeated measures ANOVA of the neutral outcome conditions, revealed main effects of reason (F(2, 40) = 11.2 p < 0.001, partial h
2 = 0.36) and truth (F(1, 41) = 92.6 p < 0.001, partial h
2 = 0.69), and no interaction. A 3 [Reason: bad vs. unspecified vs. good] ×2 [Outcome: neutral vs. bad] repeated measures ANOVA of the false belief conditions, revealed main effects of reason (F(2, 40) = 14.1 p < 0.001, partial h
2 = 0.41) and outcome (F(1, 41) = 29.7 p < 0.001, partial h
2 = 0.42), and no interaction.
For belief justification judgments, a 3 [Reason: bad vs. unspecified vs. good] ×2 [Truth: true vs. false] repeated measures ANOVA of the neutral outcome conditions, revealed main effects of reason (F(2, 37.4) = 11.2 p < 0.001, partial h
2 = 0.66) and truth (F(1, 40) = 19.0 p < 0.001, partial h
2 = 0.32). However, the interaction between effects of reason and truth, in Behavioral Experiment 1, was not replicated in Behavioral Experiment 2; even beliefs based on bad reasons were judged to be more justified when they were true than when they were false. A 3 [Reason: bad vs. unspecified vs. good] ×2 [Outcome: neutral vs. bad] repeated measures ANOVA of the false belief conditions, revealed main effects of reason (F(2, 40) = 48.6 p < 0.001, partial h
2 = 0.71) and outcome (F(1, 41) = 9.2 p = 0.004, partial h
2 = 0.18), and no interaction.
We were then able to conduct mediation analyses to look at the relationship between condition variables (reason, truth, and outcome) and participants’ judgments of moral blameworthiness and belief justification.
First, we examined the relationship between reason (good vs. unspecified vs. bad reason), moral judgments, and belief justification judgments. The conditions for a mediation analysis were met: (1) the difference in reason had a significant effect on both moral judgments and belief justification judgments, as noted above, and (2) moral judgments and belief justification judgments were themselves significantly correlated (r = −0.336, p < 0.001). As predicted, a Sobel test showed that the effect of reason on moral judgments was mediated by the effect of reason on belief justification judgments (Z = −4.78, p < 0.00001). In other words, part of the effect of reason on moral judgments was due to the effect of reason on belief justification judgments (Fig. 4).
Second, we examined the relationship between truth (true vs. false beliefs), moral judgments, and belief justification judgments. The conditions for a mediation analysis were met: (1) the difference in truth had a significant effect on both moral judgments and belief justification judgments, as noted above, and (2) moral judgments and belief justification judgments were themselves significantly correlated (r = −0.276, p < 0.001). As predicted, a Sobel test showed that the effect of truth on moral judgments was mediated by the effect of truth on belief justification judgments (Z = −2.50 p = 0.01). In other words, part of the effect of truth on moral judgments was due to the effect of truth on belief justification judgments (Fig. 4).
Third, we examined the relationship between outcome condition (bad vs. neutral), moral judgments, and belief justification judgments. The conditions for a mediation analysis were met: (1) the difference in outcome condition had a significant effect on both moral judgments and belief justification judgments, as noted above, and (2) moral judgments and belief justification judgments were themselves significantly correlated (r = −0.224, p < 0.001). A Sobel test provided no evidence for the notion that the effect of outcome on moral judgments was mediated by the effect of outcome on belief justification judgments (Z = 1.77 p = 0.08). Instead, the effect of outcome on belief justifications was mediated by the effect of outcome on moral judgments (Z = −2.58 p = 0.01). In other words, part of the effect of outcome on belief justification judgments was due to the direct effect of outcome on moral judgments (Fig. 4).
Finally, we tested our prediction that false beliefs account for more moral luck than bad outcomes. To do so, we computed two difference scores. First, for the effect of false beliefs, we calculated the difference in moral blame for extra lucky agents (false beliefs, neutral outcomes) versus lucky agents (true beliefs, neutral outcomes). Second, for the effect of bad outcomes, we calculated the difference in moral blame for unlucky agents (false beliefs, bad outcomes) versus extra lucky agents (false beliefs, neutral outcomes). Paired-samples t-tests showed that the effect of false beliefs was greater than the effect of bad outcomes (fMRI experiment, t(18) = −3.3 p = 0.004, Behavioral Experiment 2, t(41) = −1.8 p = 0.07).
fMRI Results
A whole-brain random effects analysis of the data replicated results of previous studies using the same task (Saxe and Kanwisher 2003), revealing a higher BOLD response during stories describing mental states such as thoughts and beliefs, as compared to stories describing physical (non-mental) states, in the RTPJ, LTPJ, PC, DMPFC, and VMPFC (p < 0.001, uncorrected, k > 10). ROIs were identified in individual subjects at the same threshold (Table 1): RTPJ (identified in 17 of 17 subjects), LTPJ (17/17), PC (17/17), DMPFC (13/17), and VMPFC (10/17).
Table 1 Localizer experiment results
We observed a robust response in the ROIs when the belief and the agent’s reason for the belief were presented. However, we found no effect of condition (bad vs. unspecified vs. good reason) on the response in any ROI during the belief presentation. That is, while participants were reading about beliefs, there was a large and robust response (especially in the RTPJ and LTPJ), independent of whether the belief was justified or unjustified (Fig. 5).
Later in the trial, while participants made moral judgments (and the belief information was no longer on the screen), we observed a small but significant response in the RTPJ and LTPJ for bad outcomes versus neutral outcomes (3 [Reason] ×2 [Outcome] ANOVA, main effect of outcome, RTPJ: F(1, 16) = 4.33 p = 0.05, partial h
2 = 0.21; LTPJ: F(1, 16) = 16.2 p = 0.001, partial h
2 = 0.50; Fig. 5). There was also a unpredicted effect of truth in the LTPJ only (F(1, 16) = 5.6 p = 0.03, partial h
2 = 0.26); the LTPJ response during moral judgment was higher for true beliefs than false beliefs. No significant effects of condition were observed in any of the other ROIs.