Thinking about Reasons for One’s Choices Increases Sensitivity to Moral Norms in Moral-Dilemma Judgments

Whereas norm-conforming (deontological) judgments have been claimed to be rooted in automatic emotional responses, outcome-maximizing (utilitarian) judgments are assumed to require reflective reasoning. Using the CNI model to disentangle factors underlying moral-dilemma judgments, the current research investigated effects of thinking about reasons on sensitivity to consequences, sensitivity to moral norms, and general action preferences. Three experiments (two preregistered) found that thinking about reasons (vs. responding intuitively or thinking about intuitions) reliably increased sensitivity to moral norms independent of processing time. Thinking about reasons had no reproducible effects on sensitivity to consequences and general action preferences. The results suggest that norm-conforming responses in moral dilemmas can arise from reflective thoughts about reasons, challenging the modal view on the role of cognitive reflection in moral-dilemma judgment. The findings highlight the importance of distinguishing between degree (high vs. low elaboration) and content (intuitions vs. reasons) as distinct aspects of cognitive reflection.


Robustness against Dilemma Exclusion
An item-based analysis by Gawronski et al. (2020) revealed that one of the 12 dilemmas used in the current research has low construct validity in the manipulation of moral norms (abduction dilemma). Moreover, four of the remaining 11 dilemmas can be criticized for confounding moral norms with whether the focal action requires interference with the action of someone else (transplant dilemma, torture dilemma, vaccine dilemma, and tyrantkilling dilemma). Using the pooled data, we conducted supplemental analyses with these five dilemmas excluded.
Response times to the four variants of the remaining seven dilemmas (i.e., 28 dilemmas in total) were summed and re-standardized for each individual experiment. Fortyeight cases were flagged as potential outliers based on the standardized response times to the dilemma included in the current analysis (nno-reasons = 27; nthink-about-reasons = 21). Response times did not differ across the think-about-reasons and no-reasons conditions regardless of whether potential outliers were excluded, ts > -0.82, ps > .413, dfull = 0.055, dreduced = 0.041.
Means and 95% confidence intervals of the three model parameters after exclusion of the potentially problematic dilemmas are presented in Table S1.

Robustness against Model Assumptions
Another potential issue with the CNI model pertains to its hierarchical structure.
Given the arbitrary positions of the C and N parameters in the CNI model processing tree, we SUPPLEMENT 4 re-analyzed the pooled data using an alternative model in which the positions of the C and N parameters are reversed (for the sake of brevity called NCI model). We also re-analyzed the pooled data using the CAN algorithm, which algebraically calculates the three model parameters concurrently rather than hierarchically (Liu & Liao, 2021).

Re-analyses using the NCI Model
Means and 95% confidence intervals of the three parameters obtained with the NCI model are presented in Table S2. The NCI model fit the pooled data well, G 2 (2)s ≤ 1.32, ps ≥ .517, ws < 0.01, regardless of whether response-time outliers were removed. Consistent with the CNI model analyses, significant between-group differences were found for sensitivity to moral norms regardless of whether potential response-time outliers were removed, ∆G 2 s > 16.16, ps < .001, dfull = 0.316, dreduced = 0.279. The NCI model analysis revealed no significant differences in general preference for inaction versus action both before and after outlier exclusion, ∆G 2 s < 0.43, ps > .514, dfull = 0.023, dreduced = 0.046. Contrary to the integrative CNI model analyses, the NCI model analyses revealed a significant difference in sensitivity to consequences across conditions, ∆G 2 s ≥ 4.05, ps < .044, dfull = 0.138, dreduced = 0.141, suggesting that participants in the think-about-reasons condition were more sensitive to consequences than those who responded intuitively or thought about their intuitions.

Re-analyses using the CAN Algorithm
Means and 95% confidence intervals of the model parameters obtained using the CAN algorithm are presented in Table S3. Because the parameters of the CAN algorithm tend to be highly correlated (which is not the case for the CNI model parameters), it is important to control for shared variances in analyses using the CAN algorithm to avoid false positive results for a given parameter. We therefore conducted analyses of covariance to test effects of our experimental manipulation on each of three CAN algorithm parameters while entering the remaining two parameters as covariates (Table S4). The ANCOVA predicting the N parameter SUPPLEMENT 5 revealed a significant experimental main effect, Fs > 4.31, ps < .038, 2 full = .006, 2 reduced = .005. There were no significant differences on the C parameter, Fs < 0.62, ps > .433, 2 full = .001, 2 reduced = .001, and A parameter, Fs < 0.36, ps > .551, 2 full < .001, 2 reduced < .001.

Table S1
Means and 95% confidence intervals of estimated CNI model parameters after excluding five potentially problematic dilemmas, pooled data from Experiments 1 to 3. Full sample refers to the pooled data before exclusion of response-time outliers; reduced sample refers to the pooled data after exclusion of response-time outliers. CNI model parameter scores can range from 0 to 1.

Table S2
Means and 95% confidence intervals of NCI model parameters, pooled data from Experiments 1 to 3.

No Reasons
Think   Note. Full sample refers to the pooled data before exclusion of response-time outliers; reduced sample refers to the pooled data after exclusion of response-time outliers.