The stereotype content led participants to value some lives over others in moral tradeoff scenarios. Participants most endorsed saving ingroup members—Americans and students, who seem both warm and competent. Furthermore, participants did not value different outgroup members’ lives equivalently. Targets belonging to extreme outgroups (i.e. low-warmth, low-competence targets) became targets of relative moral exclusion. It was most morally acceptable to sacrifice them and least acceptable to save them.
Surprisingly, acceptability ratings did not depend on the warmth of the sacrificed target because it was as acceptable to sacrifice high warmth, low competence targets (elderly, disabled) as it was to sacrifice low warmth, low competence targets (homeless, drug addict). It is possible that the increased acceptability associated with sacrificing disabled and elderly people was driven by participants' lay theories regarding those targets' quality of life. While it is difficult to imagine why people would endorse shoving an individual in a wheelchair or an elderly person off an overpass, many people (both college-aged individuals as well as adults over 65) believe that health impairments, which interfere with valued life activities, constitute a “fate worse than death” (Ditto et al., 1996). If participants were ambivalent about the value of high-warmth, low-competence targets’ lives, they may have deemed it most acceptable for Joe to maximize the number of lives saved (i.e. preferring that Joe sacrifice one elderly or disabled person if he could save five of anyone).
On the other hand, neither warmth nor competence alone was the best predictor of moral acceptability of saving any single group of people. Rather, in concord with SCM predictions, both warmth and competence determined that it was most acceptable to save the ingroup, and least acceptable to save extreme outgroups, who are low in both warmth and competence.
Besides predicting that participants would most endorse sacrificing extreme outgroups and saving the ingroup, we predicted that these combinations would activate regions previously associated with resolving complex tradeoffs, because people would have to override their general aversion to report that some combinations were more acceptable than others (Greene et al.
; Greene and Haidt, 2002
). This network includes lateral OFC, dorsolateral and medial PFC (Wallis, 2007
; Rangel et al.
, 2008) as well as anterior cingulate cortex (Cunningham et al.
, 2004). The only contrast that activated this network was when Joe sacrificed a low-warmth, low-competence person to save high-warmth, high-competence people ( and ). This activation dovetails nicely with the ratings data, which demonstrated that sacrificing low-warmth, low-competence people and saving high-warmth, high-competence people constituted the most acceptable classes of dilemmas.
According to recent reviews (Wallis, 2007
; Rangel et al.
, 2008), human and animal studies suggest that OFC may function to integrate multiple attributes of a decision, and compute an associated value. Lateral PFC then utilized the value to plan behavior and medial prefrontal cortex and ACC evaluate the outcome in terms of success and required effort. By some accounts, lateral OFC contributes to information processing by inhibiting neural activity associated with unrelated or distressing information and sensations (Shimamura, 2000
; Beer et al.
, 2006). Nevertheless, studies in both humans and non-human primates have demonstrated that OFC is implicated in calculating expected value of stimuli and integrating the determined value into present and future behavior (Knutson et al.
, 2005; Wallis, 2007
). In accord with this evidence, patients with OFC damage demonstrate difficulty integrating multiple attributes in making a decision (Fellows and Farah, 2005
). Note that the current data show activation in left lateral OFC, whereas others (Cunningham et al.
, 2004) observed activation in right lateral OFC. Some evidence suggests that left lateral OFC is particularly important for the suppression of threat in decision making (Bishop et al.
; Beer et al.
, 2006), though it is not clear whether distinct functions engage the right vs
left lateral OFC during inhibition (Hooker and Knight, 2006
These findings suggest that even though most people say it is unacceptable to shove a person off a bridge to save five other people, utilitarian valuation of tradeoffs are demonstrably biased by the stereotype content. Specifically, 88% of people say the act is unacceptable when the targets are unidentified (Hauser et al., 2007), indicating most people's default is moral aversion to the sacrifice. We reverse this pattern by manipulating the warmth and competence of the targets involved: 84% of our respondents say it is acceptable for Joe to push a low-warmth, low-competence person off a bridge to save five high-warmth, high-competence targets. We also have preliminary evidence that greater OFC and DLPFC activation was related to higher acceptability ratings of sacrificing a low-warmth, low-competence person to save high-warmth, high-competence people (though the correlations were not significant due to a small sample size). We propose that participants are actively overriding their moral aversion to using another person as a means to an end when they have the opportunity to save ingroup members by sacrificing extreme outgroup members.
What remains unclear is whether participants are actually exerting more cognitive control to override their moral aversion to sacrificing low-warmth, low-competence targets or whether they experience less moral aversion to override in the first place. Unfortunately, our design prevents claims about whether the observed activation was driven more by saving high-warmth, high-competence people (ingroup favoritism) or by sacrificing low-warmth, low-competence people in general (extreme outgroup derogation). The temporal proximity of the sacrificed and saved target images does not allow parsing of the independent effects of target sacrificed and targets saved. Future studies should either employ EEG to increase temporal resolution or provide sufficient time between the presentation of the sacrificed and saved targets to model them and behavioral responses separately.
An open question regards the other moral dilemma combinations in the study. One possibility is that exactly the same computation is occurring, only to a lesser extent, because the cost of utilizing people (other than low-warmth, low-competence outgroup targets) as a means to an end is so salient that the moral calculus is simpler: not acceptable. Recall that Greene et al.
) examine which brain regions respond more to difficult as compared to easy personal moral dilemmas and find a pattern of activation similar to our sacrifice low–low to save high–high results. Alternatively, a different, rule-based (as opposed to value-based) process may be taking place in the case of the other combinations. Well-practiced sequences (e.g. routine tasks) may be processed in posterior regions of PFC, whereas less predictable event sequences are thought to be represented in the DLPFC (Wood and Grafman, 2003
In sum, intergroup biases and stereotypes appear to weigh heavily on neural systems implicated in moral decision making. Exactly what strategies participants used for their judgments is a complicated matter for treatment in future studies. Nevertheless, our data suggest that perceptions of warmth and competence, irrespective of the specific social groups in question, may be potent motivators in moral decision making.
Conflict of interest