Partisanship and the trolley problem: Partisan willingness to sacrifice members of the other party

Abstract

Do partisans view members of the other party as having lower moral status? While research shows that partisans view the out-group quite poorly, we show that affective polarization extends to expressing a willingness to sacrifice an out-partisan’s life. We report the first study to consider partisanship in the classic “trolley problem” in which respondents are asked whether they would sacrifice an individual’s life in order to save the life of five individuals. We explore this issue with a nationally representative survey experiment in the United States, inquiring about politicized variants of the trolley problem case. First, we vary the political affiliations of both the group of five (to be saved by turning the trolley) and the single individual (to be sacrificed by turning the trolley). We find that individuals are less willing to sacrifice a co-partisan for the sake of a group of out-partisans. These findings go beyond earlier work by suggesting that partisans not only hold negative attitudes and judgments toward political out-groups but also they will at least signal approval of differing moral treatment. We take stock of how these results bear on normative questions in democratic theory.

What moral obligations do Americans recognize toward their political opponents? Recognizing fellow citizens—including political adversaries—as worthy partners for cooperation and legitimate political engagement has long been thought to be an important basis for democratic governance. Contemporary philosophers, with a nod to Aristotle, label this value as “civic friendship” (Rawls, 2005; Ebels-Duggan, 2010; Vallier, 2019).

Affective political polarization poses an intuitive challenge to realizing the values associated with political cooperation among adversaries. Partisanship is a primary source of moral intuitions, including moral judgments about wrongful actions (Connors, 2019; Hatemi et al., 2019; Smith et al., 2017; Walter and Redlawsk, 2019; Goren et al., 2009; McConnell et al., 2018). If people dislike members of the other party and want to avoid contact with them (Huber and Malhotra, 2017; Iyengar and Westwood, 2015), then their political attitudes dim the prospects for realizing civic friendship. Not only have partisans become more polarized (Hetherington and Rudolph, 2015; Mason, 2018), but also this trend has primarily manifested through increasing hostility toward members of the other party (Abramowitz and Webster, 2016). Hostile feelings toward opposing partisans may extend to and include counter-empathetic responses, such as favoring policies and candidates expected to harm the opposing party (Webster et al., 2022; Hudson et al., 2019).

Antagonistic feelings may also shape how individuals think about the moral status of members of opposing parties. Extreme affective polarization is associated with a greater willingness to dehumanize political opponents, ascribing to them properties of being less evolved or more animal like, and a willingness to dehumanize predicts greater authoritarianism, partisan bias, and reduced tolerance (Martherus et al., 2019; Cassese, 2019). Indeed, the COVID-19 pandemic has sadly illustrated this with partisan commentators suggesting deaths in “Blue states” are less concerning than those in “Red states,” or online posts celebrating the deaths of unvaccinated Republicans who “got what they deserved.” ¹ Furthermore, recent unrest in US cities during the summer of 2020 and the Capitol insurrection in 2021 have shown political opponents engaging in lethal violence. ²

While prior work has focused on whether partisans will ascribe negative traits to their political opponents, we are interested in whether negative affect extends to approving of different kinds of moral treatment. In recent years, negative affect has escalated to include even the prospect of lethal harm, with 10–15% of the population willing to countenance violence or threatened violence against opposing partisans (Kalmoe and Mason, 2022). We explore this issue with a nationally representative survey experiment, inquiring about politicized variants of the “trolley problem” case. With this conceptual device, we can extend earlier findings by specifically considering partisan trade-offs in violent harm. First, we vary the political affiliations of both the group of five (to be saved by turning the trolley) and the single individual (to be sacrificed by turning the trolley). As in other survey applications of the trolley case, professed willingness to save the greater number is high. We find, however, that individuals are less willing to sacrifice a co-partisan for the sake of a group of out-partisans. Second, we consider how much this reflects negative partisanship by comparing partisanship with a variety of other out-groups. Among those considered, only neo-Nazis received less support by respondents than opposing partisans and then only among Democrats. For Republicans, all out-groups were indistinguishable. These findings go beyond earlier work by suggesting that partisans not only hold negative attitudes and judgments toward political out-groups, but also they will at least signal approval of differing moral treatment. In the final section, we take stock of how these results bear on normative questions in democratic theory.

Theory

In the simplest version of the trolley problem, one must decide whether to pull a lever, which will bring about the death of one person in order to avoid the deaths of the five. Beginning with its introduction by Foot (1967), the trolley problem has prompted a great deal of debate in philosophy and psychology (Thomson, 1976, 2008; Christensen and Gomila, 2012).

Most respondents act to bring about the utility promoting outcome (Navarrete et al., 2012). There are two significant classes of exceptions. First, the more an agent must monitor their intentional activity so as to secure the death of one person, the more objectionable such acts feel. For example, subjects are more willing to accept pulling a lever to save five at the expense of one than they are to accept pushing one person in front of the oncoming trolley in order to save five (Lanteri et al., 2008; Greene and Haidt, 2002). Second, individuals are less willing to sacrifice a person with whom they share an identity, such as a relative or romantic partner (Bleske-Rechek et al., 2010).

When confronted with the trolley problem in a political context, considerations of partisan in- and out-group identities may also become salient as negative affective partisanship may extend to moral judgments of out-partisans. Motivated by theories of social identity theory, scholars of partisan identity and party affiliation have described partisanship as primarily a function of group identification, with parties operating as “diametrically opposed ‘teams’ of citizens (Kane et al., 2021).” This groups-based view of partisanship helps explain the rise of negative partisanship, which is well documented in the political science literature. Relative to the past, contemporary partisans use fewer positive traits to describe opposing candidates and experience greater negative emotion when thinking about them (Iyengar and Krupenkin, 2018). Although warm feelings toward co-partisans have not correspondingly increased, antagonism toward political adversaries has been associated with greater party loyalty and heightened political participation (Abramowitz and Webster, 2016; Iyengar and Krupenkin, 2018). Beyond attitudinal measures, negative partisanship predicts biases in behavior in the lab as well as in practice. In economic games in the lab, participants favor co-partisans and impose penalties on opposing partisans (Iyengar and Westwood, 2015). Survey experiments show respondents avoid offering scholarships or jobs to opposing partisans, even when more qualified than in-party alternatives (Iyengar and Westwood, 2015; Gift and Gift, 2015), and also more willing to date co-partisans (Huber and Malhotra, 2017).

Cassese (2019) finds that members of both major political parties regard political out-group members as less human, both in a willingness to apply more “animalistic” traits to out-group members, and also willing to associate them with “less evolved” visual representations. Dehumanization at least points in a morally worrisome direction, in particular to potential consequences for compromise, gridlock, and a willingness to believe false or misleading media about opponents (Leidner et al., 2013).

The literature on dehumanizing opposing partisans stops short of any evidence for the devaluation of an out-partisan’s life, yet there is evidence that it predicts violence and other norm violations in some contexts (Cassese, 2019). Kalmoe (2014) has recently found that some partisans are motivated by violent language, focusing on the use of metaphors comparing political activity to war, fighting, or other forms of conflict. Taken together, these results appear to support the expectation that partisanship will affect moral judgments. Yet, we are unsure if partisans use dehumanizing language and violent metaphors because they think differently about the moral status of political opponents, or merely for rhetorical or motivational purposes.

The trolley problem provides a well-established way of assessing moral judgments about members of different groups (Swann Jr et al., 2010). Although responses to the trolley problem do not directly predict behavior (Bostyn et al., 2018), they do help answer questions about inter-group moral judgments that meaningfully differ both from expressions of affect and from the other behaviors associated with negative affect. Moreover, if negative affect and moral judgment do align, that would lend additional, novel support for predicting that the findings associated with dehumanization in other contexts might also apply to the dehumanization of political out-groups.

This leads us to the development of three different hypotheses. As is discussed above, prior work reveals the utilitarian option to “lever turning” or “switch flipping” in the trolley problem to be the modal response. In the context of partisanship and the trolley problem, the utilitarian hypothesis may be formulated as follows:

H1

Partisans will be willing to sacrifice a single co-partisan (out-partisan) for the sake of saving five co-partisans (out-partisans).

There is little reason to think that one’s party membership is a relevant fact in the overall utility calculus, or at least not the kind of fact which, in isolation from anything else, could exceed the value of saving four additional lives. This is especially true when the partisanship of the six people (5 on the track and 1 to be sacrificed) is held constant.

However, if polarization influences moral judgment as well as affective responses, then we anticipate that partisanship will also affect responses. First, partisanship would predict that individuals would be less willing to sacrifice members of their own party than members of the opposing party.

H2

Partisans will be more willing to save a co-partisan group of five than opposing partisan group of five, regardless of the identity of the single individual being sacrificed.

H3

Partisans will be less willing to sacrifice a co-partisan than an opposing partisan.

If partisans are both less willing to sacrifice their own and more eager to save their own, then H2 and H3 together suggest the highest support for turning a trolley to sacrifice one out-partisan for the sake of five co-partisans. Likewise, we would expect the least support for sacrificing a co-partisan for the sake of five out-partisans.

Data and methods

To test these hypotheses, we use a single-choice conjoint survey design that was embedded in a team module of the 2018 Cooperative Congressional Election Study (CCES) (Hainmueller et al., 2015). ³ The 2018 CCES is a large-n nationally representative survey of Americans in which teams contribute survey modules of approximately 1000 respondents (N ≈ 60,000). Our questions were split across multiple modules for an initial n of 1597. We exclude pure independents, who have no co- or out-partisan comparisons for an n of 1324.

Respondents were introduced to the trolley problem with the following vignette:

In the next four questions, we will ask you to make a series of ethical judgments.

A train is running out of control down a track. In its path are five [randomization 1] who have been tied to the track by a madman. Fortunately, you can flip a switch that will lead the trolley down a different track and save the five [randomization 1]. Unfortunately, there is a single [randomization 2] tied to the other track who will die if you flip the switch.

Would you flip the switch and save the five [randomization 1]?

• Yes

• No

In each of the four rounds, the partisan identity of the five potential victims (randomization 1) was randomized with replacement. In other words, in each of the four rounds, the five individuals were independently and randomly chosen to be either five Republicans or five Democrats. The unfortunate individual who would be potentially sacrificed (randomization 2) by diverting the trolley was also randomly assigned; but without replacement. The individual who would be sacrificed by diverting the trolley in each of the four rounds was chosen at random from seven possible identities: (1) a Republican, (2) a Democrat, (3) a convicted criminal, (4) an illegal immigrant, (5) a neo-Nazi, (6) a golden retriever dog, or (7) a non-US citizen. In other words, of the six persons + dog listed above, four were chosen at random to be the potentially sacrificed person (or dog), one in each round of the vignette.

In total, we have 4823 different responses to the trolley problem. This is a much larger sample of responses than previous work testing variations of the trolley problem in other domains and is a dramatic improvement in providing precise estimates of any observed effects. And, while respondents were asked to complete four scenarios with different randomly chosen characteristics in each round, some skipped questions leading to a small amount of item nonresponse. 963 respondents completed all four vignettes while 259 completed three vignettes, 92 answered two rounds of questions, and 10 individuals only completed one round of the trolley scenario (963*4 + 259*3 + 92*2 + 10 = 4823 trolley responses). In the supplemental materials, we show that there is no correlation between various demographics and not completing all four trolley vignettes. We account for potential within-subject correlation of errors by including robust standard errors clustered by respondent. In the supplemental materials, we also address the possibility of “demand effects” or other issues that may arise from respondents completing multiple rounds of the same style of question. We note here that demand effects are rare (Mummolo and Peterson, 2019), and we find no such evidence in our data. To test for this, we replicate the results presented below while only including the first of the four vignettes presented to each respondent (so that respondents have not yet “learned” anything about the nature of the question or the possibility of randomization in the question wording). The results are substantively the same as the results presented in the main paper. Furthermore, it is unlikely that respondents will be able to determine a priori the researchers’ intentions/expectations. While the identity of the five individuals on the track always contains partisan content, only two of the seven sacrificial individuals are partisans. The remaining five contain no partisan content. Beyond that, it is hard to know if a respondent were to determine that the question was designed to test partisan effects whether the respondent would react in a more “violently partisan” or less violently partisan manner. Taken together, we are confident that our results are not driven by respondent demand effects.

Our key question is how the partisan identity of the five individuals on the track, combined with the partisan identity of the respondent and the partisanship of the individual being sacrificed affects the likelihood of making the decision to flip the switch, sacrifice the one individual, and save the five. Our hypotheses, described above, suggest that partisanship (whether shared or not between the respondent and the group or the individual sacrificed) should play a meaningful role in this decision. We test this using simple mean differences in which the dependent variable is 1 if the respondent chose to save the five individuals by sacrificing the single person (or dog in one of the seven cases). As independent variables, we include a series of dummy variables measuring the identities of the seven different possible sacrificial individuals. We conduct separate analysis depending on whether or not the five individuals to be saved are co-partisans or out-partisans.

Results

As the utilitarian hypothesis (H1) would predict, there is a high absolute level of support for turning the trolley to save five in all cases (0.80 across all observations). However, as the partisanship hypotheses predict, shared partisanship significantly affects one’s willingness to turn the trolley. The first two bars of Figure 1 display these results. There is high support for diverting the trolley in cases where a respondent is saving five co-partisans, regardless of whether the one individual being sacrificed is a co-partisan (0.90) or out-partisan (0.92). The bottom panel of Figure 1 shows these results divided by the partisanship of the respondent. Republican and Democratic respondents turned the trolley at nearly identical rates.

Figure 1.

Trolley problem and partisanship: Proportion of survey respondents choosing to sacrifice one co-partisan or one out-partisan to save five co-partisans or five out-partisans. The bottom panel shows these same results split by the party of the survey respondent. Hash marks indicate 95% confidence intervals.

Things look very different when the five individuals being saved are members of the out-party. H2 suggested that partisans will be more willing to save groups of co-partisans than opposing partisans. Comparing the first two bars of Figure 1 to the third and fourth bars of the figure shows exactly this. Support for turning the trolley drops from 90% to 59% when asked to sacrifice one co-partisan, and the identity of the individuals to be saved changes from five co-partisans to five out-partisans (diff = 0.31, p < 0.001). Likewise, the rate of trolley turning to save the five declines from 92% to 75% when asked to sacrifice one out-partisan to save five co-partisans versus five out-partisans (diff = 0.17, p < 0.001).

Evidence for H3 is somewhat mixed. H3 predicted that, all else equal, partisans would be more willing to sacrifice an out-partisan than a co-partisan. Comparing the first and second bars of Figure 1 shows equal support for sacrificing a co-partisan or an out- partisan in order to save five co-partisans, which does not align with H3. This may be due to potential ceiling effects given the very high rates of trolley turning to save a group of co-partisans. However, we do see support for H3 when looking at the third and fourth bars of Figure 1. Here, partisans are much less willing to sacrifice a co-partisan than an out-partisan, when such a sacrifice would save a group of 5 out-partisans (59% vs 75%, p < 0.001).

What might account for the effect of partisanship on willingness to turn the trolley? One possibility is that respondents regard it as morally unacceptable to sacrifice a co-partisan but regard it as morally acceptable to sacrifice an out-partisan. In other words, the former agent has a higher moral status. According to what we might call the moral status explanation, it is wrong to sacrifice co-partisans in some circumstances in which it would not be wrong to sacrifice opposing partisans.

Moral philosophers have traditionally identified personhood with relatively high moral status (Darwall, 2006). If an agent is a person, a greater variety of actions toward that agent are objectionable, and so the agent has higher moral status. Likewise, those same actions towards an agent without such status would be permissible. This moral status explanation coheres well with the partisan dehumanization finding. If co-partisans award each other a higher moral status than out-partisans, then we would expect them to be more reluctant to sacrifice fellow in-group members. Our findings do not support the interpretation that agents are never willing to sacrifice co-partisans, as Figure 1 shows that they will sacrifice one of their own to save five of their own at a very high rate (90%). Nevertheless, they might yet regard it as objectionable to sacrifice a higher status agent for a group of lower status agents, as was the case when asked to sacrifice a co-partisan to save a group of five out-partisans, which occurred only 59% of the time.

Other interpretations are possible. One alternative to the moral status explanation is what we might call the partiality explanation: It is permissible to withhold from sacrificing a co-partisan in some circumstances in which it would not be permissible to withhold from sacrificing opposing partisans.

Partiality is distinct from moral status. According to philosophical explanations of partiality, it is the relationship one has to another agent, rather than that agent’s properties, that creates a moral permission (Kolodny, 2010). For example, one might be justified in refusing to sacrifice a loved one to save others, even if they would be willing to sacrifice one stranger to save five. Such a view does not imply that their loved ones differ in moral status from anyone else (Setiya, 2014). With respect to the political science literature, partiality is associated with warm feelings toward co-partisans, whereas negative partisanship, with its associated dehumanization, would predict treating out-partisans as occupying a lower moral status.

A third interpretation is that partisans simply want to maximize the number of their fellow partisans. This might all be political consequentialism: It is morally acceptable to sacrifice a co-partisan or an out-partisan if it maximizes the number of co-partisan lives saved. Although our vignettes gave no indication that increasing the total or relative numbers of co-partisans would serve some greater political end, it is possible that partisans independently attach instrumental value to bolstering their own ranks. Although we find relatively high levels of absolute support for saving opposing partisan groups (between 59% and 75%), there is less support than for co-partisan groups (90% and 92%). As in early trolley studies (and consistent with the philosophical literature), some percentage of the population declines any nonpareto consequentialist sacrifices (Greene and Haidt, 2002: p. 519).

All three explanations can account for the preference to save co-partisans over out-partisans. Only the moral status explanation requires a negative attitude toward out-partisans, as it alone suggests that they are vulnerable to forms of treatment from which in-group members should be protected. If the moral status explanation accounted for the results observed in Figure 1, then we would expect that partisans would not only prefer co-partisans, but also that they would also regard out-partisans as comparable to groups with lower moral status. Members of some groups might be regarded as deserving less moral protection for a variety of reasons. Some (e.g., noncitizens and illegal immigrants) may be perceived as lacking a legal basis for some moral protections (Utych, 2018; Louis et al., 2013). Others (e.g., convicted criminals) may be thought to have alienated their protections against punishment (Bastian et al., 2013). Others (e.g., neo-Nazis) represent views that deprive other persons of moral status. Finally, members of some groups are not persons at all (e.g., dogs).

Figure 2 shows the proportion of trolley diverting to save 5 co-partisans and 5 out-partisans. Among all respondents, a sacrificial out-partisan is the most likely to trigger a “diversion” to save 5 co-partisans and is second only to neo-Nazis when saving 5 out-partisans.

Figure 2.

Trolley problem, partisans versus other groups: Rates of sacrificing one to save five among seven different sacrificial individuals. Stars indicating statistical significance (p < 0.05) are in reference to sacrificing an out-partisan within each panel. The top panel shows rates of sacrificing to save 5 co-partisans. The bottom panel shows rates of sacrificing to save 5 out-partisans. Vertical hash marks indicate 95% confidence intervals, with standard errors clustered by respondent.

These results suggest that among other candidates for lower moral status, out-partisans sit quite comfortably among neo-Nazis, criminals, and dogs.

Figure 3 shows whether these preferences differ by the party of the respondent pulling the hypothetical lever. The top panels show rates of sacrificing one to save five co-partisans. The bottom panels show the rates for saving five out-partisans. Republican respondents are shown on the left panels, and Democratic respondents are shown on the right panels. For Republicans, we find no statistical difference in how they respond to saving a group of fellow Republicans (top left panel). When saving a group of Democrats (bottom left panel), Republicans exhibit a strong aversion to sacrificing one of their own (60%), but otherwise show no statistical difference among other sacrificial candidates. In other words, Republicans display no partiality toward Democrats (92%), even relative to neo-Nazis (93%), convicted criminals (94%), or golden retrievers (88%).

Figure 3.

Trolley problem, partisans versus other groups, by party of respondent: Rates of sacrificing 1 and save 5 among seven different sacrificial individuals by party of survey respondent. Stars indicating statistical significance (p < 0.05) are in reference to sacrificing an out-partisan within each panel. The top panel shows rates of sacrificing to save 5 co-partisans. The bottom panel shows rates of sacrificing to save 5 out-partisans. Hash marks indicate 95% confidence intervals.

Democrats, when saving five fellow Democrats (top right panel), treat Republicans (93%) on a par with neo-Nazis (91%), convicted criminals (87%), and surprisingly, fellow Democrats (90%). They are less willing to sacrifice an immigrant (81%), non-citizen (85%), or dog (81%) to save five fellow Democrats. When saving five Republicans (bottom right panel), Democrats show more discrimination across the seven different options. Neo-Nazis are most likely to be sacrificed (83%), followed closely by a Republican (71%). A fellow Democrat as the sacrifice to save five Republicans is the least likely outcome (58%). In short, of those groups considered, Republicans care only about shared Republican identity. Democrats might as well be criminals or Nazis. For Democrats, Republicans are slightly favored to Nazis, but only to Nazis.

Discussion

Our findings offer evidence that partisan loyalties do extend to moral judgments. Negative partisan attitudes appear reactive—directed toward opposing partisans themselves, rather than merely targeting circumstances of inter-partisan interaction. Finally, these attitudes appear quite serious. People treat out-partisans comparably to other dehumanized and denigrated groups. Partiality to co-partisans cannot explain the comparison between out-partisans and the most extreme outgroups we considered. Congruent with other findings affirming the pervasiveness of negative partisanship, our results appear driven at least in part by negative attitudes toward political opponents. In our case, these negative attitudes include not only affect but also the judgment (at least, the expressed judgment) that out-partisans occupy a lower moral status.

Our result considers the total effect of partisan identity. Because stereotypes about opposing partisans are unreliable (Ahler and Sood, 2018) and negative affect may be partly driven by partisan misperception (Lees and Cikara, 2021), further work would be needed to determine how much the result results from partisanship alone—independent of overlapping identity categories.

Partisan violence is not a new phenomenon in American politics (e.g., Kalmoe (2020)). What, if anything, might justify political violence (or threats of such violence) is, of course, a further normative question. At the outset, we noted a normative aspiration to civic friendship as an ideal of shared citizenship. Our results tend toward pessimism about this normative ideal. There is little indication that partisans invest much positive value in shared citizenship. The idea that co-citizens, even of opposing political tribes, share a common project of ruling together, and further that this common project gives them special obligations to each other, is absent from our picture (Scheffler, 2010; Kolodny, 2014). Insofar as they require that opposing partisans share a valuing relationship (Scheffler, 2005; Rawls, 2005; Viehoff, 2014), normative theories of citizenship look untethered from political reality.

However, other normative theorists affirm a distinctive normative value to partisan attachment. These theorists see partisanship as an expression of a political commitment that makes ongoing political action possible (Ypi, 2016). Our results offer grounds for a more sanguine perspective on this value; however, our findings also offer a cautionary note for proponents of partisan loyalty. Such bonds appear not to be constituted merely by partiality to one’s political allies or ideas. They include, as well, a willingness to compare opponents with disliked and even reviled groups. This may extend to seeing them as less deserving of moral concern. The partisan ideal may be one about which one might be appropriately cautious—and not only when approaching a trolley crossing.

Supplemental Material

Supplemental Material—Partisanship and the trolley problem: Partisan willingness to sacrifice members of the other party

Supplemental Material for Partisanship and the trolley problem: Partisan willingness to sacrifice members of the other party by Michael Barber and Ryan Davis in Research & Politics.