Voters’ Partisan Responses to Politicians’ Immoral Behavior
Abstract
Politicians’ moral behaviors affect how voters evaluate them. But existing empirical research on the effects of politicians’ violations of moral standards pays little attention to the heterogeneous moral foundations of voters in assessing responses to violations. It also pays little attention to the ways partisan preferences shape responses. We examine voters’ heterogeneous evaluative and emotional responses to presumably immoral behaviors by politicians. We make use of moral foundation theory’s argument that people vary in the extent to which they endorse, value, and use the five universally available moral intuitions: care, fairness, loyalty, authority and sanctity. We report on a 5 × 3 between‐subjects experiment asking a random sample of 2,026 U.S. respondents to respond to politicians’ violations of different moral foundations. We randomly vary which of the five foundations is violated and the partisanship of the actor (Republic/Democrat/Nonpartisan). Results suggest that partisanship rather than moral foundations drives most of U.S. voters’ responses to moral foundations violations by politicians. These foundations seem malleable when partisan actors are involved. While Democrats in this sample show stronger negative emotional response to moral violations than Republicans, partisans of both parties express significantly greater negativity when a politician of the other party violates a moral foundation.
Immoral behavior by politicians is nothing new. The candidacy and subsequent election of Donald Trump to the U.S. presidency seems to suggest that in the current American political environment, moral violations may be more rule than exception. During his campaign, Trump was accused of sexual misconduct as a tape surfaced where he talked about “grabbing them by the pussy,” while a number of women came forward accusing Trump of inappropriate and sexually harassing behaviors that in other times would have doomed his candidacy. Trump routinely verbally violated a wide range of moral norms during his campaign, for example, mocking a reporter for his disabilities and referring to a former Miss Universe contestant campaigning for Hillary Clinton as “Miss Piggy.” Even after the election, Trump continued to violate basic moral positions that might have sunk other presidents. Recently, Stormy Daniels, a pornographic movie star, alleged that she had had an affair with Trump and he paid her to cover it up just before the November 2016 vote. Unlike John Edwards, the 2008 Democratic presidential candidate whose campaign was doomed when an affair came to light, Trump continues with little obvious impact on his favorability ratings.
Each of the above would be considered clear moral violations by moral foundation theorists, in particular, violations of the foundations of “care” and “sanctity” (Graham et al., 2011; Haidt & Graham, 2011). Among Trump’s most consistent supporters are religious conservative voters who generally perceive themselves as high in morality in everyday life, while identifying as strong Republicans when it comes to politics. This apparent conflict between moral values and partisan preferences suggests a need to examine the link between voters’ endorsements of moral foundations and their responses to moral transgressions of those foundations by politicians. If moral foundations are, in fact, innate and foundational, voter’s moral values should dominate when a foundation is violated. A voter who strongly cares about a particular moral foundation should react negatively to its violation, regardless of the party of the politician involved. And yet, as the Trump example makes clear, there is reason to question this belief. Over 90% of Republican voters supported Trump in 2016, despite his continual violations of moral foundations, and presumably in opposition to their own support of those same foundations.
We wish to examine the extent to which underlying moral values subscribed to by American voters condition responses to violations of moral foundations by politicians. We consider whether the extent to which people care about moral foundations like care, fairness, loyalty, authority, and sanctity influences their negative emotional responses to violations. Alternatively, given the partisan nature of American politics in the early twenty‐first century, it may be that partisan agreement is more important than moral foundations. We seek to answer the question of whether partisanship in America also extends to the point of overriding, or at least reducing, the effect of underlying moral values.
Partisanship is a core feature of the American political system. It acts as a perceptual screen (Campbell, Converse, Miller, & Stokes, 1960) coloring how partisans view all aspects of politics. More recent research on motivated reasoning (Lodge & Taber, 2013; Redlawsk, 2002; Taber & Lodge, 2006) provides a mechanism for this process, as existing affective evaluations—such as partisan preference—influence the cognitive processing of relevant information. As a result, in an era of increasing partisan and social sorting, partisan preference may provide a great deal of cover for politicians who violate moral foundations. Mason (2018) documents how American social and political identities have recently aligned so that previous crosscutting cleavages have all but disappeared. As a result, partisanship is now reinforced by other social identities, including religious identities. There may be good reason to think that moral foundations themselves can become subsumed within partisan identity, so that violations of foundations by the “other side” are much worse than violations by “our side.”
At the same time, studies have shown that exposure to scandals depresses voters’ candidate evaluations (Bhatti, Hansen, & Olsen, 2013; Carlson, Ganiel, & Hyde, 2000; Doherty et al., 2011) and reduces trust in political institutions and the political process (Bowler & Karp, 2004; Maier, 2011). Politicians’ moral transgressions are extensively covered by the media (Fogarty, 2013). This does not appear to have changed in recent years; the allegations against Trump were certainly front and center in media reports during the 2016 presidential campaign. Moreover, following Trump’s election, allegations of sexual harassment against other powerful men in and out of politics spurred the #metoo movement, as women recounted their experiences. Former Sen. Al Franken, thought to be in the mix for the presidential campaign in 2020, was forced to resign, while others have also lost positions of power. Yet the very same kinds of claims against Trump did not, in the end, derail his candidacy, nor so far, his presidency.
Emotions play an important role in moral foundations theory (Haidt, 2003) and are key factors in voters’ moral judgments about politicians and institutions (Ben‐Nun Bloom, 2014; Bowler & Karp, 2004). Emotions not only often guide moral judgments, but also shape voting behavior. Emotions have been found to underpin political campaigns generally (Marcus et al., 2000; Redlawsk, 2006) as voters both think and feel about politics. Negative emotions can be especially important. Among the many aspects of politics that might trigger emotional responses, we would expect violations of moral foundations by politicians to be central, with voters expressing negative emotions about violators, but only to the extent that the voters themselves care about any given foundation that has been violated.
Despite the prominent role of emotions in explaining political behavior and numerous studies examining effects of politicians’ violations of moral standards, little attention has been paid to the intersection of the two, that is, how voters respond emotionally to politicians’ moral violations (the notable exceptions are Halmburger, Rothmund, Schulte, & Baumert, 2012 and Jiang et al., 2011). In addition, to our knowledge, no study has examined how heterogeneous preferences for moral foundations condition how voters respond to politicians’ moral transgressions.
This study thus aims to answer three research questions: (1) How do American voters respond emotionally to violations of moral foundations by politicians? (2) Are voters’ emotional responses conditioned by their own moral values? and (3) Does partisanship influence the negative emotional responses voters have to violations of moral foundations? To examine these questions, we conducted a 5 × 3 between‐subjects experiment with an online random sample of about 2,000 U.S. voters. We manipulated the moral foundation violated by a politician (care, fairness, loyalty, authority, and sanctity) and the partisanship of the politician involved (Republican, Democrat and no partisan label).
We find voters express negative emotional responses to politicians’ moral transgressions, but the level of negativity is strongly conditioned by partisanship. Democratic voters have stronger negative emotional responses to many of these moral violations than do Republicans. At the same time, partisans of both parties express more negative emotions when a politician of the other party violates moral foundations, all else equal, while responding more similarly to a nonpartisan actor. Finally, while we anticipated that a voter who endorses the values of a particular foundation to a greater degree would be more negative when it is violated, this effect was and clearly less than the effects of party when partisan actors were involved.
The remainder of this article proceeds as follows. First, we discuss moral foundation theory and the role of emotions. Second, we summarize the literature explaining individual variance in response to politicians’ immoral behaviors and develop hypotheses from this literature. Third, the experimental design, analysis strategy, and operationalization of the variables are discussed. Finally, results are presented and conclusions are drawn.
Moral Violations and Emotions
We build on two strands of literature: the scandal literature from political science and literature on (moral) emotions and moral political judgements from social psychology. Moral judgment is the evaluation of an act as morally wrong or right (Ben‐Nun Bloom, 2014). Moral transgressions, that is, harm to others’ welfare, are thought to be inherently wrong since they have an intrinsic effect on the well‐being of others (Ben‐Nun Bloom, 2014). Moral transgressions by politicians can become scandals, although the word “scandal” itself does not refer to the moral transgression, but to the communicative event surrounding the moral transgression becoming public (Lee, 2015).
Moral foundation theory (MFT) sees moral judgment as an intuitive process characterized by automatic affective reactions to stimuli (Clifford, Iyengar, Cabezzam, & Sinnott‐Armstrong, 2015). This is in line with the social intuitionism model of morality (Haidt, 2001) which argues that people know intuitively whether acts are right or wrong. They are capable of swift judgment of an (im)moral act, but they take considerably more time to come up with a rationale when asked to explain their judgment (Haidt & Hersh, 2001). Haidt and Hersh (2001) argue that intuitions and emotions most often precede and guide moral emotions.
MFT categorizes moral intuitions into five foundations: care, fairness, loyalty, authority, and sanctity (Haidt & Graham, 2011). Care refers to the dislike for the suffering of others; fairness to a commitment to fairness and justice. Loyalty is seen as a commitment to one’s own group. Authority refers to respect for authority and tradition, and sanctity refers to concerns with purity and contamination. People differ in the extent to which they endorse these five values, and thus MFT also provides an understanding of moral diversity (Graham et al., 2011). MFT extends most scales used in moral psychology as it does not limit the moral domain to concerns about individuals harming or unfairly treating other individuals (Graham et al., 2011). Moreover, MFT is meant to cover the full range of moral concerns, including those found in non‐Western cultures, in religious practices, and among political conservatives (Graham et al., 2011). Studies have found that political liberals and conservatives differ in the weight that they place on the various moral foundations (Graham et al., 2009; Haidt & Graham, 2007). Specifically, liberals have been found to rate considerations of care and fairness as significantly more important moral values than loyalty, authority, or purity. To liberals, acts are perceived as immoral primarily to the extent that they harm others or treat people unfairly.
There appear to be only two studies that have examined people’s emotional responses to moral transgressions by politicians (Halmburger et al., 2012; Jiang et al., 2011). Both studies report that exposure to a political scandal generates negative emotions towards the politician involved. Halmburger et al. (2012) incorporates specific moral emotions in their study, reporting higher levels of anger and shame when subjects are exposed to a news report including a politician’s moral transgression. They also find that negative moral emotions stimulate need for retribution versus need for restoration of the moral transgressing politician (Halmburger et al., 2012). But these studies are of limited generalizability since they do not effectively account for the role partisanship plays in conditioning responses when partisan actors are involved.
Although moral foundation theory is a prominent theory, it is not uncontroversial. Various scholars criticize the assumptions underlying MFT, such as the innateness and stability of moral foundations (Smith, Alford, Hibbing, Martin, & Hatemi, 2017), the existence of five or six distinct moral foundations underlying moral judgment (Schein & Gray, 2018), and the strength and direction of the relationship between moral foundations and political predispositions (Ciuk, 2018; Smith et al., 2017). Most recently, Connors (2019) reports that political values—like moral foundations thought by most scholars to be core beliefs—are readily influence by the social environment. Even with this, the theory is well enough established with key implications for politics that call for testing it in the political context we do here, following work by Clifford et al. (2015).
Heterogeneous Responses to Politicians’ Moral Transgressions
Moral transgressions by a politician should signal to voters that he or she is an immoral candidate, which should negatively affect the candidate’s electoral prospects. If it were that simple, we would have little to examine here: Voters would simply punish those who violate moral standards, with those feeling more strongly about a given moral foundation responding more negatively. However, politicians embroiled in scandals are not always electorally punished for their moral transgressions, and individual voters’ responses to such transgressions differ in strength (Fernández‐Vázquez, Barberá, & Rivero, 2016). This has puzzled scholars and stimulated research trying to understand the psychology of the public’s heterogeneous reactions to scandals (e.g., Fischle, 2000; Halmburger et al., 2012; Lee, 2015).
Numerous factors are mentioned as potential sources for this variance in voters’ responses. Voters may respond differently to different types of scandals (Bhatti et al., 2013; Carlson et al., 2000; Doherty et al., 2011; Fernández‐Vázquez et al., 2016). Thompson (2013) distinguishes three types of scandals, namely sex scandals, financial fraud scandals, and corruption scandals. Financial scandals are punished more severely than sex scandals (Brenton, 2011; Carlson et al., 2000, Funk, 1996), although Doherty et al. (2011) notes this holds only as long as the sex scandal does not involve abuse of power. The identity of the politician involved matters as does the politician’s response to the moral transgression (Lee, 2015; Tiedens, 2001). Gender appears related to voters’ judgments (Brenton, 2011), but probably in combination with the type of scandal (Carlson et al., 2000; Smith et al., 2015).
Other research has shown that trait impressions and prior affect for the politician influence voters’ responses (Fischle, 2000; Funk, 1996). In judging a politician’s moral transgression, Funk (1996) argues that perceived competence matters more than perceived warmth, but only for the more politically knowledgeable voters (Funk, 1996). Recently, Laustsen and Bor (2017) have shown in an electoral context that warmth is the most influential candidate trait on which people judge politicians, perhaps challenging Funk. It also matters how credible voters perceive the information about a scandal—especially when there are claims that the politician committed the transgression intentionally (Anduiza, Gallego, & Munoz, 2013; Lee, 2015). The relevance and importance of the transgression also influences voters’ responses (Anduiza et al., 2013; Lee, 2015). These perceptions are also affected by how their news sources and the media in general frame the scandal (Peterson & Vonnahme, 2014; Shah, Watts, Domke, & Fan, 2002).
Finally, and especially relevant for our study, political identity in the form of partisanship may influence voters’ perceptions of politicians’ immoral behavior (Anduiza et al., 2013; Bhatti et al., 2013; Blais et al., 2010; Fischle, 2000). Partisan preferences can engage motivated reasoning processes that lead voters to discount or otherwise accept behavior from politicians who share those preferences, that they would not for politicians from another party (Kunda, 1990; Redlawsk, 2002). People selectively process information in ways that enable them to arrive at conclusions congruent and congenial to their prior beliefs, including political beliefs (Fischle, 2000). This process can readily lead to partisans rejecting information about immoral behavior by a copartisan politician as not credible. Even when they acknowledge the moral transgression, partisan voters might still bear a less negative judgment about their party’s candidate. While partisanship may not affect perceptions of the facts of the scandal, it may still affect political judgment (Blais et al., 2010).
The usual assumption is that partisan bias works both ways, so partisans perceive their own party more positively and other parties more negatively. However, Blais, Gidengil, and Kilibarda (2017) argue that the partisan effect is asymmetrical, although they note there has been little systematic investigation of how symmetric (or asymmetric) it might be. They find that partisans view their own parties as less corrupt than do nonpartisans, but they do not necessarily view other parties as more corrupt. Anduiza et al. (2013) also find an asymmetrical effect, arguing that moral transgressions are judged differently by voters depending on whether the politician involved is a member of the respondent’s party, rival party, or of an unknown affiliation. However, not all studies find this partisan effect when it comes to how voters process politician’s moral violations (Halmburger et al., 2012). Some find that political sophistication interacts with this partisan bias, and the partisan bias is absent among the more politically sophisticated.
Hypotheses
Considering the prominence that MFT has gained in social psychology, it seems surprising that political scientists have not used it yet to try to explain voters’ responses to moral violations by politicians. Certainly, moral violations occur, and voters historically have seemed to care about them, even if responses might be tempered for one’s own party. While there is evidence that partisans on different sides of the aisle see different moral foundations as salient (Haidt & Graham, 2011), examining all five foundations should let us get a better understanding of how voters respond to their violations and in particular, the extent of negative emotions generated by violations. Thus, the literature we have reviewed above leads us to propose the following four hypotheses:
H1 (Partisanship and Negative Emotions Hypothesis): Across parties, respondents will have negative emotional responses to politicians committing moral violations, all else equal. But based on work by Haidt and Graham (2011), Democrats (typically liberals) will show stronger negative emotions in response to violations of care and fairness specifically, compared to Republicans. Given no prior evidence of partisan effects, we do not have specific expectations about partisan responses to the other three foundations: loyalty, authority, and sanctity.
H2 (Moral Values Hypothesis): Negative emotional responses to violations of moral foundations by politicians will be conditioned on voters’ own endorsements of particular moral values. The more that respondents endorse a particular moral value, the stronger their negative emotional response will be when a politician violates that particular moral foundation.
H3 (Partisanship Interaction Hypothesis): Partisan respondents will be less negative about violations of moral foundations by politicians of their own party, compared to violations by out‐party and nonpartisan politicians committing the same violation.
H4 (Moral Values by Partisanship Interaction Hypothesis): Moral foundations are thought to be based on innate, evolutionarily developed intuitive ethics, where “moral judgment is caused by quick moral intuitions” (Haidt, 2001, p. 817). At the same time, in politics, we know that partisanship acts in many ways as a perceptual screen (Campbell et al., 1960), conditioning how voters respond to partisan information. Thus, when partisanship is not invoked in a moral foundation violation, we expect the strength of a given moral value to drive emotional response to it. However, when the actor is a partisan and so is the voter, we expect that partisan preference will moderate these effects.
This leads to a testable hypothesis: Respondents will express a lower level of negativity toward co‐partisans violating a given foundation, compared to a nonpartisan or other party actor, at all levels of moral values strength. That is, even though voters with stronger moral values should be more negative to violations of the corresponding moral foundation (H2), partisan‐motivated reasoning should moderate these effects. However, for a nonpartisan actor, respondents who more strongly support a given moral value will be more negative about its violation than those for whom the value is less important.
Experimental Design
To examine voters’ emotional responses to moral violations by politicians, we conducted a between‐subjects vignette experiment in a 5 (moral foundations) × 3 (levels of partisanship) design embedded in a survey of just over 2,000 U.S. voters. Using random assignment, we presented each participant with one of 15 pretested short vignettes describing a fictional, but realistic sounding scenario in which a politician’s behavior violated one of the five moral foundations. The vignette was preceded with a simple instruction: “We would like to have you consider an action of a politician that you might observe. Please read the statement and answer the questions that follow it.” We independently manipulated the moral foundation violated (care, fairness, loyalty, authority and sanctity) and the partisanship of the politician (Republican, Democrat, nonpartisan). The nonpartisan treatment allows us to estimate the effect of partisan labels on voters’ response to moral violation. Full details of the 15 conditions are shown in Table S2.1 in the online supporting information, while Section 3 of the online supporting information describes the stimuli seen by the study participants.
The five vignettes used here were chosen from a pretest of 25 vignettes using a sample of 648 U.S. respondents recruited through Amazon MTurk. Each vignette represents a violation of one of the five moral foundations. The original pool of 25 vignettes were developed by building on Clifford et al.’s (2015) standardized vignettes. Five vignettes were tested for each moral foundation. The vignettes chosen for this experiment best represented the moral foundations while also being perceived by participants as understandable and realistic scenarios. The online supporting information provides a description of the pretest and the selection process of the stimulus material. The stimuli chosen for the experiment were perceived as correctly representing the intended moral foundation by the experiment’s participants; see Table S2.2 in the online supporting information for details. All vignettes were viewed as highly credible by the pretest sample, with 71% to 85% indicating they could very easily imagine the vignette as occurring.
The experimental study was conducted online using Qualtrics software. A sample of 2,026 respondents living in the United States were surveyed through Survey Sampling International (SSI), a market research firm. The sample was recruited to closely match the adult population of the United States on age, gender, race/ethnicity, income, and region of residence, and successfully did so. Data were collected between August 11 and August 20, 2017. See Table S2.3 in the online supporting information for sample characteristics.
We presented each respondent with a single vignette and obtain comparability in our models by using statistical controls for subject characteristics in order to obtain unit homogeneity (King, Keohane, & Verba, 1994). This subsequently allows for analytical separation of the effect of each of the factors that define the vignettes on the evaluative responses of the respondent.11 On average, there is no need to include control variables in models estimating the effects of treatments on subjects who are randomly assigned to treatment groups. However, this is “on average,” which means that if we were to randomly assign an infinite number of respondents to these treatment groups, the distribution of characteristics in each condition would be perfectly equal. However, any particular study represents only one such attempt and is therefore subject to the vagaries of chance. The smaller the number of subjects and the larger the number of conditions the more likely it is that the groups exposed to different conditions differ in terms of their composition. For example, there are differences in gender across our conditions. Therefore, it is a safe strategy to include relevant characteristics in the model. We also ran the models without controls and found no significant difference in results. By presenting respondents with a single vignette, we eliminate potential response effects of earlier presented stimuli on later stimuli. After the respondents were exposed to a vignette, we asked them to report the extent they experienced feelings of anger, anxiety, enthusiasm, pride, hope, shame, disgust, contempt, admiration, sympathy, sadness, optimistic about humanity, warm‐hearted and uplift, with the order randomized to minimize order effects. Finally, we gathered sociodemographic information and information on respondents’ own partisanship and support for the values that underlie the five moral foundations. Section 5 of the online supporting information presents the complete survey question wording and order.
Analytical Design
Although all 2,026 respondents were exposed to a vignette, we did not force respondents to answer all questions, and thus we have some missing values. For our analyses, then, we excluded those respondents that had a missing value on the dependent variables measuring emotion or on the independent variables included in our analyses, such as sociodemographic variables and moral values, leaving 1,918 cases for analysis. Randomization checks confirm that conditions were balanced on pretreatment covariates (see Table S2.4 in the online supporting information).
While we collected both positive and negative emotional responses to the vignettes, we focus here on negative emotions in order to understand the extent to which violations of moral foundations repel voters. As we describe in detail below, we combined the set of six negative emotions for the purposes of this article to create a single negative‐emotions scale. Accordingly, the dependent variable has a range of 0 to 30 and thus allows us to make use of simple ordinary least‐squares regression models. The large number of observations in our sample lets us test for main effects and key interactions to test our hypotheses. Because we have three‐way interactions in our model, we will present the higher‐order interactions using predicted values showing the net differences in the negative‐emotions scale of exposure to each vignette compared to the baseline, by partisanship and moral values.
Variable Operationalization
We asked respondents about the standard nonmoral negative emotions that are part of most studies on emotions (anxiety and sadness) and added a subset of negative moral emotions drawn from Haidt (2003); contempt, anger, shame, and disgust. While not used in this article, we also asked about positive emotions—hope, enthusiasm, admiration, and three indicators of “elevation” (Algoe & Haidt, 2009)—warm‐hearted, optimistic about humanity, and uplift.
All emotions are measured as ordinal variables on a 5‐point scale, where 0 is “not at all” and 4 is “extremely.” Table S2.5 in the online supporting information reports the correlations between the emotions, while Tables S2.6 and S2.7 report the result of a Mokken (1971) scale analysis of both positive and negative emotions. The analysis shows the emotions are indices of two underlying dimensions, namely positive and negative affect, with homogeneity coefficients indicating strong scales, respectively, 0.56 and 0.68. While each specific negative emotion may have differential responses, for our purposes the focus here is on negative affect more generally. Disentangling each emotion may provide detail, but it does so at the cost of multiple complex models that do not increase our basic understanding of how moral transgressions influence negative emotions about candidates. Thus, we collapse all of the negative emotions into a single scale for the analyses in this article.
To measure participants’ own support for moral values, we use the most prominent instrument, Graham et al.’s (2011) Moral Foundations Questionnaire. This questionnaire measures the degree to which individuals endorse each of the five intuitive systems posited by the moral foundations theory (MFT). The Moral Foundation Questionnaire contains 30 items in two parts. In the first part, participants rate how relevant each of the 15 concerns are to them when making moral judgments. In the second part, participants rate their agreements with statements that embody or negate each foundation (Koleva, Graham, Iyer, Ditto, & Haidt, 2012). The resulting scale covering the full range of human moral concerns is found to be reliable and valid (Graham et al., 2011). For the analyses, we make use of the subscales of the moral foundations that were measured with six items each, the scores run between 0 and 5.22 See http://www.moralfoundations.org/questionnaires for how to combine the scores on the questions to come to the subscales. The Cronbach alpha’s of the subscales care, fair, loyalty, authority, and sanctity are respectively, 0.72, 0.68, 0.72, 0.69, and 0.79. We use the results of the moral values assessment to examine how the level of endorsement of a particular value influences the response to the violation of the underlying foundation.33 New work by Montgomery, Nyhan, and Torres (2018) has highlighted potential biasing effects of measuring moderating variables posttreatment in an experimental design, as we have done here. Measuring moral values and partisanship could not be done before treatment, given the very real risk that doing so would prime participants in their responses to the moral foundations vignette, which implicated moral values and in some cases the partisanship of the actor. However, this does not mean that our results are biased, per se. For a treatment effect to be present, the treatment, the dependent variable, and moderators need to share variance. We have examined the shared variance between these variables and find only weak relationships between these variables. In particular, the shared variance between the treatment indicators and the Moral Foundations Questionnaire are negligible, with none higher than r2 = .0025; none are statistically significant. Results of this analysis are available upon request from the first author.
Partisanship is measured using a typical 7‐point scale, constructed from a series of questions that first ask whether the respondent is a Democrat, Republican, or Independent and then asks the strength of the party preference, or for independents, whether the respondent leans one way or the other. In the analyses to come, however, we do not differentiate by strength of party, creating dummy variables representing Republicans, Democrats, and Independents.
Results
Do U.S. voters evidence negative emotional responses to politician’s immoral behavior? We begin by examining the responses across all of our participants to each of our moral values vignettes. We focus here on initial models testing our first two hypotheses: that violations of moral foundations differentially generate negative emotions for Democrats versus Republicans and that these negative emotions are greater when the underlying moral foundation that is violated is one about which voters feel strongly.44 Table S2.3 in the online supporting information describes the sample demographics while Table S2.2 describes the marginal responses to each of the five vignettes, for all three levels of partisanship.
Table 1 presents the partisan and interaction coefficients from a series of OLS models predicting our negative emotions scale. The models include a set of control variables not reported in the table (full models can be found in Table S1.1 in the online supporting information). Each model is defined by the interaction between a vignette and its associated moral value while including responses to all other vignettes and preferences on all other moral values. Table 1 ignores the partisan manipulation to examine Hypothesis 1: Do moral values expressed by voters and the moral violations presented in the vignettes generate negative responses as would be expected by MFT?
| Model 1 | Model 2 | Model 3 | Model 4 | Model 5 | |
|---|---|---|---|---|---|
| Care | Fairness | Loyalty | Authority | Sanctity | |
| Vignette Care × Foundation Care | 1.427****
p = 0.01.
(.410) |
||||
| Vignette Fairness × Foundation Fairness | −.519 (.482) | ||||
| Vignette Loyalty × Foundation Loyalty | .212 (.383) | ||||
| Vignette Authority × Foundation Authority | .921**
p = 0.05;
(.435) |
||||
| Vignette Sanctity × Foundation Sanctity | 1.444****
p = 0.01.
(.295) |
||||
| Republican | .041 (.423) | −.016 (.424) | −.020 (.424) | −.028 (.403) | −.078 (.423) |
| Democrat | 1.208****
p = 0.01.
(.393) |
1.180****
p = 0.01.
(.394) |
1.173****
p = 0.01.
(.393) |
1.169****
p = 0.01.
(.393) |
1.132****
p = 0.01.
(.391) |
| Constant | 5.471****
p = 0.01.
(1.197) |
3.796****
p = 0.01.
(1.204) |
4.379****
p = 0.01.
(1.187) |
4.995 (1.2021) | 4.528 (1.160) |
| Adjusted R square | .169 | .156 | .155 | .157 | .166 |
Note.
- Model: OLS Regression Table displays regression coefficients with standard errors in parentheses. Exposure to the vignette sanctity is the baseline category with exception of model 5. The main effects and control variables Age, White, Hispanic, African, Bachelor degree and Postgraduate degree are not displayed. Full model is available in the online supporting information. N = 1,914.
- * p = 0.05;
- ** p = 0.01.
We begin by looking at differences between Republicans and Democrats in our sample. Recall that Hypothesis 1 expects Democrats to express greater negative emotion than Republicans on the foundations of care and fairness, while anticipating that Republicans will weaker negative responses (compared to independent voters as the baseline). We did not anticipate any partisan effect for loyalty, authority, and sanctity. However, we find that in all models, Democrats express higher levels of negative emotion than Republicans and Independents to violations of all moral foundations. In general, Republicans react significantly less negatively than Democrats and are indistinguishable from independents on this set of vignettes, all else equal. Thus, we partially see the effect anticipated by our first hypothesis, where Democrats are more negatively influenced by violations of care and fairness, while Republicans are not. But in these data, the GOP respondents are also less likely to be negative about the other three foundations, compared to Democrats, an unexpected result.
We turn next to the effects of the vignettes themselves and the underlying moral values each represents. Here we focus on the interaction term for each model, which describes the added effect of viewing a moral value vignette on negative emotional response, conditioned on support for the moral value in question (see Table S1.1 in the online supporting information for main effects). We find that for three of the moral values—care, authority, and sanctity—the effect of viewing the associated vignette significantly increases negative emotional responses when the respondent cares more about that foundation.22 One anonymous reviewer suggested that our analyses controlling for partisanship may be unfair to moral values in hiding their true effects. In order to test this, we reestimated the model in Table 1 without the partisan‐control variable, but retaining all other controls. In the revised model, the coefficients for the moral values remain virtually unchanged. There are no significant differences between the coefficients with and without partisan controls. This analysis is available on request from the first author, and we thank the reviewer for suggesting this test. No such evidence is found for the other two values, loyalty and fairness. In fact, the coefficient of the interaction for fairness is negative, although not significant. Thus, we find only partial evidence for our Hypothesis 2 (Moral Foundations Hypothesis), that a stronger commitment to a moral value leads to a stronger negative emotional response when one sees that particular moral foundation violated.
Our third hypothesis tests the proposition that violations of moral foundations elicit less negative emotion when they are committed by a copartisan than when they are committed by a member of the other party or nonpartisan actor. We expect to see the equivalent for Republicans. However, we might temper this expectation given our earlier finding that Democrats express stronger negative emotions to moral foundation violations than do Republicans and Independents, across all of the values and partisan actors. Thus, it is possible that Democrats will give less leeway even to their own party than Republicans give to theirs.
Tables 2 and 3 present the results of a series of models testing Hypothesis 3 by examining the partisan nature of the vignettes without differentiating the moral foundations involved. Models 1 and 2 in Table 2 show the effects when the respondent and the candidate guilty of the moral transgression are from the same party, using the nonpartisan vignette as the baseline. Our primary interest is the interaction between the partisan nature of the vignette and the respondent’s party. We find for both sets of partisans, knowing that a politician of the same party committed the moral transgression goes a long way toward reducing negative responses, as shown by the significant and substantive negative coefficients. However, summarized across all five moral values, the effects for Democrats are not much smaller than for Republicans; Democrats may show some greater negative emotion towards the specific moral transgressions (Table 1), but they are nearly as mollified when the perpetrator is a fellow Democrat as Republicans are when the transgression is committed by a GOP politician.
| Model 1 | Model 2 | Model 3 | Model 4 | |
|---|---|---|---|---|
| Own Republican Candidate | Own Democratic Candidate | Democratic Rival | Republican Rival | |
| Republican Vignette | 1.243****
p = 0.01.
(.427) |
.157 (.360) | .163 (.360) | −.742 (.467) |
| Democratic Vignette | .487 (.367) | 1.734****
p = 0.01.
(.475) |
−.630 (.438) | −.496 (.368) |
| Nonpartisan Vignette | ||||
| Republican Voter | 2.109****
p = 0.01.
(.493) |
1.053****
p = 0.01.
(.440) |
.110 (.486) | 1.072****
p = 0.01.
(.441) |
| Democrat Voter | 1.496****
p = 0.01.
(.415) |
2.307****
p = 0.01.
(.459) |
1.519****
p = 0.01.
(.415) |
.818 (.469) |
| Nonpartisan Voter | ||||
| Republican Vignette × Republican Voter | −3.083****
p = 0.01.
(.664) |
|||
| Democratic Vignette × Democrat Voter | −2.675****
p = 0.01.
(.644) |
|||
| Democratic Vignette × Republican Voter | 3.139****
p = 0.01.
(.672) |
|||
| Republican Vignette × Democrat Voter | 1.955****
p = 0.01.
(.633) |
|||
| Constant | 13.354****
p = 0.01.
(.755) |
13.342****
p = 0.01.
(.757) |
14.095****
p = 0.01.
(.756) |
14.021****
p = 0.01.
(.761) |
| Adjusted R square | .019 | .017 | .019 | .013 |
Note.
- Model run is OLS regression. Table displays regression coefficients with standard errors in parentheses. The control variables Age, White, Hispanic, African, Bachelor degree, Postgraduate degree, are not displayed. Full model is available in Table S1.2 in the online supporting information. In the models with Nonpartisan, Republican is the baseline. N = 1,918.
- * p = 0.05;
- ** p = 0.01.
| Model 5 | Model 6 | Model 7 | Model 8 | Model 9 | |
|---|---|---|---|---|---|
| Nonpartisan Voter/Democratic Candidate | Nonpartisan Voter/Republican Candidate | Republican Voter/Nonpartisan Vignette | Democrat Voter/Nonpartisan Vignette | Nonpartisan Voter/Nonpartisan Vignette | |
| Republican Vignette | .175 (.362) | −.083 (.392) | −.309 (.372) | ||
| Democratic Vignette | .544 (.399) | .476 (.369) | .309 (.372) | .314 (.372) | |
| Nonpartisan Vignette | −.493 (.437) | −.458 (.465) | .023 (.393) | ||
| Republican Voter | 1.071****
p = 0.01.
(.501) |
1.077****
p = 0.01.
(.442) |
|||
| Democrat Voter | .407 (.348) | .415 (.349) | 1.486****
p = 0.01.
(.417) |
1.269****
p = 0.01.
(.473) |
.409 (.349) |
| Nonpartisan Voter | −.978 (.510) | −1.552****
p = 0.01.
(.512) |
−.698 (.531) | ||
| Republican Vignette × Nonpartisan Voter | 1.394 (.820) | ||||
| Democratic Vignette × Nonpartisan Voter | −.342 (.851) | ||||
| Neutral Vignette × Republican Voter | .0268 (.659) | ||||
| Neutral Vignette × Democratic voter | .614 (.631) | ||||
| Neutral Vignette × Nonpartisan Voter | −1.053 (.811) | ||||
| Constant | 14.772****
p = 0.01.
(.739) |
14.873****
p = 0.01.
(.739) |
14.196****
p = 0.01.
(.769) |
13.981****
p = 0.01.
(.774) |
14.869****
p = 0.01.
(.753) |
| Adjusted R square | .008 | .008 | .009 | .008 | .009 |
Note.
- Model run is OLS regression. Table displays regression coefficients with standard errors in parentheses. The control variables Age, White, Hispanic, African, Bachelor degree, Postgraduate degree, Republican and Democrat are not displayed. Full model is available in the online supporting information. In the models with Nonpartisan, Republican is the base line. N = 1,918.
- * p = 0.05;
- ** p = 0.01.
Models 3 and 4 in Table 2 display the effects when the moral transgression is committed by a politician from the opposite party. Now we see, as expected, that negative emotions are greatly enhanced, exactly the partisan effect we would expect. Here, Democrats seem more condemning than Republicans, showing a significantly larger negative effect in the interaction term. The remaining models, displayed in Table 3, reinforce the partisan nature of these effects. These represent the cases of either nonpartisan voters or nonpartisan vignettes. In all of these models (Models 5–9), we see no interaction effect for partisanship of voter by partisanship of vignette. Overall, these data suggest we cannot reject Hypothesis 3, the Partisanship Interaction Hypothesis.
Finally, Hypothesis 4 tests whether partisanship moderates the effects of the strength of support for a particular moral value. That is, we anticipate a three‐way interaction between the partisanship of the voter, of the political actor, and the extent to which the voter endorses a moral value, such that the increasing negativity that comes from endorsing a moral value and seeing it violated is reduced when a copartisan politician is the one doing the violating.
For purposes of interpretation, we do not present the detailed regression models with higher order interactions here, but we focus on a set of figures drawn from the full models available in Tables S1.3 through S1.8 of the online supporting information, each of which examines one of the foundations with the last examining the nonpartisan vignettes. Because of our experimental design, we do not have a nonmoral foundations, nonpartisan baseline. Thus our analyses compares each moral foundation to the average of the other four foundations, which serves as the baseline. This, plus the fact that the moral foundations themselves vary in their intensity and effects on respondents, means that our analysis focuses within each moral foundation, not across foundations. That is, we are not comparing foundations to each other, but instead the partisan and personal moral values effects within foundations. Our hypothesis does not suggest that one foundation has stronger or weaker effects than another, and so we do not address this question.
Each figure displays the mean predicted negative emotions score from exposure to a given moral foundation vignette minus the baseline negative emotions score for those exposed to the other vignettes. Thus, we chart the net increase in negative emotions from exposure to a specific moral foundation violation compared to the average of the other four vignettes. A positive value signals that exposure to the specific vignette increased negative emotions, and a negative value means it decreased negative emotions. The lines at the top of each bar are the 95% confidence intervals of the means. The size of the effect can theoretically run from − 30 to + 30; in practice it is much more constrained.-3-3 In theory, the baseline emotional response to a vignette could be as high as +30 (all negative emotions felt at the extreme level) or 0, all negative emotions felt not at all. If the baseline were at the extreme in either case, and the specific vignette were felt at the opposite extreme, the scores would be either +30 (vignette felt a the positive extreme, baseline at 0) or −30 (vignette at 0 and baseline at 30.) In practice, the actual results are constrained between +7 and −5 across all five vignettes. The result of interest is not so much the absolute size of any effect, but instead the relative differences in the effects of partisanship and moral values. In these analyses, we examine key effects: in‐party versus nonpartisan and out‐party actors and the effects of personal strength of the referenced moral value. If moral foundations are basic orientations to values, we should see the effects of the individual’s adoption of a foundational value to be stronger than the effects of party.
Turning to Figure 1, we see a clear in‐party/out‐party effect. Republican and Democratic voters both express greater negative reactions when exposed to a politician of the other party violating the foundation care, compared to a politician of their own party, regardless of the extent to which they value the care foundation. We also find that Democratic and Republican participants are equally negative when the other party candidate violated care, but Democrats are much more negative toward their own politician’s violation than Republicans are towards theirs. Republicans respond with more equanimity than Democrats to the care vignette when their own politician violates it. For both party’s voters, the nonpartisan vignette effects are somewhere between the effects for partisan actors, as we would expect.-2-2 We thank the anonymous reviewer who suggested we examine the nonpartisan vignettes more closely than we had originally done. The differences between partisan and nonpartisan vignettes help strengthen our case.
We also find that all of the effects of violating the moral foundation of care are stronger for those participants for whom that particular moral value is stronger, compared to those for whom it is less important. Generally, respondents who emphasize care, that is, have a higher score on that value, generally express more negative emotions when the care foundation is violated, whether or not the politician is a member of their party. This effect is made very clear when examining high and low care results for nonpartisan vignettes. When partisanship is removed, the effects of personal values are much clearer. High‐care Republicans are 2.0 points more negative than low‐care voters, compared to 1.47 and 1.02 points for GOP and Democratic actors, respectively. For Democrats, the difference is even greater, with high‐care voters 2.95 points more negative than low‐care voters versus 1.22 and 1.62 points for GOP and Democratic politicians. When viewing a partisan vignette about the care foundation, voters are particularly negative about the other party, compared to their own, and the strength of their personal care value is much less important.
Figure 2 displays the net difference of exposure to violations of the foundation fairness. Again we see a clear partisan effect. Both Republican and Democratic respondents express additional negative feelings relative to the baseline when exposed to an out‐party politician violating fairness. At the same time, both groups of respondents do not become more negative when their own party politician violates fairness. For Republican respondents, exposure to a violation by their own politician appears to reduce their negative emotions. But, unlike care, we do not find a foundation effect for fairness for Democrats. It does not matter how Democratic respondents score on the fairness value for how they respond to a politician’s violation of that foundation, even when the politician is not presented as partisan. But for Republican respondents we see a curious effect: Scoring high on fairness reduces the negativity associated with either party politician violating it although the effect does not occur when partisanship is not included in the vignette. As with care, the effects of partisanship on fairness are much larger (a maximum of 4.17 points comparing the high‐fairness GOP voter faced with a GOP politician to one faced with the Democratic politician) than the effects of the moral value score (maximum of 1.42 points for a GOP voter learning about a Democratic politician.).
Turning to loyalty, we again find that partisanship plays the more important role, as is seen in Figure 3. Again it is the case that both Democrats and Republicans have more negative feelings about an out‐party candidate violating loyalty than an in‐party candidate. For GOP voters feelings about a nonpartisan politician are somewhere in between. Their negative feelings in comparison to the baseline do not increase when their own candidate commits a moral transgression, and in fact the Republicans are less negative compared to the baseline. We do not find a foundation effect; the extent to which respondents care about the foundation loyalty does not significantly affect their feelings when this foundation is violated.
Figure 4 shows the net difference of exposure to a violations of the foundation authority. We see a clear negative vignette effect. Exposure to these vignettes resulted in fewer negative emotions compared to the baseline average across all other vignettes. Given that we are not comparing across vignettes in this way, the result is interesting, but probably reflects differing strengths of the vignettes themselves. More importantly, we find no effects for partisanship in the violation of this foundation: Democratic and Republican voters respond similarly across all vignette actors. We do, however, see some effects of the strength of the moral value for authority. In general, scoring higher on the authority values results in increased negative emotions on seeing it violated. However, the difference is largest (and significant) for the nonpartisan vignette, where there are no partisan cues. For Democrats only the increased negativity is also significant for their own party candidate, but it is not significant for the GOP candidate, nor is it significant for either candidate for Republican voters. When partisanship is in the mix, the effects of the strength of the authority value are more limited.
When examining the effects of politicians’ violating the moral foundation sanctity, we once more find a strong partisan effect and virtually no effect of strength of the sanctity moral value. Republican and Democratic respondents experience more negative feelings when exposed to a moral transgression of an out‐party candidate versus an in‐party candidate. The effect is stronger for Republicans than Democrats. However, as with authority, when partisanship is not a factor, voters who care more strongly about the sanctity value are much more negative about its violation. This is true for both Republicans and Democrats in our study.
In order to make these results even more clear, we turn to Table 4, where we show the predicted mean differences in the negative emotion scores comparing voters who are high on a moral value to those low on the value (first section of Table 4) and comparing Republicans to Democrats (second section.) Table 4 thus summarizes the comparisons we drew from Figures 1-5 and shows which differences are statistically significant at p < .01. The table reinforces the narrative from the figures: For the most part, partisanship has greater effects on negative emotions than does the strength of preference for a particular value. Only for our authority vignette does partisanship show no effects. Of the nine significant effects for strength of the moral value (out of 30 tests), two‐thirds occur when the actor is not a partisan Democrat or Republican. Only three of the 20 tests for partisan actors are significant.
| Care | Fairness | Loyalty | Authority | Sanctity | |
|---|---|---|---|---|---|
| Net Negative Emotions: Low on Moral Value vs. High on Moral Value | |||||
| Republican Voter | |||||
| Republican Politician | 1.47**
Significant difference, p < .01, t‐test.
|
−1.3**
Significant difference, p < .01, t‐test.
|
0.04 | 1.23 | −0.12 |
| Nonpartisan Politician | 2.00**
Significant difference, p < .01, t‐test.
|
0.18 | −0.29 | 1.96**
Significant difference, p < .01, t‐test.
|
3.53**
Significant difference, p < .01, t‐test.
|
| Democratic Politician | 1.02 | −1.42**
Significant difference, p < .01, t‐test.
|
0.27 | 1.44 | −0.1 |
| Democratic Voter | |||||
| Republican Politician | 1.22 | −0.22 | −0.35 | 0.42 | −.04 |
| Nonpartisan Politician | 2.95**
Significant difference, p < .01, t‐test.
|
−0.53 | 0.54 | 1.50**
Significant difference, p < .01, t‐test.
|
2.17**
Significant difference, p < .01, t‐test.
|
| Democratic Politician | 1.62**
Significant difference, p < .01, t‐test.
|
0.28 | 0.17 | 1.36 | −0.52 |
| Net Negative Emotion: Republican Voter vs. Democratic Voter | |||||
| Low on Moral Value | |||||
| Republican Politician | −6.42**
Significant difference, p < .01, t‐test.
|
−6.41**
Significant difference, p < .01, t‐test.
|
−3.9**
Significant difference, p < .01, t‐test.
|
−.45 | −3.41**
Significant difference, p < .01, t‐test.
|
| Nonpartisan Politician | −2.55**
Significant difference, p < .01, t‐test.
|
0.25 | −1.58**
Significant difference, p < .01, t‐test.
|
.38 | 1.55 |
| Democratic Politician | 3.38**
Significant difference, p < .01, t‐test.
|
3.36**
Significant difference, p < .01, t‐test.
|
4.41**
Significant difference, p < .01, t‐test.
|
.01 | 2.6**
Significant difference, p < .01, t‐test.
|
| High on Moral Value | |||||
| Republican Politician | −6.17**
Significant difference, p < .01, t‐test.
|
−7.49**
Significant difference, p < .01, t‐test.
|
−3.51**
Significant difference, p < .01, t‐test.
|
.36 | −3.49**
Significant difference, p < .01, t‐test.
|
| Nonpartisan Politician | −3.5**
Significant difference, p < .01, t‐test.
|
0.96 | −2.41**
Significant difference, p < .01, t‐test.
|
.84 | 2.91**
Significant difference, p < .01, t‐test.
|
| Democratic Politician | 2.78**
Significant difference, p < .01, t‐test.
|
1.66**
Significant difference, p < .01, t‐test.
|
4.51**
Significant difference, p < .01, t‐test.
|
.09 | 3.02**
Significant difference, p < .01, t‐test.
|
Note.
- Predicted values from Tables S1.3 through S1.8 in the online supporting information. Cells are difference in means for those respondents exposed to a vignette compared to the baseline of all other respondents not exposed to the vignette.
- * Significant difference, p < .01, t‐test.
Taken together, these figures and Table 4, drawn from underlying interaction models, show that when taking into account the strength which participants attach to moral values and the partisanship of an actor violating moral foundations, partisanship usually shows significantly greater effects in generating negative emotions in response to violations across four of the five moral foundations. We thus find evidence suggesting that we cannot reject our fourth hypothesis.-3-3 An anonymous reviewer raised the question of the causal relationship in our model, asking to what extent the effect of the control variables partisanship and voters’ own moral values might have a mediating instead of a moderating effect on voters’ emotional response to politicians’ moral violations. We have examined this by estimating the potential mediating effects with block recursive regression modelling. The difference between the coefficients of the models with and without mediator is not significant and negligible, suggesting that a moderation analysis is appropriate. We thank the reviewer for suggesting this analysis. Details are available from the first author on request.
Discussion and Conclusions
This study set out to answer three questions, namely (1) How do American voters respond emotionally to violations of moral foundations by politicians? (2) Are voters’ emotional responses conditioned by their own moral values? and (3) Does partisanship minimize the effects of violations of moral foundations by politicians of the voter’s own party compared to the other party? We find that in general voters respond with negative emotions to politicians’ moral violations. However, not all voters respond in the same manner; we find that Democrats tend to respond more negatively to this set of moral violations than do Republicans. We might speculate that the political environment in which our study was done could have played a role in this unexpected result. As we detailed at the beginning of the article, there have been many accusations of moral violations by President Trump, none of which seem to shake his core Republican supporters. One impact of this may have been to lessen Republican voters’ sensitivity to moral violations by politicians more generally. Unfortunately, we have no way to test this speculation. We do find that voters’ responses to these moral violations can be sometimes conditioned by their own moral values, but they are more so by their partisanship when partisan actors are involved. Partisans of both parties express more negative emotions when a politician of the other party violates moral foundations.
While we do not have a direct test of the mechanism by which partisanship conditions the effects of moral violations on emotional responses to politicians, a lengthy literature on partisanship in American politics makes clear that partisans see the political world through a very specific perceptual screen (Campbell et al., 1960). Motivated reasoning (Lodge & Taber, 2013; Redlawsk, 2002) likely becomes engaged when a voter sees a copartisan politician violating a moral foundation, leading to biased processing of the event, and the reduction of negativity about the event. But when the other party commits the violation, partisans are more than willing to express negative emotions about the event. Brain imaging studies reinforce the potential of this mechanism as distinct differences are seen between Democrats and Republicans in their processing of political information (Schreiber et al., 2013). Note that while we started this article with a brief discussion about moral violations by U.S. President Trump and his seeming imperviousness to them, it is worth recalling that it is generally only his Republican supporters who accept his behavior.
This study contributes to the literature in various ways. First of all, it is among a handful studies (Halmburger et al., 2012; Jiang et al., 2011) to study emotional responses to moral violations. A follow‐up study will examine the specific negative emotions to see whether these moral violations evoke specific discrete emotions, in particular so‐called moral emotions (Haidt, 2003). However, our main interest here is to see how partisanship and the importance of moral values for the voter affect emotional responses more generally.
Second, this study is the first to assess the role of voters’ moral values in their response to moral transgressions, as the results show they do matter. However, partisanship, more often than not, overrides the effects of moral values. This is a very interesting finding and suggests that moral foundations are maybe not as innate and foundational as might be supposed (see also, Connors [2019] on political values). The partisanship effect in this study is asymmetrical, showing that voters judge the opposing candidate more harshly for moral transgressions than their own candidate. This is in line with the trend of negative partisanship in American politics (Mason, 2018). Research has shown that while the feelings Democrats and Republicans have about their own party have not changed, their feelings about the opposing party have become much more negative (Abramowitz & Webster, 2016).
As with any research, this study is not without its shortcomings, one being the time frame in which it was conducted, namely following the 2016 presidential elections, which led to the election of a politician with frequent moral transgressions. It is not unthinkable that in a different time period we might find differences with respect to the strength of partisan responses to violations of moral foundations. Thus, we must consider that aspect of our results to be conditional, warranting further study in a different context.
Another limitation is that the vignettes used were specific. That is, the politician referred to in a vignette took a specific action. There is probably heterogeneity in people’s responses to these actions, and by the necessity of the research design, each study participant saw only one action representing one moral foundation. It is possible, for example, that a different vignette representing the same foundation might have a different impact on participants’ emotions. However, we are less interested here in the specific reactions to specific vignettes than we are in the larger story, that moral foundations appear to be readily ignored in the face of partisan actors. Across all of the vignettes we use, partisan voters were far less negative about a same‐party actor violating a foundation than they were an opposite‐party actor. At the same time, although extensively pretested, the authority vignette had the lowest homogeneity coefficient and resulted in the weakest findings. Apparently American voters did not find that specific example to be compelling.
Finally, as described in Footnote 2, new research by Montgomery et al. (2018) suggests that appropriate care must be taken in measuring moderating variables in an experimental context. We were concerned that measuring our moderators—partisanship and moral values—before the experimental treatment would prime participants as to the purpose of our study and thus influence their responses to the treatment. Thus, we measured them at the end of the study. While the very small shared variance we report between the moderators and treatment provides some confidence that our results are not biased, future studies should consider ways to separate these measures, such as the use of multiple waves. If the work by Montgomery et al. (2018) is sustained through additional research, experimentalists in general will need to consider whether the potential costs of such strategies outweigh the risks of biasing results by measuring moderators after an experimental treatment.
While recognizing the potential bias suggested by Montgomery et al. (2018) may be present in our analyses, notwithstanding the very small shared variance between our measures, any such bias would matter more if we were making inferences about the direct effects of specific coefficients in our models. But, instead, we are more interested in the patterns that we see in the data when different moral foundations are engaged across partisan voters. To the extent that these patterns of differential response by partisan voters and those who more or less adopt the moral values of the foundation are biased, we would not expect the bias to eliminate the differences we see across groups. Thus, while the effects of any given treatment may be more or less than we find here, we expect any bias would be unidirectional across the treatments, and thus the patterns we find would remain.
Notwithstanding these limitations, this study makes a significant contribution to the moral foundations literature. For the first time, we examine the intersection of partisanship and moral foundations and find that, as with so many other things in American politics in the early twenty‐first century, responses to moral violations by politicians are subject to partisan preferences. This holds even when voters themselves feel strongly about a given moral foundation. Put simply, when the effects of partisanship and strength of support for moral values are tested against each other in predicting emotional responses to violations of moral foundations, partisanship usually comes out the winner. This provides new insight into the role of moral violations in politics and helps us understand perhaps why some American politicians are able to continue to receive strong support from their own party voters even after violating what are thought to be basic moral values.
Acknowledgments
Research presented in this article has been funded with a Nottingham Research Fellowship of the University of Nottingham. We would like to thank the anonymous reviewers, participants of the 2017 APSA Morality Panel and the 2018 MPSA Political Norms and Values Panel for their helpful comments and suggestions to improve this article. In particular, we would like to thank Cees van der Eijk for his methodological support throughout this project. Correspondence concerning this article should be addressed to Annemarie S. Walter, School of Politics and International Relations, Nottingham, NG7 2RD, United Kingdom. Email: Annemarie.Walter@nottingham.ac.uk
