This is an open access article distributed under the terms of the Creative Commons
Attribution License (In 2003, Alabama’s Chief Justice, Roy Moore, violated the Establishment Clause of the United States Constitution by refusing to remove a marble monument of the Ten Commandments he erected in the courthouse rotunda. Alabama’s judicial system removed him from office as a result. Yet voters continued to support Moore; in 2012, he was reelected. In 2016, Moore defied US marriage-equality law, and was again removed from office. Yet voters still supported him; Moore won the Republican nomination to the United States Senate in 2017. However, soon after the nomination, voters learned of three sexual assault accusations against Moore. With this news, Moore lost the election and a Democratic senator was elected for the first time in Alabama in 25 years. Although Roy Moore’s transgression may have crossed a line in the sand for some Republicans, others maintained their support: Moore still earned 48.4% of the Republican vote.
In some respects, Moore’s case is not unique. Voters often continue to support favored candidates after learning that they have committed normative, legal, or moral transgressions. These transgressions have varied in severity and context, and have led to a range of political outcomes. For example, past marijuana-use by Barack Obama and a D.U.I. arrest of George W. Bush did not prevent their elections. Although former President Donald Trump was impeached twice for various offenses (Leatherby et al., 2021) and ultimately lost the presidency in 2020, Trump still maintains supporters, and only seven Republican senators crossed party lines in support of his second impeachment (Sprunt, 2021).
Such cases of voter support amidst candidate transgression raise two important questions that guide the current work. First, are there specific characteristics that motivate some voters to maintain support for transgressive candidates, and other voters to abandon such candidates? Second, is there a limit in transgression severity – a “line in the sand” – at which most voters are no longer willing to support a candidate and, if so, where is it?
To answer these questions, we introduce methodology from the decision sciences. In decision-analytic research in the moral domain, investigators have employed formal models and multi-trial within-person methodologies in which independent variables are manipulated parametrically to model functional relations between varying levels of independent variables and outcomes of interest (see Crockett, 2016a). For example, Crockett et al. (2014) parametrically varied number of shocks and levels of financial payouts in research establishing that people generally place greater value on the avoidance of pain to others than to themselves.
Drawing from such work, the current research introduces a novel paradigm that models the relationship between transgression severity and voter support. We explore why voters differ in their likelihood of supporting transgressive candidates by incorporating moderating sociopolitical factors into the model. By incorporating an existing continuous, numerical scale of transgression severity, and employing a multiple, within-person design, the current model estimates severity thresholds, or the extent of severity voters tolerate before abandoning support for transgressive candidates. For the current work, transgressions are defined as acts that break social norms and moral codes that have actually been committed by candidates, rather than mere accusations or attempts of media scandalization. Although severity can be both an objective and subjective measurement, the current work uses a pre-tested, objective measure of transgression severity.
Throughout the current work, ideological terms (liberal and conservative) and partisanship terms (Democrat and Republican) will be used interchangeably. Although ideology and partisanship possess different definitional meanings, research supports the ideological realignment thesis: the increasing tendency for people to perceive ideology and partisanship as interchangeable concepts (Abramowitz & Saunders, 2006). Indeed, the meaning of political orientation may be largely symbolic, conveying psychological attachment to a group, rather than issue orientation (Camobreco, 2016). Therefore, the current work uses partisanship and ideological terms interchangeably unless otherwise noted.
Ideological Identity
One voter characteristic we explored was strength of ideological identity. Different theoretical approaches address the potential impact of ideological identity on the relationship between transgression severity and voter choice. The current work uses the expressive approach of party affiliation. As posited by social identity theory (Tajfel et al., 1979), people perceive ingroups, such as political party membership, as representations of the self. The status of the political party impacts the perceived status and self-esteem of the individual. Thus, people are highly motivated to feel positively about their political ingroup, and behave in ways to maintain their political party’s success.
The expressive approach also posits that members of a political party tend to share group membership in additional identity domains, such as race or religion. These additional affiliations strengthen emotional identity with the ideological group (Fowler & Kam, 2007); ideological identity thus is viewed as fundamentally grounded in broader socioemotional identities. By contrast, an instrumental approach focuses on policy considerations. It contends that people affiliate with a given party because they feel strongly about that party’s ideological beliefs and policies, irrespective of the additional social identities associated with party membership.
Empirical findings that compare these views support the expressive claim that social identity concerns (i.e., is this a Republican or Democratic policy?) drive most political decisions (Huddy et al., 2015) and behaviors (Fowler & Kam, 2007). More so than attitude strength towards specific policy issues, identification with a political party positively predicted political participation, and emotions of enthusiasm or anger when faced with a political loss or victory (Huddy et al., 2015). In a dictator game, participants could distribute money to three people, whose only stated attribute was their political identity (i.e., Republican, Democrat, or Anonymous). Participants who strongly identified with their political party more frequently allocated all available money to a fellow ingroup member, simply because they belonged to the same political ingroup (Fowler & Kam, 2007).
This emotional attachment to one’s party affiliation suggests that political party membership will strongly influence voters’ choices after learning of ingroup candidate transgression. Specifically, we anticipate that people who are highly identified with their ideological group may be most likely to maintain support for transgressive ingroup candidates, as compared to those who are less strongly identified. Thus, ideological identity strength is included in our model as a potential sociopolitical factor that moderates the relation between candidate transgression and voter choice.
Current historical conditions further underscore the potential significance of political ideology. Findings document increasing ingroup political bias as well as political and affective polarization, (i.e., the increased attitude divergence and animosity between conservatives and liberals) (Mason, 2015). Polarization likely enhances the impact of political identity and ingroup bias on political decision making, increasing the voter’s tendency to engage in motivated reasoning processes that ameliorate the severity of transgression (Iyengar et al., 2019; Rogowski & Sutherland, 2016).
The impact of political ideology in decision making in contemporary American life is also illustrated by work in which an online sample read about fictional political candidates either from their political ingroup, outgroup, or neither, who each violated one of the five moral foundations in office (i.e., violations of harm, fairness, ingroup loyalty, authority, and purity/sanctity) (Walter & Redlawsk, 2019). The biggest impact on participants’ feelings about candidates’ post-transgression was not the type of violation, but whether the candidate belonged to their own political party; the more people identified with their party, the more positively they felt about the transgressive candidate. People who strongly valued the foundations of care, authority, and sanctity reported slightly more negative affect regarding candidates, but this effect was dwarfed by the power of partisanship (Walter & Redlawsk, 2019).
Ideological Identity in the Existing Literature
Displaying the importance of identity strength, a meta-analysis of all existing scandal research found that one of the strongest moderators of public opinion after candidate scandal was the level of voter-partisanship. Although factors such as the sex and race of the candidates, the type of scandal, and response of the politician can impact voter opinion, increased partisanship consistently predicts positive attitudes towards transgressive ingroup candidates (von Sikorski, 2018).
For example, some studies report that fiscal transgressions (i.e., tax evasion, embezzlement) are perceived as more severe than other forms of transgression (Dobratz & Whitfield, 1992; Doherty et al., 2011). Other studies suggest that financial transgressions are more acceptable than candidates engaged in smear campaigns (Gonzales et al., 1995) or sexual transgressions (Newmark et al., 2019). Yet others report only minimal differences between different types of scandals (Carlson et al., 2000; Cucchi & Cavazza, 2021) and find, instead that partisanship is often a stronger driver of attitudes towards candidates than scandal type (Bhatti et al., 2013; Walter & Redlawsk, 2019). These inconsistent findings regarding scandal-type may be a by-product of studies not accounting for the greater socio-political context and political bias of the voters.
Ideology can also influence the perceived intentionality of transgressive acts. Even when highly partisan voters recognize that an ingroup politician is implicated in a transgression, they tend to see the individual as less culpable than others involved in the scandal. Motivated reasoning processes may account for this (Blais et al., 2010). Strongly identified voters may be motivated to attend to details of the transgression that portray their candidate’s intentions in the best light, and thereby may maintain their support (Redlawsk et al., 2010). Such intentionality judgments are particularly significant in that intentional transgressions are perceived as more severe and punishable than accidental ones (Cucchi & Cavazza, 2021; Doherty et al., 2011).
Given the observed importance of political identity in political-decision making across the literature, as well as the current rise in political and affective polarization, it is vital to include ideological identity strength into the proposed model of voter choice after candidate transgression. Because political affiliation is part of personal identity, ideological identity strength likely plays a stronger role in voter choice after transgression compared to other features of the voter. Similar to identity strength, the political party membership of the voter may also determine how people make voting decisions after candidate transgression.
Ideological Differences
In tandem with ideological identity strength, the political identity of the voter, (i.e., whether a voter identifies as Democrat or Republican, liberal or conservative), also likely plays a prominent role in how people make voting decisions after learning of favored ingroup candidate transgression. Specifically, Republican voters may be more likely to vote for severely transgressive ingroup candidates than Democrats. Much like ideological identity strength, the existing scandal literature often does not compare differences between the ideological left and right (e.g., Funk, 1996; Smith et al., 2005; von Sikorski et al., 2020). However, work outside the scandal literature suggests conservatives may be more willing to vote for transgressive candidates than liberals because they place a higher value on ingroup loyalty (Graham et al., 2009; Van Leeuwen & Park, 2009).
According to Moral Foundations Theory, conservatives tend to value the binding foundations, including respect to authority, ingroup loyalty, and purity more so than liberals. Both liberals and conservatives tend to equally value the foundations of harm and fairness, though some work finds liberals value these individualizing foundations more than conservatives (Graham et al., 2009).
The tendency for Republicans to prioritize loyalty above harm is captured in recent work investigating perceptions of prominent politicians. When controlling for perceived ethicality of behavior, Trump supporters perceived Trump’s transgressive actions as more acceptable than liberals perceived Hilary Clinton’s transgressive actions (Morais et al., 2020). Further, after a political scandal in Spain that involved an equal representation of both liberal and conservative leaders, conservative voters were more tolerant of ingroup corruption than liberals (Anduiza et al., 2013). These findings indicate that conservatives, compared to liberals, may be more willing to tolerate transgressive behavior committed by fellow ingroup members if it means that the ingroup party will remain in power. Conservatives may weigh the cost of violating ingroup loyalty as higher than the cost of the potential harm committed by the candidate. In equivalent circumstances, a liberal may weigh the potential harm caused by the candidate as greater than ingroup loyalty.
A Novel Paradigm for Investigating Voter Support After Ingroup Candidate Transgression
We introduce a novel paradigm to identify “lines in the sand” in the relation between transgression severity and voter choice, and to examine how political party and identity strength determine where those lines are drawn. Our paradigm was inspired by research in decision sciences. As Crockett (2016b) explains, formal decision-science models have the potential to inform understanding of social choices in morally-charged circumstances. Such models are generally tested through within-person experimental designs. Rather than employing between-person methods which merely establish that an independent variable has a non-zero effect on an outcome, investigators employ multi-trial within-person paradigms in which independent variables are manipulated parametrically (Zayas et al., 2019). These paradigms enable investigators to model functional relations between varying levels of independent variables and outcomes of interest (see Crockett, 2016b; Crockett et al., 2014).
Such within-person paradigms have the potential to inform the study of transgression severity and voter choice. However, the researcher faces a practical obstacle: Unlike numbers of shocks and amounts of financial payment, moral transgressions do not come with numerical tags attached. Is a D.U.I worse than marijuana use and, if so, exactly how much worse? To harness the potential of formal models in the socially significant domain of transgressions and political choice, one needs to employ a method for quantitatively scaling transgression severity.
The current work provides a novel solution to this problem. We capitalize on data from the National Survey of Crime severity (NSCS; Wolfgang, 1985) to create an ecologically valid, continuously varying scale of transgression severity. We varied severity across experimental trials in order to model the relationship between transgression severity and voter choice, and thereby to identify specific types and levels of transgression at which people abandon ingroup candidates and vote for the outgroup.
Hypotheses
First, we hypothesize that increases in transgression severity will increase the likelihood that people will abandon ingroup candidates. Second, we predict that those with stronger ideological identities will possess higher severity thresholds than those with weaker identities – i.e., a “line in the sand” that allows for more severely transgressive behavior in candidates. Third, we predict that Republicans will possess a higher severity threshold than Democrats, and will continue to vote for transgressive ingroup candidates for increasingly severe transgressions. In order to understand how these lines in the sand translate into concrete, real-life scenarios, we will explore which types of transgressions correspond with each group’s respective severity thresholds.
Method
Participants
Five-hundred Amazon Mechanical Turk workers residing in the United States completed the study. Seven participants were excluded from analysis for failing attention checks, leaving the final sample size at 493. Participants were 40.37% female, aged 19 – 78 (M = 36.85, SD = 11.33), and 11.76% African American, 4.87% Asian American, 75.86% white, 4.06% Latinx, and 1.01% other. Participants identified as either Democrat (63.38%) or Republican (36.63%), and reported their political ideology on an 11-point scale, with 53% identifying as slightly-to-very liberal, 31.85% as slightly-to-very conservative, and 14.6% as moderate (14.60%). To create the measure for ideological identity strength, the 11-point political ideology scale was “folded” to create a 6-point scale that ranged from “moderate-middle of the road”, to “leans”, “slightly”, “moderately”, “much”, to “very much”. We did not measure political party membership strength. However, political ideology and political party were strongly correlated (r = -.79, p < .001).
The functional relationship between transgression severity and voter choice was investigated by implementing a multi-level, highly-repeated within-persons design. Voter choice was regressed across six predictors: transgression severity, political orientation, the interaction between identity strength and political orientation, identity strength, the random intercept of individual and random slope of severity. In order to obtain sufficient power to detect the relationship between voter choice and transgression severity, we selected a sample size based on two criteria. First, we consulted the theoretical literature on multilevel models. Research suggests models are able to make accurate estimations with at least 30 clusters (or trials), and when 50-100 participants are represented in each condition (McNeish & Stapleton, 2016; Moineddin et al., 2007).
Second, sample sizes in previous multilevel studies investigating moral decision-making served as models for the current work. For example, Crockett et al. (2014) varied two parameters across a series of trials; number of shocks and dollar amount, to determine how people weigh the costs of self-inflicted harm compared to inflicting harm on others. Forty-five participants completed 150 trials, resulting in approximately 75 trials and 20 participants per parameter. Because only one parameter is systematically varied in the current work (i.e., transgression severity), 70 trials were created.
Unlike shocks and dollar amounts, the TSS scale involves multiple domains of behavior, potentially leading to greater subjectivity in participant perceptions. The current work also contains a second, between-person variable: political party membership. Standards in psychology suggest collecting at least 100 participants per condition when sample size is estimated to be small to moderate (Brysbaert, 2019). In order to account for political party membership and potential error generated by the transgression severity scale, we collected at least 100 participants per political party. Because representation of Democrats outweighs that of Republicans on Amazon Mechanical Turk (Lewis et al., 2015), we sampled a larger pool of approximately 500 to ensure at least 100 Republicans and 100 Democrats were sampled.
Development of Materials: A Transgression Severity Scale
In preliminary work, we developed stimulus materials in which transgressions varied systematically along a continuous severity dimension (see Supplementary Materials). Key to this effort was the National Survey of Crime severity (NSCS; Wolfgang, 1985), in which 60,000 U.S. residents evaluated each of 204 crimes by rating “how many times more severe” was each offense than stealing a bicycle. We identified 70 NSCS crimes that included topics such as violence, theft, sex, drugs, and perjury. They varied in severity from mild (“a person played hooky from school”) to severe “a person planted a bomb in building and killed 20 people.” Transgressions were selected evenly across the full range of severity, and represented multiple domains of crime. To enhance relevancy, we rephrased a selection of crimes to reflect more modern situations (i.e., lewd texting instead of a lewd phone call).
Because the NSCS severity ratings are now decades old, and some crime descriptions were altered, a pilot study of the severity perceptions of each transgression was conducted. This pretest was designed to yield an accurate scaling of severity for the current social climate. Using the same free-response numerical rating procedure as the original NSCS, participants rated the severity of the updated 70 transgressions across two studies (see Supplementary Materials for details). The average was taken across participants for each transgression to create scores. In each scenario, the gender-neutral pronoun “they” was used to refer to the candidate, as we were not interested in testing how gender impacts perceptions of candidates. The free-response format resulted in positively skewed ratings which were log transformed, yielding a Transgression Severity Score (TSS) scale. TSS’s ranged from log values of .50 to 6.83 (M = 2.24, SD = 1.13). These 70 transgressions comprised the items used in the final TSS Scale1. The correlation between the TSS items and the original NSCS items is strong (r = .51, N = 55, p < .001).
Examples of mild transgressions include items such as “They were caught playing hooky from school before they were 16” (TSS) = 0.74), and “They stole $10 worth of office supplies from their place of work” (TSS = .88). Moderate transgressions include items such as “They broke into a display case in a store and stole $1,000 worth of merchandise” (TSS = 2.09), and “They paid another person to commit a serious crime” (TSS = 2.46). Very severe transgressions include items such as “They forcibly raped a woman. No other physical injury occurred” (TSS = 3.56), and “They owned a factory that knowingly got rid of its waste in a way that polluted the water supply of a city. As a result, 20 people die” (TSS = 4.42). See Supplementary Materials for the full list of transgressions. Due to a coding error in Qualtrics, the transgression “They sent unwanted sexually explicit text messages to coworkers” was dropped from the final analyses.
Procedures
After providing informed consent, participants read instructions explaining that the researchers were interested in “how bad transgressions have to be for you to consider voting for a candidate that is not in your party.” Participants learned that they would encounter 70 “candidate conundrums” in which they had to vote for either a Democratic or Republican candidate in an election that would determine whether “the Democrats or Republicans gain control of the House of Representatives.” The “conundrum” was that, in each trial, self-identified Democratic (Republican) participants read about transgressions by in-group Democratic (Republican) candidates. The outgroup candidate, by contrast, had not “committed any crime or transgression,” and had a “platform representative of the average candidate in their party”. After reading the transgression scenario, participants decided whether to vote for the transgressive candidate or the outgroup candidate. Once all 70 randomly presented trials were completed, participants were thanked and compensated. The research in the current study meets ethical guidelines and adheres to the legal requirements for human-subjects research in the United States.
Results
All data and codes for analyses are available at our OSF page, https://osf.io/y7xbc/. Our primary data-analytic goal was to estimate the relationship between candidate transgression severity, political identity strength, political party membership, and the likelihood to abandon the ingroup candidate after transgression. In order to investigate this relationship, we conducted a multi-level logistic regression. Each participant (Level 2 variable) made 70 voting decisions each (Level 1 variable). Thus, 70 votes were nested within each participant, creating two levels of data. Because the outcome variable was binary (i.e., ingroup or outgroup), a logistic multi-level regression was used.
An advantage of the multilevel model is that it naturally accounts for between-person variation. For example, some people likely possess higher severity thresholds than others, and vary in how consistently they make voting decisions (i.e., sensitivity). These between-person differences in thresholds can be controlled for by including the random intercept of severity in the model, and between-person differences in sensitivity can be controlled for by including the random slope of severity in the model.
In order to test whether the random slope of severity explained significant variation in the data, we compared the model with the random slope included (complete model) to the model without the random slope (initial model). If the complete model explained variation in the data significantly better than the initial model, then we could conclude that the random slope of transgression severity should be included in the final model
To create the initial model, voter choice was regressed on identity strength, the random intercept of participant, transgression severity, political party, and the interaction between transgression severity and political party. Any time an interaction was included in the model, the first order effects were also included. This can be assumed for the remainder of the paper. The random slope of transgression severity was added to create the complete model. A likelihood ratio test comparing parameters and chi-square values revealed that the complete model fit the data significantly better than the initial model, χ2(2) = 2,278.90, p < .001. Thus, the complete model was used for all further analyses. Transgression severity was grand mean centered so that intercepts and interaction terms were interpretable. Therefore, all slopes should be interpreted at the mean level of transgression severity (M = 2.24). Political party was a dichotomous variable, with Republicans coded as 0, and Democrats coded as 1.
Relationship Between Transgression Severity and Voter Choice
We investigated whether people were increasingly likely to vote for the outgroup candidate as transgression severity increased, and whether identity strength and political group membership moderated this relationship. As predicted, when controlling for all other variables, increased transgression severity decreased the likelihood a person would continue to vote for the transgressive ingroup candidate, b = 1.73, z = 16.60, p < .001, OR = 5.61, 95% CI [4.54, 7.40]. The random intercept of participant explained 5.20% (SD = 2.28) of variance, meaning some people possessed low severity thresholds, and others possessed high thresholds.
Demonstrating a low threshold, one participant switched their vote to the outgroup after learning the candidate “was drunk in public at age 16,” (slope: y = 2.31 + 1.32x). In comparison, another participant with a high threshold did not switch their vote for the outgroup until they learned that the candidate “took a bribe to give a light sentence to a criminal case when they were a county judge” (y = -3.19 + 3.77x). Thirty participants never voted for the outgroup. The random slope of transgression severity explained 1.63% (SD = 1.28) of variance, indicating differences in voter sensitivity in the relationship between transgression severity and support of the ingroup candidate. For example, the most sensitive voter consistently voted for the ingroup until a severity threshold was reached, and then consistently voted for the outgroup for remaining items (y = .84 + 8.30x). The least sensitive voter seemed to vote at random across the severity scale (y = -1.37 + -.07x). See Supplementary Materials for plots.
When examining severity thresholds, people tended to consistently vote for the ingroup until a TSS of approximately 1.00, which corresponds to transgression such as the candidate “stole merchandise worth $100 from a department store” (TSS = .98), and the candidate “dressed up in blackface for a Halloween costume when they were in college” (TSS = 1.13). See Figure 1. The model begins to asymptote at TSS = 4.00, indicating that participants have an almost 100% chance of voting for the outgroup when they learn the candidate “robbed a victim at gun point”, and the “victim struggled and was shot to death”. However, 42 participants actually voted for the ingroup candidate across the whole scale of severity, indicating some voters may vote for the ingroup candidate regardless of transgression.
Figure 1
Relationship Between Transgression Severity (Increasing Left to Right) and the Percent Likelihood of Voting for the Outgroup Political Candidate
Note. Error bars represent 95% confidence intervals. The * (TSS = 3) corresponds to the following transgression “Took a bribe of $10,000 from a company to vote for a law favoring the company when they were a legislator.”
Relationship With Moderators
We predicted that stronger ideological identity would increase the likelihood that people would continue to vote for highly transgressive candidates. When controlling for all other variables, increases in ideological identity strength increased the likelihood that a voter would support a transgressive candidate, supporting predictions, b = -.07, z = -7.05, p < .001, OR = .93, 95% CI [.91, .95]. Thus, people with stronger ideological identities were more likely to vote for the transgressive candidate than those with weaker identities.
We also predicted an interaction between political party and transgression severity, such that for increasingly severe transgressions, Republicans would be more likely to vote for the ingroup candidate than Democrats. Supporting predictions, the interaction between political party and transgression severity was significant, β = .60, z = 4.57, p < .001, OR = 1.81, 95% CI [1.33, 2.53] (see Figure 2), and the main effect of political party was not, β = .40, z = 1.79, p = .073, OR = 1.45, 95% CI [.79, 2.11]. Because transgression severity was grand mean-centered, the interaction slope describes the differences between Democrats and Republicans at the mean value of transgression severity (M = 2.24). At this specific level of severity, there is no significant difference between Democrats and Republicans, β = .42, z = 1.89, p = .058, OR = 1.48, 95% CI [.79, 2.30]. However, a visual inspection of the graph, along with the significant interaction slope, indicates that as transgression severity increases, Republicans may be more likely to remain loyal to the ingroup for more severe transgressions than Democrats (see Footnote 1).
Figure 2
Relationship Between Transgression Severity (Increasing Left to Right), Political Party Identification and the Percent Likelihood of Voting for Outgroup Political Candidate
Note. Error bands represent between-person variability in severity thresholds, and voting consistency. Predicted probability of likelihood to vote for outgroup is significantly different for Republicans and Democrats at *, and corresponds to the following transgression: “They paid a witness to give false testimony in a criminal trial” (TSS = 2.6).
When examining Figure 2, there is a widening ideological difference in likelihood to vote for the outgroup at a TSS of approximately 2.6, which corresponds to transgressions such as the candidate “paid a witness to give false testimony in a criminal trial”. Specifically, it appears that Democrats have a 75% chance of voting for the outgroup, and Republicans have a 61% chance of voting for the outgroup. When comparing party slopes at a severity score of 2.6, Democrats were more likely to vote for the outgroup than Republicans, supporting hypothesis 2, β = .64, z = 2.55, p = .011.
To create a stronger test of our hypothesis and determine which party is more likely to have nearly 99% of all party members abandon a candidate first, we tested the differences between the Republican and Democratic curves at the asymptote for Democrats (where there is a 99.9 percent chance of voting for the outgroup). A visual inspection of the graph suggests that the asymptote occurs for Democrats when they learn their candidate “robbed a victim at gunpoint, the victim was shot to death” (TSS = 4.2). Further supporting hypothesis 2, the majority of Democrats were significantly more likely to abandon their ingroup candidates than Republicans at this level of severity (β = .60, z = 4.57, p < .001).
Overall, as transgression severity increases, people are more likely to abandon ingroup candidates. Regarding the moderating factors, people who were strongly identified with their ideology were more likely to continue to vote for severely transgressive candidates compared to those who were less identified. Lastly, Republicans were more likely to vote for their ingroup candidate for more severe transgressions than Democrats, and these differences became significant as transgression severity increased.
Exploratory Analyses
Although not directly hypothesized, we explored the three-way interaction between transgression severity, political orientation, and ideological identity strength. The three-way interaction was significant, β = .06, z = 3.71, p < .001, OR = 1.06, 95% CI [1.03–1.11], as well as the two-way interaction between identity strength and transgression severity, β = -.05, z = -4.16, p < .001, OR = .94, 95% CI [.92–.97]. Follow-up interaction plots reveal that highly identified Republicans had the highest severity thresholds: Republicans and highly identified voters were most likely to vote for highly transgressive candidates (see Figure 3).
Figure 3
Interaction Between Transgression Severity (Increasing Left to Right) and Political Identity Strength on the Percent Likelihood of Voting for Outgroup Political Candidate by Political Party
Note. Error bars represent confidence intervals.
Discussion
The current research introduced a novel paradigm for investigating the relationship between candidate transgressions and voter choice. By creating a numerical scale of transgression severity and employing a highly-repeated within-person design, we were able to estimate the location of ‘lines in the sand’: levels of transgression severity at which people are no longer willing to vote for their ingroup candidate. Our paradigm moves beyond a simple detection of “a main effect of transgression severity” to a more precise specification of the exact types of transgression that voters classify as “crossing the line.”
Supporting Hypothesis 1, we found that increased transgression severity increased the likelihood people abandoned ingroup candidates. Also as hypothesized, people who were strongly identified with their ideology were more likely to vote for transgressive candidates than voters who were less strongly identified. Last, Republicans demonstrated a higher severity threshold than Democrats, and voted for more severely transgressive ingroup candidates, potentially indicating greater ingroup party loyalty.
The novel paradigm is a critical first step in the development of more complex computational models, as additional parameters can be added to more accurately estimate the intricacies of human decision making and behavior in the political realm. For example, future research can explore the mechanistic processes that lead highly identified, and conservative voters to be more likely to support transgressive candidates compared to those who are less identified, and less conservative. With additional parameters, the model could test whether people primarily engage in motivated reasoning processes that ameliorate the severity of transgression, or whether they simply prioritize loyalty to the ingroup above potential costs of supporting a transgressive leader.
A second implication of the current work is that it fills a theoretical gap in the scandal literature by expanding the understanding of how ingroup bias impacts political decision making after transgression. Although the extant literature explores many important components of candidate transgression, a significant portion of the existing work does not incorporate the power of ideological ingroup bias. As observed, the incorporation of political identity tends to significantly minimize the effects of other variables (e.g., Bhatti et al., 2013; Blais et al., 2010; Walter & Redlawsk, 2019). Ultimately, transgressions cannot be separated from the identity of the transgressor, or the current polarized socio-political context. Thus, to most closely approximate the psychological process that occurs when people learn of transgressive candidates, the political identity of the voter must be incorporated into the literature and future study.
Third, the current work potentially informs future policy-making in the U.S, and specific findings from our model may shed light on contemporary elections. First, on average, people possessed over a 50% chance of voting for candidates who committed crimes, such as smuggling drugs into the country for resale and exposing themselves in public. Second, Republicans were more likely to vote for transgressive ingroup candidates than Democrats as transgression severity increased. For example, Republicans had a 24.5% chance of voting for a candidate who “enticed a minor into their car for immoral purposes” (TSS = 3.0), while Democrats had a 12.5% chance. In the most recent Presidential election, outcomes in three states (whose combined electoral college votes determined the winner of the election), were decided by margins of less than 1.3%. (The American Presidency Project). Thus, some election outcomes may be the result of Republicans and Democrats voting differently when they learn of ingroup candidate transgressions. For example, in the 2016 election, Trump was frequently attacked by political opponents for committing tax fraud (Mangan, 2016). It may be that Trump’s supporters did not perceive the act of not paying taxes to be severe enough to abandon him as a candidate. Centering a debate around a more severe transgression on the severity scale (TSS) may have been sufficient to sway Trump supporters.
In sum, the stakes are high for politicians in the United States—policy decisions made by elected officials today will largely determine how the nation recovers from the trauma wrought by the pandemic, economic crises, and racial tensions in the future. Thus, it is imperative that the most capable leaders are elected into office to help shape these influential decisions. This observed tendency for highly identified voters to cling to problematic ingroup leaders may hinder the creation of successful policy, and threaten the wellbeing of the nation. In order to develop solutions, this potentially problematic phenomenon needs to be documented, and causative agents recognized. The current work and proposed model takes an additional step at investigating the parameters that may lead to the current polarization and political gridlock, opening the possibility for future research to develop more targeted interventions.
Limitations
Although a strength of the current work is the objective measure of transgression severity (TSS), this is also, in part, a limitation. In order to investigate and compare severity thresholds, we had to use the same, objective, numerical parameter of transgression severity for all participants. Thus, we developed a scale that used transgressions from a pre-validated measure of transgression severity, and then measured the perceived severity again in a pilot study to ensure each item’s severity ranking was still accurate. However, like other scales, participants varied from each other, at least to some extent, in their perceived severity for each item.
In order to ensure our results were not solely driven by ideological differences in perceived severity, the model was run using a TSS that removed all transgressions in the pilot data that Democrats and Republicans did not perceive to be equal. Even when removing these transgressions, the same patterns of results were found. Thus, it is likely that differences in perceived severity would not have largely altered the outcome of the study. However, an additional question asking each participant to rate the severity of each transgression in the main study would have been an even more stringent test of the severity scale’s accuracy. Other micro-level data, such as perceived characteristics of the transgressor, would also have provided fruitful directions for investigation, and potentially would have explained even more variance in the relationship between candidate transgression and voter choice.
Another possible limitation in the current work is the generalizability of our results: Typically, when deciding who to vote for, participants do not learn about 70 possible candidates, each who has committed a unique transgression. Participants typically make a single voting choice, and only learn about two – three candidates. However, many people spend multiple hours each day either watching or reading about news on television or online. Indeed, some studies document that adults spend up to 145 minutes on social media every day (Currey & Nazir, 2021). A significant portion of this time is spent reading negative news stories, especially during elections (Enli, 2017) or pandemics (Anand et al., 2022). Thus, learning about 70 different transgressions, or pieces of bad news, may not actually be a far departure from daily life for a significant portion of the population.
Furthermore, our single Mturk sample may not present the best representation of the American public. Because the majority of Turkers identify as young and liberal (Lewis et al., 2015), it is possible our conservative sample is limited in scope. Although we found that conservatives tended to possess higher severity thresholds, it is possible older, more religious conservatives are more sensitive to transgressions than younger, less religious conservatives. Thus, thresholds for conservatives in the current work may be artificially high. It would improve the validity and reliability of the current study if additional samples, from different sources, obtained similar results.
One way to increase the ecological validity of our results would be to include the impact of media scandalization on perceptions of transgression. The media is an integral piece of the relationship between transgression severity and voter choice, as the public only learns about the transgressions the media chooses to disseminate. Depending on the ideological slant of the media, a reporter can illustrate a transgressive candidate as worthy of sympathy or guilt. Thus, the media largely impacts how people perceive transgressive candidates. For a review of the impact of media on scandals, see: von Sikorski, 2018.
Last, the dichotomous dependent variable, Republican or Democrat, may have also challenged generalizability of results. In the study, participants could only vote for the ingroup or outgroup candidate, an “abstain from voting” option was not provided. However, studies find that many people primarily vote because they feel it is there responsibility, or civic duty, as a citizen, not because they have a strong preference for one candidate over the other (Harder & Krosnick, 2008; Knack 1992). Thus, people who make the decision to vote, or typically vote in elections, are likely to still vote for one candidate or the other, regardless of the quality of both candidates. Therefore, we believe that this “forced choice” study format is, for much of the US population, not significantly deleterious to external validity.
Conclusion
By combining methods from political psychology and the decision sciences, the proposed model introduces a novel methodology for investigating the relationship between favored candidate transgression and voter choice. The creation of an objective measure of transgression severity and employment of a multi-trial within person design enabled us to estimate severity thresholds: the point of severity at which people are no longer willing to vote for transgressive ingroup candidates. We found that as transgression severity increased, people were more likely to abandon ingroup candidates, and vote for the outgroup. Ideological differences emerged as well: those who identified strongly with their political identity, as well as more Republican leaning participants, were more likely to continue to vote for highly transgressive ingroup candidates. Future studies should investigate the mechanisms beyond transgression severity in order to more accurately reflect how people vote in real life when faced with transgressive ingroup candidates.