Using the online survey tool MTurk, we recruited adults who had voted in the 2016 U.S. presidential election. We sought a total sample in excess of 300 to allow the identification of small effect sizes with power at .80 and p < .05 (Cohen, 1992). We obtained 159 individuals who reported that they had voted for Donald Trump and 154 individuals who reported that they had voted for Hillary Clinton. Although research indicates that MTurk volunteers are representative of American adults (Levay et al., 2016), a limitation of the study was that we could not determine whether the present sample was representative of all U.S. voters in the 2016 election.

The sample included 182 males and 131 females, and this ratio was similar for Clinton and Trump voters in the sample, χ²(1) = .01, ns. The age range was 18 to 70 years, and the average age of respondents was 35.04 years (SD = 10.6). No difference in age between Clinton and Trump voters was found, t(311) = .07, ns. In the overall sample, ethnic group identification was weighted toward non-Hispanic Whites (75%) in comparison with African American, Hispanic American, and others (25% total). A contemporaneous accounting of U.S. voters yielded 69% non-Hispanic Whites and 31% Hispanic, Black, and Other registered voters (Gramlich, 2020). In our data, more European Americans than non-European American individuals voted for Trump than for Clinton, χ²(1) = 5.65, p = .019. Frequencies of the types of highest level of schooling were: high school (12%), some college education (31%), a four-year college degree (40%), and advanced degree (13%). A significant chi-square result was obtained for schooling by endorsed candidate, χ²(5) = 11.61, p = .04, which was accounted for by more high school graduates and individuals with some college education being counted among Trump voters and more individuals with four-year college degrees among Clinton voters.

Participants were recruited within the MTurk online survey website for a study named “political views about today’s society” in the two weeks following Donald Trump’s presidential inauguration on January 20, 2017. Demographic variables were assessed but no name or other identifying information was obtained. Completion of the entire survey took about 20–30 minutes, and participants were paid $3 for their time. Ethics approval was obtained from the first author’s host institution.

Data, copies of surveys, and the codebook for both studies are available at osf.io in a folder titled “Perceptions of psychopathy in U.S. presidential candidates” (se Jose, 2025) Data were analyzed using SPSS and Mplus for this report. This study's design and its analyses were not preregistered.

Means, standard deviations, and correlations for the two groups of voters are reported in Table 1.

Less than 1% missingness was noted, Little’s chi-square test yielded p > .30, and EM imputation was used to obtain a full dataset. No outliers were noted. Means for self-reported SDO and RWA scores were higher for Trump voters compared with Clinton voters, consistent with expectation. Means for attributions of psychopathic tendencies for own vs. other candidate seem to be mirror images.

To analyze the data for Hypothesis 1, we performed a repeated measures MANOVA in which attributions of psychopathic tendencies were the repeated measures across the two groups of voters. Further, we covaried out gender, ethnicity, highest schooling obtained, and age. A power analysis (Cohen, 1992) based on power at .80, an alpha level of .05, and an intention to identify medium effect sizes, suggested that a sample size of 125 would be needed. Our total sample of about 300 satisfied this initial basic criterion, but covarying out four covariates and planned analyses of interactions increased the recommended total sample to 325. Thus, our obtained sample fell slightly short of this more rigorous sample target. Caution is recommended for interpreting small effect size differences.

Contrary to prediction, we obtained a non-significant difference between the two candidates on perceived psychopathy, Wilk’s Λ = .998, F(1, 307) = 0.61, p = .436, ηp2 = .002: collapsing across the two groups of voters, In addition, a non-significant effect for Voter was obtained, F(1, 307) = 2.71, p = .101, ηp2 = .009, suggesting that Trump and Clinton voters attributed almost identical amounts of psychopathy, collapsing across the two candidates.

By far and away, the largest effect size obtained in this analysis was the significant Candidate × Voter interaction, Wilk’s Λ = .405, F(1, 307) = 451.37, p < .001, ηp2 = .60. Estimates of partial eta squared greater than .14 are considered “large” (Gray & Kinnear, 2012), so the effect size of this interaction can be considered to be extremely large. Figure 1 graphically depicts the interaction effect, and although it conforms to a classic cross-over interaction, the largest divergence occurred among Clinton voters considering the two candidates. Although both groups of voters rated their disfavored candidate as higher on psychopathic tendencies, Clinton voters’ perceptions more widely diverged than did Trump voters.

Simple effects analyses indicated that all four pairwise comparisons were significantly different. In sum, this pattern is consistent with Hypothesis 2, which suggested that voters would attribute psychopathic tendencies to the two candidates in a mirror image fashion: lower for their own candidate and higher for the opposing candidate.

A path model was constructed in Mplus (Muthén & Muthén, 1998–2015) in which background ideological beliefs (SDO and RWA) predicted perceptions of psychopathic tendencies of both candidates and these, in turn, predicted vote choice (see Figure 2). Since the dataset was concurrent, this mediation analysis illuminated amounts of shared variance among the three variables, not temporal or causal relationships discernible with experimental or longitudinal datasets (Jose, 2013). In such a case, sequential placement of variables in the model can be uncertain. We positioned these variables in this particular order for theoretical reasons (i.e., background personological variables were expected to explain variance in attributions of psychopathy to candidates, which, in turn, were expected to explain variance in vote choice behavior). To obtain a generalizable result, covariates of the model were gender, age, highest educational attainment, and ethnicity (not depicted in the figure). The fully saturated path model featured a dichotomous dependent variable in voting choice, so theta parameterization in Mplus (Muthén & Asparouhov, 2002) was used to accommodate this variable type.

The numerical values in Figure 2 are beta weights, and every path was significant in the expected direction, except for one path which was marginally significant. Several significant direct effects can be noted in the model: (1) high RWA and high SDO predicted higher perceptions of psychopathic tendencies in Clinton; (2) in contrast, high RWA and high SDO predicted lower perceptions of psychopathic tendencies in Trump; (3) higher perceptions of psychopathic tendencies in Trump and lower perceptions of psychopathic tendencies in Clinton predicted voting for Clinton (and the reverse pattern for voting for Trump); and (4) RWA and SDO weakly and directly predicted voting for Trump.

Four significant indirect effects were identified: 1) RWA to psychopathic tendencies in Trump to Vote: a*b = -.29, SE = .09, 95% CI [-.47, -.15], p < .001; 2) RWA to psychopathic tendencies in Clinton to Vote: a*b = -.26, SE = .08, 95% CI [-.44, -.13], p < .001; 3) SDO to psychopathic tendencies in Trump to Vote: a*b = -.11, SE = .05, 95% CI [-.23, -.03], p = .027; and 4) SDO to psychopathic tendencies in Clinton to Vote: a*b = -.21, SE = .08, 95% CI [-.38, -.08], p = .005. These indirect effect sizes fell between medium and large (Kenny, 2024). In sum, these findings supported Hypothesis 3: We found that voters reporting higher authoritarian beliefs perceived fewer mental health problems in Trump’s behavior while at the same time they perceived greater psychopathic tendencies in Clinton. In contrast, individuals reporting lower levels of authoritarian beliefs evidenced associations in the opposite direction.

As before, using the online survey tool MTurk, we recruited adults who had voted in the 2016 U.S. presidential election, and we obtained data from 149 individuals who reported that they had voted for Donald Trump and 148 individuals who reported that they had voted for Hillary Clinton. The sample included 147 males and 150 females, and this ratio was similar for Clinton and Trump voters in the sample, χ²(1) = 2.79, p > .10. The age range was 20 to 78 years, and the average age of respondents was 40.97 years (SD = 18.13). The age difference between Clinton and Trump voters was small, t(295) = 0.86, p = .589; Trump voters (M = 41.6 years) were slightly older than Clinton voters (M = 40.3 years).

Cronbach’s alphas of all key constructs for Study 2 were high and very similar to Study 1: SDO (α = .94), RWA (.94), perceived psychopathic traits of Trump (.96), and perceived psychopathic traits of Clinton (.94).

A chi-square test of gender yielded a significant difference between the two studies: χ²(1) = 4.59, p = .032, revealing that the male/female gender ratio was more equal for Study 2 (49%) than Study 1 (58%). A similar chi-square test of white vs. non-white ethnicity yielded a non-significant difference, χ² = 1.63, p = .201. The sample of Study 2 was found to be older (M = 40.1) compared with Study 1 (M = 35.0), t(608) = 4.97, p < .001. A chi-square test of highest level of schooling yielded a significant difference: χ²(5) = 17.6, p = .003, revealing that Study 2 included fewer individuals with high school degrees and more individuals with professional degrees than Study 1. These results led us to covary out gender, age, and level of schooling in comparing mean levels and associations between the two samples in subsequent analyses.

Table 2 reports the correlations, means, and standard deviations of the four key variables divided by Trump vs. Clinton voters within Study 2.

We next analyzed the data from Study 2 alone; as before, we performed a repeated measures ANOVA in which attributions of psychopathic tendencies was the repeated measure across the two groups of voters. Further, we covaried out gender, ethnicity, highest schooling obtained, and age.

First, we obtained a significant (albeit small) difference between the two candidates for perceived psychopathy, Wilk’s Λ = .986, F(1, 291) = 4.17, p = .042, ηp2 = .014. Collapsing across the two groups of voters, Trump was perceived, on average, in Study 2 to display higher psychopathic tendencies, M = 3.67, SE = .045, 95% CI [3.58, 3.76], than Clinton, M = 2.78, SE = .041, 95% CI [2.70, 2.86]. Thus, support for Hypothesis 1 was obtained, however statistical significance was barely achieved and the effect size was small.

As before, a non-significant effect for Voter was obtained, F(1, 291) = 0.63, p = .427, ηp2 = .002, suggesting that Trump and Clinton voters attributed almost identical amounts of psychopathy, collapsing across the two candidates.

Again, a very large effect size was obtained in this analysis for the Candidate × Voter interaction, Wilk’s Λ = .443, F(1, 291) = 365.40, p < .001, ηp2 = .56. As with Study 1, this effect size was extremely large. The pattern of the interaction strongly supports Hypothesis 2. Figure 3 graphically depicts the interaction effect for Study 2, and it bears a remarkable similarity to the same graph generated for Study 1.

Next, to explicitly compare the ratings between Study 1 vs. Study 2, we performed a three-way mixed model MANOVA in which Study (1 vs. 2) was crossed with Voter, and the repeated measures factors were perceived psychopathic traits for the two candidates, and four demographic variables were entered as covariates. The main effect for Study yielded a significant difference, F(1, 602) = 11.45, p < .001, ηp2 = .02 (albeit a small effect size), and it was accounted for by slightly higher average ratings of perceived psychopathic traits for Study 2, M = 3.23, 95% CI [3.18, 3.29], compared with Study 1, M = 3.11, 95% CI = [3.06, 3.16]. These means were determined after collapsing across candidates and voters, and covarying out demographic variables. Importantly, no interaction term involving Study yielded significance: Study × Voter (p = .73); Study × Candidate (p = .21); or Study × Candidate × Voter (p = .65). The single significant univariate result was a main effect for Study for perceived psychopathic traits for Trump, F(1, 602) = 7.91, p = .006, ηp2 = .012. Notably, both Trump and Clinton voters (i.e., no significant three-way interaction was found) evidenced a small increase in these perceptions from the time of the election (M = 3.50) to the first impeachment (M = 3.69). A change of .19 unit constituted about 18% of the standard deviation for this variable and, interestingly, Trump voters increased slightly more than Clinton voters. Against the backdrop of overwhelming similarity in means for variables over time, the one key difference was a modest increase in the perceived psychopathy of Trump by both Clinton and Trump voters after three years of his presidency.

The final replication effort was to compare the path model findings of Study 1 with the same model constructed with Study 2 data. To statistically evaluate similarity, equality constraints were applied to all paths (direct paths as well as mediations) one by one, and all eight direct tests and four indirect tests yielded non-significance, ps > .05. Thus, Hypothesis 3 was supported in that perceptions of candidates’ psychopathy were found to significantly share variance with RWA/SDO scores as well as the decision to vote for Trump or Clinton. More broadly, this result, similar to the MANOVA findings, provides support for the proposition that these findings were almost universally replicated with a different set of participants at a different point in time.

Mental health literacy, as defined by Jorm (2000), is the skill and knowledge base of laypeople to accurately label instances of behavior as conforming to particular psychiatric syndromes. Little research evidence has been obtained on this issue, but the available evidence suggests that laypeople are poor at correctly labelling even well-known mental disorders such as depression and schizophrenia (Furnham et al., 2011; Jorm et al., 1997; Regier et al., 1988). Research is scarce on whether laypeople are able to correctly identify instances of psychopathic behavior (a type of personality disorder) and research is further lacking on whether voters are able to accurately identify psychopathic behavior in political candidates and officeholders. One significant contribution of the present work is that it broaches the question of how skilled laypeople are in the ability to identify disqualifying psychological characteristics of candidates. The present work does not tackle the issue of how accurate these perceptions were (we had no means to make this determination), but future work should explore that issue.

A notable finding in the study was that the cross-over statistical interaction depicted by Figure 1 revealed an almost mirror-image pattern of voters attributing absence of psychopathy in their preferred candidate at the same time as attributing presence of psychopathy in the rejected candidate. This dramatic and opposite bifurcation of views is consistent with ingroup vs. outgroup perceptions held by Democrats and Republicans in U.S. politics (Greene, 2004) and the phenomenon described by Wright and Tomlinson (2018), that political orientation profoundly shapes one’s attributions of personality to candidates. We did not assess political affiliations of voters in these data collections, but voter preferences in these data likely well represent voters’ political loyalties. The interaction pattern strongly suggests that voters see no flaws in their preferred candidate and see many disqualifying problems in their disfavored candidate.

However, the pattern was not perfectly symmetrical: Clinton voters more strongly attributed psychopathy to Trump and lack of psychopathy to their own candidate than did Trump voters. In other words, Clinton voters more strongly felt that their candidate lacked psychopathic traits relative to the opposing candidate. The widespread and intense incredulity and dismay of Clinton voters to Trump’s electoral victory is consistent with the highly polarized perceptions of psychopathic behavioral tendencies obtained in this study. Clinton voters were surprised and dismayed that the candidate they perceived as exceedingly unsuitable for the office of president was elected by the American public to the highest office in their country (DeGroot & Carmack, 2021; Teng, 2017).

The set of results obtained in the path model may provide some tentative answers as to why so many voters chose to vote for Donald Trump. Our prediction was that individuals who held high levels of authoritarian/SDO beliefs would be likely to perceive Trump’s behaviors as consistent with the authoritarian mindset and thus view him more positively than Clinton (based on findings by Crowson & Brandes, 2017). In short, authoritarian-leaning voters attributed psychopathic traits, for example, dishonesty and a tendency to con other people, to Clinton and did not attribute these pernicious traits to Trump, whereas voters lacking in authoritarian beliefs attributed psychopathic traits to Trump and not to Clinton. And last, unsurprisingly, the path model showed that individuals who attributed high levels of psychopathy to Trump tended to vote for Clinton, and those individuals who attributed high levels of psychopathy to Clinton tended to vote for Trump.

The results of this study support the “beauty is in the eye of the beholder” theory of perceptions of political actors (Killian, 2023). Clinton voters were aghast at the stance of Trump voters explaining away the Access Hollywood recording concerning Trump’s claim of “pussy grabbing” as “locker room talk,” while at the same time many Trump voters wanted (and still want) Clinton to be jailed for perceived criminal mishandling of her e-mails. This extreme polarization of political views may be explained in part by the range of authoritarian beliefs held by voters. If a voter holds strong authoritarian/SDO views, Trump’s actions and words are not considered to be psychologically disturbed or psychopathic; instead, they are perceived as evidence of his strong leadership skills and socially dominant status. Thus, a voter who endorses authoritarian views is likely to view Trump’s treatment of minorities as legitimate and appropriate. Someone who rejects authoritarianism interprets Trump’s behavior differently, noticing the motives for self-aggrandizement and deception, imposition of power over other people, and joy in the misery of others (Lee, 2017).

These findings are consistent with Caprara and Zimbardo’s (2004) congruency model of political preference, which argues that “individual characteristics of voters, such as their traits and values, become decisive for political choice” (p. 581). Their findings reveal that people tend to vote for candidates whose personality traits accord with the ideology of their preferred political party. They also select politicians whose “traits match their own traits” (p. 581). The findings of the present study show that individuals who held authoritarian beliefs tended to vote for Donald Trump, the more overtly authoritarian candidate relative to Hillary Clinton, in agreement with Caprara and Zimbardo’s model. The theory of “ideology as motivated social cognition” similarly claims that personality characteristics inform and drive, to a substantial extent, political positions and choices (Carney et al., 2008; Fortunato et al., 2018; Jost, 2006; Jost & Amodio, 2012). Thus, the data suggest that an authoritarian ideology motivated Trump supporters to vote for him, and, in service of this behavior, attributed lower levels of aberrant behavior to him.

A significant caveat that we need to highlight is that the authors of the present paper have no authoritative evidence on the mental health status of Donald Trump or Hillary Clinton. However, it is relevant that empirical researchers, similar to us, have asked political experts, voters, and even politicians themselves about their perceptions of candidates (e.g., Dutton, 2016; Nai & Maier, 2018) But lacking objective, authoritative evidence, we could not, within the present study, ascertain whether voters accurately perceived the presence of psychopathic tendencies in the two U.S. presidential candidates.

Another significant limitation is that we do not know the direction of the influences of these perceptions and behavior over time, but, still, we ordered the three sets of variables within the path models to reflect a likely temporal sequencing. Beliefs in authoritarianism/SDO are stable belief systems partially based on personality, so they were placed as exogenous variables in the path model. Attributions of psychopathic characteristics were likely based upon the voters’ perceptions of the candidates’ behavior unfolding over months and years leading up to the day of the election when the voting choice was recorded. Longitudinal data would have provided a more nuanced picture of how these variables affected each other over time.

Another limitation in the present dataset is that party affiliation was not empirically assessed. Much research in political psychology has shown that party loyalty and identification exert considerable influence upon political perceptions and behavior (e.g., Folke & Rickne, 2020), and it would have been helpful to have this information to determine whether and how party affiliation interacts with authoritarian political beliefs in determining candidate preference. In particular, to what extent did Trump voters ignore or discount reports of his unsavory behavior out of party loyalty? At the same time, boundaries of political parties move over time to accommodate potential new members and new political leaders (Habersack, 2024). An illuminating future research direction would be to determine how party loyalty affects perceptions of candidates over time as critical events unfold.

In addition, future work should examine the mediators between authoritarian belief systems and attributions of psychopathy. A hypothesis worthy of further examination is to what extent do authoritarian voters publicly approve candidates who use extreme (and possibly criminal) behaviors to exert dominance over marginalized outgroups. Another useful direction would be to include measures of constructs antithetical to right-wing authoritarianism and social dominance orientation such as belief in democratic systems and racial equality (“wokeness”; Jose, 2024). And last, further investigation of the willingness of Trump supporters to excuse and dismiss his overtly antisocial behavior is needed. It is not clear at present whether they are uninformed, informed but consider this behavior as irrelevant, informed but consider his behavior as proof of his dominant and superior status, or something else. We would also like to extend the examination of this attributional bias to other politicians’ supporters to determine whether supporters across the full range of political affiliations show similar or dissimilar biases.

The 2016 U.S. presidential election raised a number of important questions about how voters in the American democratic system consider candidates’ behavior and estimate future political performance of candidates. One such question is whether the average American voter possesses a sufficient knowledge base to discern significant psychological pathologies in candidates and uses those judgements to guide their choice among candidates. We need to improve our understanding of how background ideological beliefs, whether benign or malignant to the body politic, may influence perceptions of candidates and consequently lead individuals to cast their vote for one individual or another. The consequences of misperceiving psychopathic tendencies in candidates are potentially immense, and our democratic institutions need to identify and minimize such dangers.