Elections are usually considered a mechanism through which voters decide in which direction a polity should be heading policy-wise: What measures should be taken to boost the economy? How should the problem of social inequality be addressed? How can the environment be protected, and climate change effectively tackled? What policies should be implemented to protect the country from foreign threats? But elections are also the time when voters choose political leaders. Often there are large - sometimes even dramatic - differences between candidates in terms of their (perceived) skills (e.g., competence, leadership) and image (e.g., charisma). More fundamentally, most candidates differ with respect to their personality - “who we are as individuals” (Mondak, 2010, p. 2). The recent U.S. presidential elections provide a clear example that voters were asked not only to make a choice between competing sets of policies, but also between different personalities (e.g., Visser et al., 2017; Nai and Maier, 2018; Book et al., 2020).

Choosing leaders with a particular personality profile can potentially lead to serious political consequences. For instance, the personality of political leaders has been shown to drive their accomplishments once in office in terms of, e.g., policy accomplishments, relationships with the legislative branch, use of executive orders, and likelihood of unethical behavior (e.g., Rubenzer et al., 2000; Lilienfeld et al., 2012; Watts et al., 2013; Joly et al., 2019).

Voters often display low motivation and information about politics (e.g., Delli Carpini and Keeter, 1996), and tend thus to rely on cognitive heuristics when making up their mind on political matters (e.g., Sniderman et al., 1991; Lau et al., 2001). Since the personality profile of candidates is hard to hide (and is often explicitly showcased for electoral purposes), it provides ready-to-use cues for voters to gauge what they can expect from a given candidate if elected.

Of course, the personality profile of voters is equally likely to matter for their choices (e.g., Chirumbolo and Leone, 2010; Mondak, 2010; Nai and Maier, 2020a). including when it comes to candidate perception. Most notably, consistent evidence exists that candidate and voter traits are systematically linked to each other, in such a way that that voters are more likely to support candidates with personalities that “match” their own (e.g., Caprara et al., 2003; Caprara and Zimbardo, 2004; Fortunato et al., 2018). However, as this is also the case with respect to partisanship – voters strongly prefer candidates of “their” party – disentangling the specific effect of personality from the effects of partisanship is not a trivial task. Indeed, much evidence exists that the perception of candidates' personality traits is a direct function of partisan preferences(e.g., Hyatt et al., 2018; Nai and Maier, 2019; Fiala et al., 2020).

In this article, we attempt to contribute to a better understanding of how candidates’ (perceived) personality traits influence their likeability, and the role of voters’ individual differences and partisanship. Using an innovative survey experiment among U.S. respondents we demonstrate that 1) the public at large dislikes politicians scoring higher on “nefarious” personality traits; 2) voters that themselves score higher on those “dark” personality traits tend to like dark candidates; 3) the effects of candidates’ personality traits are, in some cases, stronger for respondents with weak partisan attachments.

There is a long tradition that aims to conceptualize, measure, and describe individual personality traits. The Big Five Inventory (BFI; McCrae and John, 1992) is the most studied personality inventory, and the most widely used to study the effects of personality on political attitudes and behavior (e.g., Mondak, 2010). The inventory identifies five “general” personality traits (extraversion, agreeableness, conscientiousness, emotional stability, and openness). More recent studies suggest that humans, in addition to the rather positively valenced traits assessed via the BFI, can have socially aversive - yet non-pathological - traits (Moshagen et al., 2018). The so-called “Dark Triad” identifies three “malevolent” components: narcissism, psychopathy, and Machiavellianism (Paulhus and Williams, 2002). These components have been shown to be associated to political attitudes and behaviors (e.g., Arvan, 2013; Jonason, 2014). In a nutshell, psychopathy is “the tendency to impulsive thrill-seeking, cold affect, manipulation, and antisocial behaviors” (Rauthmann, 2012, p. 487), narcissism is “the tendency to harbor grandiose and inflated self-views while devaluing others [… and to] exhibit extreme vanity; attention and admiration seeking; feelings of superiority, authority, and entitlement; exhibitionism and bragging; and manipulation” (Rauthmann, 2012, p. 487) and Machiavellianism is the tendency to harbor “cynical, misanthropic, cold, pragmatic, and immoral beliefs; detached affect; pursuit of self-beneficial and agentic goals (e.g., power, money); strategic long-term planning; and manipulation tactics” (Rauthmann, 2012, p. 487).

There are good reasons to expect that voters tend to dislike candidates with such dark traits. Individuals higher in psychopathy tend to have a more lenient approach to anti-social behaviors, which they often lack the ability to recognize. They tend furthermore to be impulsive and prone to callousness, and often show a strong tendency towards interpersonal antagonism (Jonason, 2014). Indeed, candidates scoring higher on psychopathy tend to display a “confrontational, antagonistic and aggressive style of political competition” (Nai and Maier, 2020b, p. 2). Like psychopathy, narcissism has been shown to predict more successful political trajectories (Watts et al., 2013), also in part due to the prevalence of social dominance intrinsic in the trait. This being said, narcissism is often linked to overconfidence and deceit (Campbell et al., 2004), a marked preference for hypercompetitiveness (Watson et al., 1998), reckless behavior and risk-taking (Campbell et al., 2004). Narcissists tend to go to great lengths to promote themselves and have indeed been shown to likely engage in angry/aggressive behaviors and general incivility in their workplace (Penney and Spector, 2002). Like psychopathy, Machiavellianism also has an aggressive and malicious side (Rauthmann and Kolar, 2013). People higher in Machiavellianism tend to display “cynical and misanthropic beliefs, callousness, a striving for argentic goals (i.e., money, power, and status), and the use of calculating and cunning manipulation tactics” (Wisse and Sleebos, 2016, p. 123), and in general show a proclivity to engage in malevolent behaviors intended to “seek control over others” (Dahling et al., 2009). Indeed, behavioral evidence suggests that higher Machiavellianism is associated with bullying at work (Pilch and Turska, 2015) and the use of more aggressive forms of humor (Veselka et al., 2010).

All in all, candidates higher in the Dark Triad should be more likely to adopt more aggressive behavioral patterns, as shown for instance in Nai and Maier (2020b) with respect to the use of a harsher communication style. Since all three components of the Dark Triad - narcissism, psychopathy, and Machiavellianism - point towards the direction of anti-social behavior, and voters tend to prefer leaders with a positive personality (Aichholzer and Willmann, 2020), we expect that voters tend, on average, to dislike “dark” candidates. We therefore expect:

H1. Exposure to candidates with dark personality traits reduces positive feelings for the candidate.

Importantly, we do not expect this general effect to exist across the board. Recent advances in the literature on elite cues and electoral behavior have clearly demonstrated that individual differences matter. For instance, Weinschenk and Panagopoulos (2014) show that respondents higher in agreeableness can be discouraged to turn out when exposed to negative campaigning messages. Similarly, the usage of “aggressive metaphors” tend to mobilize voters with “aggressive traits” and demobilizes strong partisans lower in aggression (Kalmoe, 2019). Mutz and Reeves, (2005) show that exposure to uncivil content lowers political trust in respondents that dislike conflicts, Nai and Maier, (2020a) present several instances in which darker personality traits of voters meaningfully moderate the effectiveness of negative and uncivil campaign messages. Beyond communication dynamics, Bakker et al., (2016) show that it is especially voters scoring lower on agreeableness that tend to appreciate populist candidates (who themselves score particularly lower on agreeableness, Nai and Martinez i Coma, 2019).

All in all, we have strong reasons to expect individual differences in voters to moderate the effect of candidates’ personality traits. First, we expect that the detrimental role of the dark personality profile of candidates, expected to exist in general (H1), does not exist among a specific set of respondents: those who themselves score higher on those dark traits. The rationale supporting this expectation is twofold. On the one hand, increasing evidence exists that voters with “darker” personality profiles tend to like darker politics - be it in terms of exposure to more negative and uncivil campaigns (Weinschenk and Panagopoulos, 2014; Nai and Maier, 2020a), or in terms of support for more confrontational and aggressive candidates (e.g., Bakker et al., 2016). On the other hand, this mechanism perfectly overlaps with the general “homophily” (or “congruence”) effect - that is, the established notion that voters are often more likely to support candidates with personalities that “match” their own (Caprara et al., 2003; Caprara and Zimbardo, 2004; Caprara et al., 2007; but see; Klingler et al., 2018). As summarized by Caprara and Vecchione (2017), personality “traits represent important elements through which the similarity-attraction principle may operate in politics because they allow voters to organize their impression of politicians, to link politicians’ perceived personalities to their own, and ultimately to justify their preferences on the assumption that similarity in traits carries similarity in worldview and values. Therefore, the more voters acknowledge their own pattern of behavior in a political leader, the more they may assume that the leader in question also shares their own principles” (Caprara and Vecchione, 2017, p. 236). We thus expect the following:

H2. Exposure to candidates with dark personality traits increases positive feelings for the candidate among respondents with dark personality traits.

We also expect the attitudinal profile of respondents to play a moderating role - more specifically, the strength of their partisan identification. Countless studies have shown that strong partisan affiliation (strong partisanship) is a central factor in determining how voters receive, accept, sample and process (new) political information. Voters unconsciously act as motivated reasoners (Kunda, 1990) and tend to reject information that is inconsistent with their attitudes and previously held beliefs (Druckman, 2012; Taber and Lodge, 2016). Because strong partisanship helps voters navigate the complex and treacherous waters of contemporary politics, it is no surprise that party attachment is one of the most important cognitive heuristics in their toolbox (Lau and Redlawsk, 2001; Schaffner and Streb, 2002; Fortunato and Stevenson, 2019). What happens when this navigation tool is absent? For voters that do not rely on (strong) partisanship to guide their political perceptions – a continuously increasing slice of the population in Western democracies (e.g., Dalton 2019) – we argue the following: exposure to the personality of candidates can act as “thin slices” - that is, “brief excerpt[s] of expressive behavior sampled from the behavioral stream” (Ambady et al., 2000, p. 203; see also; Spezio et al., 2012) - and heuristically provide them with schemata on which they develop their judgment. Voters heuristically compensate the lack of information they suffer from when they make judgments about political candidates (e.g., Huckfeldt et al., 2005). They use “evaluative impression formation of candidates by organizing and summarizing a diverse body of information in relatively simple terms [… which] ultimately determine voters’ likes and dislikes of candidates” (Caprara et al., 2002, p. 78). In other terms, we expect the effect of exposure to personality vignettes to be generally more effective, that is, more strongly associated with differences in candidate perception, for voters with weak partisan attachment.

H3. Candidates personality traits have stronger effects on candidate likeability among respondents with weak party attachment.

The main objective of this article is to assess the effect that (dark) personality profiles of political candidates have on shaping how voters perceive them - both directly, and as a function of individual differences in voters themselves (personality, partisanship). Unfortunately, disentangling the effects of candidates’ personality on voters’ perceptions is an arduous task. Voters’ perception of political figures is likely to reflect their underlying partisan preferences. For instance, there is consistent evidence that liberals have a much more critical perception of Donald Trump than conservatives. The former mostly highlight Trump’s lower agreeableness, lower conscientiousness, and lower emotional stability, whereas the latter rate the President higher on all the Big Five, and especially on openness and conscientiousness (e.g., Hyatt et al., 2018; Nai and Maier, 2019; Fiala et al., 2020). In this case, assessing how voters perceive specific personality traits - and the effects of such perceptions - is contaminated by their (pre-existing) political opinions about Trump refracted through the lens of partisanship.

In this article we tackle this issue via innovative experimental data. We present what is, to the best of our knowledge, the first study that manipulates the personality profile of a candidate along well-established personality inventories - and assesses its subsequent effects in terms of voters’ perceptions (however, see Rehmert, 2020 and de Geus et al., 2020, for examples of studies that use conjoint experiments to manipulate other salient aspects of the personal profiles of candidates, such as gender or socio-economic background). The design, embedded in an online survey distributed to a convenience sample of US respondents (MTurk, N = 1,971), exposed respondents randomly to one of eight different “vignettes” presenting personality cues for a fictive candidate - one vignette for each of the five general traits (Big Five) and one for each of the three “nefarious” traits of the Dark Triad. Respondents were asked to rate the personality of the candidate they were exposed to using the traditional abbreviated personality measures (the “TIPI” for the Big Five and the Dirty Dozen for the Dark Triad) and were subsequently asked to give an overall assessment of the candidate (thermometer).

Via this innovative experimental setup – a research design able to disentangle the effects of candidate personality, perceived traits, and voter’s preferences in such a way that their partisan preferences do not come into play - our analyses provide rather consistent support for our hypotheses. Our results will show that 1) the public at large dislikes “dark” politicians, and rates them significantly and substantially lower in likeability; 2) voters that themselves score higher on “dark” personality traits (narcissism, psychopathy, Machiavellianism) tend to like dark candidates, in such a way that the detrimental effect observed in general is completely reversed for them; 3) the effects of candidates’ personality traits are, in some cases, stronger for respondents displaying a weaker partisan attachment.

All materials, data, and syntaxes are available for replication in the following OSF repository: https://osf.io/wxruy/

In May 2020 we fielded a survey among a convenience sample of 2,010 US respondents via Amazon’s Mechanical Turk (MTurk; Paolacci and Chandler, 2014), an online crowd-sourced data platform. MTurk provides convenience samples, which should not be assumed to be representative of the general US population. In this sense, they are ill-suited to provide information to project general trends to the population at large (e.g., electoral predictions based on voting intentions). Nonetheless, MTurk surveys have been shown to perform quite well when compared to other convenience samples (Berinsky et al., 2012), because they tend to mirror the psychological divisions of liberals and conservatives in the US general population (Clifford et al., 2015). MTurk samples seem thus to represent a cheap and reliable way to collect systematic data from convenience samples (Hauser and Schwarz, 2016) - but see, for a more critical take, Harms and DeSimone (2015) and Ford (2017).

MTurk participants were invited to fill in a short online survey against a small compensation ($0.7). The questionnaire included an “attention check” (Berinsky et al., 2014) where specific instructions - select the option “other” and write a keyword in the entry box - were embedded within a long and digressing question. Respondents that failed such attention check (N = 39, 1.9%) were assumed to only skim through the questions and were excluded. The analyses are run on a final sample of N = 1,971 respondents. The final sample is composed of 49% of female respondents, and the average age is 42 years. The sample is mostly composed of white/Caucasian respondents (75%), followed by blacks/African-Americans (12%). 41% of respondents declare being “very interested” in politics, and only 2% declare “no interest at all”. The average self-reported left-right position is 4.8 (SD = 3.1) on a 0–10 scale.

The survey included an experimental component in which we “simulated” the personality traits of a fictive candidate. We created eight imaginary magazine interviews with a fictive candidate - independent Paul A. Bauer, running for a seat in the US House of Representatives for Minnesota's 9th Congressional district. ¹ Each mock interview was set up to cue respondents towards a specific personality trait of the fictive candidate, using both the framing of the journalist conducting the interview and the candidate response. For instance, the introductory paragraph the interview intended to cue higher extraversion (henceforth: “extraversion vignette”), reads as follows (excerpt):

“Bauer is a rising star in politics but is still relatively unknown to the public at large. Acquaintances describe him as enthusiastic and outgoing, but also as extremely talkative. I asked him three short questions, and found him to be extraverted and warm.”

After this initial introduction, tailored to the specific trait we wanted to cue, all mock interviews (“vignettes”) were set up as a series of questions and answers about what their usual day looks like and their perception of what politics is, similar to interviews that one might encounter reading the back page of a magazine like Newsweek. For instance, the “emotional stability vignette” reads as follows for the answer to the journalist question “what is politics to you?”:

“Politics is being able to take the best decision in the most calm and nuanced way possible. Impulsivity cannot have a place in politics. At the end of the day, only nuanced and rational decisions matter.”

Finally, the fictive candidate was asked to identify which “fictional character” he would like to be “for just a single day.” The use of fictional character to illustrate personality traits and facets is relatively common in the literature. For instance, Jonason et al. (2012) refer, to illustrate the dark traits of narcissism, psychopathy and Machiavellianism, to the fictive characters of James Bond, Hannibal Lecter, and House, M.D. Similarly, Schumacher and Zettler (2019) contrasts the two opposed personas of the fictive US presidents Josiah Bartlett (The West Wing) and Frank Underwood (House of Cards) to illustrate higher and lower scores on the “Honesty-Humility” trait in the HEXACO inventory. Drawing inspiration from these works, the fictional candidate refers in the interview to two fictive characters he would like to be for one day, with the idea that such characters reflect his personality, thereby amplifying the cueing potential of the vignette. ² The mock magazine interview included a picture of the fictive Paul A. Bauer; in actuality a portrait of former Swiss federal councilor Didier Burkhalter, who reflects, in our opinion, a perfectly generic stereotype of the political norm: a “normal” white, middle-aged male candidate.

After random exposure to one of the eight “personality vignettes”, respondents were asked to rate the candidate using two “short” personality batteries: the “TIPI” for the Big Five (Gosling et al., 2003) and the “Dirty Dozen” for the Dark Triad (Jonason and Webster, 2010). The former is set up as a battery of 10 statements about the candidate (e.g., “the candidate might be someone who is extraverted, enthusiastic,” “anxious, easily upset”), which respondents had to evaluate; pairs of statements yield scores on the five traits in the Big Five inventory. The latter is a battery of 12 statements (e.g., “the candidate might be someone who tends to want others to pay attention to him,” “… tends to be cynical”); the average of three sets of four statements yield scores for each trait in the Dark Triad. Using abbreviated measures of personality traits is not without its critics. Very brief measures (e.g., 1-item and 2-item scales, like the TIPI) have been shown to substantially underestimate the role personality traits appear to play when it comes to political behaviour, thereby increasing the odds of generating Type I and Type II errors (Credé et al., 2012). Bakker and Lelkes (2018) also show that abbreviated measures of personality traits tend to underestimate the relationship between ideology and personality traits and that researchers should ideally utilize more elaborate measures (e.g., 20-item or 50-item batteries). We have nonetheless chosen to use the 10-item “TIPI” battery in this research for pragmatic reasons: as it occupies the proverbial “middle ground” between the (extremely abbreviated) measures critiqued by Credé et al. (2012) and the ideal yet unwieldy measures proposed by Bakker and Lelkes (2018), it therefore represents an acceptable trade-off between feasibility and reliability for the purposes of our study.

A series of t-tests shows that respondents that were exposed to a vignette for a specific trait (e.g., extraversion) systematically rated the candidate as significantly higher on that trait when compared to the average of the other seven traits: t(1,969) = −13.77, p < 0.001 (extraversion), t(1,969) = −13.56, p < 0.001 (agreeableness), t(1,969) = −5.61, p < 0.001 (conscientiousness), t(1,969) = −11.23, p < 0.001 (emotional stability), t(1,969) = −7.85, p < 0.001 (openness), t(1,969) = −11.48, p < 0.001 (narcissism), t(1,969) = −13.81, p < 0.001 (psychopathy), and t(1,969) = −16.81, p < 0.001 (Machiavellianism). On average, thus, the “personality vignettes” were quite successful: they evoked in the mind of the respondents the personality profile that we intended to manipulate in the first place. Supplemental Figure SA in the Appendix illustrates the average score on all the personality traits for the fictive candidate as estimated by the respondents, depending on which vignette they were exposed to (bars in each panel).

Randomization checks indicate a successful random distribution of respondents according to their age, party identification, and personality traits (even if some marginal differences exist for some traits). Our tests indicate that female respondents were more likely to be exposed to a positive treatment and male more likely to be exposed to a negative treatment; the difference is statistically significant, χ2(1, N = 1964) = 9.87, p = 0.002. To exclude any confounding effects, we will replicate all analyses discussed below controlling for the gender of the respondents; see robustness checks discussed in Robustness Checks.

To what extent is the (perceived) personality of political candidates associated with their likeability by the public at large? Are agreeable candidates more likeable? Are narcissists disliked? Table 1 regresses the scores on the feeling thermometer for the candidate (0–100) on the personality vignette respondents were exposed to. Because conscientiousness represents an ideal trait for political leaders (and is also the trait that is more likely to drive better electoral results for competing candidates; Nai, 2019), we use exposure to the conscientiousness vignette as the reference category - that is, the effects of exposure to the other vignettes are computed against exposure to this vignette. Model M1 shows that candidates framed as higher in agreeableness and emotional stability receive somewhat higher ratings on the feeling thermometer, whereas candidates framed as higher in openness receive lower scores. But it is for the Dark Triad that we see the most impressive effects. Compared to candidates framed higher in conscientiousness, candidates framed with narcissistic, psychopathic, and, especially, Machiavellian traits receive significantly and substantially lower thermometer scores - up to 30 points less for Machiavellianism. The average thermometer score associated with all vignettes is illustrated in Figure 2.

Model M2 then estimates the thermometer score of the candidate as a function of respondents’ exposure to a “socially desirable” personality vignette (one of the Big Five, reference category) or rather to a “socially nefarious” vignette (one of the Dark Triad traits). Exposure to a dark trait, compared to exposure to a Big Five trait, reduces positive feelings for the candidate up to 26 points. Models M1 and M2 confirm, in other terms, that darker personality traits are detrimental for the likeability of competing candidates. The public at large, it seems, dislikes dark politicians.

Model M3 controls for respondents’ partisan identification and adds an interaction term between partisan identification and type of personality vignette (Dark Triad or Big Five) respondents were exposed to. There is no direct effect of party identification – which makes sense as the fictional candidate has been introduced as an Independent. The significant interaction in Model M3 shows that exposure to a “dark” vignette yields slightly higher thermometer scores for respondents identifying as a (strong) Republican. This reflects results in the literature showing that dark personality traits are more likely to be expressed among (strong) conservatives (e.g., Jonason, 2014). The effect is however not particularly strong.

Beyond simple exposure to personality cues, perceived personality traits of the candidates are likely to matter. For instance, it would not matter that a respondent is exposed to a narcissist candidate if they do not perceive the candidate as particularly higher on that trait. With this in mind Figure 3 plots, for each trait, the marginal effect of trait perception on the candidate likeability (feeling thermometer). For each panel in Figure 3, the models estimate how respondents feel about the candidate (thermometer) as a function of how high they perceive the candidate to score on the trait, depending on which vignette they were exposed to. Thus, for instance, the top-right panel is only run for respondents exposed to the “conscientiousness” vignette and estimates the marginal effects of perceived candidate conscientiousness (1-7 scale on the x-axis) on the feeling thermometer.

As the figure shows, for all personality traits - excluding extraversion (top-left panel) - the more respondents perceive the candidate as scoring higher on the trait in question, the stronger its effects on the thermometer. Full results are in Supplementary Table SA1 in the Appendix.

Results discussed above regarding the partisan identification of respondents - that is, that exposure to a “dark” vignette yields slightly higher thermometer scores for respondents identifying as (strong) Republican(s) - support the idea that the personality of candidates does not play uniform roles across the electorate. Evidence discussed in Bakker et al. (2016), for instance, shows that it is especially voters scoring lower on agreeableness that tend to appreciate populist candidates (themselves scoring particularly lower on agreeableness, Nai and Martinez i Coma, 2019). Similarly, recent experimental evidence shows “darker” forms of political communication, such as negativity and incivility, are appreciated by voters with specific personality profiles (e.g., Weinschenk and Panagopoulos, 2014; Nai and Maier, 2020a). With this in mind, the question is then: to what extent is the effect of candidates’ personality traits on their likeability a function of the personality of the respondents themselves? Table 2 tests for the moderating role of respondent’s personality (dark core) on the effects of exposure to dark personality vignettes on the thermometer scores. M1 shows a significant interaction term, substantiated with marginal effects in Figure 4.

As Figure 4 shows clearly, not only does the respondents’ (dark) personality moderate the effects of the candidate’s personality, but it reverses the negative effect shown across all respondents. This means that higher scores on the feeling thermometer are a function of increasing levels of dark personality of respondents themselves (dark core, representing the average scores on narcissism, psychopathy, and Machiavellianism) - but only for respondents exposed to a “dark” vignette. For respondents scoring lower on the dark core it is exposure to positive personality traits (Big Five) that drives higher thermometer scores. Simply put: dark voters like dark candidates. This effect does not seem to be further moderated by the partisan affiliation of the respondent (M2).

We also expected respondents with a weaker party identification to be more likely to be affected by the candidate’s personality cues - because they are more likely to use such cues heuristically. Supplementary Table SA2 in the Appendix reports a series of models where we have regressed, for each personality vignette, the candidate thermometer scores on the interaction between perceived candidate personality and the respondent strength of partisanship (binary variable, 0 low, 1 strong). Figure 5 substantiates all interaction effects in Supplementary Table SA2, with marginal effects. Each panel represents respondents exposed to a specific personality vignette (e.g., extraversion in the top-left panel), and the graph reflects the estimated marginal thermometer scores as a function of perceived trait (x-axis) for respondents with weak party attachment (white circles) and strong party attachment (black diamonds). We find significant interaction terms in three cases - agreeableness, emotional stability, and psychopathy (respectively, models M2, M4, and M7 in Supplementary Table SA2). In all three cases, the effect of the personality vignettes shown before for all respondents (Table 1) is stronger for respondents with a weak party attachment compared to those with a strong attachment. The effect is particularly visible for agreeableness and emotional stability. Put otherwise, for these three traits we can confirm the expectation that respondents with weak party attachment use cues related to the personality of candidates to make up their mind about the likeability of said candidates - much more so compared to respondents with strong party attachment.

Table 3 reports results of a simplified test for the moderating effect of party strength, contrasting only exposure to a “socially desirable” personality vignette (one of the Big Five, reference category) instead of a “socially nefarious” vignette (one of the Dark Triad traits). Model M1 illustrates the absence of interaction effects between this simplified measurement of simulated personality and intensity of party strength - confirming the idea, discussed above, that this interaction exists only for specific traits and not across the board. Furthermore, M2 suggests that the moderating role of party strength is also a function of respondents’ dark personality traits. As substantiated in Figure 6 with marginal effects, the three-way interaction shows that it is especially among respondents scoring higher on the “dark core” that weak party attachment increases the effect of dark personality cues (left-hand panel), and that this effect exists also, and more strongly so, among respondents with high party attachment. In other terms, if strength of party attachment seems to have specific effects for specific traits, it does not moderate the effectiveness of personality cues across the board. Furthermore, its effect is clearly overshadowed by the strong moderating role of the dark personality traits of respondents.

All results presented above resist models with alternative specifications. The same results are found in models that exclude respondents living in Minnesota (thus potentially privy of the deception in our experimental manipulation; N = 26), and in models that do not exclude “shrinkers” that failed the attention check (N = 39). Replication materials in the OSF repository include all specifications for these additional robustness checks. Finally, all results resist controlling the models by the gender of the respondent; the fact that male and female respondents were not randomly distributed across experimental conditions, as described beforehand, does not seem to affect the results.