Teflon Trump? Shifting Public Perceptions of Donald Trump’s Personality Traits across Four Studies over 18 Months

Democrats and Republicans have divergent views about Trump’s personality. Are these perceptions stable over time? We describe the evolution of Trump’s perceived personality (Big Five and Dark Triad) between November 2018 and May 2020 by comparing four samples of American respondents (Mechanical Turk). Trump’s perceived personality is on average extremely stable. Liberals have a much more critical perception of Trump than conservatives. Yet, strong Republicans have an increasingly more negative perception over the time—perhaps because the scandals that have marred the Trump presidency are at odds with conservative “moral foundations.” We also detected a potential demobilization trend for strong Republicans; the more they perceive the president negatively, the lower their likelihood to turn out. The electoral implications of these trends are important. Politics, Personality and Individual Differences, Presidential Studies Quarterly, Public Opinion Quarterly and more. His current research focuses on the content and the impact of political (campaign) communication with a specialization on televised debates, negative campaigning, political scandals, experiments, and real-time response measurement.


Introduction
Beyond pocketbook, retrospective, and partisan considerations, it is undeniable that how voters perceive the character and personality of political figures matters (e.g., Bittner, 2011;Costa & Ferreira da Silva, 2015;Garzia, 2013). This article contributes to our current understanding of the importance of perceived personality traits of political leaders by tackling two key issues: (a) how these perceptions are formed, and (b) their possible behavioral consequences. We do so by presenting novel evidence from four studies with American samples, in which respondents were asked to rate the personality of the 45th U.S. president.
During the first term of the Trump administration, the character and personality of the 45th President of the United States has attracted the attention of the academic literature (e.g., Ashcroft, 2016;Visser, Book, & Volk, 2017;Hyatt et al., 2018;Nai & Maier, 2018;Nai, Martínez i Coma, & Maier, 2019). Unsurprisingly, the issue of Donald Trump's mind is a topic that fascinates also outside of academia. In a widely read story published in the The Atlantic a few months before the 2016 election, psychologist Dan P. McAdams (2016) famously described Trump's personality as exhibiting "sky-high extroversion combined with off-the-chart low agreeableness [. . . and] grandiose narcissism." The "spectacle of Trump presidency" (Lynch, 2017, p. 620) is, however, far from being a consensual matter. In the general public, there is indeed only "modest agreement" (Hyatt et al., 2018, p. 1) about the character and personality of Donald Trump. Is he a charismatic leader that "tells it like it is," or an insecure hotheaded egocentric? Several studies have documented that voters of different partisan identifications have radically opposing perceptions (Fiala, Mansour, Matlock, & Coolidge, 2020;Hyatt et al., 2018;Wright & Tomlinson, 2018). These studies, broadly, show that liberals (or voters that supported Clinton in 2016) tend to have a much more critical perception than conservative (Trump voters). The former mostly pinpoint Trump's (very) low agreeableness, low conscientiousness, and low emotional stability, whereas the latter rate the President much higher on all the Big Five personality traits, especially on openness and conscientiousness. There seem to be a greater consensus about Trump's extraversion, which is rated high across all voters; some evidence also exists that different voters converge to see Trump as relatively high in narcissism (Hyatt et al., 2018;Williams, Pillai, Deptula, Lowe, & McCombs, 2018). All in all, however, where some voters overwhelmingly see in Trump the evil Mr. Hyde, others see the polished and well-made Dr. Jekyll (Hyatt et al., 2018). In light of the current levels of affective polarization (Iyengar, Sood, & Lelkes, 2012), such opposing views are no "strange case."

Scandals, Moral Foundations, and Shifting Perceptions of the President
What these studies have been unable to address so far, however, is the extent to which these perceptions are stable over time. Trump seems to benefit from a blanket approval by part of the U.S. electorate, and indeed a central element of Trump's first term has been his low but surprisingly stable approval rate (Dunn, 2018), as shown in Figure 1.
This stability is particularly noteworthy, especially in the face of the many allconsuming controversies that marred Trump's first years in office, some of which are also featured in Figure 1. From the onset, the Trump administration (and the President himself) were criticized for their relaxed approach to factual truth, mediatized spats with foreign leaders (and dictators), alleged affairs with adult film stars, numerous accusations of sexual misconduct, repeated refusals to provide documents to auditing authorities, alleged campaign finance violations, overt conflict with federal investigative authorities and personal feuds with "top cops," a multi-year investigation into potential collusions with Russian operatives (the "Mueller probe"), a pressure campaign towards Ukraine's President Zelensky leading to Trump's impeachment in the House of Representatives for abuse of power and obstruction of Congress, and, of course, a 2 I N T E R N A T I O N A L J O U R N A L O F P U B L I C O P I N I O N R E S E A R C H rather suboptimal response to the coronavirus pandemic throughout 2020, leading into the Presidential election.
Scandals and political controversies, defined as a violation of "general moral and ethical standards" (Maier, 2011, p. 284), can alter the way people perceive political figures. In line with Schlenker's (1980) "Impression Management" theory, Sigal, Hsu, Foodim, and Betman (1988) show that voters adapt their perception of candidates' character after exposure to the media coverage of political scandals-and adjust their voting patterns accordingly (for a summary see von Sikorski, 2018). Of course, this readjustment is a function of partisan preferences, as voters tend to process information in a motivated manner (Fischle, 2000). They are more critical of candidates for which they have an already negative opinion (Bhatti, Hansen, & Olsen, 2013). Nonetheless, scandals can also affect partisan supporters. We argue here that the scandals and controversies that have marred the first years of the Trump presidency might have especially played a negative role on a group of voters: strong conservatives.
The "moral foundations theory" (Haidt & Graham, 2007) can hold clues to why this might be the case. The theory, in a nutshell, argues that liberals and conservatives "often hold divergent opinions on moral issues" (Kugler, Jost, & Noorbaloochi, 2014, p. 414). If liberals generally tend to align more closely to moral values and principles related to fairness and harm avoidance, conservatives tend to assign a higher moral status to ingroup loyalty, deference and respect to authority, and purity of actions (e.g., Graham, Haidt, & Nosek, 2009). The conservative moral preferences are "focused less on the protection of individuals and more on restraint and the integration of Trump approval rate during his first three years in office, timeline of our four studies, and selected controversies and events. Note. Data from FiveThirtyEight. The trends reported are estimates, based on aggregated polls. See https://projects.fivethirtyeight.com/trump-approval-ratings/voters/. communal life" (Federico, Weber, Ergun, & Hunt, 2013, p. 590). It is rather straightforward to see how the numerous controversies that have characterized the first years of Trump's presidency are particularly at odds with the three "conservative" moral foundations. First, at the heart of both the "Mueller probe" and the President's impeachment was the alleged inclination to put the personal and political interests of the President ahead of the interests of the country, including via alleged covert forms of contact with foreign powers to harm domestic political opponents; this strongly contrasts with the patriotic principle of "country over party" that is directly associated with (national) ingroup loyalty, "exemplified by virtues like loyalty, patriotism, and self-sacrifice" (Clifford, 2017, p. 534). Second, Trump has shown much overt contempt for federal and investigative authorities-from his numerous public spats with the FBI to his unwillingness to comply with subpoenas and turn in documents to investigative authorities-which is undoubtedly at odds with the conservative moral imperative of deference to and respect for authority. Finally, allegations of sexual misconduct, extramarital affairs, and inappropriate intimate relationships with adult film stars particularly contrast with the conservative moral principles of purity and sanctity of actions.
All in all, the sum of all controversies that have marred the first years of the Trump presidency seems to be particularly at odds with key conservative moral principles. Therefore, given the accumulation of controversies, we might expect to find an increasingly critical view of the president among (hard) conservatives. Given the nature of the controversies, the more time passes, the higher the chances that conservatives will be offended in one of their core values, if not all of them. This expectation seems especially reasonable in light of the "ideo-attribution effect" theory (e.g., Morgan, Mullen, & Skitka, 2010), suggesting that conservatives tend to blame the culprits of misdeed much more personally than liberals, who rather tend to look for situational justifications. As such, conservatives should be much more likely than liberals to draw a direct link between the controversies and their perception of Trump himself. The fact that we do not advance specific expectations for each perceived personality trait should not be seen as the implicit assumption that all traits are equivalent, or interchangeable. Not all perceived traits are likely to be affected equally. For instance, a negative effect of (perceived) violation of moral foundations should be particularly apparent in evaluations of conscientiousness-because this trait is directly associated with responsible behavior and dependability (Judge, Higgins, Thoresen, & Barrick, 1999), and, as such, is likely to be affected by professional misconduct and improper behavior. Similarly, accusations of unethical behaviors such as campaign finance violations and abuse of power should reinforce the image of the president as entitled, manipulative, and remorseless-facets closely associated with the three dark traits (Rauthmann, 2012). In contrast, for other traits, such as openness, an effect is harder to clearly foresee.
With this in mind, in this article we present evidence suggesting that the "Teflon" image of Donald Trump, a hard surface to which nothing sticks, has probably reached its limits-especially for conservatives. The fact that we expect shifts in such perceptions over time might seem at odds with the idea that the personality of individuals is a construct that tend to be stable over the individuals' lifetime (e.g., McCrae & Costa, 1994). However, aggregated personality ratings of political figures from voters are unlikely to reflect "true" personality profiles of those figures in settings where (affective) polarization is especially high. In polarized environments, voters are likely to perceive all political issues exclusively through the lens of their partisan preferences, and as such, they cannot be expected to provide ratings of political leaders that are "uncontaminated" with those preferences. Therefore, our article does not claim to discuss any measurement of Trump's personality per se, but rather, and more simply, how voters perceive it to be.

Our study
We describe the evolution of how voters perceive Trump's personality traits-both in terms of the general traits in the Big Five inventory (extraversion, agreeableness, conscientiousness, emotional stability, and openness) and the "socially aversive" traits of the Dark Triad (narcissism, psychopathy, Machiavellianism)-over a period of a year and a half. We do so by comparing results of four surveys fielded at intervals of $6 months between November 2018 and May 2020 in the United States (Figure 1), each using a convenience sample of respondents surveyed through Amazon's Mechanical Turk (MTurk) platform. In each survey, respondents were asked their partisan identification and to rate the U.S. President using two established short personality batteriesrespectively, the Ten Items Personality Inventory (TIPI; Gosling, Rentfrow, & Swann, 2003) for the Big Five, and the Dirty Dozen (D12; Jonason & Webster, 2010) for the Dark Triad.
We use the data gathered in these four consecutive studies to test whether, and to what extent, there have been shifts in public's perception of Trump's personality. In line with what discussed above regarding the violations of conservative moral imperatives, we particularly expect more negative perceptions of the President among strong conservatives over time. We will also explore the behavioral consequences of such perceptions. Evidence suggests that politicians' violations of established norms can lead to higher abstention (e.g., Stockemer, LaMontagne, & Scruggs, 2011). Structural shifts in party support over the past decades created a situation where "dealigned electors are more susceptible to short-term factors such as the influence of political leaders and consideration of their personalities" (Costa & Ferreira da Silva, 2015, p. 1228. With this in mind, we will also test (a) whether negative perceptions of the president might be associated with lower intended turnout, and (b) whether the relationship between perceptions and turnout varies according to party identification. Data and materials are available for replication at the following Open Science Foundation (OSF) repository: https://osf.io/xrbnq/? view_only¼bed195b56f51427da1323ffcda83ad15

Samples
Below we compare trends found in four surveys administered to as many different convenience samples of U.S. citizens, fielded respectively on November 5, 2018 (study 1, initial N ¼ 1,218), May 10, 2019 (study 2, initial N ¼ 1,508), December 6, 2019 (study 3, initial N ¼ 1,106), and May 12-13, 2020 (study 4, initial N ¼ 2,010). In all studies respondents were recruited through Amazon's MTurk (Paolacci & Chandler, 2014), an online crowd-sourced data platform, and offered a small monetary incentive to complete a short survey (respectively, $0.6 for study 1, and $0.7 for studies 2-4). MTurk provides convenience samples, which cannot be assumed to be representative of the population. Yet, a number of studies have found that MTurk produces results that are similar to 5 S H I F T I N G P U B L I C P E R C E P T I O N S O F T R U M P ' S P E R S O N A L I T Y more traditional surveys. For instance, Berinsky, Huber, and Lenz (2012) find that MTurk samples tend to be more representative of the U.S. population than other types of convenience samples, and Clifford, Jewell, and Waggoner (2015) report that MTurk samples tend to mirror the psychological divisions of liberals and conservatives in the U.S. general population. All in all, evidence exists that MTurk offers a good "bang for the buck" when it comes to diverse convenience sample (Hauser & Schwarz, 2016; for a more critical take, see Ford, 2017;Harms & DeSimone, 2015).
The four questionnaires included additional batteries, which we discuss elsewhere (e.g., Nai & Maier, 2020). Most notably, studies 2-4 included an experimental component, where respondents were exposed, for example, to mock campaign messages mixing humor and negativity (study 2). The batteries about Trump personality appeared at the end of the questionnaires, after these experimental components. Supplementary Appendix A discusses the existence of "treatment biases." Broadly speaking, there are only a handful of significant differences driven by exposure to these treatments, and these effects are often rather marginal (but see Supplementary Appendix Table A3). Especially given that these effects are arbitrarily spread across all partisan groups due to experimental randomization, we can confidently exclude any concerns over biases due to experimental conditions. It is also important to note that these four studies represent as many independent samples, that are not linked within a panel design. As such, we do not claim to measure perception shifts at the individual level, but rather at the aggregate level across party identification groups-much in the same way as approval rate of political leaders is often plotted over time across independent polls (see, e.g., Supplementary Figure B9).

Measures
Trump Perceived Personality. All four questionnaires asked respondents to rate President Trump's personality using two short scales usually employed for the selfassessment of respondents: the TIPI (Gosling et al., 2003) for the Big Five (extraversion, agreeableness, conscientiousness, emotional stability, and openness), and a simplified version of the D12 inventory (Jonason & Webster, 2010) for the Dark Triad (narcissism, psychopathy, Machiavellianism). Both batteries asked respondents whether they agree or disagree with a series of statements that might or might not apply to Donald Trump (e.g., "In your opinion, Donald Trump might be someone who. . . is extraverted, enthusiastic," ". . . is critical, quarrelsome," ". . . tends to want others to admire him"). Pairs of statements were combined to measure the Big Five and the Dark Triad traits (8 traits in total); all personality variables vary between 1 "very low" and 7 "very high." To be sure, with only a handful of items per trait these "short" batteries cannot capture all nuances and facets of complex personality constructs ( Trump's average personality profile across all respondents is illustrated in Supplementary Appendix Figure B1, showing an extremely consistent profile across the four studies (run on four independent samples at $6 months intervals). The average opinion of respondents is that Donald Trump scores very high on extraversion and the Dark Triad (especially on narcissism), and low(er) on openness, conscientiousness, emotional stability, and (especially) agreeableness. These trends are broadly in line with what has been discussed in several academic studies relying on expert ratings of Trump's personality (Nai & Maier, 2018;Nai, Martínez i Coma, & Maier, 2019;Visser et al., 2017). Of course, the scores in Supplementary Appendix Figure B1 represent only an average assessment across all type of respondents. Also in alignment with several previous studies (Hyatt et al., 2018;Wright & Tomlinson, 2018), we find profound differences in how Trump's personality is perceived exist across ideological lines.
Partisanship, Affective Polarization, and Turnout. In the four studies we measured party ID following the protocol used in the American National Election Study. Respondents were first asked whether they usually think of themselves as a Democrat, a Republican, or an independent; in the first two cases respondents were then asked if they would call themselves a strong Democrat/Republican (strong/not very strong), whereas respondents initially identifying as independents were asked whether they would think of themselves as closer to the Republican Party or to the Democratic Party (or neither). Combining these different questions yields a 5-point scale, from Strong Democrat to Strong Republican. In parallel, all respondents were also asked to rate themselves on the standard 0-10 left-right scale. Supplementary Appendix Table B1 presents the distribution of respondents in the four samples on these two scales.
In study 3, we asked respondents to what extent they agree or disagree with a series of statements describing "people who are Republicans" (and, separately, to describe "people who are Democrats"). The statements they had to evaluate were "patriotic," "closed-minded," "intelligent," "hypocritical," "selfish," "honest," "open-minded," "generous," and "mean". As described in Iyengar et al. (2012), the overall evaluation of these statements in light of the partisan affiliation of the respondents provides measures of the extent to which respondents hold "negative stereotypes for the out-party," on the one hand, and "positive stereotypes for the in-party," on the other hand-that is, to what extent they really like their party and really dislike their opponents, beyond partisanship. Respondents identifying as independents or neither as Republicans nor as Democrats (N ¼ 114) are excluded. The measures perform well in reliability (a ¼ 0.91 throughout), and are, of course, negatively correlated, r(965) ¼ À0.32, p < .001. Averaging these two variables provides us with a unified measure of affective polarization that is independent of respondents' partisanship; the variable ranges from 1 "very low affective polarization" and 7 "very high affective polarization" (M ¼ 4.92, SD ¼ 0.88). Interestingly, the magnitude of affective polarization decreases when respondents move towards a (strong) Republican identification, r(965) ¼ À0.20, p < .001. In other words, strong Democrats are holding the highest negative stereotypes towards the other party and the strongest positive feelings towards their own party.
Finally, in study 3, we also asked respondents how likely it is that they would participate in the election for the House of Representative "if such election would happen tomorrow and they would be eligible to vote" (from 1 "extremely unlikely" to 7 "extremely likely"). In general, respondents are quite likely to participate (M ¼ 6.0, SD ¼ 1.5), especially respondents leaning towards a (strong) Democratic party identification, r(1,039) ¼ À0.09, p ¼ .004.

Results
Partisanship and Perceived Trump Personality U.S. politics is increasingly defined by affective polarization (Iyengar et al., 2012), and it should thus come as no surprise that the way people perceive the current, controversial President is a strong function of their partisan affiliation. If we regress the eight traits of Trump's perceived personality on the respondents' self-assessed left-right position, the trends are very clear. For instance, for study 1, the more the respondents position themselves towards the right the more they will rate the President high on the Big Five (except for extraversion, on which most respondents overall agree that the President scores very high), and low on the Dark Triad. See Supplementary Appendix Figure B4. The zero-order correlations are especially high for conscientiousness, r(1,183) ¼ 0.63, p < .001, psychopathy, r(1,183) ¼ À0.60, p < .001, and Machiavellianism, r(1,183) ¼ À0.59, p < .001. Very similar trends exist also in studies 2-4 (Supplementary Appendix Figures B5-7). All in all, right-leaning respondents tend to have a more "positive" image of Trump's personality (higher on the Big Five, and lower on the Dark Triad), whereas the opposite is true for left-leaning respondents.
Supplementary Appendix Figure B2 does the same but decomposes the sample across the five categories of party identification (study 1). Except for Extraversion, the drastic differences between (strong) Democrats and (strong) Republicans appear very clearly. The gap is especially remarkable for conscientiousness, psychopathy, and Machiavellianism; strong Democrats rate the President low on conscientiousness (M ¼ 2.2, SD ¼ 1.6) while strong Republicans rate him quite high on this trait (M ¼ 5.6, SD ¼ 1.5); the difference between the two averages is considerable, t(461) ¼ À21.13, p < .001, d ¼ 2.17. Inversely, strong Democrats rate the president high in Machiavellianism (M ¼ 6.5, SD ¼ 1.1), whereas strong Republicans rate him lower on this trait (M ¼ 3.8, SD ¼ 1.7). Again, the difference is substantial, t(461) ¼ 19.83, p < .001, d ¼ 2.04. A similar trend exists also for psychopathy. Interestingly, these differences are great enough that strong Democrats and strong Republicans have a reversed image of the President across these traits; for the former Trump scores much higher in Machiavellianism and psychopathy than in openness, whereas for the latter the reverse is true. These trends are, broadly speaking, in line with what found in other recent studies (Hyatt et al., 2018;Wright & Tomlinson, 2018).
On all eight traits, the effect of partisanship is significant at p < .001 even in multivariate models controlling for respondent's socio-demographic profile (gender, age, education, and race), interest in politics, and personality traits (see materials in the OSF repository).
We argued in the introduction that the diverging opinions about Trump's personality across different partisan identifications are not surprising, given the current levels of affective polarization in the U.S. electorate. Study 3 included batteries allowing us to quantify respondents' affective polarization (strong negative stereotypes of the outparty and strong positive stereotypes for the in-party; Iyengar et al., 2012). Unsurprisingly, the contrasts in perceptions about Trump personality between Democrats and Republicans are especially strong for respondents high in affective polarization. Supplementary Appendix Figure B8 illustrates this trend. All panels in the Supplementary Appendix Figure B8 substantiate the perceived personality of Trump as a function of the interaction between partisanship and affective polarization, via marginal effects. All models are controlled by the respondents' profile (gender, age, education, race/ethnicity, and interest in politics); full results are in Supplementary Appendix Table B4.

The Evolution of Trump Perceived Personality
The trends discussed so far find confirmation in recent work done by us and other scholars. However, the existing literature has not yet shown whether, and if yes, to what extent, public perceptions of Trump's personality shift over time. Shifts could be expected given the sheer number of scandals and controversies that the Trump administration has been engulfed in since its first days. The repeated measures in our four subsequent surveys allow us to test this intuition. Figure 2 plots the evolution of Trump's perceived personality for strong Democrats and Strong Republicans, and for each of the eight traits. In all panels, the evolution across the four studies is presented chronologically in decreasing levels of color intensity. See Supplementary Appendix Figure B3 for other partisan affiliations (leaning Democrats, leaning Republicans, and independents).
As shown in the figure, strong Democrats have a rather consistent perception of Trump's personality over time, especially on the Dark Triad. Their perceived levels of Trump's agreeableness, conscientiousness, and emotional stability is perhaps slightly less critical in more recent data (studies 3 and 4), but these differences are minimal. Leaning Democrats also show a rather consistent perception of Trump, even if they tend to become somewhat more critical over time in terms of Trump's conscientiousness. Overall, however, the position of respondents identifying with the Democratic party does not seem to fluctuate over time. The position of respondents that either identify as independents or as neither Democrats nor Republicans (Supplementary Appendix Figure B3b) fluctuates a bit over time-for instance, the level of perceived agreeableness is slightly higher in study 4 when compared with the other three studies-but not in a substantial way, nor following a consistent trend. The same can be said for respondents leaning towards the Republican party (Supplementary Appendix Figure B3c). The clearest trend, as shown in the figure, can be found for respondents strongly identifying with the Republican party. Contrasting with the classical "Rokeachean" narrative that portrays hardcore Republicans (and conservatives, more in general) as closed-minded and dogmatic (Costin, 1971;Di Renzo, 1968;Rokeach, 1960; but see Conway et al., 2016), it is in this category of respondents that we find the strongest perception shifts over time. Over time, if we compare their scores for study 1 (November 2018) and study 4 (May 2020), strong Republicans rate Trump as significantly lower in extraversion and conscientiousness. Even more notably, their perception of Trump becomes more critical when it comes to the "dark" traits; they rate him higher in narcissism, and substantially higher in psychopathy and Machiavellianism. Supplementary  Table B2 reports the results of series of t-tests that compare, for each of the five partisan identifications, the respondents' perception of trump personality in studies 1 and 4. As the Supplementary Appendix Table B2 shows, it is especially among strong conservatives that we find the greatest shifts in perceptions across the two studies.
Supplementary Appendix Table B2 also shows some significant reversed results for strong Democrats: comparing the first and last samples, they seem somewhat slightly less critical of the President. However, the magnitude of the effects for strong Democrats is substantially more marginal. Because our studies were relatively highpowered, even more than significance of the effects (which are more likely in bigger samples) what matters is their magnitude. Thus, Supplementary Appendix Table B2  also reports, for each t-test, its effect size (Cohen's d). It is unquestionably among strong Republicans that we find the biggest adjustments, as indicated by effect sizes that are up to d ¼ 0.50, for extraversion.
Because of the potential of capitalization on chance due to the presence of multiple comparisons, Supplementary Appendix Table B3 reports the detailed results of nonparametric pairwise multiple comparisons using a Dunn's test that applies a Bonferroni correction (Dinno, 2015). Importantly, the significant differences between studies 1 and 4 for strong Republicans are still significant, even in this more demanding testing environment. To test whether the evolution across the four studies is statistically meaningful even controlling for the respondent's profile, and to compare the shifts across all partisan identifications at once, we have pooled the data of all studies together into a stacked dataset. Table 1 regresses the respondents' perception of Trump personality traits on their profile and their partisan identification; an interaction term between their partisan identification and the study number (from 1 to 4) tests for the changing effect of partisanship across the four studies. As the Table 1 shows, the changing effect of partisan identification on Trump's perceived extraversion, conscientiousness, narcissism, psychopathy, and Machiavellianism across the four studies is statistically significant even when controlling for the respondents' profile. More specifically, as shown earlier with bivariate analyses, even controlling for the respondent profile, more conservative respondents have an increasingly more negative perception of Trump's personality over time.

Turnout
Is the way respondents perceive Trump relevant for their political participation? In light of the trends discussed above, respondents that have a more negative perception of Trump could potentially have lower political mobilization. Table 2 provides some insights into this, using data again from study 3. The Table 2 presents four models; models M1 and M3 regress the self-assessed likelihood to participate if an election for the House would happen tomorrow on Trump's perceived Big Five personality traits. Models M2 and M4 do the same, but for Trump's perceived Dark Triad personality traits. Given the overlaps between these inventories, we decided to run separate models. For each inventory, the first model tests for its direct association on estimated participation (M1 and M2), whereas the second one tests whether this effect is moderated by partisanship (interaction effects in models M3 and M4).
Models M1 and M2 confirm some known trends. Higher electoral participation is associated with greater political interest, education, and age. Assessments of the president as more extraverted, neurotic and narcissistic are associated with higher intentions to participate. But, what matters the most for us here is whether partisan differences intervene, moderating the association of Trump's perceived personality and intentions to turn out. Models M3 and M4 include interaction effects to test for this assumption. Starting with the Big Five, model M3 shows a significant interaction between partisanship and Trump's perceived conscientiousness. Respondents that strongly identify with the Republican party are associated with a higher intention to participate if they perceive Trump as high in conscientiousness, whereas strong Democrats are associated with a lower intention. The opposite trend exists for Trump's perceived psychopathy (M4); respondents that strongly identify with the Republican party are less likely to

Note.
All dependent variables vary between 1 "very low" and 7 "very high." a The variable takes the value 1 for "study 1", 2 for "study 2", 3 for "study  Table   2.
Likelihood of Electoral Participation (Study  have higher intentions to participate if they perceive Trump as high in psychopathy, and vice-versa. These two effects are substantively relevant in light of the trends discussed in the previous section. As shown in Figure 2, over time strong Republicans are (a) increasingly less likely to perceive Trump as being high in conscientiousness, and (b) increasingly more likely to perceive him as high in psychopathy. With this in mind, the fact that in study 3 intention to participate is higher for those who rate him higher in conscientiousness and low in psychopathy could suggest the existence of a potential demobilization potential for strong Republicans.
Of course, what is tested in Table 2 is not the effect of increasingly more negative perceptions of Trump over time-only analyses with panel data could allow us to test that-but rather how perceptions on a given moment in time are associated with intended participation. Because we cannot test for the causality between perceptions and intentions to vote, the effects shown could also reflect the fact that strong Republicans who do not intend to turn out subsequently develop a more negative image of Trump's personality. Especially in light of what discussed above in terms of blame attribution, it is not impossible that frustrated Republicans that decided not to participate resolve their cognitive dissonance by building up a more negative perception of their party leader. Further analyses combining panel and experimental data should keep this alternative causal mechanism in mind.

Weighted Analyses
Some scholars argue for the use of weights for a better interpretation of results when using MTurk samples (e.g., Levay, Freese, & Druckman, 2016). Supplementary Appendix C replicates all analyses discussed in this article but using a weight for age groups by gender-that is, accounting for the differences between the sample distributions and the U.S. general population on the combination of these two factors (data from U.S. census 2018). Most notably, Supplementary Appendix Table C2 replicates the "stacked" analysis discussed above in Table 1 and shows that even in the weighted data conservatives have, over time, a more critical perception of president Trump: lower extraversion (but not conscientiousness), and higher narcissism, psychopathy, and Machiavellianism. Results for turnout (Supplementary Appendix Table C3) are also, broadly speaking, consistent with the trends in the unweighted samples.

Conclusion
Especially in times of heightened affective polarization (Iyengar et al., 2012), the perception of political leaders is a highly partisan phenomenon-and Trump is not an exception (Dunn, 2018). Indeed, consistent evidence across several studies shows that voters of different partisan identifications hold divergent perceptions of the President's personality and character (Hyatt et al., 2018;Williams et al., 2018;Wright & Tomlinson, 2018); liberals tend to highlight his low agreeableness, low conscientiousness, and low emotional stability, whereas conservatives tend to rate the President much higher on all the Big Five, and especially on openness and conscientiousness. The existence of these partisan differences is unquestionable. Yet, in light of the many controversies that have marred the Trump presidency that were particularly at odds with conservative "moral foundations" (Graham et al., 2009;Haidt & Graham, 2007), a more negative perception of the president over time could have been expected. We described in this article the evolution of Trump's perceived personality traits-both in terms of the Big Five and the Dark Triad-over a period of $18 months (November 2018 to May 2020), via data from four subsequent convenience samples of American respondents (MTurk). Our results show that (a) across the four studies, the perceived personality of Trump is, on average, extremely stable. Aggregating all individual preferences and thus controlling out any ideological differences, respondents rate Donald Trump very high on extraversion and the Dark Triad (especially on narcissism), and low(er) on openness, conscientiousness, emotional stability, and (especially) agreeableness. These trends are broadly in line with what discussed in several academic studies relying on expert assessments (Nai & Maier, 2018;Nai, Martínez i Coma, & Maier, 2019Visser et al., 2017; (b) Liberals have a much more critical perception of Trump (lower on the Big Five and higher on the Dark Triad) than conservatives. This is hardly a surprise given the current levels of affective polarization, and confirms trends discussed in other studies (Hyatt et al., 2018;Wright & Tomlinson, 2018); (c) Over time-that is, over the course of 18 months-only one group significantly shifts their opinion about Trump: strong Republicans. This group has a consistently more negative perception of the President over time: they increasingly score Trump lower on conscientiousness and extraversion, and higher on the Dark Triad. Furthermore, (d) a potential demobilization trend exists for strong Republicans; the more they perceive the president as high in psychopathy (and low in conscientiousness), the less they are likely to report a strong intention to turn out.
These results discussed are not only interesting from a descriptive standpoint, or for the American case. Perceived personality matters. In contrast to characteristics related to a candidate's professional record-for example, competence and strength of leadership-personal characteristics such as candidates' character and personality can be quite easily assessed without necessarily needing a deeper understanding of their performance or policy agendas. Indeed, several studies have shown that voters are able to assess the personality of politicians within milliseconds even based on scattered nonverbal cues (e.g., Olivola & Todorov, 2010). Since the early studies on public opinion and electoral behavior (e.g., Campbell, Converse, Miller, & Stokes, 1960), a sizable scholarship shows that candidate evaluation is one of the most important drivers of voting pattern. The perception of personality traits is closely associated with the broader image that voters form of the competing candidates. For instance, Williams et al. (2018) show an association between the perception of leader's narcissism and their perceived charisma, leadership performance, and effectiveness. Even more fundamentally, consistent evidence shows that voters' perceptions of the leaders' traits are a powerful driver of their electoral behavior (Garzia, 2013). All in all, a narrower focus on Trump's perceived personality traits is both theoretically relevant and potentially practically salient. The identified shifts in (aggregate) perceptions of the President's personality across different political groups could potentially help prognosticate subsequent alterations in his electoral support and could be seen as the first fissures in his Teflon coating.
The trends discussed here are informative about the interplay between personalization of politics, scandals, and voters' perceptions of leaders-also beyond the Trump case. Taking a step back, the trends discussed here for Donald Trump could even be seen as a particularly conservative test for the potential detrimental effects of political scandals on leaders' perceptions, and subsequent electoral behavior. Trump's 16 I N T E R N A T I O N A L J O U R N A L O F P U B L I C O P I N I O N R E S E A R C H personality traits have consistently been shown as somewhat "extreme" (e.g., Visser et al., 2017), and the high levels of affective polarization in the United States today make that voters likely stick to their political opinions. In this sense, the moderate but substantive negative opinion shifts shown here could be found in a much more dramatic way in countries where leaders have not such an extreme personality in the first place and voters are less obfuscated by hyperpartisanship.
Finally, the results presented in this article have a clear methodological implication for the study of public opinion. Our two sets of results-that is, not all traits are assessed more negatively over time (there is variation in perceptions) and not all perceptions affect turnout equally (there is variation in the relevance of trait perceptions for subsequent behavioural attitudes)-strongly indicate that personality perceptions are not simply functional proxies of general attitudes towards the president (e.g., job approval ratings or feeling thermometer scores). Perceived personality traits reflect a richer measure of the leaders' image in the mind of voters, and as such are likely to foster a more nuanced and deeper understanding of the interplay between how voters "feel" about political leaders and their subsequent behavior. As such, the inclusion in public opinion surveys of personality perception batteries for competing candidates or elected officials is recommended.