Adaptive and Dark Personality in the COVID-19 Pandemic: Predicting Health-Behavior Endorsement and the Appeal of Public-Health Messages

Who embraces directions to socially distance, boost hygiene, and protect others during a pandemic of contagious respiratory disease? Do differently phrased public-health messages appeal to different people? I based predictions on the five-factor, triarchic psychopathy, and Dark Triad models of normal-range and dark traits; the extended parallel process model (EPPM); and schema-congruence theory. In a survey of 502 online participants, normal-range traits (esp agreeableness and conscientiousness) predicted endorsement of social distancing and hygiene, as well as the appeal of health messages in general. Consistent with the EPPM, conscientiousness and neuroticism had an interaction. Dark traits (esp psychopathy, meanness, and disinhibition) predicted low endorsement of health behaviors and the intent to knowingly expose others to risk. Most participants preferred a message appealing to compassion (“Help protect the vulnerable…”), but dark traits predicted lower appeal of that message. Personality appears relevant to epidemiology and public-health communication in a contagious-disease context.

More narrowly, personal hygiene's links to personality have hardly been studied. Personality-bound processes like disease worrying (Liao et al., 2011) seemed to predict hygiene during an influenza pandemic, implicating neuroticism. Extroversion and (low) neuroticism may predict oral hygiene (Kressin et al., 1999).
Research on nonrespiratory communicable diseases and the FFM may inform predictions about responses to contagious SARS. Aspects of neuroticism, (low) conscientiousness, and (low) agreeableness may predict HIV risk behavior (Trobst et al., 2002) and sexually transmitted infection (Mõttus et al., 2012). Would this generalize to SARS?
Furthermore, conscientiousness and agreeableness may predict adaptive behavior in a pandemic synergistically, as they interact in academic and job performance (Witt et al., 2002). Additionally, the extended parallel process model (EPPM) of persuasion in PHMs suggests a synergy between threat sensitivity (entailed in neuroticism) and self-efficacy (linked to conscientiousness) in predicting people's fear-appeal susceptibility (Witte, 1992).

Health Behavior and Dark Personality
Apart from the FFM, clinical conceptualizations of maladaptive personality may explain people's diverse reactions. Sensation seeking and neuroticism's clinical manifestations predict many risk behaviors, including HIV-risk sexual behavior (Kalichman et al., 1998;Razei et al., 2017), so do antisocial and borderline personality disorder features (Adams et al., 2016;Kelley & Petry, 2000;Ladd & Petry, 2003). Because it relates to these disorders and sensation seeking, such findings implicate psychopathy.
Psychopathy and, to some extent, narcissism and Machiavellianism predict not only antisocial tendencies (including aggression and violence) but also various health behaviors and outcomes (Hudek-Knezevic et al., 2016;Jonason et al., 2015;Malesza & Kaczmarek, 2019), including HIV-risk sexual behavior (Hudek-Knezevic et al., 2007). Like the FFM's normalrange traits, the Dark Triad has not been studied in relation to PHMs or health-related behaviors in a SARS pandemic.

Dark Personality and Harm to Others' Health
If maladaptive personality partially explains reactions to the pandemic and to PHMs, it matters to public health. For example, it may explain who intentionally spreads, or threatens to spread, disease-which raises societal concern (Dryer, 2020). Sensation seeking's link to HIV-risk behavior in infected persons has such implications (Kalichman et al., 2008;Shuper et al., 2014). Findings link psychopathy to deliberately misleading sex partners about one's HIV-positive status (Benotsch et al., 2012) and narcissism to knowingly putting others at risk of HIV (Martin et al., 2013). More generally, psychopathy and the Dark Triad correlate with endorsement of unethical behavior (Pletti et al., 2017;Roeser et al., 2016). During a pandemic, dark personality may predispose people to dismiss or act contrary to PHMs and endanger others.

Personality and Health Communications
Above, I theorized links between personality and health-related behavior during the COVID-19 pandemic. Additionally, personality may have applied implications (Ferguson, 2013), particularly for communicating health advice (Dutta-Bergman, 2003). It may partly explain what PHMs are persuasive to different people. According to schema-congruence theory (Brock et al., 1990), people prefer and find more persuasive messages that align with their views of themselves. This has been suggested in commercial advertising (Chang, 2005;Matz et al., 2017), antismoking campaigns (Chang, 2009), diabetes education (Lawson et al., 2010), college drinking (York et al., 2012a), dental-hygiene promotion (Sherman et al., 2008), and condom use (Noar et al., 2006). If personality is reflected in their self-concept, people may favor messages that match their self-reported traits.

The Current Study
Given the paucity of research on personality in relation to SARS-protective and risk behaviors during a pandemic, I tested such relationships using a survey. Measuring actual behaviors would be preferable, yet self-reported current and intended behaviors provide a starting point to probe the boundaries of personality theory.
Among the constructs discussed above, conscientiousness, agreeableness, (low) extroversion, neuroticism, (low) boldness, and (low) disinhibition emerged as candidates to predict selfreported ongoing SD. Conscientiousness, agreeableness, neuroticism, low boldness, disinhibition, and psychopathy may predict endorsement of intent for future SD (presumably because of growing knowledge and PHMs about the pandemic). Conscientiousness, agreeableness, (low) disinhibition, and (low) meanness are likely to predict intent for future hygiene (to protect not only the self but also others). Psychopathy, meanness, disinhibition, and Machiavellianism may predict endorsing the willingness to infect others.
Additionally, conscientiousness and agreeableness, and conscientiousness and neuroticism would interact in predicting current SD and future intent for SD (as per prior research cited above and the EPPM).

Method
After preregistering the study on March 20, I collected data from March 20 to 23, just after California and New York (followed by other states) issued shelter-at-home orders. At that time, Americans had varied and evolving opinions about the pandemic, whether it posed risk to them and whether the public was overreacting (Rakich, 2020). Participants U.S. Mechanical Turk (MTurk) workers with master status (N ¼ 502) received US$2.50 to complete a 15-min survey. Expecting small effect sizes and limited by funding, I planned to obtain 500 cases for .85 power (Algina & Olejnik, 2003) to detect partial correlations of .15 at p ¼ .01 (one-tailed) after data cleaning. I anticipated careless (possibly automated) responding on MTurk and planned to screen the data for duplicate, automated, and invalid responding before testing hypotheses. Of 652 people who reached the survey's end, 616 passed effort checks. Of them, 562 passed validity checks, 540 passed subsequent inconsistency checks, 503 passed a completiontime check, and 502 had fully completed all questionnaires (Supplemental Material Tables S1-S4). No duplicates were evident. Thus, 23% of response sets were excluded as noncompliant or suspicious. Participants reported being 51% male, 77.5% White, 47.2% single, and 60.4% full-time employed. Age ranged from 21 to 76 years (M ¼ 41, SD ¼ 11), and 72.5% denied having health conditions that increase risk from COVID-19 (Table S8).

Measures
Based on WHO and Centers for Disease Control and Prevention advice for preventing COVID-19's spread, I listed 10 health behaviors coinciding with widely popularized advice (Nierenberg, 2020). Conceptually, they represent SD and hygiene. From that list, I wrote two measures to capture selfreported current and intended future health behaviors (FHBs). A third questionnaire, with broader content, was intended to measure covertly participants' willingness to deliberately put others at risk (vs. following advice to protect others).
Current health behavior (CHB; 10 items) asked participants how much they currently did each behavior as a precaution against COVID-19 (e.g., "Limit contact with people as much as possible . . . ") on a 4-point scale from not at all or never to extremely or always. I expected the data to be reducible to CHB: SD and CHB: hygiene. A principal component analysis (PCA) corroborated this (Table S5). Thus, although CHB's internal consistency was high, a ¼ .89, I reduced its data to two dimensions. Table S10 contains descriptives.
FHB (14 items) asked participants to rate on a 5-point scale from extremely likely to not likely at all their likelihood to do each behavior over the next 2 weeks. Scales were keyed opposite to CHBs (and later reversed) for data control purposes. Ten items paralleled the CHBs. Four captured more venturous interpersonal contact, from going to a large gathering to having sex with someone new. I expected the data to be reducible to FHB: SD, FHB: hygiene, and venturous behavior. Internal consistency was acceptable, a ¼ .76, and PCA supported the intended structure (Table S6).
Carrier scenario (CS; 18 items) asked participants to rate on a 4-point scale (from not at all to definitely) the likelihood of specific behaviors if they had good reason to think they were carrying COVID-19. The behaviors ranged from calling ahead prior to seeing a physician, to wearing a mask, to shaking hands in different situations, and to purposefully trying to infect others. CS (a ¼ 71) was readily reducible to two components (Table S7): harmful behavior (HB) and protecting others (POs).
PHM appeal. Participants viewed, in a randomized order, five PHMs written to appeal to different personalities: selfcentered, responsible, compassionate, avoidant, and sociable (Appendix SA). They rated from 0 to 100 how much each message (a) was persuasive to them, (b) was likely for them to take seriously, and (c) was likely to affect their behavior. Each PHM's composite score was internally consistent (a ¼ .93 to .95). Participants also viewed distinctive lines from each message with instructions to rank order the PHMs based on personal appeal.
Personality dimensions. The Mini-International Personality Item Pool (IPIP) (MIP; Donnellan et al., 2006) is 20-item form of the IPIP (Goldberg, 1992(Goldberg, , 1999 developed to maximize short-form validity and factor separation. It captures the FFM's general trait domains (agreeableness, conscientiousness, etc.). Participants responded on a 5-point scale from very false to very true. For exploratory purposes, I computed personality superfactors: stability, plasticity, and the general P factor (Rushton & Irwing, 2011; Tables S11 and S15).
The Abbreviated Measure of Psychopathy (AMP; Semel, 2018) operationalizes the triarchic psychopathy model with 33 items as a parsimonious alternative to Patrick's (2010) longer instrument. It has a 4-point scale from true to false and yields scores on boldness (e.g., "I have a very strong and dominating personality"), meanness (e.g., "It sometimes gives me pleasure to see someone in pain"), and disinhibition (e.g., "The saying 'plan ahead' is definitely not for me"). For exploratory purposes, I also employed the overall score.
The Short Dark Triad (SD3; Jones & Paulhus, 2014) is a 27item measure yielding scores on narcissism (e.g., "Many group activities tend to be dull without me"), Machiavellianism (e.g., "I like to use clever manipulation to get my way"), and psychopathy (e.g., "Payback needs to be quick and nasty"). Participants responded on a 5-point scale from disagree strongly to agree strongly. I also utilized its overall score to estimate the dark factor, D (Moshagen et al., 2020).

Additional considerations.
To minimize the effects of global response sets, I keyed the measures' scales both ways. Additionally, about half of the MIP and multiple AMP and SD3 items were reverse keyed. Personality questionnaire names were disguised.
I planned to control for gender, age, and presence of health conditions that increase risk from COVID-19. These could be expected to covary with both personality and pandemicrelated behavior. Extreme responses are of interest when studying maladaptive behavior, so I did not plan deleting outliers. I adopted a ¼ .01 (one-tailed) for 37 partial-correlation hypotheses (as noted earlier) and a ¼ .05 for four interaction hypotheses .
The hypothesis that conscientiousness and agreeableness interact to predict FHB: SD received support at a ¼ .05  (b ¼ .09, p ¼ 2.9 À2 ), but the 95% CI for b included 0, rendering this inconclusive. It raises the possibility that conscientiousness and agreeableness may predict FHB: SD synergistically, conscientiousness predicting FHB: SD slightly less well at higher (b ¼ .12, p ¼ 7.4 À2 ) than lower (b ¼ .17, p ¼ 4.3 À3 ) agreeableness levels. The hypothesis that conscientiousness and agreeableness would interact in predicting CHB: SD did not receive support (b ¼ À.05, p ¼ 2.9 À1 ).
The hypotheses that PHM2: Responsible's appeal would correlate with conscientiousness (r p ¼ .15, p ¼ 4.6 À4 ) and disinhibition (r p ¼ À.11, p ¼ 7.1 À3 ) received support. However, as was generally the case for PHMs, its appeal also correlated with other traits: positively with adaptive and negatively with maladaptive ones.

Discussion
This research extends personality theory to differences in people's reactions to a pandemic of communicable SARS. The five-factor, triarchic psychopathy, and Dark Triad models informed such questions as: What traits predict intent for SD and hygiene, endorsement of behavior that risks others' health, and the appeal of differently phrased PHMs? Such knowledge may aid individual risk prediction (Chapman et al., 2019). It may have applied implications when a minority of people dismisses conventional PHMs (de Bruin et al., 2020).
Twenty-four of the 40 hypotheses received support. This was true particularly for hypotheses using trait theory to predict people's self-reported current and intended FHB, as well as risks they may create for others. Although most correlations were small (a few were medium), they may reflect meaningful patterns in behavior over time. Self-reported traits predict traitrelevant behavior (Fleeson & Gallagher, 2009), including health behaviors sampled in real time (Kroencke et al., 2019). Furthermore, behavioral intentions likely predict actual behavior (Webb & Sheeran, 2006). The results probably reflect true trait-behavior connections.

Normal-Range Personality
Regarding the FFM, 8 of the 11 hypotheses received support. In planned and EAs, agreeableness and conscientiousness predicted endorsement of SD and hygiene. This agrees with prior literature on health (Strickhouser et al., 2017), health behaviors (Roberts et al., 2005), and nonrespiratory contagions (Mõttus et al., 2012). Neuroticism may play a small role in hygiene intent, whereas extroversion's negative links to SD and hygiene vanished after controlling for other traits. Consistent with the EPPM (Witte, 1992), conscientiousness and neuroticism interacted in predicting adaptive responses to the pandemic (perhaps by increasing susceptibility to PHMs). Somewhat consistent with prior research on work and school behavior, conscientiousness and agreeableness showed a trend toward a synergistic interaction.
If such findings replicate, future research should address mechanisms behind them. Such mechanisms may include the empathic concern for others and flexibility in interpersonal behavior entailed in agreeableness (Graziano & Tobin, 2017) and the industrious self-efficacy and self-regulation involved in conscientiousness (Jackson & Roberts, 2017).

Dark Personality
Regarding triarchic psychopathy and the Dark Triad, 9 of the 12 hypotheses received support. In planned and EAs, meanness and disinhibition (and overall psychopathy) as well as Machiavellianism (to a lesser extent) predicted lower intent for SD and hygiene. Together with boldness, the psychopathy traits predicted endorsement of risky, venturesome behavior (when participants were asked to imagine being disease carriers). This agrees with research on dark personality in relation to health risk (including nonrespiratory communicable disease; Malesza & Kaczmarek, 2019).
As predicted, meanness and disinhibition (and overall psychopathy, but not boldness) predicted endorsement of behavior that puts others at risk of infection (knowingly, and perhaps deliberately). This corroborates the link between psychopathy and unethical everyday behavior (Benotsch et al., 2012;Pletti et al., 2017). In EAs, disinhibition related to all undesirable behaviors, meanness was particularly implicated in putting others at risk, and boldness may be linked to (adaptive) intent for hygiene. This parallels research (in nonforensic samples), whereby boldness either does not predict maladaptive or predicts adaptive outcomes (Berg et al., 2017;Sörman et al., 2016).
Future research may ask whether similar mechanisms explain the links between the two sets of traits (maladaptive vs. adaptive) and health outcomes. Beyond this project's scope (see Limitations section), the ability of dark traits to predict health risk behavior incrementally (over and above normal traits) merits investigation.

Appeal of PHMs
Knowledge of the kind presented above becomes useful if it can inform public-health interventions, for example, to limit a pandemic. As noted earlier, researchers are testing personality theory's ability to inform the tailoring of effective communication in applied contexts. I tested whether personality predicted participant's ratings of differently phrased PHMs.
This was, indeed, the case, but not always in the predicted manner. Only 8 of the 17 predictions received support. This may reflect methodological limitations. For example, the PHMs may not have been congruent with the self-schemas they were meant to match, or schema-congruence theory cannot encompass self-reported traits and trait-congruent PHMs (York et al., 2012b).
Agreeableness and conscientiousness predicted the appeal of the compassionate PHM ("Help protect the vulnerable!"), conscientiousness of the responsible message ("Take personal responsibility!"), and neuroticism of the avoidant message ("Avoid the disease!"). Psychopathy, meanness, and Machiavellianism negatively predicted the compassionate message's appeal. However, such findings were message-nonspecific: Overall, adaptive traits predicted PHM's appeal, and maladaptive traits predicted their nonappeal.
Participants typically chose the compassionate message over others (including the self-centered one: "Keep yourself healthy!"). This parallels finding that appeals to altruism improve hygiene in analog (Betsch et al., 2013) and real-life (Grant & Hofmann, 2011) experiments. Yet, ranking the compassionate message as most effective was linked to lower meanness, disinhibition, narcissism, and Machiavellianism. Thus, appeals to altruism may work for most people but might backfire in antagonistic individuals.

Limitations
The influential personality models that informed this study are not free of potentially controversial assumptions (e.g., Watts et al., 2017). For example, Machiavellianism and psychopathy may not be truly separable (Miller et al., 2016). Indeed, their scores were highly intercorrelated and had similar associations with health-behavior endorsement. The models imply separation between normal-range versus maladaptive traits. However, much of the personality variation they encompass may be due to a smaller set of underlying traits with wide ranges of intensity (Hyatt et al., 2020;Lynam & Miller, 2019).
The nonprobability sample and potentially unusual characteristics of MTurk participants may raise concerns. Reviews of studies on MTurk in personality research (e.g., Chandler & Shapiro, 2016;Miller et al., 2017) have been favorable regarding the measurement, factor structures, and construct validity of normal-range and maladaptive traits (including from five-factor, Dark Triad, and psychopathy-related models).
The sample included participants from across the United States. However, most represented were California (13%) and New York (6%), already heavily affected by the pandemic. In conjunction with the study's timing, self-selection into research on COVID-19, and the participants' relatively high education, the sample's composition may have affected the results. The most likely impact is range restriction of the health and risk behavior scores. Indeed, consistent with polls, most participants highly endorsed health-protective behaviors. Thus, I may have underestimated the magnitude of trait-behavior correlations, perhaps promoting false negative results. This was countered by a respectably large sample. Future research, if using similar methods, may benefit from more sensitive rating scales to mitigate range restriction.
Screening out noncompliant and suspicious responses likely increased measurement reliability, thus promoting power to obtain true positive results. Conversely, screening may have limited the range of irresponsible, disinhibited, antagonistic, and deceptive (i.e., psychopathic) traits, making it harder to detect their links to other variables.
The self-report, cross-sectional, survey method using convenience sampling enabled rapid data collection at a unique historic moment. It also poses limitations, some of which (e.g., measuring intent instead of observing behavior; potential noncompliant responding) have been noted. I used brief personality measures for practicality; however, by limiting content validity and measurement reliability (Crede et al., 2012), this may have caused inaccurate estimation (most likely underestimation) of some correlations. Also by necessity, the health-behavior and PHM-appeal measures had not been pilot tested or previously validated. Besides age, gender, and health status, the study did not include potentially relevant predictors or suppressors, such as COVID-19 knowledge accuracy, misinformation susceptibility, political affiliation (Allyn & Sprunt, 2020), and regional policies. Furthermore, the measures were acontextual vis-à-vis interpersonal factors like kinship, in-group/out-group membership, and situational factors that may interact with personality.

Conclusion
In summary, early in the pandemic, prosocial traits likely facilitated SD and hygiene, and antagonistic traits likely detracted from adaptive behavior and promoted harm to people's health. Distinctly antagonistic persons may have disregarded or acted contrary to public-health appeals for altruistic behavior. This is an early demonstration that personality may matter in understanding communicable respiratory disease. More generally, personality has implications for contagious diseases and public-health communication. The findings, if replicated, may find applications in public-health messaging, individual risk prediction, and doctor-patient communication.
The results do not mean that it is mostly irresponsible and inconsiderate people who spread viruses. The correlations were often small, and the traits' scientific conceptualizations are not quotidian judgments about character. The results do not mean that people who contract a disease like COVID-19 have maladaptive traits. The findings do invite further research on personality in public health.

Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.

Supplemental Material
The supplemental material is available in the online version of the article.