Individuals differ not only in their physical characteristics, such as height, weight, facial attractiveness, and skin color, but also in their psychological makeup, including personality and intelligence. Both intuition and research suggest that these individual characteristics influence important aspects of a person’s life course, including health and longevity. For example, considerable evidence suggests that various aspects of personality are associated with mortality, as well as with health behaviors like smoking, drug use, and alcohol consumption (Smith and MacKenzie 2006; Ozer and Benet-Martinez 2006). Although controversial, the bulk of research indicates that cognitive ability, or intelligence quotient (IQ), affects somatic and mental health, as well as mortality (Gottfredson 2004; Deary et al. 2010). In addition, evolutionary psychology’s “good genes theory” posits that physical attractiveness predicts health and longevity because physical beauty accurately advertises health status (Thornhill and Gangestad 1993); in other words, according to this theory, physical attractiveness is a reliable indicator of underlying genetics which humans have evolved to distinguish and prefer.

Nonetheless, past studies have been limited in three important respects. First, these traits – personality, intelligence, and physical attractiveness – have considerable genetic heritability(Kern and Friedman 2011) and, therefore, are inseparable in a relationship with health outcomes. However, most past literature has examined the effect of only one or two traits on health. For a more comprehensive understanding of health risk factors, these individual features should be integrated in a model.

Second, most past studies have examined one personality trait at a time, or a single health outcome; thus, it is difficult to distinguish the relative importance of these personality domains or to determine whether they affect all health outcomes equally. Furthermore, the use of varying personality instruments has led to inconsistent findings across studies, and, although health is equivalent to neither a mere absence of disease nor a long life (Friedman et al. 2010), extant literature tends to focus on a single dimension of health. Thus, a more comprehensive approach should be pursued by employing not only objective, but also subjective, measures of health.

Third, it remains unclear how personality, intelligence, and physical attractiveness are correlated with each other and with health behaviors. These traits are closely interrelated (Langlois et al. 2000) and have a complex causal chain in the subsequent development of one’s life course, including socioeconomic status and health behaviors (Almlund et al. 2011; Heckman et al. 2006). Some studies have shown that health behaviors play a mediating role between personality and health (e.g., Schnittker 2004), while other studies have demonstrated a moderating role (e.g., Weiss et al. 2009; Chapman et al. 2010).

Using the Wisconsin Longitudinal Study (WLS) – a unique dataset that includes facialattractiveness ratings based on high school yearbook photos, measures of the Big Five personality traits, and IQ scores, along with rich sociodemographic and health information – the present study explores (a) which, and to what extent, personality traits are associated with health outcomes (i.e., all-cause mortality, self-rated health, depression, and number of physical symptoms); (b) whether IQ has protective effects on health; (c) whether facial attractiveness is associated with health; (d) whether these variables are mediated by health behaviors (i.e., smoking, alcohol consumption, and physical exercise); and (e) whether the foregoing relationships differ by gender. Moreover, this study compares and contrasts results from the WLS with results from the Americans’ Changing Lives (ACL) study, which sampled from a considerably more diverse population and which provided a similar set of predictors.

Interest in the link between personality and health can be traced back to the ancient Greeks, who believed that an imbalance among the four bodily fluids (blood, phlegm, yellow bile, and black bile) – called humors – could result in physical and psychological illness (Friedman and Booth-Kewley 1987; Kern and Friedman 2011). For example, the Greeks thought that an excess of black bile led to depression and eventually physical illness.

While the humors have been replaced with hormones, efforts to find a link between personality and health have continued in the modern era. A scientific approach to the personality-health relationship began with Friedman and Rosenman, who found that men with an “intense, sustained drive for achievement and being continually involved in competition and deadlines” (Friedman and Rosenman 1959: 1286) were seven times more likely to have clinical artery disease than matched controls. This type of behavioral pattern, later known as type-A personality, spawned a substantial number of studies examining the role of personality as a health risk factor (Deary et al. 2010).

Although several models of personality exist, Roberts defined personality traits as “the relatively enduring patterns of thoughts, feelings, and behaviors that reflect the tendency to respond in certain ways under certain circumstances” (Roberts 2009: 149). He suggested that personality traits should be distinguished from mental abilities, although both are affected by genetic predispositions. According to Roberts’ conceptual framework, personality traits and cognitive abilities, along with preferences (e.g., goals, interests, life tasks) and narratives (e.g., stories, significant memories), affect one’s identity and reputation, in turn, shaping individuals’ roles in society. Roberts also envisioned feedback processes among all components, such as the possibility that identity may affect personality traits and/or mental abilities.

With respect to the taxonomy of personality traits, a growing consensus amongpsychologists supports the five-factor model (FFM) of personality, or the Big Five traits (McCrae and John 1992; Goldberg 1993). The FFM is based on the notion that basic dimensions of personality can be discovered by decoding trait terms because all important individual differences have been noted in regular language (McCrae and John 1992). The FFM is comprised of extraversion, agreeableness, conscientiousness, neuroticism, and openness to experience or intellect (John and Srivastava 1999). Extraversion captures gregarious, energetic, and expressive features of behavior and includes traits such as sociability, activity, assertiveness, and positive emotionality. Agreeableness captures a prosocial orientation toward others and includes traits such as altruism, tender-mindedness, trust, and modesty. Conscientiousness describes multiple elements of persistence and impulse control in task- and goal-oriented settings, such as following norms and rules and planning, organizing, and prioritizing tasks. Neuroticism reflects emotional instability and negative emotionality, such as feeling anxious, nervous, sad, or tense. Finally, openness to experience refers to traits such as imaginativeness, curiosity, creativeness, and susceptibility to absorbing experience (John and Srivastava 1999; Costa and McCrae 1994; Goldberg 1993).

Of these five factors, neuroticism is perhaps most important to public health because it is robustly correlated with a wide array of both mental and physical health problems (Lahey 2009). Empirical evidence indicates that the predictive power of neuroticism on health outcomes is not trivial. For example, using data from the National Survey of Midlife Development in the United States (MIDUS), Chapman et al. (2010) found that 15.9 percent of the population mortality was attributable to low socioeconomic status (SES), 11 percent to high neuroticism, and 8.6 percent to the combination of low agreeableness and conscientiousness. Hence, the effect of neuroticism on mortality at the population level rivals the effect of SES.

More recently, conscientiousness has also been identified as a significant predictor of health outcomes, and its implications for public health also appear to be significant. Roberts and his colleagues (2007) reviewed evidence from 34 studies on the predictive power of the Big Five personality traits compared to that of IQ and SES for longevity. Their meta-analysis showed that conscientiousness was a stronger predictor of longevity than any other Big Five trait and was also a stronger predictor than either IQ or SES. Likewise, using data from the Terman Longitudinal Study, Savelyev (2012) demonstrated that, even after accounting for measurement error and reverse causality, omitting conscientiousness from a model was comparable to the bias from omitting all other control variables, including parental education, occupation, and origin, as well as many other variables that predict mortality.

Unlike the literature on neuroticism or conscientiousness, existing literature on the effects of agreeableness, extraversion, and openness to experience on mortality is inconsistent (Roberts et al. 2007). In the Terman sample, Martin et al. (2002) found that cheerful children were more likely to engage in risky health behaviors, which increased their risk of premature death. On the other hand, Taylor (2007) suggested that extraverted individuals tend to havestronger social support networks, which have protective effects on mortality. Similarly, Friedman et al. (2010) showed that agreeableness was associated with better physical health and subjective well-being for men, but not for women. The authors noted that agreeable individuals are more likely to build strong social relationships with others, but health may deteriorate if agreeableness surpasses a certain threshold.

Some researchers have found a positive effect of openness on mortality. For instance, in a population-based prospective study, openness predicted all-cause mortality after accounting for medical risk factors, health behaviors, and individual differences in hostility and neuroticism (Surtees et al. 2003). In contrast, Savelyev (2012) claimed that, in the Terman sample, the latent factors of early openness and extraversion did not contribute to explanatory power for mortality. In addition, Maier and Smith (1999) showed that the coefficient for openness became insignificant after controlling for demographic factors.

Personality may affect health outcomes through four potential pathways (Smith and MacKenzie 2006). First, the health behavior model posits that personality influences healthrelevant habits, such as smoking, diet and exercise, and medical check-ups. A variety of health behaviors are associated with the major domains of personality (Bogg and Roberts 2004; Terracciano and Costa 2004), and these health behaviors may mediate the association between personality and development of subsequent diseases and illnesses.

Second, the interactional stress moderation model suggests that personality influences the appraisal of stressful life events, as well as coping strategies. Individual differences in appraisal and coping strategies can result in differences in physiological responses to stressors – for example, greater cardiovascular reactivity, higher morning levels of cortisol, or prolonged suppression of secretory immunoglobulin – (Lahey 2009), which in turn contribute to the development of diseases (Sutin et al. 2010; Phillips et al. 2010; Hutchinson and Ruiz 2011).

Third, the transactional stress moderation model posits that personality influences an individual’s exposure to potential stressors and stress-reducing interpersonal resources like social support. Through decisions to enter or not enter particular situations, and intentional or unintentional alteration of interpersonal contexts, individuals influence the social environment that they encounter in their daily lives. Differing exposure to negative life events could enhance the contribution of personality to the level of everyday stressors (Lahey 2009; Hutchinson and Ruiz 2011).

Finally, the constitutional predisposition model and the confounding model from Deary et al. (2010) suggest that an underlying genetic or other factor determines both physical and mental health and behavioral, emotional, and cognitive manifestations of personality traits.

Like personality, intelligence is also associated with various health behaviors and outcomes. Gottfredson (2004) argued that a general intelligence factor, g, is the fundamental cause ofhealth disparities. According to Gottfredson, a social epidemiological model of health disparities that emphasizes material resources (e.g., income and access to healthcare) cannot explain why health differentials by social class continue to exist across diverse times, places, and diseases, or why these differentials have grown despite improvements to healthcare systems.

Intelligence appears to be correlated with mental health. For instance, using data from two large British birth cohort studies, Gale et al. (2009) found that general cognitive ability measured at ages 10 and 11 was inversely associated with psychological distress at ages 30 and 33, even after accounting for pertinent predictors. In a subsequent study using Swedish military conscription data for more than one million men, Gale et al. (2010) demonstrated that IQ was inversely associated with the risk of hospital admission for all categories of mental disorders.

There is also some evidence that intelligence is linked to physical health outcomes. Batty et al.’s (2007) review of nine independent longitudinal cohort studies showed that each study reported an inverse association between premorbid intelligence test scores and risk of all-cause mortality in adulthood. Using Swedish conscription data, Batty et al. (2010) found that IQ was negatively associated with major groups of injury mortality, such as poisonings, fire, falls, drowning, and road injuries. More striking is that IQ and injury mortality were associated in a gradient fashion. Similar results were reported in Australia by O’Toole and Stankov (1992), who used the Australian Veterans Health Studies and discovered that a general intelligence test score was a good predictor of mortality for men ages 22 to 40.

Intelligence appears to be associated with specific causes of death as well. For instance, Hart et al. (2004) matched 938 participants from the Midspan prospective cohort studies with the Scottish Mental Survey 1932. Their data followed the subjects for approximately three decades and showed that a 1 SD disadvantage in intelligence at age 11 was associated with an 11 percent increase in the risk of hospital admission or death due to cardiovascular disease up to age 65, but not to these events occurring after age 65.

Furthermore, studies examining personality and intelligence jointly have found a significant relationship between intelligence and health outcomes. Maier and Smith (1999) used Berlin Aging Study data to relate measures of intellectual functioning, as well as personality traits such as neuroticism, extraversion, and openness, with all-cause mortality. Intellectual functioning was the strongest and most robust predictor for mortality. Similarly, Weiss et al. (2009) reported a significant interaction between neuroticism and intelligence when predicting mortality among the Vietnam Experience Study (VES) cohort. However, they noted that the effect of intelligence on mortality appeared to be indirect through SES, whereas the direct effect of neuroticism on mortality was robust.

There are four pathways through which intelligence and mortality may be associated (Deary 2008). First, high intelligence leads to educational success, placement in high status occupations, and, ultimately, high income (Deary et al. 2010). Second, lower intelligence isassociated with negative health behaviors, such as smoking, excessive alcohol consumption, poor diet, and less physical activity. These behaviors may result in various physical illnesses, including diabetes, coronary heart disease, some cancers, and, ultimately, death. Third, intelligence scores may reflect injuries to the brain that occurred before the test date (Deary 2008). Finally, intelligence is associated with individual differences in the integrity of a person’s underlying physiological makeup, which may affect health outcomes (Deary 2008; Deary et al. 2010). In other words, intelligence may be linked to system integrity, an ability to respond efficiently to environmental challenges and return to equilibrium.

Recently, a growing body of research has found null or very weak effects of intelligence on health. For example, using the WLS data and survival analysis, Hauser and Palloni (2011) found that the relationship between IQ and mortality was significant after controlling for social background factors. However, the coefficient for IQ was reduced substantially and became insignificant when high school rank was included in the model. Furthermore, the effect of high school rank on mortality was three times larger than the effect of IQ. Hauser and Palloni suggested that high school rank is a cumulative measure of cognitive ability plus responsible, compliant, and consistent behavior over the long term and, thus, is a better predictor of mortality than IQ alone. In addition, Hauser and Palloni claimed that conscientiousness is a superior measure to IQ given the similarity between conscientiousness and high school rank. Another study using the cumulative 1974-2000 General Social Survey showed that the association between intelligence and self-rated health was very sensitive to SES controls, although the coefficient for intelligence was significant (Schnittker 2005). Moreover, contrary to Gottfredson's (2004) claim, intelligence did not account for health disparities between whites and blacks.

An individual’s overall attractiveness is multidimensional and only approximated by assessments of facial or physical attractiveness (Berry 2000). Three major components make a face beautiful (Rhodes 2006): averageness, symmetry, and sexual dimorphism.

First, although it is somewhat counterintuitive, faces that are close to the population’s average are consistently judged attractive (Rhodes 2006; Berry 2000). For instance, Langlois and Roggman (1990) created composite faces by digitally averaging pictures of young adults using 2, 4, 8, 16, and 32 pictures. Composite faces created with 2, 4, or 8 photos were not rated as attractive; however, raters did assess those composite faces that were based on 16 or 32 pictures – and so were closer to the population’s average – as attractive. Second, like many other animals, humans prefer symmetrical faces (Rhodes 2006; Berry 2000). Finally, sexual dimorphism (or exaggerated features) refers to masculine features in a male face and feminine features in a female face (Rhodes 2006). Evolutionary psychology literature suggests that the features that make a face attractive are rooted in humans’ evolutionary history. That is,averageness may indicate developmental stability, resistance to disease, and optimum functionality, which, ultimately, signal aspects of mate quality.

The association between physical attractiveness and health is discussed mostly within the good genes theory. The good genes theory suggests that humans’ judgments of physical attractiveness, particularly when selecting a mate, have evolved to respond to heritable cues reflecting underlying health status (Weeden and Sabini 2005; Gallup and Frederick 2010). Considerable evidence supports the hypothesis that both facial and bodily attractiveness are health certifications and accurately represent genetic quality. For example, body symmetry is positively associated with attractiveness (Geary et al. 2004) and predictive of health (Grammer et al. 2003; Singh 2004) and fertility (Jasienska et al. 2006). Low waist-to-hip ratio (WHR) is associated with health and fertility among women (Sarwer et al. 2003; Singh 2004), while wide shoulders and tallness predict both attractiveness and health among men (Grammer et al. 2003; Mueller and Mazur 2001). In addition, facial symmetry is associated with attractiveness and health (Fink et al. 2004; Geary et al. 2004; Geary 2005; Kramer and Ward 2011).

Despite the evidence that attractiveness is associated with other physical traits, a relatively small number of studies has explored whether facial attractiveness is associated with mortality or morbidity. The evidence is mixed. Using the Intergenerational Studies data, a longitudinal study of people born in the 1920s in California, Kalick et al. (1998) found that the overall relationship between health (measured annually by physicians from age 3 to 18 and in the patients’ 30s) and facial attractiveness (measured with photographs taken at about age 17) was insignificant for both males and females after controlling for SES. As a result, the authors argued that little evidence supports the good genes theory.

In contrast, Shackelford and Larsen (1999) studied 100 undergraduates in the U.S. who reported daily physical symptoms including headaches, runny or stuffy nose, nausea or upset stomach, muscle soreness or cramps, sore throat or cough, backache, and jitteriness, as well as a measure of cardiovascular recovery. The authors correlated attractiveness with these health measures and found that attractive participants displayed greater cardiovascular health and complained less often of a headache or a runny or stuffy nose compared to their less attractive counterparts.

Among Canadian undergraduates, Hume and Montgomerie (2001) showed that facial attractiveness was positively associated with a health rating based on the incidence and seriousness of past diseases for women, but not for men. Also in Canada, Henderson and Anglin (2003) used high school yearbook photos from the 1920s to examine the association between ratings of facial attractiveness and year of death. They found a positive relationship between attractiveness and longevity for men and women but failed to find a significant link between perceived health and longevity (Langlois et al. 2000).

In addition to direct effects, physical attractiveness may also affect mortality and morbidity via SES. For example, a physically attractive person is more likely to be employed and earnmore (Hamermesh and Biddle 1994; Mobius and Rosenblat 2006; Judge et al. 2009), marry earlier, enjoy a more stable marriage, have more children (Jæger 2011; Jokela 2009), perform better academically in high school (French et al. 2009), and achieve a higher socioeconomic status (Hauser 2009). Given that SES is one of the most significant and important predictors for mortality and morbidity (Link and Phelan 1995), it may be an indirect channel through which physical attractiveness affects health outcomes.

Although evolutionary theories generally suggest that physical attractiveness is positively associated with health for both sexes, based on results from a meta-analysis, Weeden and Sabini (2005) argued that the empirical evidence to date indicates the effect may be limited to females in most Western societies (see Grammer et al. 2005 and Geary 2005 for response). Moreover, Weeden and Sabini asserted that the magnitude of the effect may differ depending on the nature of the measured physical attractiveness. However, Geary (2005) suggested that, although the correlation between physical attractiveness and health is small in absolute terms, as Weeden and Sabini (2005) argued, when viewed from the standpoint of human evolution, small relationships can be important because their effects can snowball across generations. Furthermore, Geary (2005) asserted that the life history trade-off among men – which refers to the phenomenon that an advantageous trait in reproductive competition early in life (e.g., larger muscles in the upper body) may come with an increased risk of later morbidity and early mortality (e.g., greater accumulation of body fat) – complicates analysis of the relationship between attractiveness and health and may lead to mixed findings.

Data for this study were drawn from the Wisconsin Longitudinal Study (WLS). The WLS is a study of a random sample of 10,317 men and women who graduated from Wisconsin high schools in 1957, as well as their randomly selected siblings (Sewell et al. 2003). Survey data were collected from the original respondents or their parents in 1957, 1964, 1975, 1993, 2003, and 2010 (graduate sample) and from selected siblings in 1977, 1994, 2005, and 2010 (sibling sample). This study uses only the graduate sample because facial attractiveness ratings were conducted only for this sample. It should be noted that, although WLS respondents are generally representative of non-Hispanic White women and men with a high school education (who constitute over two-thirds of Americans of retirement age), they are not a random sample of the U.S. population (Sewell et al. 2003).

To check that findings from the WLS are not idiosyncratic to the particular dataset, I supplemented the analysis with an examination of the Americans’ Changing Lives (ACL) survey. The ACL is a stratified, multistage area probability sample of non-institutionalizedadults ages 25 and older living in the contiguous U.S. (House et al. 1994; House et al. 1990). Wave 1 of the ACL survey was administered in 1986 via face-to-face interviews with 3,617 respondents (male = 37.5%; female = 62.5%). Three additional waves were conducted in 1989, 1994, and 2002. One of the strengths of the ACL is that it over-sampled Blacks and those ages 60 or older. Along with sociodemographic and health measures, the ACL offers comparable measures of personality traits, physical attractiveness, and intelligence. Using these measures, analyses of the WLS data were replicated. In the ACL, respondents’ deaths were matched to the National Death Index recorded through June 7, 2008.

In the present study, the dependent variable is the waiting time until the occurrence of an event (i.e., death). Since respondents must have been alive on the 1992-94 interview dates (when the personality instrument was administered for the first time), the dependent variable was the years of life from the 1992-94 interview dates through December 2011. The year and month of death for the WLS respondents were updated periodically using the Social Security Administration’s Death Index. By the end of 2011, 2,033 (19.7%) respondents were reported to be deceased. Those who were alive at that time were treated as censored. Those whose death date was not ascertained, i.e., individuals deceased between two waves but their exact year of death was not confirmed, were censored at the most recent interview date (n = 5). Those with an unknown date of death were excluded from the analysis. For the ACL analysis, the dependent variable was the years of life from 1986 through 2008.1

In addition to mortality, three health outcomes were used as dependent variables in this study. First, mental health was measured with a depression scale (the modified CES-D). Controlling for depression is important when examining the effect of personality traits on health outcomes because instruments for depression overlap personality items (Lahey 2009). Thus, failure to account for depression overestimates the effect of personality. Second, selfrated health (SRH) was measured by the following question: “In general, would you say your health is excellent, very good, good, fair, or poor?” It is well established that self-rated health is closely associated with mortality and actual health (Idler and Benyamini 1997). Self-rated health was collapsed into a dichotomous variable (“1” if SRH equals poor or fair and “0” otherwise). Finally, the total number of illnesses was the sum of physical illnesses that medical professionals had ever diagnosed in the respondent.

Facial Attractiveness

During summer 2004, 33 judges who participated in the Madison Senior Scholars program were recruited to rate the facial attractiveness of WLS respondents’ yearbook photos. The mean age of the judges was 78.5 years, and judges’ ages ranged from 63 to 91. A total of 3,007 WLS respondents’ yearbook photos from 1957 and a sub-sample of 258 WLS respondents’ yearbook photos from 1956 were randomly selected and rated (Meland 2002). Six men and six women rated each yearbook photo using an 11-point rating scale (1 = not at all attractive, 11 = extremely attractive).2 Due to the large volume, the photos were divided into ten groups of approximately 300 photos each, plus an additional group consisting of photos from 1956. The judges rated 300 photos per coding session and were required to take at least 12 hours off between sessions. Although several judges coded multiple sets of photos, only a few coded all 11 groups. During summer 2008, an additional 5,606 yearbook photos were selected and rated using the same procedures as in 2004. Cases with fewer than 11 ratings and ratings with minor errors were removed, leaving a final sample of facial attractiveness that included 8,625 individuals. Mean ratings across the 12 judges were computed and standardized (i.e., mean = 1 and SD = 0).

It should be noted that measuring physical attractiveness through photographs may be influenced by the quality of the picture (Meland 2002). For instance, some individuals may look better in a photo than in reality. Also, given that the photos were in black and white, the effect of complexion may be obscured.

At the end of the first and second waves of the ACL, interviewers rated respondents’ appearance and attractiveness using a 5-point scale (1 = very attractive or beautiful, 5 = very unattractive), which was reverse-coded so that higher values represent greater attractiveness. For those interviewed in the first two waves, the mean of the two attractiveness ratings was used as an overall measure of physical attractiveness. Available values were taken for those who missed one of the first two rounds. Note that although the second wave of the ACL was three years after the first wave, there was a statistically significant relationship between the two attractiveness measures (χ² = 785.71, 16 d.f., p < 0.001). Also, although ratings of attractiveness in the ACL should not have been affected by the extent to which the respondent was photogenic, the respondent’s friendliness, clothes, accessories, or household tidiness could have had an influence.

The Big Five Personality Traits

The WLS administered the BFI-54, a relatively short instrument that assesses the Five-FactorModel of Personality dimensions (Hauser et al. 1999), in telephone surveys (only two items for each dimension) and mail surveys (five or six items for each dimension) during 1992-93, as well as in mail surveys during 2003-05. Responses were coded using a 6-point scale (1 = strongly agree, 6 = strongly disagree). Fourteen of the 29 items were reverse-coded when constructing the personality measures so that a higher total score represents a greater personality trait (e.g., “talkative,” “full of energy,” and “a lot of enthusiasm” items for extraversion were reverse-coded).

The ACL measured only neuroticism and extraversion, and items for each trait were recoded, if necessary, so that higher scores on the index reflect a higher level of the trait. Note that the ACL measured these personality traits only on the baseline survey in 1987. The reliabilities were 0.71 and 0.70 for neuroticism and extraversion, respectively.

Intelligence

In the WLS data, the Henmon-Nelson (H-N) mental ability test measured cognitive ability (IQ) during the respondents’ junior year of high school (1956). The H-N test, consisting of 90 verbal and quantitative items, was administered in all Wisconsin high schools from the 1930s through the 1960s. Evidence suggests that the test is highly reliable (Hauser et al. 1983).

The ACL administered Lorge-Thorndike (L-T) sentence completion items that asked respondents to fill in an appropriate word to complete a sentence (House 2010); the total number of right answers was conceived as an indicator of verbal intelligence, which is used as a proxy for IQ.3

Although the H-N and L-T tests can reliably measure general mental ability (Strenze 2007; Almlund et al. 2011), respondents’ age at the time these tests were administered differs across the two datasets. That is, in the WLS, the H-N test was administered when the respondents were in high school, whereas the L-T tests were conducted when participants were in middle age in the ACL.

Behavioral Risk Factors

Smoking status was grouped into three categories: never smoked (reference category), former smoker, and current smoker. BMI was calculated by dividing weight in pounds by height in inches squared, then multiplying by a conversion factor of 703. BMI was categorized into four groups: underweight (below 18.5), normal weight (18.5-24.9), overweight (25.0-29.9), and obese (30.0 and above). However, because there were few underweight cases in the sample, the underweight category was combined with the normal weight category. Physical activity was assessed by the number of times respondents reported participating in light physical activity,such as walking, dancing, gardening, golfing, or bowling. Physical activity was grouped into four categories – less than once a month, 1-3 times a month, 1-2 times per week, and more than 3 times per week.

Control Variables

As a proxy for parents’ socioeconomic status, father’s education was measured by the years of schooling that a respondent’s father completed. Farm background was coded as “1” if the respondent’s father was a farmer and “0” otherwise. Fathers’ occupation information was drawn from tax return reports in 1957. For total household income in 1957, I used imputed values of family income. Religiosity was measured by frequency of attendance at religious services, which was grouped into 11 categories from 1-2 times a year to once a day or more.

The effect of personality on mortality was assessed with a series of parametric proportional hazard regressions. Cleves et al. (2010) suggested that, if the proper distributional assumptions are made, parametric analysis is more efficient than Cox models. Since the current analysis concerned the hazard of death, proper functional form of the baseline force of mortality can be specified. Among the parametric models, the Weibull distribution was used to represent the baseline h₀(t) and the parametric form was specified as

where a is the level parameter and p is the shape parameter. Goodness-of-fit statistics (i.e., AIC and BIC) for Cox and parametric models indicated that Weibull model fits best for the data (results not shown here).

1Although the ACL re-interviewed 1,427 original participants (81% of survivors, including 108 proxy interviews) in 2011-12, public data were not available at the time of writing this paper.  2Alwin (1997) suggested that 11-point scales are more reliable and no more vulnerable to shared method variance than 7-point scales.  3One example of questions is “Not every cloud gives [blank]” and respondents were given “weather,” “shade,” “sky,” “climate,” and “rain” as appropriate words to complete the sentence.

Summary statistics for the variables used in the event history analysis for the WLS are presented by survival status in Table 1. Neuroticism was significantly higher for deceased men (p = 0.06) and women (p = 0.04) than for their living counterparts. Openness was significantly higher for deceased males than for living males (p = 0.001), whereas conscientiousness was marginally higher for living women than for deceased women (p = 0.08). Although there were no differences in physical attractiveness ratings between living and deceased men, a higher proportion of unattractive women were among the deceased compared to the living. IQ scores were slightly lower among the deceased men and women than their living counterparts, but thedisparities were not statistically significant. Also, there were no significant differences in terms of age at baseline, father’s years of education, or frequency of participation in light physical activity between the deceased and the living. Compared with deceased respondents, living respondents were more educated, had greater total family income, were more likely to have a farm background, and were less likely to be depressed, less likely to be either a former or current smoker, and less likely to be overweight or obese.

Figure 1 shows the Nelson-Aalen cumulative hazard estimates of mortality among WLS graduates with low (more than 1 SD below the mean), high (greater than 1 SD above the mean), and medium (1 SD ± mean) levels of neuroticism and openness. For these two personality traits, the high group had a significantly higher cumulative mortality, but there was no significant difference between the low and medium groups. Disparities between the high group and the other two groups arose ten years after the baseline interview date for neuroticism, while the gap was observed throughout the life course for openness.

The estimated parameters of the Weibull models of survival are presented in Table 2. Model 1 included measures of personality traits, physical attractiveness, and IQ. For men, only two personality traits – neuroticism and openness – were significantly associated with the hazard of mortality: a 1 SD increase in neuroticism was associated with a 17.3 percent increase in the hazard ratio of mortality, while a 1 SD increase in openness was associated with a 38.3 percent increase in the hazard ratio of mortality. For women, however, none of the personality traits was statistically associated with the hazard of mortality.

On the other hand, as physical attractiveness increased, mortality decreased only among women. Very attractive women had 32.3 percent lower hazard ratio relative to unattractive women. For men, there were no differences in the risk of mortality based on ratings of physical attractiveness. However, the protective effect of IQ was observed only among men: a 1 unit increase in IQ was associated with a 0.9 percent decrease in the hazard of mortality. For women, although IQ was negatively associated with mortality, the coefficient failed to reach statistical significance.

Model 2 added family background and achieved socioeconomic status. As expected, education significantly reduced the risk of mortality for both men and women: a one-year increase in education was associated with a 9.3 and 11.0 percent decrease in the hazard of mortality among men and women, respectively. While men in the highest quartile of total household income showed a significantly reduced mortality risk, the results for women indicated a gradient with respect to income levels. That is, women in the second through fourth quartiles showed a mortality advantage over women in the first quartile. Furthermore, consistent with Hauser and Palloni (2011), a farm background appeared to reduce the hazard of mortality, although the effect was observed only among females. Finally, frequency of attending religious services significantly reduced the hazard of mortality (by about 4 percent) for men and women.

In model 2, the coefficients for personality traits and physical attractiveness changed little compared with the results from model 1. One exception was the coefficient for men’s IQ, which became insignificant in model 2. This suggests that the effect of IQ on mortality for men is mediated by socioeconomic achievement.

Model 3 added health behaviors. As expected, being obese was associated with a 59 and 46 percent increase in the hazard ratio of mortality for men and women, respectively, compared to those of normal weight. However, being overweight did not appear to increase the hazard of mortality relative to those of normal weight. Past studies have reported that, partly due to decreases in stature and changes in body composition associated with aging, being overweight may not be a significant predictor for all-cause mortality among the elderly, unlike in younger populations (e.g., Visscher et al. 2000). Current smoking elevated the hazard ratios of mortality by 2.4 times for men and 2.0 times for women compared with their counterparts who never smoked. Being a former smoker increased the mortality risk by 51 percent relative to those who never smoked, but the effect was significant only for men. Contrary to expectations, frequency of physical activity was not associated with mortality in either men or women.

Again, the coefficients for personality traits, physical attractiveness, and IQ changed little compared with model 2. One exception was that the coefficient for neuroticism became insignificant for men, suggesting that the effect of neuroticism on all-cause mortality for men is mediated by health behaviors.

Results of regression analyses predicting self-rated health, depression, and number ofdoctor-diagnosed illnesses are presented by sex in Figure 2. All of the regressions controlled for SES, social background, and health behaviors. For men, among the variables of interest, only neuroticism (OR = 1.38) was significantly and positively associated with the odds of reporting poor or fair health. For women, in contrast, conscientiousness (OR = 0.81) was negatively associated with the odds of reporting poor or fair health, while neuroticism (OR = 1.24) had a positive association. Furthermore, physical attractiveness reduced the odds of reporting poor or fair health for women.

With respect to depression among men, extraversion and conscientiousness significantly reduced the CES-D score while neuroticism increased it. For women, in addition to extraversion, conscientiousness, and neuroticism, agreeableness was also negatively associated with depression. Furthermore, very attractive women experienced less depression than unattractive women. In terms of number of doctor-diagnosed illnesses, only neuroticism was significantly and positively associated for men, whereas neuroticism, openness, and extraversion were significantly and positively associated for women.

Results from logistic regressions predicting selected physical diseases (i.e., cancer, diabetes, and hypertension) indicated that, consistent with past studies (e.g., Stürmer et al. 2006), there is no significant effect of personality on cancer for either men or women.4 Also, neither physical attractiveness nor intelligence was associated with the odds of cancer. Although none of the personality traits was associated with the odds of diabetes, very attractive women were significantly less likely to have diabetes than unattractive women. This result is consistent with Reither et al. (2009), who predicted adult diabetes from high school yearbook pictures in the WLS sample. Hypertension was also significantly and positively associated with neuroticism for men and women.

Taken as a whole, the results for men indicate that neuroticism is a risk factor for mortality, and that this relationship is mediated by behavioral risk factors such as smoking, depression,and being obese. Somewhat unexpectedly, openness also appears to significantly elevate the odds of mortality for males, a result which also appears to be related to health behaviors. Figure 3, which shows regression results for selected health behaviors, indicates that, for a 1 SD increase in men’s openness, the odds of having smoked cigarettes increase by a factor of .17. Also, the odds of participating in physical activities were 12 percent smaller if women’s openness increased by 1 SD. Openness to experience may encompass cognitive and behavioral flexibility, which may enhance the capacity to avoid and manage health problems (Chapman et al. 2011). At the same time, curiosity – one facet of openness – may be linked to boredom (Lundberg 2011), which may inhibit continuous efforts to adjust behaviors over the long run.

Descriptive statistics by survival status and gender for the ACL sample are presented in Table 3. In order to increase comparability across the ACL and WLS datasets, the ACL sample was restricted to whites, and independent variables were selected or constructed to be as similar aspossible to those in the WLS. No significant differences in neuroticism by mortality status existed for men (p = 0.50) or women (p = 0.17). However, deceased females (p < 0.005) showed significantly lower scores for extraversion than their surviving counterparts. Also, deceased men and women were considerably older and had significantly lower IQ scores (p <0.000) than their living counterparts. Deceased respondents were less educated, had fathers with lower levels of education, reported less total household income, were more likely to be former smokers, drank alcohol more frequently, and participated in physical activity less frequently than those who survived.

Table 4 presents the estimated parameters of the Weibull models of survival in the ACL cohort. As in the WLS cohort, models were estimated hierarchically; that is, the first model included only personality traits, physical attractiveness, and intelligence, along with age and race; the second model added socioeconomic factors; and the last model added health behaviors. Results from model 1 indicate that neuroticism and extraversion are not associated with the hazard of mortality among men or women in the ACL cohort. However, higher physical attractiveness ratings appear to decrease the hazard of mortality for both men and women. For instance, men in the highest quintile of physical attractiveness were 87.8 percent less likely to die than their counterparts in the lowest quintile. Similarly, the most attractive women were 74.2 percent less likely to die compared with their least attractive counterparts, holding other characteristics constant. On the other hand, although intelligence was negatively associated with the hazard of death, the coefficient for intelligence failed to reach statistical significance.

Model 2 added measures of social background and socioeconomic status, but the coefficients for physical attractiveness changed little, implying that physical attractiveness affects the hazard of mortality independent of SES. In model 3, health behaviors (i.e., smoking status, alcohol consumption, and physical activity) were added. Among women, physical attractiveness remained a significant predictor of mortality; for example, the hazard of mortality for women in the highest quintile of physical attractiveness was 70.4 percent smaller than for those in the lowest quintile. However, the effect of physical attractiveness on men’s mortality lost statistical significance in model 3.

As expected, results for men’s health behaviors indicate that, compared to individuals who never smoked, being a former smoker elevated the hazard of mortality roughly 61 percent, while current smoking increased the hazard almost 2.1 times. Although being a former smoker did not have a significant effect on women’s mortality, currently smoking women had a 2.4 times higher risk of mortality compared with those who never smoked. For both men and women, alcohol consumption and depression were not associated with the hazard of mortality. However, participation in physical activities significantly reduced the hazard of mortality among both men and women. Specifically, a one unit increase in the frequency of engaging in physical activities was associated with 25 and 13 percent reductions in the hazard of mortality among men and women, respectively.

4Results are not presented here due to space constraint, but available upon request.

The current study explored the effect of personality traits, physical attractiveness, and intelligence on longevity and health among the WLS and ACL samples. The central proposition of this study is that an individual’s life course outcomes are determined by the psychological, biological, and socioeconomic influences accumulated over one’s life. The contribution of this research is that it goes beyond the prevailing paradigm of proximate determinants for social behaviors by examining more fundamental causes and their functional relationships with other features of the whole individual.

This study found that, even after accounting for SES and health risk factors, two personality traits – neuroticism and openness – mattered for men’s mortality in the WLS. The effect of neuroticism on all-cause mortality for men seemed to be mediated by health behaviors like smoking and frequency of physical activity. In contrast, openness remained significant after controlling for health behaviors. Even so, the causal link between openness and mortality is not entirely clear; it appears to be related to smoking behavior in those high in openness.

While no personality traits significantly affected all-cause mortality among WLS females, higher ratings of women’s physical attractiveness were associated with significantly diminished hazard of mortality. Conscientiousness, which has been linked to lower risk of all-cause mortality in a number of past studies (Chapman et al. 2011; Kern and Friedman 2011), did not increase the hazard of mortality when physical attractiveness and intelligence were controlled. Similarly, extraversion and agreeableness were not significant predictors of mortality.

However, among both men and women in the WLS, personality traits were more strongly associated with self-rated health and depression than with mortality. It makes sense that personality traits, especially neuroticism, exert stronger influence on psychological well-being than on somatic health outcomes, which are engendered by a complex web of causes, including genetic predispositions, environmental hazards, and diet, among other factors.

In contrast to results from several studies from cognitive psychology (e.g., Gottfredson 2004; Batty et al. 2010), intelligence had a trivial effect on all health outcomes for both sexes in the WLS. However, the vast majority of past studies examining the effect of intelligence on health outcomes have not accounted for physical attractiveness, which may provide accurate information about underlying health (Shackelford and Larsen 1999; Hume and Montgomerie 2001; Henderson and Anglin 2003). Although both intelligence and physical attractiveness may be strongly influenced by genetics, physical appearance may be a more proximal determinant of health outcomes because the effect of intelligence may be mediated by behavioral factors.

In general, the patterns observed in the WLS were replicated in the ACL. Two personality traits – extraversion and neuroticism – were not associated with all-cause mortality among males or females in the ACL. However, higher physical attractiveness ratings, which were measured by interviewers at the end of each interview, significantly reduced the hazard of mortality for both sexes. As in the WLS, intelligence was not associated with mortality risk.

The present study is not without limitations. First, the WLS measured intelligence using the Henmon-Nelson test of mental ability when respondents were in their junior year of high school. In contrast, the ACL measured intelligence using verbal intelligence tests when most of the respondents were in their mid-40s or 50s. Thus, not only did respondents’ ages when the intelligence tests were administered differ, but also the WLS measured more fluid aspects of intelligence while the ACL measured more crystalized intelligence (Almlund et al. 2011). To the extent that the effect of the two aspects of intelligence and the respondents’ age at measurement on health outcomes differs, the null findings of intelligence in both datasets should be interpreted with caution. Second, the current study did not examine effects of changes in personality on mortality. Some studies using growth curve modeling (e.g., Mroczek and Spiro 2007) have found that both changes in personality traits as well as absolute levels are significantly associated with longevity among the elderly. Finally, the age range of samples in the current analyses is somewhat narrow. More diverse samples should be used for crossvalidation of the current results.

To conclude, the effects of personality traits, physical attractiveness, and intelligence on health outcomes differ by sex. Analyses of the WLS suggest that personality traits, especially neuroticism and openness, have a significant impact on men’s survival, while physical attractiveness matters for women’s longevity. Results from the ACL revealed that physical attractiveness is strongly and inversely associated with mortality risk for men and women. In addition, the results indicated that personality traits, especially neuroticism, extraversion, and conscientiousness, substantially affect psychological well-being, including self-rated health and depression.