Content uploaded by Leslie R Martin
Author content
All content in this area was uploaded by Leslie R Martin
Content may be subject to copyright.
Personality and Mortality Risk Across the Life Span: The Importance of
Conscientiousness as a Biopsychosocial Attribute
Leslie R. Martin
La Sierra University
Howard S. Friedman
University of California, Riverside
Joseph E. Schwartz
Stony Brook University, State University of New York
Objective: This study addressed whether personality in childhood and personality in adulthood are indepen-
dent predictors of mortality risk and the extent to which behavioral and other psychosocial factors can explain
observed relationships between personality and mortality risk. Design: This was a prospective longitudinal
cohort study of 1,253 male and female Californians over 7 decades (1930 –2000). Proportional hazards
regressions were the principal analyses. Main Outcome Measures: Mortality risk (in the form of relative
hazards) was the primary outcome. Additional tests of mediators and moderators ascertained whether
associations between personality and mortality risk remained significant when psychosocial and behavioral
variables were statistically controlled. Results: The findings, including a new 14-year additional follow-up in
old age, revealed that conscientiousness, measured independently in childhood and adulthood, predicted
mortality risk across the full life span. The link from childhood remained robust when adult conscientiousness
and certain behavioral variables were controlled. Psychosocial and behavioral variables partly explained the
adult conscientiousness–longevity association. Conclusion: The findings demonstrate the utility and com-
plexity of modern personality concepts in understanding health and point to conscientiousness as a key
underexplored area for future biopsychosocial studies.
Keywords: personality, health behaviors, life-span mortality risk, conscientiousness
Personality, a relatively stable set of cognitive–motivational and
socioemotional traits and behavior patterns, is increasingly recog-
nized as a meaningful contributor to the prediction of a wide range
of important human outcomes. Because it captures a combination
of genetic, familial, and sociocultural elements, personality can be
an important factor for understanding long-term biopsychosocial
processes. In the decades since the Type A behavior pattern was
recognized as a possible risk factor for cardiovascular disease,
accumulating evidence has suggested that personality, broadly
construed, is related to important health outcomes. Increasingly,
the five-factor approach to personality, coupled with attention to
trait– environment interactions, has drawn some focus away from
narrower and more pathology-based concepts like Type A, anxi-
ety, repression, reactivity, inhibition, and delinquency. That is,
conscientiousness, extraversion, agreeableness, and neuroticism
(and to some extent openness), in the context of situational exi-
gencies, have been shown to be relevant to later important out-
comes, including life-span mortality risk (Caspi et al., 1997; Chris-
tensen et al., 2002; Danner, Snowdon, & Friesen, 2001; Friedman,
Tucker, Schwartz, Martin, et al., 1995; Friedman, Tucker,
Schwartz, Tomlinson–Keasey, et al., 1995; Friedman, Tucker,
Tomlinson–Keasey, Schwartz, Wingard, & Criqui, 1993; Martin et
al., 2002; McCann, 2005; Smith, Gallo, & Ruiz, 2003; Wilson,
Mendes de Leon, Bienias, Evans, & Bennett, 2004).
Initially surprising but now increasingly well documented is the
finding that conscientiousness may be a key biopsychosocial attribute
(Friedman, Tucker, Schwartz, Martin, et al., 1995; Friedman et al.,
1993; Martin & Friedman, 2000; McCann, 2005; Roberts & Bogg,
2004; Roberts, Caspi, & Moffitt, 2003; Weiss & Costa, 2005), in part
because of its demonstrated links to healthy behaviors (e.g., Hamp-
son, Goldberg, Vogt, & Dubanoski, 2006; Roberts & Bogg, 2004;
Tucker et al., 1995). Despite the promise conscientiousness holds for
aiding understanding of patterns of individual difference and their
associations with health, a number of interrelated questions about
conscientiousness have not been addressed. First, there is the question
as to whether conscientiousness predicts mortality risk independently
in both childhood and adulthood. If childhood conscientiousness has
far-reaching, independent effects, then its associations with health and
Leslie R. Martin, Department of Psychology, La Sierra University;
Howard S. Friedman, Department of Psychology, University of California,
Riverside; Joseph E. Schwartz, Department of Psychiatry and Behavioral
Science, Stony Brook University, State University of New York.
This research was supported by National Institute on Aging Grants
AG08825 (Howard S. Friedman, principal investigator) and AG15188
(Leslie R. Martin, principal investigator). Part of the data was made
available from the Terman Life-Cycle Study, and assistance was provided
by Eleanor Walker of the Terman project. We bear full responsibility for
death certificate collection and coding, data refinements, analyses, and
interpretations. This article is one in a series from our multiyear, multidis-
ciplinary project on psychosocial predictors of health and longevity. Pre-
vious publications from the project, including previous studies of person-
ality and longevity using a shorter time period and less complete death
data, are cited as appropriate, and care should be taken not to include
overlapping findings in meta-analyses or other reviews.
Correspondence concerning this article should be addressed to Leslie R.
Martin, Department of Psychology, La Sierra University, 4500 Riverwalk
Parkway, Riverside, CA 92515-8247. E-mail: Lmartin@Lasierra.edu or
Howard.Friedman@ucr.edu
Health Psychology Copyright 2007 by the American Psychological Association
2007, Vol. 26, No. 4, 428 –436 0278-6133/07/$12.00 DOI: 10.1037/0278-6133.26.4.428
428
mortality may be more reflective of fundamental psychophysiological
phenomena, such as serotonin function (Williams et al., 2004), than if
the prognostic value of childhood conscientiousness is due mostly to
its association with later-life (adult) conscientiousness and associated
healthy adult behaviors. Second, there is the question of whether the
prognostic value of conscientiousness continues across the full life
span, into old age. If it does, this would be additional evidence for a
fundamental process. Third, there is the related question of the extent
to which risk taking (and death from injury) and health behaviors,
such as smoking, drinking, and overeating, can explain any lifelong
relationships between personality (especially conscientiousness) and
mortality risk. Longitudinal data are required to address such ques-
tions but are difficult to obtain (e.g., Hampson et al., 2001). No
longitudinal studies to date have assessed personality from childhood
through the full life span with rich life-course behavioral data.
The present study used new data we collected to follow up the
Terman life-cycle cohort, born on average in 1910 (Terman et al.,
1925; Terman & Oden, 1947), to predict and explain mortality
from 1930 to 2000. Specifically, this article addresses three issues:
(a) whether childhood and adult personality variables (especially
conscientiousness) are independent predictors of later mortality
risk, or whether childhood personality predicts longevity only
through its association with adult personality; (b) the associations
of personality, in both childhood and adulthood, with mortality
risk and cause of death through the year 2000 (when the average
participant was deceased or well into his or her 80s); and (c)
certain potential mediators and moderators as explanations for
observed relationships between personality and mortality risk.
Method
Participants
The participants were drawn from the original group of 1,528
individuals recruited by Stanford psychologist Lewis Terman and
colleagues in the 1920s. Nominated by their teachers and others as
excellent students, the participants included by Terman had IQs of
at least 135. Consistent with procedures of previous personality
studies using this data set to produce a relatively age-
homogeneous, school-age initial sample, we excluded participants
not born from 1904 to 1915 (n ⫽ 155), those who died or were lost
to follow-up prior to 1930 (n ⫽ 74), and those missing all child-
hood personality information (n ⫽ 46). This resulted in a final
sample of 1,253 (706 male) participants, 99% White and mostly
middle class, who were born on average in 1910. The 275 ex-
cluded individuals did not differ systematically from those retained
for analysis on the dimensions of sex, intelligence, parents’ socio-
economic status, or ethnicity.
In addition to addressing new questions regarding the persis-
tence of personality effects when certain variables are controlled,
the data herein represent a significant expansion on our previous
work (Friedman et al., 1993), as the time period for predicting
mortality risk is now extended an additional 14 years (1986 –
2000). Further, we have now identified some individuals who were
previously lost to the study but were still presumed alive; that is,
we have verified some of them as still living and documented date
and cause of death for others. (As of 2000, 789 participants in this
sample had been verified as deceased.)
Personality Variables
We developed a set of reliable personality measures from the first
of Terman’s assessments, made in 1922 (based on parent and teacher
reports, it is described elsewhere; Friedman et al., 1993). Of these
childhood measures, four (conscientiousness–social dependability,
cheerfulness– humor, sociability, and permanency of mood) were
utilized in the present study because of their documented association
to the five-factor model (John, 1990; McCrae & Costa, 1987; McCrae
& John, 1992) constructs of Conscientiousness, Agreeableness, Ex-
traversion, and Neuroticism (Martin & Friedman, 2000).
1
Table 1
reports the reliabilities for these scales.
Using factor analysis, testing for measurement invariance, and
rational analysis, we also created adult personality scales for four of
the Big Five (Conscientiousness, Agreeableness, Extraversion, and
Neuroticism) from data collected in 1940 (average participant age ⫽
30 years; Martin & Friedman, 2000). The item content seems insen-
sitive with regard to measurement of openness, and this is the most
likely explanation for the failure of openness to emerge as a factor
(Martin & Friedman, 2000). These 1940 personality scales were then
used as models for selecting items to construct similar personality
scales from self-report data collected in 1950.
2
Table 1 reports the
alpha reliabilities of these 1940 and 1950 adult personality scales as
well as their test–retest correlations.
We averaged the 1940 and 1950 scale scores for each of the four
personality scales (standardized and adding a constant to eliminate
negative values) for those who had personality data available at both
time points (n ⫽ 731). For participants who had personality data for
only one assessment (n ⫽ 291 with 1940 data only; n ⫽ 50 with 1950
data only), the available decade was used to indicate adult personality.
The decision to define adult personality in this way was made to
maximize the number of participants with personality data (given the
high correlations between 1940 and 1950) while also minimizing
1
The childhood conscientiousness scale included prudence–forethought,
freedom from vanity– egotism, conscientiousness, and truthfulness; the
cheerfulness– humor scale included cheerfulness– optimism and sense of
humor; sociability included fondness for large groups, popularity, leader-
ship, preference for playing with others, and preference for social activi-
ties; permanency of mood was a single item.
2
The adulthood conscientiousness scale included being thrifty and careful
about making loans, making excuses (⫺), enjoyment of planning work in
detail, driving self steadily in work, being impulsive (⫺), persistence, and
having definite purposes for apportioning time and energy; the agreeableness
scale included trying to avoid arguments, trying to get one’s own way (⫺),
upbraiding those who don’t complete tasks on time (–), being critical of others
(⫺), avoiding hurting others’ feelings, ignoring the feelings of others (⫺),
having an extra good opinion of self (⫺), losing one’s temper easily (⫺), being
dominant with the opposite sex (⫺), being easy to get along with, and being
vain and egotistic (⫺); the extraversion scale included taking the lead to
enliven a party, preferring a play to a dance (⫺), being a “stay-at-home” versus
a “gadabout” type (⫺), preferring a quiet mate (⫺), preferring to have many
women friends, enjoyment of social contacts, and having exclusive friendships
(⫺); the neuroticism scale included being very affected by praise/blame,
feeling miserable, being touchy, having periods of loneliness, feeling lonesome
with others, being bothered with useless thoughts, feeling burdened with
remorse/regret, lacking self-confidence, worrying over humiliating experi-
ences, alternating between happiness and sadness without apparent reason,
having easily hurt feelings, finding it difficult to be serene/cheerful, being
moody, being self-confident (⫺), being emotional, having sensitive feelings,
and suffering from feelings of inferiority.
429
PERSONALITY AND MORTALITY RISK
measurement error for those with data at both 1940 and 1950. Less
than 5% had data at 1950 only.
Potentially Explanatory Variables
Alcohol use. Participants answered questions about their own
alcohol use in 1950 and 1960. For each decade, we categorized
participants as (a) never drink or only rarely, (b) drink moderately
(never or seldom intoxicated), or (c) heavier drinkers. For individ-
uals with alcohol data at both time points, we used the report that
indicated heavier drinking as our measurement.
Smoking. Data on smoking were not collected as part of Ter-
man’s original study. Our team did, however, collect this infor-
mation (retrospectively) in 1991–92 through self- or next of kin
report. Number of pack years smoked was calculated as (years
participant smoked ⫻ average number of cigarettes smoked per
day during the time participant smoked) / 20. The number of pack
years ranged from 0 (never smoked) to 180. These data likely
underestimate the effect of smoking, because information on heavy
smokers who died very young appears more likely to be missing.
Body mass index (BMI). Participants reported their weight and
height as part of the 1940 assessment. We used this self-report
information to calculate BMI using Quetelet’s formula: BMI ⫽
weight in kg / height in m
2
.
Risk taking. Participants reported on their hobbies and leisurely
pursuits (by listing them in an open-ended format) several times from
1922 to 1940. These activities were compiled and rated independently
by two coders (trained psychology graduate students) according to
whether the activity involved risk (e.g., piloting small airplanes was
considered risky, whereas stamp collecting was not); participants
were then categorized according to whether they engaged in at least
one risky activity (Martin et al., 2002). This method was utilized,
rather than a continuous format, because no information on frequency
was available. Thus, a summation of risky activities might yield a high
score for someone who had tried a number of risky endeavors only
once, as opposed to a person who regularly engaged in a single
high-risk activity. The dichotomous categorization offers less preci-
sion (and effects associated with this measure are likely attenuated),
but it also minimizes misclassifications of the type described above.
Mental health. In 1950, Terman and his research team classi-
fied participants into one of three categories based on their re-
sponses to questions in 1936, 1940, 1945, and 1950, as well as
personal interviews. These classifications were (a) showing serious
maladjustment, (b) showing some maladjustment, and (c) satisfac-
torily adjusted (Martin et al., 1995).
Education. Cumulative level of education was reported by
participants in 1950, and this was used as a measure of socioeco-
nomic status. Information on income was also available in the data
set but included only earned income, although some participants
reported other forms, such as room and board. In addition, earned
income underestimates socioeconomic status for women who did
not work outside the home.
Cause of death. Cause-of-death data were recorded from death
certificates, most of which our research team gathered from state
and county agencies. Each certificate was coded by a certified
nosologist for underlying cause of death, using the ninth revision
of the International Classification of Diseases (U.S. Department of
Health and Human Services, 1980). These detailed codes were
collapsed into five broad categories: (a) cardiovascular disease, (b)
cancer, (c) accident or injury, (d) other, and (e) unknown. In cases
where death certificates were not available (n ⫽ 81), information
from next of kin was used by our physician expert on death
classification to assign the deaths to one of the above categories.
Statistical Analyses
Proportional hazards regressions, which assess mortality risk
while taking age into account, were used to test the associations of
childhood and adult personality with longevity and cause of death.
We then explored variables that might help to explain these asso-
ciations and tested for gender interactions. Tuma’s RATE program
was used for these analyses (Tuma, 1980).
3
3
In this article we present results from Gompertz models, having cross-
checked the analyses against the more widely used Cox model (which
makes no assumptions about the survival curve’s shape, assuming instead
that the ratio of the hazard functions for participants, whose values of the
covariates differ, is not age dependent and that a log-linear function is
therefore adequately able to represent the effects of these covariates on
hazard rates). Results from the comparable Gompertz and Cox models
were very similar in each case.
Table 1
Personality Scales, Alpha Reliabilities, and Descriptive Statistics
Personality variable
␣ reliability Test–retest
1922 1940
a
1950
b
1940/1950
Childhood
Conscientiousness .76 (4)
Cheerfulness-humor .52 (2)
Sociability .65 (5)
Permanency of mood single item
Adulthood
Neuroticism .85 (17) .68 (8) .87
Agreeableness .72 (11) .66 (4) .75
Extraversion .65 (7) .65 (3) .66
Conscientiousness .65 (7) .71 (5) .71
Note. The number of items per scale is shown in parentheses.
a
n ⫽ 1,022.
b
n ⫽ 781.
430
MARTIN, FRIEDMAN, AND SCHWARTZ
Correlations between personality and potential explanatory vari-
ables (alcohol use, smoking, BMI, risk taking, mental health, and
education) were examined, and those demonstrating significant
relationships were included in survival analyses to see if they
might elucidate mechanisms by which personality is related to
mortality risk. To better understand possible pathways, we also
tested whether predictive strength varied across causes of death. A
constrained Gompertz proportional hazards model, in which the
independent variable predicted equally to all causes of death, was
compared with an unconstrained model, in which the independent
variable was allowed to predict differentially to individual causes
of death. In addition, the well-documented relationship between
sex and mortality risk was evident in this sample, and so sex was
controlled for in all analyses.
Because the relative hazard represents the increase or decrease
in risk associated with a 1-point change on a predictor variable, we
present the interquartile relative hazard (InterQ rh) for variables
with a range greater than 3 points. In these cases, every individual
with data was included in the analyses, but the estimated effect
indicates the relative risk associated with being at the 75th versus
the 25th percentile of the predictor.
Results
Personality and Mortality Risk
As of the year 2000, 492 (70%) of the men and 297 (51%) of the
women in this sample had verified deaths. The median age of
death, calculated with Kaplan–Meier estimates, was 72.6 years for
men and 74.5 years for women.
Childhood personality. Across 7 decades (1930 –2000), child-
hood conscientiousness, InterQ rh(1252) ⫽ 0.81, p ⬍ .001, and
cheerfulness, InterQ rh(1252) ⫽ 1.19, p ⬍ .01, were both signif-
icantly related to life span mortality risk, with conscientious chil-
dren living longer and cheerful children dying sooner. Childhood
sociability, InterQ rh(1252) ⫽ 0.96, ns, was unrelated to mortality
risk. Permanency of mood (lower neuroticism) was related to
decreased mortality risk, although this association was driven by
the effect for males. There was no significant gender interaction,
and this was the only case in which a gender difference in the
direction of the relative hazards was found, male InterQ rh(705) ⫽
0.89, p ⬍ .05; female InterQ rh(546) ⫽ 1.05, ns. These results
replicate through the year 2000 our earlier findings (Friedman et
al., 1993) that predicted deaths through 1986 (and now include
Terman participants who have been newly located by us during the
past dozen years as dead or alive).
Adult personality. Conscientiousness was the only adult per-
sonality variable significantly related to mortality risk from 1950
to 2000, InterQ rh(1071) ⫽ 0.82, p ⬍ .001 (see Figure 1). Adult
neuroticism, InterQ rh(1071) ⫽ 1.00, ns; agreeableness, InterQ
rh(1071) ⫽ 0.94, ns
4
; and extraversion, InterQ rh(1071) ⫽ 1.04,
ns, all failed to predict mortality risk from 1950 to 2000.
Child Versus Adult Personality
Both child and adult conscientiousness were shown, in the first
set of survival analyses, to be significant predictors of mortality
risk. The two scales were reliably but only modestly correlated,
r(1071) ⫽ .15, p ⬍ .01. Although this is consistent with some
recent reports on the measurement of conscientiousness over time
(e.g., Shiner, Masten, & Roberts, 2003), other studies have shown
considerable stability of the construct over time, and heritability
coefficients for conscientiousness are generally robust (e.g., Jang,
Livesley, & Vernon, 1996; Loehlin, McCrae, Costa, & John, 1998;
Luciano, Wainwright, Wright, & Martin, 2006). The construct of
personality change is complex (Caspi & Roberts, 1999), and the
more modest correlation shown here likely reflects the fact that the
child and adult conscientiousness variables differed in a number of
ways, including method (parent and teacher ratings vs. self-report),
emphasis on different facets of the construct (the childhood scale
was more focused on self-discipline), and developmental stage at
the time of assessment (child vs. adult, allowing for real change).
Given that conscientiousness was predictive of mortality risk at
both time points, a survival analysis was conducted in which both
indicators were entered into the same equation. In this analysis,
each indicator of conscientiousness retained statistical significance
and with equal strength; each InterQ rh(1071) ⫽ 0.85, p ⬍ .01.
There was no evidence of a gender interaction with either measure.
This analysis indicates that childhood conscientiousness did not
predict mortality risk solely through its link with adult conscien-
tiousness (see Figure 2, which plots the predicted probability of
death by childhood conscientiousness while controlling for adult
conscientiousness).
Explanatory Pathways
Because both child- and adult-measured conscientiousness were
predictive of decreased mortality risk, we examined whether rel-
atively high versus low long-term conscientiousness and increas-
ing versus decreasing pathways were differentially related to mor-
tality outcomes (as measured by conscientiousness scores at
childhood and adulthood). We thus plotted four predicted survival
curves, one for each of the following: (a) an individual in the upper
quartile in both childhood and adulthood [⫹, ⫹], (b) an individual
in the upper quartile in childhood and the lower quartile in adult-
hood [⫹, ⫺], (c) an individual in the lower quartile in childhood
and the upper quartile in adulthood [⫺, ⫹], and (d) an individual
in the lower quartile in childhood and adulthood [⫺, ⫺]. Figure 3
shows that the highest mortality risk was for those with low
conscientiousness at both time points, whereas those who scored
high at both time points had the lowest risk. Individuals who were
low at one time point and high at the other were nearly identical in
terms of risk, regardless of direction of change, and this risk was
about midway between the other two.
Causes of death may also be informative regarding the biopsy-
chosocial pathways leading to death, and so analyses treating each
4
In an earlier article (Martin & Friedman, 2000), the relative hazard (rh)
associated with agreeableness in the Terman sample was reported as 0.84
( p ⬍ .01). Because the current study found a weaker relationship (rh ⫽
0.94, now not significant), the following differences between that study and
this one should be noted: (a) The earlier analysis predicted deaths from
1940 to 1991, whereas the current study predicted from 1950 to 2000; (b)
The earlier analysis included only participants who had 1940 personality
data available, whereas the current study combined information from 1940
and 1950 when both were available and used either 1940 or 1950 data if
only one was available (resulting in different samples for the two studies).
Future researchers might note this indication that agreeableness may have
some protective effect.
431
PERSONALITY AND MORTALITY RISK
cause of death as a separate dependent variable were conducted
next. In this sample, 34% of deaths were from cardiovascular
disease, 35% were from cancer, 8% were from injuries (trauma),
15% were from other causes, and 7% were from unknown causes.
Although there were some slight differences across our five cause-
of-death categories in the risk associated with low conscientious-
ness, these differences were not statistically significant. The most
interesting of these differences was that less conscientious indi-
viduals were more likely to die from injuries than were those
higher on conscientiousness, but perhaps because so few individ-
uals overall died of injury, this trend was not significant.
Next, behavioral and psychosocial variables were examined to
better understand the observed personality–mortality associations.
These variables were identified by an examination of correlations
with the personality predictors. Adult alcohol use, smoking, BMI,
risk taking, education, and mental health were all related to one or
0.
00
0.
10
0.20
0.
30
0.40
0.
50
0.60
0.70
0.80
0.90
50 55 60 65 70 75 8
0
85
9
0
Age
Probability of Death
Low Conscientiou
sness (25th %-ile)
High Conscientiousness (75th %-ile)
Men
Wo
men
Figure 1. Estimated probability of a Terman participant’s dying (1950 –2000) by a given age, by sex and adult
conscientiousness.
0.
00
0.
10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
50 55 60
65 70 75
80 85 90
Age
Probability of Death
Low Conscientiousness (25t
h %-ile)
High Conscientiousness (75th %-ile)
Men
Women
Figure 2. Estimated probability of a Terman participant’s dying (1950–2000) by a given age, by sex and
childhood conscientiousness, controlling for adult conscientiousness.
432
MARTIN, FRIEDMAN, AND SCHWARTZ
more of the relevant personality variables (childhood conscien-
tiousness, cheerfulness, and permanency of mood, and adulthood
conscientiousness; see Table 2). Therefore, these behavioral and
psychosocial variables were entered as controls into proportional
hazards regression equations along with the personality variables.
As can be seen in Table 3, the predictive power of childhood
conscientiousness was unchanged when alcohol use, smoking, and
education were taken into account (compare Models 1 and 2).
Adulthood conscientiousness was examined next, along with al-
cohol use, smoking, BMI, education, and mental health. In the
subsample of individuals with data available on all of these di-
mensions, and with these variables controlled, adult conscientious-
ness was no longer a significant predictor of mortality risk (com-
pare Models 3 and 4 in Table 3). Childhood cheerfulness also
failed to maintain its predictive ability when alcohol use, smoking,
risk taking, and mental health were included in the equation.
5
The
analysis of permanency of mood for males, controlling mental
health and education, indicated that inclusion of these variables in
the model did not diminish the predictive power of childhood
mood permanency.
Discussion
This study, which assessed the associations of personality char-
acteristics with mortality risk across 7 decades (1930 –2000), con-
firmed the importance of conscientiousness, measured in both
childhood and adulthood, for predicting life span mortality risk. Of
importance for understanding process, each characteristic had an
independent predictive effect.
The child and adult conscientiousness variables themselves
were somewhat different for the two time points, suggesting both
the complexity and the power of this basic dimension of person-
ality for understanding health. The childhood measure was derived
from parent and teacher ratings and most closely corresponded
with the facets of self-discipline and dutifulness as described by
the well-validated NEO Personality Inventory—Revised (NEO
PI–R; Costa & McCrae, 1992), whereas the adulthood measure of
conscientiousness was based on self-report data and was most
closely related to the NEO PI–R facets of achievement striving,
self-discipline, and deliberation (Martin & Friedman, 2000). Dif-
ferences in method of assessment, measurement unreliability, and
the substantial time lag from 1922 to 1940 and 1950 (including
real personality change) are likely contributors to the relative
independence of these two conscientiousness measures. Their
comparable strength suggests that a full range of conscientiousness
facets may be relevant to increased longevity. Their relative inde-
pendence raises the possibility that childhood personality dimen-
sions change, perhaps becoming more complex, as they encounter
a multitude of environments and situations over time, complicating
the efforts to ascertain the mediators of their long-term sequelae.
These possibilities merit further research.
Childhood conscientiousness remained significantly related to
mortality risk when (adult) health behavior and education were
controlled. The adulthood conscientiousness relation to mortality
risk, however, was meaningfully reduced when behavioral and
psychosocial variables were controlled, perhaps due in part to
more contemporaneous measurement. In particular, smoking be-
havior and education level were most relevant, at least for the
portion of the sample with complete data on the adulthood control
5
Because of the dramatic decrease in sample size due to missing data on
smoking, Models 2, 4, and 6 were also calculated with smoking eliminated
from the equation. For Model 2 (n ⫽ 1,011 vs. 642), the relative hazard for
conscientiousness remained unchanged at 0.81. For Model 4 (n ⫽ 965 vs.
527), the relative hazard for conscientiousness was 0.89 (vs. 0.90) and was
marginally significant ( p ⬍ .10). For Model 6 (n ⫽ 1,018 vs. 578), the
relative hazard for cheerfulness was 0.93 (vs. 0.91) and nonsignificant.
0.00
0.10
0.
20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
40 45 50 5
56
0657075808
59
0
Age
Probability of Death
Consistent Conscientious
ness Scores
Inconsistent Conscientiousness Scores
Low-High High-
Low
Low-Low High-High
Figure 3. Estimated probability of a Terman participant’s dying (1950–2000) by a given age, by conscien-
tiousness at two time points.
433
PERSONALITY AND MORTALITY RISK
variables. Given the well-established links between smoking and
mortality risk, it is not surprising that smoking (inversely corre-
lated with conscientiousness) explains a portion of the
conscientiousness–mortality link in this sample. With regard to
education, the fact that achievement striving constituted part of the
adult conscientiousness assessment may have reduced the capacity
of conscientiousness to relate to mortality risk when education was
controlled. Alternatively, those with less education may have
experienced more stress throughout adulthood, with a cumulative
effect on mortality risk.
Aside from conscientiousness, other personality characteristics
gave hints of their possible relevance. Childhood cheerfulness,
although now shown to predict increased risk across 7 decades, did
not remain a reliable risk factor in the smaller subsample with data
on potential adult explanatory variables, either before or after those
variables were controlled (this is consistent with our prior findings;
see Martin et al., 2002). For boys, permanency of mood remained
a significant predictor (with more labile males being at higher risk)
and did not appear to function through the pathways of mental
health or education. Permanency of mood was not significantly
correlated with smoking, alcohol use, risk taking, or BMI.
We confirmed and extended earlier analyses that showed no
simple relation between mortality risk and midlife neuroticism or
extraversion. This may be because these personality dimensions
interact with other aspects of life situations. For example, when
people are surrounded by a social circle of smokers and drinkers,
the extraverted may be more likely to partake. Similarly, neurot-
icism may sometimes make one either helpfully cautious (worry-
ing and running to the doctor) or unhelpfully stress prone (over-
encountering and overreacting to life’s tribulations; see Friedman,
2000, 2007). Agreeableness, which had previously been related to
decreased mortality risk through 1991, demonstrated only a
weaker and nonsignificant trend when examined through 2000
(rh ⫽ 0.94, ns). Thus, the Terman data provide some tantalizing
hints about the possible relevance of agreeableness, but these
effects are weaker and perhaps more complicated than those of
conscientiousness.
This sample, employing our multiyear follow-ups on the Ter-
man data, provides the first detailed life span prospective look at
personality and longevity. Few longitudinal studies of individual
differences and health are based on a true random sample of the
population, and the important question of generality that arises
should not be whether any results are fully generalizable (as they
usually are not); rather, the focus should be on the likely limits on
generality and the purposes for which the results are best used. The
Terman data set is especially valuable in helping to avoid over-
looking important relationships (i.e., making Type II errors); that
is, it helps uncover phenomena that might not be recognizable in
cross-sectional studies or in studies of shorter duration. Further,
the Terman sample is relatively homogeneous on dimensions of
intelligence and social class; a resulting advantage is that these
people had the ability to understand medical advice, had opportu-
nities for physical activity, and had routine health care. The sample
thus allows a clearer focus on the effects of certain psychosocial
variables than would studies on more heterogeneous samples.
The homogeneous nature of the sample with respect to intelli-
gence obviously restricts the range on intellect (openness), but our
published studies of the past decade show that there is a more than
adequate range of most individual differences; that is, the sample
was not selected on the basis of personality, and it is not homo-
geneous with regard to personality. Analyses by P. S. Sears (1979)
showed that attrition has not significantly changed the makeup of
the group with respect to a variety of demographic characteristics,
including age, income, and education, and other research has
confirmed that participants in the Terman sample are similar to
other bright, middle-class cohorts (R. R. Sears, 1984; Subotnik,
Table 2
Correlations of Personality Predictors With Potentially Explanatory Variables
Variable Child con Child cheer Child perm mood Adult con
Alcohol ⫺.13
***
(1,136)
.07
*
(1,136)
⫺.03 (1,136) ⫺.15
***
(1,068)
Men ⫺.11
**
(642)
.08 (642) ⫺.02 (642) ⫺.22
***
(571)
Women ⫺.12
**
(494)
.06 (494) ⫺.06 (494) ⫺.13
**
(497)
Smoking ⫺.13
***
(691)
.19
***
(691)
.02 (691) ⫺.17
***
(657)
Men ⫺.15
**
(359)
.22
***
(359)
⫺.01 (359) ⫺.20
***
(341)
Women ⫺.10 (332) .15
**
(332)
.05 (332) ⫺.16
**
(316)
BMI ⫺.03 (1,064) .03 (1,064) .02 (1,064) .06 (1,041)
Men .04 (590) .00 (590) .02 (590) .03 (577)
Women ⫺.03 (474) .04 (474) ⫺.01 (474) .01 (464)
Risk taking ⫺.03 (1,253) .07
*
(1,253)
⫺.01 (1,253) .01 (1,072)
Men ⫺.01 (706) .08
*
(706)
⫺.01 (706) .01 (600)
Women ⫺.04 (547) .04 (547) ⫺.04 (547) ⫺.01 (472)
Mental health .04 (1,248) .09
**
(1,248)
.07
*
(1,248)
.20
***
(972)
Men .09
*
(711)
.08 (711) .11
*
(711)
.26
***
(544)
Women ⫺.02 (537) .09 (537) .01 (537) .10
*
(428)
Education .15
***
(1,089)
.01 (1089) .10
**
(1,089)
.17
***
(1,054)
Men .16
***
(614)
.01 (614) .08 (614) .12
**
(593)
Women .17
***
(475)
⫺.02 (475) .11
*
(537)
.20
***
(461)
Note. Numbers in parentheses are ns. Higher values for personality predictors indicate more of the construct
or behavior. Child con ⫽ childhood conscientiousness; Child cheer ⫽ childhood cheerfulness; Child perm
mood ⫽ childhood permanency of mood; Adult con ⫽ adult conscientiousness; BMI ⫽ body mass index.
*
p ⬍ .05.
**
p ⬍ .01.
***
p ⬍ .001.
434
MARTIN, FRIEDMAN, AND SCHWARTZ
Karp, & Morgan, 1989). Caution is needed in generalizing from
this or any single sample, especially when cultural or cohort-
sensitive variables are likely to affect a particular relationship; this
data set is most valuable when used as a complement to the
findings of other research, and its effect sizes are not directly
generalizable to the current U.S. population.
Personality characteristics were assessed differently in child-
hood and adulthood, and indexes were constructed after the fact,
using the best available information. These factors almost certainly
attenuate the associations of these variables with one another and
with mortality risk. If comprehensive personality measures, such
as are available today, had been administered at each time point, it
is quite possible that the observed associations between conscien-
tiousness and mortality risk would be even stronger.
In terms of explanatory pathways, missing data on one or more
of the control variables resulted in various exclusions from the
final analyses. In some cases, the full models incorporating these
control variables included only about half of the participants from
the initial models. These exclusions decrease the power and per-
haps the generalizability of the analyses, particularly in cases
where four or five control variables are present. The present
findings, although based on hard-to-obtain data, are therefore most
useful for orienting future theorizing and research and for focusing
shorter term biopsychosocial explorations of personality, behavior,
adaptation, and health.
The ways in which personality relates to longevity are not
simple, but this study provides striking evidence of the importance
of the personality dimension of conscientiousness over a 70-year
period. Complementing the recent results of other studies (e.g.,
Caspi et al., 2003; Christensen et al., 2002; Roberts & Bogg, 2004;
Roberts et al., 2003; Weiss & Costa, 2005), our finding that a key
individual-difference variable predicts a key outcome (longevity)
many decades into the future strongly suggests that attention to the
genetic, psychosocial, life-pathway, and physiological aspects of
conscientiousness is now warranted. In terms of underlying psy-
chophysiological third variables, studies hint that serotonin func-
tion may be relevant (Manuck et al., 1998; Wand et al., 2002;
Williams et al., 2004), as serotonergic functioning influences vari-
ations in impulse control. Conscientiousness clearly has far-
reaching implications and is likely one of the core health-relevant
personality dimensions in humans.
References
Caspi, A., Begg, D., Dickson, N., Harrington, H., Langley, J., Moffitt,
T. E., & Silva, P. A. (1997). Personality differences predict health-risk
behaviors in young adulthood: Evidence from a longitudinal study.
Journal of Personality and Social Psychology, 73, 1052–1063.
Caspi, A., & Roberts, B. W. (1999). Personality change and continuity
across the life course. In L. A. Pervin & O. P. John (Eds.), Handbook of
personality: Theory and research (pp. 300 –326). New York: Guilford
Press.
Caspi, A., Sugden, K., Moffitt, T. E., Taylor, A., Craig, I. W., Harrington,
H., et al. (2003, July 18). Influence of life stress on depression: Mod-
eration by a polymorphism in the 5-HTT gene. Science, 301, 386 –389.
Christensen, A. J., Ehlers, S. L., Wiebe, J. S., Moran, P. J., Raichle, K.,
Ferneyhough, K., & Lawton, W. J. (2002). Patient personality and
mortality: A 4-year prospective examination of chronic renal insuffi-
ciency. Health Psychology, 21, 315–320.
Costa, P. T., Jr., & McCrae, R. R. (1992). The Revised NEO Personality
Inventory professional manual. Odessa, FL: Psychological Assessment
Resources.
Danner, D. D., Snowdon, D. A., & Friesen, W. V. (2001). Positive
emotions in early life and longevity: Findings from the Nun Study.
Journal of Personality and Social Psychology, 80, 804 – 813.
Friedman, H. S. (2000). Long-term relations of personality and health:
Dynamisms, mechanisms, tropisms. Journal of Personality, 68, 1089 –
1108.
Friedman, H. S. (2007). Personality, disease, and self-healing. In H. S.
Friedman & R. C. Silver (Eds.), Foundations of health psychology (pp.
172–199). New York: Oxford University Press.
Friedman, H. S., Tucker, J. S., Schwartz, J. E., Martin, L. R., Tomlinson–
Keasey, C., Wingard, D. L., & Criqui, M. H. (1995). Childhood con-
scientiousness and longevity: Health behaviors and cause of death.
Journal of Personality and Social Psychology, 68, 696 –703.
Friedman, H. S., Tucker, J. S., Schwartz, J. E., Tomlinson–Keasey, C.,
Martin, L. R., Wingard, D. L., & Criqui, M. H. (1995). Psychosocial and
behavioral predictors of longevity: The aging and death of the “Ter-
mites.” American Psychologist, 50, 69 –78.
Friedman, H. S., Tucker, J. S., Tomlinson–Keasey, C., Schwartz, J. E.,
Wingard, D. L., & Criqui, M. H. (1993). Does childhood personality
predict longevity? Journal of Personality and Social Psychology, 65,
176 –185.
Table 3
Personality Predictors Controlling for Behavioral and
Psychosocial Variables
Personality Predictor brhp
Model 1 (n ⫽ 1,253)
Sex ⫺0.31 0.74 ⬍.001
Child conscientiousness ⫺0.21 0.81 ⬍.001
Model 2 (n ⫽ 642)
Sex ⫺0.33 0.72 ⬍.01
Child conscientiousness ⫺0.21 0.81 ⬍.01
Alcohol use 0.01 1.01 ns
Smoking 0.30 1.35 ⬍.001
Education ⫺0.13 0.88 ns
Model 3 (n ⫽ 1,072)
Sex ⫺0.27 0.76 ⬍.001
Adult conscientiousness ⫺0.20 0.82 ⬍.001
Model 4 (n ⫽ 527)
Sex ⫺0.30 0.74 ⬍.05
Adult conscientiousness ⫺0.10 0.90 ns
Alcohol use 0.07 1.07 ns
Smoking 0.30 1.35 ⬍.001
BMI 0.05 1.05 ns
Education ⫺0.17 0.85 ns
Mental health ⫺0.04 0.95 ns
Model 5 (n ⫽ 1,253)
Sex ⫺0.31 0.74 ⬍.001
Child cheerfulness 0.12 1.13 ⬍.05
Model 6 (n ⫽ 578)
Sex ⫺0.22 0.80 ns
Child cheerfulness ⫺0.10 0.91 ns
Alcohol use 0.04 1.04 ns
Smoking 0.37 1.45 ⬍.001
Risk taking ⫺0.16 0.85 ns
Mental health ⫺0.10 0.90 ns
Model 7 (n ⫽ 706, males only)
Child permanency of mood ⫺0.12 0.89 ⬍.05
Model 8 (n ⫽ 522, males only)
Child permanency of mood ⫺0.23 0.80 ⬍.01
Mental health ⫺0.19 0.83 ns
Education ⫺0.24 0.78 ⬍.001
Note. Ns for analyses differ because of missing data per variable (espe-
cially smoking). rh ⫽ relative hazard; interquartile rh is presented.
435
PERSONALITY AND MORTALITY RISK
Hampson, S. E., Dubanoski, J. P., Hamada, W., Marsella, A. J., Mat-
sukawa, J., Suarez, E., & Goldberg, L. (2001). Where are they now?
Locating former elementary-school students after nearly 40 years for a
longitudinal study of personality and health. Journal of Research in
Personality, 35, 375–387.
Hampson, S. E., Goldberg, L. R., Vogt, T. M., & Dubanoski, J. P. (2006).
Forty years on: Teachers’ assessments of children’s personality traits
predict self-reported health behaviors and outcomes at midlife. Health
Psychology, 25, 57– 64.
Jang, K. L., Livesley, W. J., & Vernon, P. A. (1996). Heritability of the Big
Five personality dimensions and their facets: A twin study. Journal of
Personality, 64, 577–591.
John, O. (1990). The “Big Five” factor taxonomy: Dimensions of person-
ality in the natural language and in questionnaires. In L. A. Pervin (Ed.),
Handbook of personality: Theory and research (pp. 66 –100). New
York: Guilford Press.
Loehlin, J. C., McCrae, R. R., Costa, P. T., Jr., & John, O. P. (1998).
Heritabilities of common and measure-specific components of the Big
Five personality factors. Journal of Research in Personality, 32, 431–
453.
Luciano, M., Wainwright, M. A., Wright, M. J., & Martin, N. G. (2006).
The heritability of conscientiousness facets and their relationship to IQ
and academic achievement. Personality and Individual Differences, 40,
1189 –1199.
Manuck, S. B., Flory, J. D., McCaffery, J. M., Matthews, K. A., Mann,
J. J., & Muldoon, M. F. (1998). Aggression, impulsivity, and central
nervous system serotonergic responsivity in a nonpatient sample. Neu-
ropsychopharmacology, 19, 287–299.
Martin, L. R., & Friedman, H. S. (2000). Comparing personality scales
across time: An illustrative study of validity and consistency in life-span
archival data. Journal of Personality, 68, 85–110.
Martin, L. R., Friedman, H. S., Tucker, J. S., Schwartz, J. E., Criqui, M. H.,
Wingard, D. L., & Tomlinson–Keasey, C. (1995). An archival prospec-
tive study of mental health and longevity. Health Psychology, 14,
381–387.
Martin, L. R., Friedman, H. S., Tucker, J. S., Tomlinson–Keasey, C.,
Criqui, M. H., & Schwartz, J. E. (2002). A life course perspective on
childhood cheerfulness and its relation to mortality risk. Personality and
Social Psychology Bulletin, 28, 1155–1165.
McCann, S. J. H. (2005). Longevity, Big Five personality factors, and
health behaviors: Presidents from Washington to Nixon. Journal of
Psychology, 139, 273–286.
McCrae, R. R., & Costa, P. T., Jr. (1987). Validation of the five-factor
model of personality across instruments and observers. Journal of Per-
sonality and Social Psychology, 52, 81–90.
McCrae, R. R., & John, O. (1992). An introduction to the five-factor model
and its applications. Journal of Personality, 60, 175–215.
Roberts, B. W., & Bogg, T. (2004). A longitudinal study of the relation-
ships between conscientiousness and the social environmental factors
and substance use behaviors that influence health. Journal of Personal-
ity, 72, 325–353.
Roberts, B. W., Caspi, A., & Moffitt, T. E. (2003). Work experiences and
personality development in young adulthood. Journal of Personality and
Social Psychology, 84, 582–593.
Sears, P. S. (1979). The Terman studies of genius, 1922–1972. In A. H.
Passow (Ed.), The gifted and talented: Their education and development
(pp. 75–96). Chicago: University of Chicago Press.
Sears, R. R. (1984). The Terman gifted children study. In S. A. Mednick,
M. Harway, & K. M. Finello (Eds.), Handbook of longitudinal research
(pp. 398–414). New York: Praeger.
Shiner, R. L., Masten, A. S., & Roberts, J. M. (2003). Childhood person-
ality foreshadows adult personality and life outcomes two decades later.
Journal of Personality, 71, 1145–1170.
Smith, T. W., Gallo, L. C., & Ruiz, J. M. (2003). Toward a social
psychophysiology of cardiovascular reactivity. In J. Suls & K. A. Wall-
ston (Eds.), Social psychological foundations of health and illness (pp.
335–366). Malden, MA: Blackwell.
Subotnik, R. F., Karp, D. E., & Morgan, E. R. (1989). High IQ children at
midlife: An investigation into the generalizability of Terman’s genetic
studies of genius. Roeper Review, 11, 139 –144.
Terman, L. M. (with Baldwin, B. T., Bronson, E., DeVoss, J. C., Fuller, F.,
Goodenough, F. L., Kelley, T. L., et al.). (1925). Genetic studies of
genius: Vol. I. Mental and physical traits of a thousand gifted children.
Stanford, CA: Stanford University Press.
Terman, L. M., & Oden, M. H. (1947). Genetic studies of genius: Vol. IV.
The gifted child grows up. Stanford, CA: Stanford University Press.
Tucker, J. S., Friedman, H. S., Tomlinson–Keasey, C., Schwartz, J. E.,
Wingard, D. L., Criqui, M. H., & Martin, L. R. (1995). Childhood
psychosocial predictors of adulthood smoking, alcohol consumption,
and physical activity. Journal of Applied Social Psychology, 25, 1884–
1899.
Tuma, N. (1980). Invoking RATE. Unpublished manuscript, Stanford Uni-
versity, Stanford, CA.
United States Department of Health and Human Services. (1980). Inter-
national Classification of Diseases, Ninth Revision, Clinical Modifica-
tion (2nd ed.). Washington, DC: Author.
Wand, G. S., McCaul, M., Yang, X., Reynolds, J., Gotjen, D., Lee, S., &
Ali, A. (2002). The mu-opioid receptor gene polymorphism (A188G)
alters HPA axis activation induced by opioid receptor blockade. Neuro-
psychopharmacology, 26, 106 –114.
Weiss, A., & Costa, P. T., Jr. (2005). Domain and facet personality
predictors of all-cause mortality among Medicare patients aged 65 to
100. Psychosomatic Medicine, 67, 724 –733.
Williams, R. B., Kuhn, C. M., Helms, M. J., Siegler, I. C., Barefoot, J. C.,
Ashley-Kocy, A., et al. (2004, March). Central nervous system (CNS)
serotonin function and NEO–PI personality profiles. Paper presented at
the 62nd annual meeting of the American Psychosomatic Society, Or-
lando, FL.
Wilson, R. S., Mendes de Leon, C. F., Bienias, J. L., Evans, D. A., &
Bennett, D. A. (2004). Personality and mortality in old age. Journals of
Gerontology Series B: Psychological Sciences and Social Sciences, 59,
P110 –P116.
436
MARTIN, FRIEDMAN, AND SCHWARTZ
A preview of this full-text is provided by American Psychological Association.
Content available from Health Psychology
This content is subject to copyright. Terms and conditions apply.
