Content uploaded by Bryan J Pesta
Author content
All content in this area was uploaded by Bryan J Pesta on Oct 17, 2018
Content may be subject to copyright.
This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright
Author's personal copy
Differential epidemiology: IQ, neuroticism, and chronic disease by the
50 U.S. states
Bryan J. Pesta
a,
⁎, Sharon Bertsch
b
, Michael A. McDaniel
c
,
Christine B. Mahoney
a
, Peter J. Poznanski
a
a
Cleveland State University, United States
b
University of Pittsburgh at Johnstown, United States
c
Virginia Commonwealth University, United States
article info abstract
Article history:
Received 12 February 2011
Received in revised form 14 December 2011
Accepted 24 January 2012
Available online 15 February 2012
Current research shows that geo-political units (e.g., the 50 U.S. states) vary meaningfully on
psychological dimensions like intelligence (IQ) and neuroticism (N). A new scientific discipline
has also emerged, differential epidemiology, focused on how psychological variables affect
health. We integrate these areas by reporting large correlations between aggregate-level IQ
and N (measured for the 50 U.S. states) and state differences in rates of chronic disease (e.g.,
stroke, heart disease). Controlling for health-related behaviors (e.g., smoking, exercise) re-
duced but did not eliminate these effects. Strong relationships also existed between IQ, N, dis-
ease, and a host of other state-level variables (e.g., income, crime, education). The nexus of
inter-correlated state variables could reflect a general fitness factor hypothesized by cognitive
epidemiologists, although valid inferences about causality will require more research.
© 2012 Elsevier Inc. All rights reserved.
Keywords:
Intelligence
Neuroticism
Epidemiology
50 U.S. states
1. Introduction
The study of individual differences —differential psychology
—has recently expanded to include the study of differences
across groups of people categorized by shared geography (e.g.,
states or nations). Aggregate-level measures now exist for both
intelligence (IQ) and the Big Five personality traits (Lynn &
Vanhanen, 2002; McDaniel, 2006; Rentfrow, Gosling, & Potter,
2008). These aggregate-level measures seem to consistently pre-
dict important geo-political outcomes, as reviewed below. The
goal of the present study is to illustrate the unique capacity
aggregate-level psychological variables possess in predicting dis-
ease rates across populations (here, the 50 U.S. states). These re-
lationships persist even after controlling for state income levels,
and for various health-related behaviors (smoking and exercis-
ing) that epidemiologists typically study as disease antecedents.
Because we consider both dispositional and cognitive traits, we
term this area differential epidemiology (as opposed to either dis-
positional or cognitive epidemiology —for the latter, see, e.g.,
Deary, 2010). We begin by reviewing the predictive value of
both IQ and the personality trait, neuroticism (N), measured for
individuals and for geo-political units.
1.1. Individual and aggregate-level intelligence
Intelligence tests presumably measure individual differ-
ences in the brain's ability to efficiently process information
(Jensen, 1998). Though controversial as a construct outside
psychology, a massive literature shows that individual IQ
scores predict real-world outcomes, from income levels and
socioeconomic status (Strenze, 2007), to job and school per-
formance (Kuncel, Ones, & Sackett, 2010; Schmidt & Hunter,
1998), to health and mortality (Batty, Deary, & Gottfredson,
2007; Deary, 2008; Deary, 2010; Gottfredson & Deary,
2004). For many outcome variables, IQ scores emerge as the
single best (but not the only) predictor (see, e.g., Jensen,
1998).
Intelligence 40 (2012) 107–114
⁎Corresponding author at: Cleveland State University, Department of
Management, 2121 Euclid Avenue, Cleveland OH 44115, United States.
E-mail address: b.pesta@csuohio.edu (B.J. Pesta).
0160-2896/$ –see front matter © 2012 Elsevier Inc. All rights reserved.
doi:10.1016/j.intell.2012.01.011
Contents lists available at SciVerse ScienceDirect
Intelligence
Author's personal copy
In explaining links between IQ and epidemiology,
Gottfredson (2004) argued that healthcare is a type of intelli-
gencetest(seealsoGottfredson, 1997). Namely, health mainte-
nance involves active participation in a series of tasks (e.g.,
learning health-related information), duties (e.g., dealing with
health emergencies) and responsibilities (e.g., adhering to treat-
ment). These behaviors require cognitive resources to manage
effectively. Individuals (or groups of people) with high IQ
would likely be in the best position to handle the complex spec-
trum of knowledge and behavior needed for good health.
Whether geographical units (versus individuals) differ in IQ
has drawn increased attention from psychologists. In the aggre-
gate, IQ scores have now been calculated for nations across the
world (Lynn & Meisenberg, 2010; Lynn & Vanhanen, 2002), and
for the 50 U.S. states (McDaniel, 2006). Both national and U.S.
state IQs predict many of the things that individual IQ scores
do, including socio-economic status (Pesta, McDaniel, &
Bertsch, 2010), education (Lynn & Meisenberg, 2010), and
crime (Pesta et al., 2010). Particularly relevant are recent stud-
ies showing links between aggregate IQ and epidemiologic out-
comes (e.g., global state health: Pesta et al., 2010; life
expectancy, mortality and fertility rates: Reeve, 2009;positive
and negative health indicators: Reeve & Basalik, 2010).
1.2. Individual and aggregate-level neuroticism
Personality is the set of psychological traits or constructs
that create consistency in how people think, act and feel
(John, Robins, & Pervin, 2008). A highly regarded theoretical
perspective on personality is the “Big Five”model (Costa &
McCrae, 1992). The model assumes that five factors explain
most of the variance in one's personality: neuroticism, extra-
version, openness, agreeableness and conscientiousness. We
focus here on just neuroticism, as it emerged as the only con-
sistent Big Five predictor of epidemiologic outcomes (e.g.,
rates of heart disease or high blood pressure) and health-
related behaviors (e.g., rates of smoking or exercise). Individ-
uals scoring high on N tend to be anxious, stressed, and
worry-prone, while those scoring low tend to be the opposite
(Costa & McCrae, 1992).
Among individuals, N correlates with many health-related
variables, including depression and anxiety disorders (Jyhla
& Isometsa, 2006), mortality (Deary et al., 2008; Wiebe,
Drew, & Croom, 2010), coping skill (John et al., 2008), death
from cardiovascular disease (Shipley, Weiss, Der, Taylor, &
Deary, 2007), and whether one smokes tobacco (Munaf,
Zetteler, & Clark, 2007). Recent research also shows a strong
relationship between N and metabolic syndrome; a chronic
complex of health symptoms associated with increased
heart disease and mortality (Phillips et al., 2010). To explain
this relationship, Phillips et al. (2010) suggest that N “may
be a marker of central nervous system (CNS) excitation,
with higher levels leading to biological senescence, thus, in-
creasing susceptibility to disease”(p. 193).
As with aggregate-level IQ, psychologists have recently
focused on how personality traits vary across geographical
units. Estimates now exist of the Big Five personality traits
for each of the 50 U.S. states (Rentfrow, 2010; Rentfrow et
al., 2008). State personality predicts many interesting aspects
of American culture, including political preference and voting
patterns (Rentfrow, Jost, Gosling, & Potter, 2009). Consistent
with research on individuals, N seems to be the best predictor
(among the Big Five traits) of health outcomes for the 50 U.S.
states (as reviewed by Rentfrow et al., 2008).
Why do geo-political units differ meaningfully on psycho-
logical dimensions? One possibility is the attraction/similarity
paradigm, where people are drawn to others who closely re-
semble them in characteristics like cultural background, per-
sonality, or shared demographics (Lydon, Jamieson, & Zanna,
1988). Both social (e.g., religious beliefs and customs) and ge-
netic (e.g., IQ and personality, in part) factors characterize the
settlers of a particular geographic area. Settler characteristics
then become the basis for local beliefs and behaviors, which ei-
ther attract or repel future residents from assimilating a com-
munity's culture. These specific characteristics likely still
remain represented genetically and culturally in local popula-
tions in a non-random fashion (Rentfrow et al., 2008).
1.3. Explaining links between aggregate IQ/N and health
Arden, Gottfredson, and Miller (2009) proposed four pos-
sible explanations for links between individual-level IQ and
health. We generalize their discussion here to include rela-
tionships between aggregate-level IQ, N and the health of
populations:
(1) IQ/N and health could be influenced by common ge-
netic factors.
(2) IQ/N and health could be influenced by common envi-
ronmental factors.
(3) Health could influence IQ/N.
(4) IQ/N could influence health (Arden et al., 2009, p. 581).
Explanations (1) and (2) contrast genetic and environ-
mental factors. In explanation (1), genes and genetic muta-
tions affect health, IQ and N. This explanation is preferred
by Arden et al. (2009), who argued for the existence of a gen-
eral fitness factor, determined by genetics. The fitness factor
subsumes IQ, N and health outcomes. Links between IQ/N
and health are mediated by differences in lifestyle behaviors
(e.g., smoking, exercising), which then lead to differences in
disease rates across individuals or populations. In explana-
tion (2), the relationship between IQ/N and health is caused
by environmental variables. Examples include prenatal care,
social stress, and pathogen loads.
The last two explanations differ on the direction of pre-
sumed causality. In explanation (3), health influences IQ/N,
while the reverse holds in explanation (4). For the former,
perhaps good health increases brain efficiency (as measured
by IQ) and reduces stress (as measured by N); whereas dis-
ease decreases brain efficiency and increases stress. For ex-
planation (4), high IQ/low N individuals might be more
likely to engage in behaviors (e.g., exercise, eating healthy)
conducive to good health. Though similar to explanation (1)
in terms of what it predicts, explanation (4) does not necessar-
ily implicate genetics. For example, high IQ might indirectly af-
fect health by improving educational and career opportunities
(Arden et al., 2009).
All four explanations probably contribute to the relation-
ship between IQ/N and health (Arden et al., 2009). Consistent
with this conclusion, most important socio-political variables
(including health outcomes) are strongly inter-correlated at
the aggregate level. For example, Pesta et al. (2010) identified
108 B.J. Pesta et al. / Intelligence 40 (2012) 107–114
Author's personal copy
a robust general factor of state “well-being,”comprised of the
following sub-domains: intelligence, crime, education, income,
health, and religious fundamentalism. Inter-dependence
among state-level outcome variables seems to be the rule, rath-
er than the exception (Pesta et al., 2010). Interestingly, the
well-being nexus identified by Pesta et al. (2010) might largely
reflect the general fitness factor hypothesized by Arden et al.
(2009). The nexus may also result from the joint effects of all
four explanations reviewed above. More research is needed in
order to make inferences about the relative importance of
each explanation for links between IQ/N and health.
Toward that end, we provide an examination of how IQ
and N link to the behavioral antecedents of disease (e.g.,
smoking, exercising) and to disease itself. Then, we examine
how these variables correlate with other important sub-
domains (e.g., income, crime) of state well-being. Finally,
we describe challenges for researchers interested in the caus-
al mechanisms (i.e., the four explanations reviewed above)
that best explain these relationships.
Studying IQ and N at the aggregate level allows researchers
to capitalize on large, reliable data bases maintained by the U.S.
federal government. One example is the Behavioral Risk Factor
Surveillance System (BRFSS). The BRFSS tracks health and wel-
fare across the 50 U.S. states. The system is updated annually,
via population-representative surveys of residents in each
state. Data exist on both the incidence of health behaviors
(e.g., exercise; smoking) and chronic conditions (e.g., heart dis-
ease; high cholesterol) by U.S. geography.
We coded data comprising ten variables reported in a cur-
rent BRFSS Surveillance Summary (BRFSS, 2010). We selected
this specific summary because it is timely and includes many
common, chronic health problems impacting the well-being
of millions of people. Via regression, we first tested whether
IQ/N predict chronic disease, and then whether these effects
are attenuated by including health behaviors and state in-
come levels in the model. Thereafter, we add other state-
level variables illustrating a nexus of inter-correlated psycho-
logical, epidemiological, and environmental outcomes.
2. Method
2.1. State IQ and state N
State IQ estimates come from McDaniel (2006), and have
a mean of 100.3 with a standard deviation of 2.70. State N es-
timates come from Rentfrow et al. (2008) and are reported as
Zscores (mean= 0; σ= 1). All remaining variables were
coded from an on-line, Behavioral Risk Factor Surveillance
System Summary (2010). Unless noted, all BRFSS variables
represent percentages for residents age 18 years or older in
each state.
2.2. BRFSS variables
2.2.1. Health behaviors
We coded four BRFSS variables representing behaviors
that epidemiologists traditionally study when predicting dis-
ease rates across populations. These included: (1) Activity
and Exercise (created via factor analysis on BRFSS variables
measuring light, moderate, and vigorous activity —the
Appendix displays factor loadings and alpha reliabilities for
all scaled variables used in this study), (2) Smoking (every
day or occasionally), (3) Alcohol Consumption (consumption
of more than one [women] or two [men] alcoholic beverages
per day), and (4) Healthy Eating (consuming at least five
servings of fruits and/or vegetables per day). These four vari-
ables were highly correlated, so we also combined them into a
single factor, Health Behaviors, for use in regression analyses.
2.2.2. Chronic disease
We coded the following measures of chronic disease or im-
pairment (as a percentage of state residents) from the BRFSS
summary: (1) Obesity (BMI> 30), (2) Diabetes, (3) High Blood
Pressure, (4) High Cholesterol, (5) Coronary Heart Disease, (6)
Stroke. We report data on these variables separately, and then
together as scaled into a single factor (Chronic Disease)viafactor
analysis. Finally, we also created factor scores for state Metabolic
Syndrome (using the first four of six variables representing the
Chronic Disease factor) to see if recent results by Phillips et al.
(2010) replicate at the U.S. state level.
2.2.3. Data analyses
We first report descriptives and simple correlations for all
variables. Next, we report two regressions —one features
Chronic Disease as the dependent variable, the other features
Metabolic Syndrome. For each regression, IQ and N were en-
tered in Step 1, and Health Behaviors was entered in Step 2.
Lastly, we incorporate state income levels and additional var-
iables (i.e., the sub-domains of well-being reported by Pesta
et al., 2010) to see where Health Behaviors and Chronic Dis-
ease fit within the U.S. state well-being nexus.
3. Results
3.1. Descriptives and Pearson correlations
Table 1 shows rankings by U.S. state for IQ, N and the
three BRFSS factor scores (Health Behaviors, Chronic Disease,
and Metabolic Syndrome). Note for example that West Vir-
ginia has the highest incidence of Chronic Disease. It is also
the most neurotic state in the U.S. Conversely, Utah is the
least neurotic state; whereas, Massachusetts has the highest
IQ, and Vermont ranks first in Health Behaviors.
Table 2 shows means, standard deviations, and zero-order
correlations for IQ, N, all BRFSS variables, and the three BRFSS
factor scores (with N= 50, a correlation of .28 is significant at
pb.05, one tailed). In the table, state IQ and N are essentially
uncorrelated (r=−.08). However, IQ significantly predicts
two of the four behavioral variables (activity/exercise, and
smoking), and correlates .45 with the Health Behaviors factor
score. Similarly, N significantly predicts the same two behav-
ioral variables that IQ does, and N correlates −.40 with
Health Behaviors. Both IQ and N, however, failed to correlate
with either alcohol consumption or healthy eating.
State IQ correlated moderately with four of the six disease
variables (IQ predicted neither high cholesterol nor heart dis-
ease). IQ also correlated −.51 and −.53 with Chronic Disease
and Metabolic Syndrome, respectively. Relative to state IQ, the
correlations between N and disease were stronger and more
consistent. For example, state N correlated significantly with all
six disease variables. It also correlated .59 and .62 with the
Chronic Disease and Metabolic Syndrome, respectively.
109B.J. Pesta et al. / Intelligence 40 (2012) 107–114
Author's personal copy
An unexpected finding in Table 2 is the relationship be-
tween alcohol consumption, Health Behaviors and Chronic
Disease. At the state level, drinking alcohol correlates posi-
tively with exercising and eating fruits and vegetables;
whereas it correlates negatively with rates of smoking and
many of the tabled chronic diseases. These data are consis-
tent with a growing but mixed literature showing that alco-
hol consumption correlates inversely with chronic disease
rates (see e.g., Holahan et al., 2010; see also, Arden et al.,
2009, for mixed results of alcohol consumption on a battery
of health-related variables).
In sum, the zero-order correlations in Table 2 show IQ and
N to be relatively consistent predictors of both behaviors as-
sociated with chronic diseases, and the diseases themselves.
Next we test whether these relationships are attenuated by
considering behaviors like smoking and exercising.
3.2. Multiple regressions
To avoid multi-colinearity, we used factor scores for the
BRFSS behavioral and disease variables. High scores on the
Health Behaviors factor correspond to higher rates of exercising,
Table 1
State ranks and mean values for IQ, N, and the BRFSS factor score composites.
State IQ rank/score N rank/score Health behaviors
rank/score
Chronic disease
rank/score
Metabolic syndrome
rank/score
Alabama 45.5/95.7 30/−0.26 44/−1.24 4/1.74 4/1.84
Alaska 36/99.0 46/−1.2 11/0.77 47/−1.18 45/−0.88
Arizona 43/97.4 44/−1.09 16/0.65 24/−0.17 44/−0.82
Arkansas 42/97.5 10/1.01 41/−0.84 8/1.06 7/1.17
California 48/95.5 36/−0.53 3/1.21 40/−0.82 42/−0.80
Colorado 20/101.6 49/−1.97 10/0.82 49/−2.08 49/−2.16
Connecticut 9/103.1 15/0.54 4/1.18 46/−0.96 39/−0.63
Delaware 28/100.4 20/0.21 24/0.18 17/0.47 11/0.65
Florida 38.5/98.4 35/−0.5 22/0.26 19/0.31 22/0.09
Georgia 40/98.0 32/−0.39 30/−0.03 9/0.96 9/0.91
Hawaii 47/95.6 39/−0.74 2/1.36 30/−0.30 21/0.13
Idaho 22/101.4 31/−0.36 18/0.36 32/−0.38 35/−0.55
Illinois 31/99.9 19/0.21 26/0.11 22/0.07 24/0.04
Indiana 19/101.7 13/0.88 38/−0.64 18/0.44 19/0.24
Iowa 8/103.2 22/0.15 35/−0.32 27/−0.24 39/−0.22
Kansas 12/102.8 33/−0.44 37/−0.51 29/−0.27 30/−0.26
Kentucky 34/99.4 7/1.17 50/−2.09 7/1.24 10/0.82
Louisiana 49/95.3 8/1.14 46/−1.37 6/1.25 5/1.39
Maine 6.5/103.4 12/0.9 7/0.95 20/0.15 20/0.21
Maryland 32/99.7 17/0.45 19/0.33 23/−0.03 16/0.34
Massachusetts 1/104.3 11/0.98 8/0.94 41/−0.82 38/−0.61
Michigan 27/100.5 26/−0.09 28/−0.01 15/0.57 12/0.53
Minnesota 5/103.7 40/−0.8 34/−0.22 48/−1.68 48/−1.89
Mississippi 50/94.2 4/1.5 49/−1.82 3/1.85 1/2.17
Missouri 25/101.0 25/−0.09 42/−0.93 12/0.70 14/0.48
Montana 6.5/103.4 38/−0.71 15/0.69 42/−0.85 46/−0.92
Nebraska 15/102.3 43/−1.0 25/0.12 34/−0.46 33/−0.49
Nevada 44/96.5 41/−0.83 29/−0.01 26/−0.24 27/−0.19
New Hampshire 2/104.2 14/0.7 9/0.86 35/−0.49 34/−0.50
New Jersey 12/102.8 5/1.47 20/0.30 21/0.07 23/0.06
New Mexico 45.5/95.7 29/−0.2 33/−0.19 33/−0.46 36/−0.58
New York 26/100.7 3/1.55 21/0.27 31/−0.32 26/−0.12
North Carolina 29/100.2 24/−0.06 43/−0.94 14/0.60 13/0.51
North Dakota 3.5/103.8 42/−0.84 32/−0.12 38/−0.68 37/−0.59
Ohio 18/101.8 9/1.1 36/−0.47 11/0.75 15/0.47
Oklahoma 35/99.3 27/−0.15 48/−1.77 5/1.48 6/1.38
Oregon 23/101.2 47/−1.27 5/1.11 37/−0.58 31/−0.35
Pennsylvania 21/101.5 6/1.22 27/0.10 13/0.61 18/0.29
Rhode Island 33/99.5 2/1.61 13/0.73 28/−0.26 25/−0.02
South Carolina 38.5/98.4 16/0.53 39/−0.71 10/0.81 8/0.91
South Dakota 12/102.8 48/−1.68 40/−0.71 36/−0.57 41/−0.74
Tennessee 41/97.7 23/0.11 45/−1.35 2/1.94 2/1.95
Texas 30/100.0 28/−0.17 31/−0.10 16/0.56 17/0.30
Utah 24/101.1 50/−2.52 12/0.75 50/−2.11 50/−2.32
Vermont 3.5/103.8 18/0.43 1/1.57 45/−0.94 47/−1.00
Virginia 16.5/101.9 21/0.18 17/0.56 25/−0.23 28/−0.20
Washington 16.5/101.9 45/−1.1 6/0.99 39/−0.76 40/−0.70
West Virginia 37/98.7 1/2.36 47/−1.71 1/1.99 3/1.91
Wisconsin 10/102.9 34/−0.45 14/0.72 43/−0.86 32/−0.48
Wyoming 14/102.4 37/−0.59 23/0.21 44/−0.86 43/−0.81
Note. IQ (intelligence) has a mean of 100.3 and a standard deviation of 2.7. N (neuroticism) is a Zscore.
The remaining variables are factor scores.
110 B.J. Pesta et al. / Intelligence 40 (2012) 107–114
Author's personal copy
eating fruits/vegetables, and drinking alcohol, but lower rates of
smoking. For both the Chronic Disease and Metabolic Syndrome
factors, high scores indicate higher disease rates (as a percent-
age of residents) across states.
Table 3 shows results of hierarchical regressions predicting
Chronic Disease and Metabolic Syndrome from IQ, N (entered
at Step 1) and Health Behaviors (entered at Step 2). At Step 1,
the linear combination of IQ and N alone explained 57% and
61% of the variance in Chronic Disease and Metabolic Syn-
drome, respectively. Both IQ and N remained significant (but at-
tenuated) predictors of disease, after entering Health Behaviors
at Step 2. Not surprisingly, Health Behaviors itself explained
large amounts of variance (over IQ and N) in both Chronic Dis-
ease and Metabolic Syndrome. Note that the variance explained
at Step 2 is unusually large for social science research. Fully 80%
of the variance in Chronic Disease (77% in Metabolic Syndrome)
was explained by the combination of IQ, N and Health Behav-
iors. The size of the effects here, though, could exemplify the
"high resolution" that aggregate-level data offer, relative to
studies that use individuals (see Arden et al., 2009, p. 582).
Any number of third variables could be included in the re-
gression models above. Perhaps most obvious, socioeconomic
differences across states may largely explain why IQ and N
emerge as strong predictors of disease. For various reasons,
poorer states might score lower on IQ and higher on N. To test
this hypothesis, we included a composite measure of state in-
come from Pesta et al. (2010). The state income measure was
a factor derived from U.S. census data. Variables included: in-
come per capita, disposable income per capita, the percentage
of families living in poverty, and the percentage of individuals
living in poverty (Pesta et al., 2010). We re-conducted the re-
gression analysis reported above (predicting Chronic Disease
in Table 3) and included state income levels at Step 3. However,
IQ (Beta =−.18), N (Beta =.35) and Health Behaviors (Beta=
−.53) all remained significant as predictors of chronic disease,
even after controlling for state income (Beta=−.12, ns).
3.3. nexus of state outcome variables
Adding additional third variables to the regression model
seems arbitrary, given the strong correlations between most
state-level measures. Also, showing that a variable predicts
over and above another variable does not necessarily mean
that the chosen variable is the best explanation for the data.
Differences in construct validity, base rates, or variance
across variables would affect conclusions reached via regres-
sion, independent of a variable's true effect on health. In-
stead, we believe the most compelling aspect of these
results is that any given state-level outcome is consistently
and non-trivially correlated with nearly every other outcome.
Table 3
Predicting chronic disease and metabolic syndrome from IQ, N, and health behaviors.
Variable Chronic disease Metabolic syndrome
βBSEB βBSEB
Step 1
Intelligence (IQ) −.47 −.168 .035 −.48 −.174 .033
Neuroticism (N) .55 .534 .093 .58 .559 .089
R/R
2
.75/.57 .78/.61
Step 2
Intelligence (IQ) −.22 −.079 .027 −.28 −.101 .029
Neuroticism (N) .33 .322 .070 .40 .385 .076
Health behaviors −.60 −.637 .086 −.49 −.523 .093
R/R
2
.90/.80 .88/.77
Note: IQ (intelligence) is a scaled score with a mean of 100 and a standard deviation of 2.7. N (neuroticism) is a Zscore, and all other variables are factor scores.
Table 2
Descriptive statistics and correlation matrix for state IQ, N, and all BRFSS variables.
MSD23456789101112131415
1. I 1 IQ 100.3 2.7 −.08 .51 −.29 .12 .19 −.37 −.57 −.52 −.10 −.20 −.49 .45 −.51 −.53
2 N 0.0 1.0 –−.50 .39 −.07 −.04 .28 .57 .62 .53 .61 .40 −.40 .59 .62
3 Activity and Exercise
1
51.6 4.4 –−.59 .54 .52 −.68 −.83 −.80 −.28 −.73 −.76 .92 −.85 −.82
4 Smoking 20.1 3.2 –−.37 −.57 .68 .60 .65 .53 .73 .73 −.78 .74 .68
5 Alcohol Consumption 5.1 1.3 –.52 −.49 −.37 −.23 .01 −.36 −.37 .69 −.36 −.28
6 Healthy Eating 23.7 3.5 –−.65 −.28 −.35 −.23 −.43 −.43 .74 −.42 −.37
7 Obesity 26.9 2.9 –.67 .71 .46 .59 .69 −.78 .77 .75
8 Diabetes 8.2 1.5 –.87 .48 .75 .81 −.77 .94 .91
9 High Blood Pressure 28.5 3.0 –.57 .73 .80 −.74 .95 .99
10 High Cholesterol 37.8 2.2 –.60 .49 −.36 .61 .60
11 Heart Disease 6.5 1.3 –.77 −.76 .84 .76
12 Stroke 2.6 0.5 –−.78 .91 .83
13 Health Behaviors
2
0.0 1.0 –−.83 −.77
14 Chronic Disease
2
0.0 1.0 –.98
15 Metabolic Syndrome
2
0.0 1.0 –
Note: N (neuroticism) is a Zscore. The remaining variables (except as noted) represent % means for the 50 U.S. States.
1
The reported mean is the average for three variables comprising this factor —see the Appendix A.
2
Created via Maximum likelihood factor analysis —see the Appendix A.
111B.J. Pesta et al. / Intelligence 40 (2012) 107–114
Author's personal copy
To illustrate, Table 4 shows correlations between the vari-
ables presented here and various sub-domains of well-being
reported by Pesta et al. (2010). Of the 36 correlations pre-
sented in the table, eight (22%) have values of r=.70 or
higher; fifteen (42%) have values between r= .50 and
r=.69; and eight (22%) have values between r= .30 to
r=.49. Only 5 (14%) correlations are non-significant (four
of these occur for N predicting other variables in the table).
The correlations in Table 4 could reflect the existence of the
general fitness factor, as proposed by Arden et al. (2009).How-
ever, they are also consistent with the interplay of all four expla-
nations for the IQ/N and health link, reviewed above. Isolating
causality for highly-correlated, aggregate-level variables (that
are not experimentally manipulated) is a daunting task. What
is clear, though, is that a nexus of inter-correlated variables ex-
ists, and it reliably measures psychological, environmental and
epidemiologic differences across the 50 U.S. states.
4. Discussion
The present data show that psychological variables
uniquely predict differences in chronic disease rates across
the 50 U.S. states. Moderate to strong relationships exist be-
tween IQ/N and a variety of chronic health problems, togeth-
er with the behavioral antecedents of these problems. These
relationships persist, even after controlling for income, and
for many behaviors (e.g., exercising, smoking) epidemiolo-
gists typically study as the causes of disease.
4.1. Causality
We reviewed four possible explanations for links between
IQ/N and health. Arden et al. (2009) argued that each explana-
tion partly contributes to these relationships. However, measur-
ing the relative importance of each explanation is complicated
by the correlational nature of the data. One strategy is to seek
replication across different units of analysis. Consistent patterns
across individuals, states, and nations would strengthen infer-
ences about a variable's role in affecting health.
A second strategy would be to isolate key third variables
derived from testable theories. For example, Arden et al. ar-
gued for the existence of a genetic fitness factor that influ-
ences IQ/N and health. A reliable measure of "mutation
load" for the 50 U.S. states would offer an informative (but
non-conclusive) test of this hypothesis (see Arden et al.,
2009). Unfortunately, we know of no such measure. Howev-
er, aggregate-level genetic data do exist, using US schools as
the unit of analysis. Beaver and Wright (2011) examined
data for 132 middle and high schools across the USA. They
showed that genetic variation (i.e., allelic distributions of do-
paminergic polymorphisms) predicted verbal IQ scores for
different schools, even after controlling for race.
A second example of a key third variable is parasite preva-
lence (a measure of biological stress caused by infectious dis-
ease; see, e.g., Eppig, Fincher, & Thornhill, 2011). Eppig et al.
(2011) showed that this variable uniquely predicted U.S. state
IQ, even after controlling for income and educational differ-
ences across the states. As a final example, income inequality
(the wealth difference between the richest and poorest mem-
bers of a population) seems to be the third-variable of choice
for economists (see, e.g., Diener & Oishi, 2000).
Nonetheless, we caution against making strong causal in-
ferences just because a specific variable "won" by explaining
the most unique variance in a regression model (for a discus-
sion, see Gottfredson, 2009). For example, education and IQ
are strongly correlated (Kuncel et al., 2010). Often compel-
ling theoretical reasons exist for controlling education when
testing the effects of IQ (or vice versa) on some outcome.
But, if education and IQ are co-causal, the resulting attenua-
tion of IQ (by partialing out education) would give other in-
dependent variables an unfair advantage in potential to
explain unique variance (thereby leading researchers to
faulty inferences about presumed causality). Methods be-
yond regression are needed that triangulate possible causal
mechanisms for the relationships observed here.
4.2. Other limitations
Beyond issues with correlation and causation, the present
study has other limitations. First, the substantial positive mani-
fold of the variables makes it hard to identify any single mea-
sure's contribution to health differences across the 50 U.S.
states. Second, the potential for committing the ecological fallacy
(Robinson, 1950) exists when interpreting these data. The eco-
logical fallacy sometimes occurs when making inferences about
people (or groups) from data derived from groups (or people).
For example, it does not follow that all residents of Massa-
chusetts (a state with a high IQ) are smarter, less neurotic and
healthier, relative to all residents of Mississippi (a state with a
low IQ). Nor does it follow that group-level effects necessarily
apply to individuals comprising the groups. Finally, the expla-
nations that link IQ/N and health among individuals may be dif-
ferent from those that link these variables in aggregate-level
Table 4
A nexus of inter-correlated variables potentially representing a general fitness factor.
123456789
1IQ –−.08 .45 −.51 −.53 −.76 .41 −.55 .57
2N –−.40 .59 .62 .01 −.30 .05 −.12
3 Health behaviors –−.83 −.77 −.47 .74 −.78 .66
4 Chronic disease –.98 .49 −.72 .65 −.62
5 Metabolic syndrome –.48 −.67 .61 −.55
6 Crime
1
–−.26 .51 −.42
7 Education
1
–−.62 .66
8 Religiosity
1
–−.72
9 Income
1
–
1
These variables are factor scores borrowed from Pesta et al. (2010).
112 B.J. Pesta et al. / Intelligence 40 (2012) 107–114
Author's personal copy
data. Crossing levels of analysis —from states to individuals
within a state —could be an invalid extrapolation.
5. Conclusion
Epidemiologists should regularly employee psychological
variables when studying disease patterns. The present data
show that psychological variables are non-trivially correlated
with health differences across the 50 U.S states, and with be-
haviors epidemiologists study as disease antecedents. Including
psychological variables could offer epidemiologists increased
leverage when predicting, interpreting and explaining differ-
encesindiseaseratesacrosspopulations.
Coding variables by geo-political units (versus individ-
uals) allows researchers to capitalize on the power of aggre-
gation (see, e.g., Lubinski & Humphreys, 1996). Specifically,
IQ, N and Health Behaviors jointly explained 80% of the vari-
ance in Chronic Disease (77% for Metabolic Syndrome). The
amount of variance explained by these variables is unusually
large for social science research. These effects could reflect
the potential (i.e., "high resolution," Arden et al., 2009) that
comes by aggregating data. Continued use of aggregate-level
data could help researchers design powerful tests of compet-
ing theories, thereby identifying the relative importance of
proposed explanations for these relationships.
Appendix A
Maximum likelihood factor analyses, percentage of
variance explained, and alpha reliabilities for selected BRFSS
variables.
References
Arden, R., Gottfredson, L., & Miller, G. (2009). Does a fitness factor contribute
to the association between intelligence and health outcomes? Evidence
from medical abnormality counts from 3654 US veterans. Intelligence,
37, 581–591.
Batty, G., Deary, I., & Gottfredson, L. (2007). Premorbid (early life) IQ and
later mortality risk: A systematic review. Annals of Epidemiology,17,
278–288.
Beaver, K., & Wright, J. (2011). School-level genetic variation predicts
school-level verbal IQ scores: Results from a sample of American middle
and high schools. Intelligence,39, 193–197.
Behavioral Risk Factor Surveillance System (2010). Retrieved September 1,
2011 from. www.cdc.gov/brfss/
Behavioral Risk Factor Surveillance System Summary (2010). Retrieved Sep-
tember 1, 2011, from. http://www.cdc.gov/mmwr/preview/mmwrhtml/
ss5901a1.htm
Costa, P., & McCrae, R. (1992). Normal personality assessment in clinical
practice: The NEO Personality Inventory. Psychological Assessment,4,
5–13.
Deary, I. (2008). Why do intelligent people live longer? Nature,456,
175–176.
Deary, I. (2010). Cognitive epidemiology: Its rise, its current issues, and its
challenges. Personality and Individual Differences,49, 337–343.
Deary, I. J., Batty, G. D., Pattie, A., & Gale, C. R. (2008). More intelligent, more
dependable children live longer: a 55-year longitudinal sudy of a repre-
sentaitive sample of the Scottish nation. Psychological Science,19,
874–880.
Diener, E., & Oishi, S. (2000). Money and happiness: Income and subjective
well-being across nations. In E. M. Suh (Ed.), Culture and subjective
well-being (pp. 185–205). Boston: MIT Press.
Eppig, C., Fincher, C., & Thornhill, R. (2011). Parasite prevalence and the dis-
tribution of intelligence among the states of the USA. Intelligence,39,
155–160.
Gottfredson, L. S. (1997). Why g matters: The complexity of everyday life. In-
telligence,24,79–132.
Gottfredson, L. S. (2004). Intelligence: Is it the epidemiologists' elusive “fun-
damental cause”of social class inequalities in health? Journal of Person-
ality and Social Psychology,86, 174–199.
Gottfredson, L.S. (2009). Pursuing patterns, puzzles and paradoxes. Unpub-
lished manuscript.
Gottfredson, L. S., & Deary, I. J. (2004). Intelligence predicts health and lon-
gevity, but why? Current Directions in Psychological Science,13,1–4.
Holahan, C., Schutte, K., Brennan, P., Holahan, C., Moos, B., & Moos, R. (2010).
Late-life alcohol consumption and 20-year mortality. Alcoholism, Clinical
and Experimental Research, doi:10.1111/j.1530-0277.2010.01286.x.
Jensen, A. (1998). The g factor: The science of mental ability. Santa Barbara, CA:
Praeger Publishers.
John, O. P., Robins, R. W., & Pervin, L. A. (2008). Handbook of Personality, 3rd
edition: Theory and research. New York, NY: Guilford Press.
Jyhla, P., & Isometsa, E. (2006). The relationship of neuroticism and extraver-
sion to symptoms of anxiety and depression in the general population.
Depression and Anxiety,23, 281–290.
Kuncel, N. R., Ones, D. S., & Sackett, P. R. (2010). Individual differences as pre-
dictors of work, educational, and broad life outcomes. Personality and In-
dividual Differences,49, 331–336.
Lubinski, D., & Humphreys, L. G. (1996). Seeing the forest from the trees:
When predicting the behavior or status of groups, correlate means. Psy-
chology, Public Policy, and Law,2, 363–376.
Lydon, J. E., Jamieson, D. W., & Zanna, M. P. (1988). Interpersonal similarity
and the social and intellectual dimensions of first impressions. Social
Cognition,6, 269–286.
Lynn, R., & Meisenberg, G. (2010). National IQs validated for 108 nations. In-
telligence,38, 353–360.
Lynn, R., & Vanhanen, T. (2002). IQ and the Wealth of Nations. Santa Barbara,
CA: Praeger\Publishers.
McDaniel, M. A. (2006). Estimating state IQ: Measurement challenges and
preliminary correlates. Intelligence,34, 607–619.
Munaf, M. R., Zetteler, J. I., & Clark, T. G. (2007). Personality and smoking sta-
tus: A meta-analysis. Nicotine & Tobacco Research,9, 405–413.
Pesta, B. J., McDaniel, M. A., & Bertsch, S. (2010). Toward an index of well-
being for the fifty U.S. states. Intelligence,38, 160–168.
Phillips, A. C., Batty, G. D., Weiss, A., Deary, I., Gale, C. R., Thomas, G. N., &
Carroll, D. (2010). Neuroticism, cognitive ability, and the metabolic syn-
drome: The Vietnam Experience Study. Journal of Psychosomatic Re-
search,69(2), 193–201.
Reeve, C. L. (2009). Expanding the g nexus: Further evidence regarding the
relationship among national IQ, religiosity, and national health out-
comes. Intelligence,37, 495–505.
Factor BRFSS item factor
loading
% Variance
explained
Alpha
reliability
Activity and exercise Light activity
1
.92
moderate activity
2
.98
vigorous activity
3
.95
90% .94
Health behaviors Activity and exercise .77
smoking −.73
alcohol consumption .64
healthy eating .74
52% .88
Chronic disease Obesity .75
diabetes .92
high blood pressure .93
high cholesterol .59
heart disease .82
stroke .89
68% .86
Metabolic syndrome Obesity .73
diabetes .89
high blood pressure .97
high cholesterol .58
65% .85
Note. Factors appear in capital letters; individual BRFSS variables appear in
lower case.
1
State residents (%) participating in light activity or exercise (e.g.,
calisthenics) during the preceding month.
2
State residents (%) participating in moderate activity or exercise (e.g., brisk
walking, bicycling, or anything else that causes a small increase in breathing
and heart rate) on at least 5 days per week for at least 30 min each day.
3
State residents (%) participating in vigorous activity or exercise (e.g.,
running, aerobics, or anything else that causes a large increase in breathing
and heart rate) on 3 or more days per week for at least 20 min each day.
113B.J. Pesta et al. / Intelligence 40 (2012) 107–114
Author's personal copy
Reeve, C. L., & Basalik, D. (2010). Average state IQ, state wealth and racial
composition as predictors of state health statistics: Partial support for
‘g’as a fundamental cause of health disparities. Intelligence,38, 282–289.
Rentfrow, P. (2010). Statewide differences in personality: Toward a psycho-
logical geography of the United States. American Psychologist,65,
548–558.
Rentfrow, P., Gosling, S., & Potter, J. (2008). A theory of the emergence, per-
sistence, and expression of geographic variation in psychological charac-
teristics. Perspectives on Psychological Science,3, 339–369.
Rentfrow, P., Jost, J., Gosling, S., & Potter, J. (2009). Statewide differences in
personality predict voting patterns in 1996–2004 U.S. presidential elec-
tions. In J. T. Jost, A. C. Kay, & H. Thorisdottir (Eds.), Social and psycholog-
ical bases of ideology and system justification (pp. 314–347). Oxford
University Press.
Robinson, W. S. (1950). Ecological correlations and the behavior of individ-
uals. American Sociological Review,3, 351–357.
Schmidt, F., & Hunter, J. (1998). The validity and utility of selection methods
in personnel psychology: Practical and theoretical implications of
85 years of research findings. Psychological Bulletin,124, 262–275.
Shipley, B. A., Weiss, A., Der, G., Taylor, M. D., & Deary, I. J. (2007). Neuroticism,
extraversion, and mortality in the UK health and lifestyle survey: A 21-
year prospective cohort study. Psychosomatic Medicine,69,923–931.
Strenze, T. (2007). Intelligence and socioeconomic success: A meta-analytic
review of longitudinal research. Intelligence,35, 401–426.
Wiebe, D. J., Drew, L. M., & Croom, A. (2010). Personality, health and illness.
In D. French, K. Vedhara, A. Kaptein, & J. Weinman (Eds.), Psychology
(pp. 294–302). New York: Wiley & Sons.
114 B.J. Pesta et al. / Intelligence 40 (2012) 107–114
