Conference PaperPDF Available

County-level USA: No Robust Relationship between Geoclimatic Variables and Cognitive Ability

Authors:
  • Ulster Institute for Social Research
  • Ulster Institute for Social Research

Abstract and Figures

Using a sample of ~3,100 U.S. counties, we tested geoclimatic explanations for why cognitive ability varies across geography. These models posit that geoclimatic factors will strongly predict cognitive ability across geography, even when a variety of common controls appear in the regression equations. Our results generally do not support UV radiation (UVR) based or other geoclimatic models. Specifically, although UVR alone predicted cognitive ability at the U.S. county-level (β = -.33), its validity was markedly reduced in the presence of climatic and demographic covariates (β = -.16), and was reduced even further with a spatial lag (β = -.10). For climate models, average temperature remained a significant predictor in the regression equation containing a spatial lag (β = .35). However, the effect was in the wrong direction relative to typical cold weather hypotheses. Moreover, when we ran the analyses separately by race/ethnicity, no consistent pattern appeared in the models containing the spatial lag. Analyses of gap sizes across counties were also generally inconsistent with predictions from the UVR model. Instead, results seemed to provide support for compositional models.
Content may be subject to copyright.
48
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/jgr.v4i1.2765
Journal of Geographical Research
https://ojs.bilpublishing.com/index.php/jgr
ARTICLE
County-level USA: No Robust Relationship between Geoclimatic
Variables and Cognitive Ability
Bryan J. Pesta1 John G. R. Fuerst1* Emil Kirkegaard2
1. Cleveland State University, United States
2. Ulster Institute for Social Research, Denmark
ARTICLE INFO ABSTRACT
Article history
Received: 2 January 2021
Accepted: 21 January 2021
Published Online: 31 January 2021
Using a sample of ~3,100 U.S. counties, we tested geoclimatic explanations
for why cognitive ability varies across geography. These models posit that
geoclimatic factors will strongly predict cognitive ability across geography,
even when a variety of common controls appear in the regression equations.
Our results generally do not support UV radiation (UVR) based or other
geoclimatic models. Specically, although UVR alone predicted cognitive
ability at the U.S. county-level (β = -.33), its validity was markedly reduced
in the presence of climatic and demographic covariates (β = -.16), and
was reduced even further with a spatial lag (β = -.10). For climate models,
average temperature remained a significant predictor in the regression
equation containing a spatial lag (β = .35). However, the effect was in the
wrong direction relative to typical cold weather hypotheses. Moreover,
when we ran the analyses separately by race/ethnicity, no consistent pattern
appeared in the models containing the spatial lag. Analyses of gap sizes
across counties were also generally inconsistent with predictions from the
UVR model. Instead, results seemed to provide support for compositional
models.
Keywords:
Cognitive ability
Ultraviolet radiation
Climate
Geography
1. Introduction
It is well-established that cognitive ability varies across
geopolitical divisions such as nations, states, and counties
(e.g., nations: [1-2]; Vietnamese provinces: [3]; U.S. states: [4];
U.S. counties: [5]; Argentinian provinces: [6]). These cog-
nitive ability differences have frequently been quite large.
Using the fifty U.S. states as an example, the difference
between the lowest (Mississippi) and highest (Massa-
chusetts) scoring state was found to be 10.1 IQ-metric
points (henceforth just IQ points) [4]. Moreover, these ag-
gregate-level cognitive ability differences have often cor-
related strongly with other important outcomes including
income [7] and education levels [8], health and wellness [9],
and rates of various crimes [10].
Although aggregate cognitive scores are potent pre-
dictors of important social, economic, and political out-
comes [11], consensus about why these relationships exist
and for why cognitive ability varies across geography has
been lacking. Notably, a recently conducted survey of
researchers in this area revealed belief in several potential
causes for aggregate cognitive variation including differ-
*Corresponding Author:
John G. R. Fuerst,
Cleveland State University, United States;
Email: j.g.fuerst@vikes.csuohio.edu
49
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
ences in education (both quantity and quality), genetics,
health, and wealth [12]. Of particular interest for the present
study, the surveyed experts generally considered current
climate and geography to have relatively small causal ef-
fects (only 1 to 3% of the total); nearly all of the experts
seemed to dismiss the contemporary effects of climate as
a major contributing variable for geospatial differences in
cognitive ability.
Despite this, cognitive ability and other behavioral
traits show a latitudinal gradient. This led Van der Vliert
and Van Lange [13] to propose “latitudinal psychology”;
as they note, there are “north-south gradients in cognitive
ability, creativity, ingroup-outgroup dynamics, aggressive-
ness, life satisfaction, and individualism versus collectiv-
ism” (p. 43) which need to be accounted for. Indeed, in
their review, Lynn et al. [11] reported that 12 of 15 countries
exhibited a positive association between absolute latitude
and cognitive ability. These intra-country cognitive clines
in latitude mirror an international one [14]. Related latitudi-
nal behavioral clines can be found among geographically
dispersed non-human animals [13].
Geo-climatic models are in line with the traditional
view that latitude-related differences in human behavior
are in part caused by the direct effects of ecological and
geoclimatic factors [15]. In line with this paradigm, geo-
climatic variables have been offered to explain differenc-
es in cognitive outcomes (e.g., cognitive ability, future
orientation, innovation, state intelligence, educational
attainment). Specific latitude-associated causal factors
include cold weather ([16-17]; see, relatedly, [18,19,13,20]), lati-
tude-dependent infectious disease ([21-23]), and ultraviolet
radiation (UVR) ([21,24-27]; also [13]). These variables may
share overlapping causal pathways. Unlike typical socio-
cultural factors (e.g., socioeconomic status, family values,
quality of school curriculum), they are proposed primarily
to account for regional variation, and less so for individual
variation within regions since the effects of the proposed
causal factors are geographically stratied.
The most extensively developed model with respect to
cognitive ability specically is that of Federico León and
colleagues. They have argued that geographic differences
in UVR have important effects on cognitive ability which
are unmediated by genetics. Unlike other UVR models
where UVR indexes behaviorally-relevant evolutionary
pressures (e.g., [28-29]), this is an environmental model.
León et al. have proposed three complementary pathways
through which UVR might act on aggregate cognitive
ability ([21,24,25-27]). The first pathway involves high UVR
exposure exerting an amplifying effect on sex hormone
production and fertility which then reduces parental in-
vestment in offspring cognitive capital accumulation. The
second pathway supposes that UVR exposure increases
oxidative stress which is purported to be related to both
cognitive impairment and fatigue. The final pathway
invokes a supposed immunosuppressive effect of UVR
which is claimed to increase disease susceptibility. The
implication is that in high UVR regions it may be nec-
essary to divert energy from brain development to the
immune system, although it’s also possible that the direct
effect of developmental insults from disease via increased
exposure and vulnerability could be explanatory. Based
on a literature review, Meisenberg [30] determined that the
UVR model was plausible. However, it is notable that,
contrary to this model, low Vitamin D, rather than high is
associated with cognitive problems [31].
León and colleagues have tested their model in
cross-sectional designs using regression and/or path mod-
els globally [32], across Europe [33], in the U.S. ([21,27,34,35]),
Brazil [35], Italy [35], and Peru [36]. To date, their analyses
have indicated UVR has predictive validity for cognitive
ability and socioeconomic outcomes even in the presence
of several plausible confounders such as ethnicity, abso-
lute latitude, and temperature.
This geoclimatic research programme has several no-
table shortcomings. First, all analyses thus far have been
conducted at the national or subnational level, not the in-
dividual level. It has not been shown that increased UVR
exposure is associated with decreased cognitive ability
for individuals. Second, the regional and national sample
sizes have typically been small (though not always; see,
e.g., [36]). For example, in the ve U.S. studies examining
the UVR-cognitive ability relationship at the state level,
the Ns ranged from 48 to 50. For Italy, Brazil, Europe,
and globally, sample sizes were 19, 26, 32, and 194, re-
spectively. However, multivariate statistics were used to
analyze the data. This could result in imprecise parameter
estimates when the variables are strongly intercorrelated,
as they usually are with highly aggregated data [10]. Third,
spatial autocorrelation (SAC) issues are abundant in na-
tional and subnational geographic data ([37-39]) but León
and colleagues have not taken these into account (except-
ing one case; [34]). Unmodeled SAC has the potential to
bias results due to unmeasured spatially autocorrelated
confounders. SAC diminishes the precision of studies
since OLS standard errors assume independent data points
whereas SAC induces dependencies among them such that
errors can be correlated when autocorrelated causes are
unmodeled (assuming the causes themselves are autocor-
related). Finally, some research has found that results may
be discordant across levels of analysis [40]. For example,
U.S. state-level results may not match U.S. county-level
results [41]. For this reason, in their review of regional dif-
DOI: https://doi.org/10.30564/jgr.v4i1.2765
50
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
ferences in intelligence, Lynn et al. [11] urged authors to
examine data at multiple levels in order to ensure robust-
ness.
The main goal of the present study was to alleviate the
shortcomings described above, in part by analyzing data at
the U.S. county-level. There are many more U.S. counties
than there are U.S. states, which allowed us to conduct
multivariate analyses avoiding sample size concerns. Ad-
ditionally, we were able to compare state- and county-lev-
el results, and we were able to include spatially lagged
variables which allowed us to address the issue of SAC.
An advantage of this dataset was that, owing to replace-
ment migration, geography was less confounded with
evolutionary history [42]. Thus, geoclimatic effects can be
more readily interpreted as representing contemporaneous
effects, as opposed to evolutionary ones (e.g., [43]). Impor-
tantly, however, these sorts of relationships can also result
from processes aligning demography with evolutionarily
familiar or novel environments, or from migratory self-se-
lection ([34,44-45]).
A final goal of the present study was to evaluate the
UVR and other geoclimatic models (i.e., latitude and cold
weather), as advocated by León [21] and others, versus an
ethnic composition model, as suggested by, for example,
McDaniel [4]. McDaniel [4] argued that U.S. state cognitive
differences were in part a result of demography, conjectur-
ing that the regional differences would be stable so long
as the racial demographics (and, also, mean self-identied
race and ethnicity (SIRE) differences) were. Conversely,
León [22] argued that the association between state cogni-
tive ability and racial composition was spurious. That is,
the association was due to the distribution of whites in
states with low levels of UVR. The reason for testing a
racial/ethnic compositional model is that León and Has-
sall [34] clearly specied this as an alternative to their geo-
climatic model for U.S. regional differences. To be clear,
though, racial/ethnic compositional models only attempt
to account for regional differences in terms of demo-
graphics given known racial/ethnic trait differences. They
do not attempt to account for the origins of racial/ethnic
differences, which ultimately could be due to culture,
genetics, or other factors (for expert opinion on cognitive
ability differences see: [12], [46]). The point of these analyses
is to see if previously found associations between geocli-
matic variables and cognitive ability, in the U.S., can be
statistically explained by demographic confounding.
2. Method
The analytic strategy involves running regression
models with geoclimatic factors (average temperature
and UVR) and proxies for these (latitude, longitude, and
elevation) as predictors of county-level cognitive ability.
While geoclimatic effects can be interpreted as represent-
ing contemporaneous effects on cognitive ability, they
could also represent evolutionary effects on ability (e.g.,
[28],43]) because migration and settlement patterns in the U.S.
and other New World countries have not been random
([34]). Moreover, since ancestral populations differential-
ly adapted to geoclimatic effects over evolutionary time
(e.g., pigmentary, thermoregulatory, and disease-related
adaptations in response to UVR, climatic, and parasite
load-related effects [28-29]), contemporaneous geoclimatic
effects may be modied by the racial/ethnic composition
of a population. For example, in the UVR model, UVR is
proposed to act through hormones, including vitamin D;
however, perhaps owing in part to skin tone differences,
there are well-known vitamin D level differences between
U.S. racial/ethnic groups [47]. Thus, we included race/
ethnicity variables as predictors, since these act as crude
measures of genetic ancestry and the related evolutionary
environments ([48-49]); we also run analyses separately by
SIRE group.
Beyond race/ethnicity, we added a spatial lag to cap-
ture effects of both SAC and unobserved variables. In the
supplement, we detailed with simulations how including
a spatial lag confers the added benefit of controlling for
unmodeled variables ([50-51]). We do not include socioeco-
nomic status as an independent variable in our regres-
sion models since these add no analytic leverage when it
comes to evaluating geoclimatic models. This is because
geographic differences in socioeconomic status is another
outcome which geoclimatic models are invoked to explain
(e.g., [13,27]). Also, with the current dataset, it is difcult to
disentangle the causal relation between socioeconomic
status and cognitive ability. As such, we report results
with socioeconomic status as the dependent variable for
the main analysis. This is because socioeconomic status
could be treated as an alternative measure of county-level
functioning. Since our primary concern, following León et
al. is with measured cognitive ability that is our primary
focus.
We used R 3.6.1 for the analyses. All code and data
have been made publicly available in the supplementary
materials.
2.1 Measures
2.1.1 County Cognitive Ability
We used data from the Stanford Education Data Ar-
chive (SEDA v3.0; [52]), which was publicly available at
https://cepa.stanford.edu/seda/overview. This resource
contained cognitive testing data from many sources in-
DOI: https://doi.org/10.30564/jgr.v4i1.2765
51
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
cluding NAEP and state tests which had been normed to
the same scale. The data were available at the U.S. county
level for the years 2009-2015. These scores were based
on low-stakes math and reading/language tests given to
students in grades 3-8. We used the pooled le which had
precalculated scores averaged across subjects (math and
language), year (2009-2015), and grade (3-8) (seda_coun-
ty_pool_CS_v30). A detailed description of the method
used to compute these is provided by Fahle et al. [53].
The SEDA cognitive scores are based on national and
state-level achievement tests. The national tests are The
National Assessment of Educational Progress (NAEP)
exams. These have been found to relate to measures of
intelligence, though they seem to have a greater affinity
for crystallized intelligence measures. Regarding these
measures, Rindermann and Thompson [54] noted: “Both
NAEP scales together measure a mixture of general intel-
ligence and specic knowledge, covered by the construct
cognitive ability…. However, compared to gural scales
as the Ravens, NAEP scales are more measures of crystal-
lized knowledge.” These scores have frequently been used
in the intelligence literature as measures of state-level
cognitive ability (e.g., [4]). Each state additionally admin-
isters state-level assessments (e.g., California Assessment
of Student Performance and Progress, Iowa Test of Basic
Skills, Ohio’s State Test, and Washington Assessment of
Student Learning). These have been evaluated, both quali-
tatively and quantitatively, by the U.S. Department of Ed-
ucation, for the purpose of linking state and national data
([55-59]). Results for the NAEP state assessment mapping
analyses can be accessed at https://nces.ed.gov/nation-
sreportcard/studies/statemapping. The specific methods
used by the SEDA for linking the NAEP and state-level
tests are detailed on the Educational Opportunity Project
website, which can be accessed at https://edopportunity.
org/methods. In their validity report, Reardon, Kalogrides,
and Ho [60] report correlations of > .90 between the linked
district-level scores based on state tests and those based
on the NAEP for those school districts involved in the Tri-
al Urban District Assessment and Measure of Academic
Progress.
To note, we were unable to assess measurement in-
variance for these instruments so we cannot make strong
psychometric claims about the differences. These may
represent general cognitive ability differences or differ-
ences in verbal and math abilities (which are stratum I
abilities in the three-stratum Cattell–Horn–Carroll model)
independent of g. The issue is not immediately relevant
to the hypotheses being investigated and it is unlikely that
there is bias given the consistent lack of bias in other U.S.
samples.
Figure 1 is a map of the distribution of average cog-
nitive ability in our dataset, with zero as the mean for all
counties and each unit increase representing an increase
in one county-level standard deviation (equivalent to 3.6
individual-level IQ points). Consistent with state-level re-
sults, the preponderance of low-scoring counties could be
found in the southeast and southwest. Additionally, there
were low-scoring counties scattered across the Midwest
and west which corresponded to Indian reservations and
other counties with high percentages of native Americans.
DOI: https://doi.org/10.30564/jgr.v4i1.2765
Figure 1. Map of county-level cognitive ability
Note: The scale refers to county-level standardized units, where zero is the mean for all counties.
52
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
2.1.2 County Socioeconomic Status
The SEDA dataset also included several important co-
variates for research use. Among these were precomputed
measures of socioeconomic status based on six indicators,
including: (1) median family income, (2) the proportion of
adults with a bachelor’s degree or higher, (3) the propor-
tion of unemployed adults, (4) the household poverty rate,
(5) the proportion of households receiving SNAP benets,
and (6) the proportion of households with single mothers.
The component loadings and descriptive statistics for the
SES indicators are shown in Table 1 below.
Table 1. Descriptive statistics and component loadings for
the SES indicators.
Variable Loadings Mean SD
Median Family Income 0.904 10.90 0.33
Adults with BA or higher 0.721 0.28 0.14
Unemployed adults -0.921 0.20 0.11
Household poverty rate -0.925 0.12 0.72
Households receiving SNAP -0.778 0.10 0.04
Households with single mothers -0.805 0.20 0.08
Importantly, the SES scores provided here were com-
puted for each race/ethnicity as well as for the overall
population in the same way as the cognitive scores were,
thus allowing for comparison of values within groups.
Figure 2 is a map of the distribution of average SES in our
dataset. As seen, the distribution of county-level SES par-
allels that of cognitive ability.
2.1.3 Demographics
The SEDA covariate files (SEDA v3.0; [52]) provided
self-identied race and ethnicity (SIRE) composition data
for students (e.g., “percent Whites in the grade”). These
proportions are based on the 2006-2010 Common Core of
Data (CCD). The CCD is an annual survey of all public
elementary and secondary schools. These percentages
were somewhat different from the county population
percentages based on the American Community Survey
(ACS). This was because they represented the percent of
students in public schools, not the percentage of adults in
the county. We used the CCD values since the percentage
of students was the more relevant indicator for controlling
the effect of school demographics on student test scores.
2.1.4 Cognitive and SES SIRE Gaps
The SEDA data file provided precomputed cognitive
and composite socioeconomic SIRE standardized differ-
ences for each county. Black/White, Hispanic/White, and
Asian/White d values were available. Standard errors for
the d values were also provided, the inverse of which were
used as analytic weights. Note, SEDA’s SES d values
DOI: https://doi.org/10.30564/jgr.v4i1.2765
Figure 2. Map of county-level socioeconomic status
Note: The scale refers to county-level standardized units, where zero is the mean for all counties.
53
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
were based on SES composite variables.
2.1.5. Ultraviolet Radiation and Climate Data
Our primary geoclimatic variables are UVR, average
temperature, latitude, longitude, and elevation. The Na-
tional Cancer Institute [61] provided county UVR levels
measured in units of Wh/m². These were based on a 30-
year average (1961-1990). Figure 3 is a map of the distri-
bution of average UVR, standardized at the U.S. county
level. Additionally, the Centers for Disease Control and
Prevention [62] provided averaged yearly county-level tem-
perature (yearly mean °C). These were based on data col-
lected from 1979-2011.
León and colleagues have pointed out that absolute
latitude is an imperfect proxy for UVR. While these
variables covary very strongly at the national level (r =
-.89, unweighted, our calculation), there are some sizable
deviations, especially when looking at subnational data.
Case in point, in our dataset, while the correlation be-
tween UVR and latitude was strong (r = -.74, unweighted;
absolute values were unneeded because all values had the
same sign), New Orleans county (which contained the city
of New Orleans) in Louisiana lies at latitude 30.1, and has
a UVR level of 0.54 (i.e., a bit above average), while Salt
Lake county (which contained Salt Lake City) in Utah
lies at latitude 40.9 but has a UVR level of 0.63. There
are several reasons for the discrepancies between latitude
and UVR including cloud coverage, ozone layer thickness
(which partially blocks UVR) and altitude (with higher
UVR levels at higher altitudes because of less atmo-
spheric air for the sun’s rays to pass through). In the U.S.,
these factors varied longitudinally and, as a result, Rocky
Mountain regions tended to have higher UVR levels than
eastern ones at the same latitude [21].
To capture unmeasured geoclimatic factors we includ-
ed latitude, longitude, and elevation (altitude). Latitude
is included since Van der Vliert and Van Lange [13] argued
that latitude gradients explained geographic variability in
behavior and were an important tool for the behavioral
sciences. Longitude was included since León [24] argued
that it was yet another dimension along which UVR acts.
Elevation was included since this was a component in
Cabeza de Baca and Figueredo’s [63] brumal (i.e., cold)
factor, which was constructed based on temperature, lati-
tude, and altitude. The brumal factor played a central role
in Cabeza de Baca and Figueredo’s [63] human cognitive
ecology model, according to which cold weather and
higher altitudes were to be positively associated with cog-
nitive ability. Additionally, according to León and Avilés
[36], higher altitude should be related to cognitive ability,
though negatively so, owing to increased UVR. Finally,
for each county’s latitude and longitude, we coded the
U.S. census internal point. This is approximately the same
as the centroid of the geographical unit, except for cases
DOI: https://doi.org/10.30564/jgr.v4i1.2765
Figure 3. Map of county-level UVR
Note: The scale refers to county-level standardized units, where zero is the mean for all counties
54
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
where the centroid does not lie inside the polygons(s)
of the unit, in which case the closest internal point was
chosen. Van der Vliert and Van Lange [64] additionally pro-
posed steady rain as a “remote climatic predictor”. How-
ever, the type of mediating effects noted (e.g., droughts,
ooding, landslides) are not realistic causes of social and
behavioral differences between U.S. counties so we did
not include them in our analyses. As a robustness test, we
ran the model with additional variables from the Center
for Disease Control’s reported major communicable dis-
eases (tuberculosis, HIV, respiratory infections, hepatitis,
meningitis, and diarrheal diseases) since León et al. some-
times include them in their models. However, as these
variables did not substantially alter the other relations, and
as they suffer from endogeneity problems (being a partial
consequence of cognitive differences), we did not report
these results but nonetheless provided them in the supple-
ment.
2.1.6 Spatial Lag
Hassall and Sherratt [38] raised concerns about con-
founding due to spatial autocorrelation (SAC). Thus, we
calculated a spatial lag term for each county by averaging
the cognitive ability scores for each of the county’s three
closest counties (termed k-nearest spatial neighbor regres-
sion with k = 3). We used the three nearest neighbors as
this was shown in a prior study to produce the most inter-
pretable results [65]. For the SIRE specic regressions, the
lag variable was computed based on the cognitive scores
for the specic SIRE groups.
3. Results
3.1 Descriptive Statistics, Bivariate Correlations
and Main Regression Results
The descriptive statistics for the variables used in the
main regression analyses are reported in Table 2. When
noted, we reported the descriptives for the original vari-
ables, before standardizing them for the regression analy-
ses. This allowed comparison with individual differences
since county and individual cognitive differences were on
the same scale. For example, Figure 1 shows a range of 6
county-level standard deviations; this is equal to a range
of 6 county level SD x .24 (i.e., the SD of CA_all) or 1.44
individual level ones (21.6 IQ points). To note, the avail-
ability of cognitive and SES scores varied by SIRE group.
This is because scores were suppressed if the total number
for a subgroups was less than 95% of the total reported for
all students.
Table 2. Descriptive statistics for variables used in tables
3, 4, 5, and 7
N Mean SD Median Min Max
CA_all 3134 -0.03 0.24 -0.02 -1.20 0.66
CA_Asian 1483 0.34 0.33 0.35 -1.60 1.40
CA_Black 2135 -0.42 0.21 -0.43 -1.20 0.26
CA_Hispanic 2647 -0.25 0.20 -0.25 -0.87 0.57
CA_White 3112 0.10 0.21 0.11 -0.97 0.94
SES_all 3124 -0.08 0.69 -0.03 -3.60 1.86
SES_Black 2108 -1.95 0.86 -2.00 -4.60 1.10
SES_Hispanic 2624 -0.81 0.50 -0.82 -3.60 1.31
SES_White 3099 0.36 0.55 0.37 -2.20 2.42
% White 3124 0.72 0.25 0.81 0.00 1.00
% Black 3124 0.12 0.20 0.02 0.00 1.00
% Hispanic 3124 0.12 0.17 0.05 0.00 1.00
% Asian 3124 0.01 0.03 0.01 0.00 0.59
% Amerindian 3124 0.03 0.10 0.00 0.00 0.99
UVR 3106 4304.05 420.88 4300.00 3000.00 5722.54
Avg temp 3105 17.94 4.92 18.00 3.90 30.61
Latitude 3140 38.45 5.29 38.00 20.00 69.45
Longitude 3140 -92.27 12.90 -90.00 -180.00 -67.61
Elevation 3075 383.34 443.33 240.00 0.00 3096.16
Note:
1The descriptive statistics for the original variables are reported; in the
regression models, these were standardized.
Table 3 shows bivariate correlations between all study
variables. The unweighted correlations are reported below
the diagonal. The correlations weighted by the square
root of county population size are reported above. Mod-
erate relationships existed between climatic variables and
cognitive ability. UVR, by itself, correlated at r = -.33
(weighted) with cognitive ability. These correlations were
in the directions predicted by the respective geoclimatic
models. The correlations for latitude and temperature with
cognitive ability were r = .33 and r = -.42 (weighted),
respectively. All three geoclimatic variables were strong-
ly correlated (r > |.70|). To be clear, these were aggre-
gate-level or ecological correlations, which are usually
inated relative to individual-level ones [40].
To clarify the predictive validity of the variables we
DOI: https://doi.org/10.30564/jgr.v4i1.2765
55
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/jgr.v4i1.2765
Table 3. Correlation matrix (weighted above the diagonal and unweighted below)
CA_All CA_Asian CA_Black CA_Hispanic CA_White SES_All SES_Black
CA_all (3134) 1.00 .45 .67 .60 .76 .73 .43
CA_Asian (1483) .48 1.00 .31 .39 .53 .25 .25
CA_Black (2135) .65 .34 1.00 .54 .47 .46 .50
CA_Hispanic (2647) .61 .39 .59 1.00 .42 .24 .24
CA_White (3112) .69 .59 .40 .35 1.00 .58 .40
SES_all (3124) .75 .33 .43 .25 .61 1.00 .65
SES_Black (2108) .42 .31 .55 .24 .41 .65 1.00
SES_Hispanic (2624) .34 .29 .27 .35 .33 .56 .55
SES_White (3099) .40 .44 .21 .08 .76 .73 .56
UVR (3106) -.37 -.05 -.17 -.16 -.11 -.20 .12
Avg temp (3105) -.39 .07 -.19 -.03 -.13 -.34 .02
Latitude (3140) .31 -.13 .13 -.03 .10 .33 .01
Longitude (3140) .20 .31 .12 .31 .11 -.06 -.15
Elevation (3075) .05 -.18 .11 -.07 -.02 .12 .08
SES_Hispanic SES_White UVR Avg temp Latitude Longitude Elevation
CA_all (3134) .28 .37 -.33 -.42 .33 .08 .13
CA_Asian (1483) .24 .33 -.07 .05 -.13 .33 -.14
CA_Black (2135) .22 .20 -.25 -.32 .27 .06 .14
CA_Hispanic (2647) .28 .10 -.17 -.07 .04 .25 -.09
CA_White (3112) .29 .65 -.10 -.20 .19 -.01 .11
SES_all (3124) .50 .73 -.19 -.41 .39 -.15 .24
SES_Black (2018) .50 .50 -.02 -.12 .15 -.09 .14
SES_Hispanic (2624) 1.00 .46 .05 .01 .00 -.05 .05
SES_White (3099) .50 1.00 .08 -.12 .20 -.21 .20
UVR (3106) .10 .06 1.00 .73 -.73 -.34 .30
Avg temp (3105) .08 -.05 .77 1.00 -.92 .06 -.34
Latitude (3140) -.07 .08 -.75 -.93 1.00 -.29 .23
Longitude (3140) -.09 -.08 -.42 -.10 -.09 1.00 -.56
Elevation (3075) .02 .02 .25 -.27 .18 -.41 1.00
Note: N in parentheses; pairwise deletion
56
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
t several regressions, as shown in Table 4. Results were
weighted by the square root of population size and stan-
dardized betas (β) were used. We ran seven models. Mod-
el 1 contained UVR only, while Model 2 had SIRE only.
Model 3 included both UVR and SIRE. Model 4 added
covariates including temperature, latitude, longitude, and
elevation. Model 5 added the spatial lag variable. Model
6 added a spline for UVR to capture nonlinear effects (re-
stricted cubic using the rcs() in rms package; Harrell [68]).
Finally, Model 7 added interaction terms between UVR
and SIRE since León and Hassal [34] predicted them due to
differences in pigmentation between groups.
Although UVR had a moderate relationship with cog-
nitive ability by itself (Model 1, β = -.33), the relation
shrank by about 50% when demographic and climatic
covariates were added (Model 4, β = -.16). It dropped
further in Model 5 when the spatial lag variable was in-
cluded (Model 5, β = -.10). In this model, temperature
had a moderate effect (β temperature = .35), however, it was in
the wrong direction relative to contemporaneous climatic
model predictions according to which cold climate is hy-
pothesized to be causally associated with higher cognitive
ability. Additionally, latitude, longitude, and elevation had
small to medium positive effects (β = .17 to β = .24).
Allowing for nonlinear effects via a spline of UVR did
not add much to the model (Model 5→6, R2 gain = .001).
Adding interaction terms for UVR and demographics
resulted in a small model improvement (Model 5→7, R2
gain = .019), but also resulted in a positive main effect for
UVR (Model 7, β = .09). This pattern of results suggested
DOI: https://doi.org/10.30564/jgr.v4i1.2765
Table 4. County-level regression results for cognitive ability
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7
β β Β β Β β Β
Intercept .09*** .13*** .10*** .09*** .07*** .00 .08***
UVR -.33*** -.11*** -.16*** -.10*** (nonlinear) .09*
% Black -.56*** -.53*** -.59*** -.52*** -.53*** -.57***
% Asian .14*** .18*** .24*** .21*** .21*** .23***
% Hispanic -.33*** -.29*** -.25*** -.27*** -.26*** -.24***
% Amerindian -.39*** -.35*** -.31*** -.28*** -.29*** -.29***
Avg_temp .39*** .35*** .34*** .19***
Latitude .27*** .17*** .22*** .13**
Longitude .35*** .24*** .27*** .23***
Elevation .30*** .22*** .21*** .15***
CA_lag .28*** .27*** .26***
UVR * % Black .10***
UVR * % Asian -.08***
UVR * % Hispanic -.03**
UVR * % Amerindian -.03
R2 adj. 0.135 0.426 0.455 0.524 0.574 0.575 0.593
N3099 3122 3093 3062 3062 3062 3062
Note:
Weighted by the square root of population size. Values in parentheses are standard errors. * <.01, ** < .005, *** <.001. Model 1: UVR; Model 2:
SIRE groups; Model 3: UVR + SIRE groups; Model 4: Model 3 + average temperature & latitude, longitude, and elevation; Model 5: Model 4 + spa-
tial lag; Model 6: Model 5 + spline of UVR; Model 7: Model 6 + UVR*SIRE interactions.
57
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
that UVR was either not a cause of cognitive ability, its
effects were modied by the included covariates, or it had
heterogeneous and difcult to isolate causal pathways.
The models without the spatial lag predictor showed
some degree of SAC in the residuals, indicating the pres-
ence of unmodeled covariates possibly biasing estimates.
This was removed after the addition of the spatial lag
variable. From the pattern in the model R2 values, it ap-
peared demography was the main source of validity. This
conclusion was conrmed by calculating partial R2 values
for the models and then calculating the proportion of total
R2 attributed to the variables. About half was attributed
to demographics and small amounts to the other variables
(Model 5: SIRE = .57, climate = .079, UVR = .003). Note,
the variance importance metrics for the regression models
were made available in the supplement.
To see if the geospatial variables (latitude, longitude,
and elevation) were leading to underestimation of the
effects of temperature and UVR we ran Model 5 without
them. In Model 5b, the β for UVR was not signicant (β
= -.04); contrariwise, the β for temperature was signifi-
cant, but again in the wrong direction (β = .11). Thus, the
inclusion of the other geospatial variables was not likely
to be the reason for the results we found.
A reader suggested that we use county-level general
socioeconomic status, instead of test scores, as a measure
of county-level “intelligence.” The reader cited a concep-
tion of societal-level “intelligence” by sociologists Talcott
DOI: https://doi.org/10.30564/jgr.v4i1.2765
Table 5. County-level regression results for general socioeconomic status
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7
Β β Β β Β β β
Intercept .11*** .08*** .06*** .07*** .06*** .24 .10***
UVR -.17*** -.03 .15*** .20*** (nonlinear) .41***
% Black -.49*** -.48*** -.45*** -.39*** -.41*** -.44***
% Asian .21*** .26*** .27*** .24*** .23*** .24***
% Hispanic -.15*** -.15*** -.13*** -.15*** -.11*** -.07***
% Amerindian -.25*** -.25*** -.26*** -.24*** -.24*** -.25***
Avg_temp .16** .13 .10 -.01
Latitude .40*** .32*** .41*** .32***
Longitude .21*** .12*** .14*** .13***
Elevation .14*** .07** .05 .01
CA_lag .23*** .22*** .21***
UVR * % Black .05
UVR * % Asian -.05***
UVR * % Hispanic -.09***
UVR * % Amerindian -.04
R2 adj. 0.038 0.376 0.403 0.432 0.470 0.479 0.489
N3093 3122 3093 3062 3062 3062 3062
Note:
Weighted by the square root of population size. Values in parentheses are standard errors. * <.01, ** < .005, *** <.001. Model 1: UVR; Model 2:
SIRE groups; Model 3: UVR + SIRE groups; Model 4: Model 3 + average temperature & latitude, longitude, and elevation; Model 5: Model 4 + spa-
tial lag; Model 6: Model 5 + spline of UVR; Model 7: Model 6 + UVR*SIRE interactions. The dependent, general socioeconomic status, is described
in Section 2.1.2.
58
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
Parsons and Gerald Platt [66] which aligns with this idea.
These results are reported in Table 5. Since León et alia
argue that UVR acts on cognitive ability partially through
socioeconomics (e.g.,[21]), these results are germane to
their models. As seen, using county-level general socio-
economic status instead of cognitive ability did not sub-
stantially change the interpretation regarding the effect of
the climatic or other variables. In Model 5, UVR was sig-
nicant but in the wrong direction, while temperature was
not significant and also in the wrong direction. Latitude
was positively associated with socioeconomic outcomes
just as it was with cognitive ones (Table 4, Model 5).
3.2. County vs. State Results
In order to replicate the results from León [21] and León
and Hassall [34], we aggregated county data to the state lev-
el and then retted all the models. This result was placed
in Table 6. In the initial model, UVR had a stronger effect
on the state level (Model 1, β = -.51) than on the coun-
ty-level (Model 1, β = -.33). In Model 5 with the spatial
lag variable, the magnitude of the effect increased (β =
-.82).
There seemed to be an aggregation effect wherein
higher-level results based on a small dataset (n = 49) gave
markedly different results than those based on a much
larger set (n ~3,100) of lower-level units. This pattern of
results can happen due to zonation effects; these are ef-
fects resulting from how spatial areas are divided [67], or
simply from chance given the small sample size and large
standard errors.
DOI: https://doi.org/10.30564/jgr.v4i1.2765
Table 6. State-level regression results for cognitive ability
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7
Β β Β β β β Β
Intercept .01 .01 .01 -0.2 -.02 -.09 .16
UVR -.51*** -.42 -.80** -.82** (nonlinear) .20
% Black -.65*** -.49** -.70*** -.68*** -.70*** -.77***
% Asian .14 .02 .40** .42** .35 .32
% Hispanic -.44*** -.06 -.18 -.19 -.09 -.06
% Amerindian -.23 -.14 -.13 -.12 -.14 .06
Avg_temp 1.77** 1.83** 1.82** .20
Latitude 1.11 1.07 1.43 .34
Longitude .85*** .84*** .94*** .67**
Elevation .90*** .92*** .80** .22
CA_lag .10 -.02 -.07
UVR* % Black -.08
UVR* % Asian -.26**
UVR* % Hispanic -.11
UVR* %
Amerindian -.37
R2 adj. 0.315 0.376 0.657 0.438 0.65 0.637 0.742
N49 49 49 49 49 49 49
Note:
Weighted by the square root of population size. Values in parentheses are standard errors. * <.01, ** < .005, *** <.001. Alaska and Hawaii excluded,
D.C. included. Model 1: UVR; Model 2: SIRE groups; Model 3: UVR + SIRE groups; Model 4: Model 3 + average temperature & latitude, longitude,
and elevation; Model 5: Model 4 + spatial lag; Model 6: Model 5 + spline of UVR; Model 7: Model 6 + UVR*SIRE interactions.
59
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
One way to test the zonation hypothesis is to examine
pseudo-states (i.e. counterfactual state border maps that
could have existed) and ret the regression model in the
new state-level dataset. We did this using a custom al-
gorithm that began by randomly assigning 48 states one
county each before looping over states at random, assign-
ing them one random neighboring (shared borders) county
if possible (not already assigned). The algorithm nished
when it was no longer possible to assign any more coun-
ties to states (meaning that all were assigned). We created
1,000 pseudo-states in this way. We then t the regression
models of interest (models 1-3 from Table 2) to the data.
Full summary statistics and more details can be found in
the supplement.
DOI: https://doi.org/10.30564/jgr.v4i1.2765
Table 7. County-level regression results for cognitive ability decomposed by White, Black, Hispanic, and Asian Sire
SIRE: White SIRE: Black
Model 1 Model 2 Model 3 Model 1 Model 2 Model 3
Β β β β Β β
Intercept .18*** .31*** .17*** -.20*** -.15*** -.19***
UVR -.02 .30*** .02 -.05 -.03 -.04
Avg_temp .02 -.17 -.04 .05 .03 .06
Latitude .27*** -.03 .08 -.03
Longitude .22*** .01 .24*** .14***
Elevation -.10* -.03 .28*** .17***
CA_lag .62*** .62*** .42*** .40***
R2 adj. 0.416 0.042 0.418 0.202 0.064 0.217
N3079 3049 3049 2124 2095 2095
SIRE: Hispanic SIRE: Asian
Model 1 Model 2 Model 3 Model 1 Model 2 Model 3
Β β β β Β β
Intercept -.14*** -.16*** -.14*** .28*** .37*** .28***
UVR -.12*** -.37*** -.21*** -.15*** -.05 -.09
Avg_temp .16*** .28*** .13 .19*** -.21 -.06
Latitude -.19*** -.17** -.37*** -.23*
Longitude .29*** .12*** .15*** .05
Elevation .28*** .14*** -.17*** -.06
CA_lag .45*** .35*** .43*** .39***
R2 adj. 0.306 0.242 0.333 0.211 0.102 0.224
N2628 2600 2600 1464 1443 1443
Note:
These are the county-level results by SIRE subgroups. Weighted by the square root of population size. Values in parentheses are standard errors. * <.01,
** < .005, *** <.001. Model 1: UVR + average temperature + spatial lag; Model 2: UVR + average temperature + latitude + longitude + elevation;
Model 3: Model 2 + spatial lag.
60
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
Results show that across pseudostates, UVR generally
has the largest effect measured in partial R2 (.12). UVR
was largest in the comparison but 44% of the time demo-
graphics combined had the larger partial R2. The results
indicated that while zoning could inuence results at the
state level, it was unlikely to do so in the present case.
Generally, there seemed to be an aggregation effect such
that geoclimatic predictors had validity at the state but not
county level. The possibility of such paradoxes was the
rationale for Lynn et al.’s [11] call for authors to examine
data at multiple levels as a robustness check.
3.3 Separate SIRE Regression Results
After failing to find a latitudinal cline in cognitive
ability for African and Hispanic Americans, León and
Hassall [34] analyzed cognitive scores for non-Hispanic
Whites separately and reported a signicant effect for this
group. They speculated that African and Hispanic Amer-
icans were protected from the adverse effects of UVR by
darker skin color. Following León and Hassall’s [34] lead,
we ran separate regressions for Whites, Asians, Hispanics,
and Blacks. For these analyses, we used the SIRE-specic
cognitive scores. Since the dependent variable was the
SIRE-specific cognitive score we did not include SIRE
percentages as covariates. Model 1 had UVR, tempera-
ture, and a spatial lag, while Model 3 added geospatial co-
variates. Model 2 was an alternative that repeated Model
3 without the spatial lag. Results were placed in Table 7.
Among Whites in Models 1 and 3 (spatial lag includ-
ed), none of the geoclimatic variables were significant.
Among Blacks, only longitude and elevation were sig-
nificant (both positive). For Asians and Hispanics, tem-
perature was either in the wrong direction (Model 1) or
not significant (Model 3). For Hispanics, UVR was a
significant predictor in the correct direction. However,
this negative association was explicitly predicted to not
exist by León and Hassall [34], who noted “the explanatory
strength of UV radiation is shown not only by its ability
to account empirically for [north-south cognitive decline
among Whites] but also by its capacity to explain the
absence of the north-south cognitive decline among non-
White communities.” For Asians, the effect of UVR was
also in the correct direction and significant in Model 1,
though it was not significant in Model 3. Similarly, lat-
itude was not consistently in the predicted direction for
any group. Overall, our results suggested either no notable
role for the geoclimatic variables and UVR on cognitive
ability, or, perhaps, very complex, heterogeneous causal
paths of an unpredicted nature in the case of temperature,
and overall insignicant effects in the case of UVR.
3.4 Gap Analysis
León [21] argued that the association between racial
composition and cognitive ability was due to the higher
distribution of Whites in states with low UVR. To evalu-
ate this conjecture, we computed the average within-coun-
ty Black/White, Hispanic/White, and Asian/White gaps,
and then compared them to the national cognitive gaps
in NAEP for the same years. To do this, county d values
were weighted by the inverse of the standard error of the
achievement gaps and then averaged. The SEDA did not
include national averages, so we computed these ourselves
using the NAEP explorer. Specifically, we computed ef-
fect sizes for each grade for the years 2009 to 2017. These
years corresponded to those used in the SEDA database
when mapping county achievement scores to state ones
using NAEP math and reading results.
When computing effect sizes for the national differenc-
es, we used the White standard deviation, since this was
the largest group and since sample sizes were not report-
ed. In total there were 13 effect sizes (5 for Grade 4, 5 for
Grade 8, and 3 for Grade 12) for each subject (math and
reading). Following the method used in SEDA, we aver-
aged across grades and years within subjects and then av-
eraged across them. The average d values within counties
and on the national level were given in Table 8 alongside
the percentage of the national differences within counties.
As seen, 64-73% of the gaps were within counties, sup-
porting McDaniel’s [4] proposition.
Table 8. Average within-county gaps versus national-level
gaps
Within-county dNational d% (Within-county/national)
B/W 0.60 0.87 69%
H/W 0.44 0.69 64%
A/W -0.16 -0.22 73%
Note:
The effects sizes are Cohen’s ds. The within county ds are weighted by
the inverse S.E. of the county-level ds.
Another way to approach the issue is to examine the predictors of the
SIRE gaps.
Since ancestry groups are differentially adapted to
climate ([28-29], [43]), if contemporaneous climatic factors
affect cognitive ability and socioeconomic status they
should have a differential effect across SIRE groups. For
example, León and Hassall [34] conjectured that the greater
melanin levels of Blacks and Hispanics “by absorbing
DOI: https://doi.org/10.30564/jgr.v4i1.2765
61
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
and dissipating light, prevent the occurrence of radiation’s
cognitive effects among these populations at U.S. lati-
tudes.” If White but not Black and Hispanic Americans
are affected by UVR, the magnitude of the Black/White
and Hispanic/White differences should be smaller at
higher UVR levels. Table 9 shows the correlation matrix
for UVR, SIRE cognitive differences, SES differences,
and county average cognitive ability and socioeconomic
status. As seen in Table 9, there is no nontrivial negative
association between higher UVR and the Black/White or
Hispanic/White cognitive or SES gaps. The county-level
cognitive gaps, instead, were better predicted by SIRE-spe-
cic socioeconomic status gaps and overall county-level
socioeconomic status. Other geographic variables (average
temperature, latitude) likewise showed trivial correlations
with SIRE gap sizes.
Finally, we directly assessed whether SIRE composi-
tion was contributing to the differences between coun-
ties using the method detailed by Fuerst and Kirkeg-
aard [69]. This method involved correlating the percentage
of students of a SIRE group and difference scores across
counties. These difference scores were the differences
between the actual county average scores and what the
county scores would have been in the absence of a specif-
ic SIRE group. Since the overall county scores were the
weighted sum of the SIRE scores, it was readily deter-
minable if a higher proportion of one group was leading
to higher or lower cognitive ability scores, so long as one
had both SIRE percentages and scores by SIRE groups.
Since this method relied on within-county differences it
was not confounded by unmeasured factors which varied
between counties. The Pearson correlations for counties
were: rAsian % = .63 (N = 1,473), rWhite % = .25 (N = 3,102),
rHispanic % = -.87 (N = 2,637), and rBlack % = -.94 (N = 2,125).
For school districts, which are nested within counties, the
correlations were: rAsian % = .87 (N = 4,683), rWhite % = .37
(N = 12,762), rHispanic % = -.71 (N = 8,832), and rBlack % = -.87
(N = 6,197). Here, a positive correlation indicated that the
DOI: https://doi.org/10.30564/jgr.v4i1.2765
Table 9. Correlation matrix of group gaps and other variables
CA CA
Black
CA
Hisp
CA
White
CA
d bw
CA
d hw
SES
all
SES
Black
SES
Hisp
SES
White
SES
d bw
SES
d hw UV
CA 1.00 .65 .61 .69 .13 .15 .75 .42 .34 .40 -.24 -.04 -.37
CA Black .67 1.00 .59 .40 -.46 -.09 .43 .55 .27 .21 -.53 -.11 -.17
CA Hisp .60 .54 1.00 .35 -.16 -.48 .25 .24 .35 .08 -.25 -.32 -.16
CA White .76 .47 .42 1.00 .63 .65 .60 .41 .33 .76 .00 .28 -.10
CA d bw .17 -.46 -.09 .57 1.00 .71 .24 -.07 .08 .57 .46 .37 .02
CA d hw .19 -.02 -.50 .57 .61 1.00 .36 .19 .02 .65 .20 .52 .01
SES .73 .46 .24 .58 .22 .35 1.00 .65 .56 .73 -.33 -.06 -.20
SES Black .43 .50 .24 .40 -.06 .18 .65 1.00 .55 .56 -.84 -.15 .12
SES Hisp .28 .22 .28 .28 .09 .02 .50 .50 1.00 .50 -.35 -.69 .10
SES White .37 .20 .10 .65 .49 .54 .73 .49 .46 1.00 -.06 .22 .06
SES d bw -.27 -.45 -.22 -.07 .35 .13 -.36 -.87 -.31 -.05 1.00 .35 -.15
SES d hw -.02 -.08 -.22 .20 .27 .39 -.06 -.16 -.72 .16 .31 1.00 -.11
UVR -.33 -.25 -.17 -.10 .05 .01 -.19 -.02 .05 .08 -.02 -.05 1.00
Note:
CA = overall county cognitive ability, CA Black = Black county-level cognitive ability, CA hisp = Hispanic county-level cognitive ability, CA White
= White county-level cognitive ability, SES = overall county SES, SES Black = Black county-level socioeconomic status, SES Hisp = Hispanic coun-
ty-level socioeconomic status, SES White = White county-level socioeconomic status, CA d bw = county Black/White cognitive gap, CA d hw = county
Hispanic/White cognitive gap, SES d bw = county Black/White SES gap, SES d hw = county Hispanic/White SES gap. Correlations above the diago-
nal are weighted by the square root of population size. N = 1981 to 3132 (Ns in notebook).
62
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
SIRE group’s presence was raising the county or school
district scores relative to what it would have been without
that group. Thus, McDaniel’s [4] conjecture was consistent-
ly supported.
4. Discussion
We analyzed a large dataset of U.S. counties to test
whether UVR levels and other geoclimatic variables could
account for geographic variation in cognitive ability. We
found that although UVR, temperature, and latitude cor-
related with cognitive ability, these relationships were
generally neither robust nor consistent. In contrast, varia-
tion in cognitive ability across U.S. counties were strongly
and robustly related to variation in the demographic com-
position of the counties. While it has been found that low
Vitamin D levels are associated with cognitive deficien-
cies on the individual level [31] there appears to be no such
association on the regional level. Indeed, if UVR can be
taken as an index of Vitamin D levels, then these results
would suggest a slightly, though inconsistently so, nega-
tive association between Vitamin D levels and cognitive
ability.
Results from analysis of county-level data conflicted
with results from the state level reported in the literature.
This suggests an aggregation effect or modiable unit area
problem (MUAP; [70]). We found that when we simulated
random pseudo-states roughly similar to the actual ones to
test for a MUAP this level discrepancy was often replicat-
ed (i.e., a variable which was unimportant when analyzed
at the county level turned out to be important when ana-
lyzed at state level and vice versa).
We found that all variables showed substantial SAC.
Some of this was also seen in model residuals. SAC in the
residuals suggested either causal variables that themselves
are spatially autocorrelated were omitted from the models
or that the variables were measured with considerable
error. As expected, the addition of a spatial lag variable
removed the evidence for SAC in the residuals.
SAC in residuals is regarded as a problem because
it can result in spurious associations and it can lead to
overestimated precision of model estimates because the
data points are not fully independent. Thus, in line with
previous studies ([37-38]), we recommend that researchers
employ spatial statistics in their regressions when using
aggregated data. The supplement includes spatial lag
variables computed for this study (for counties and states)
which can be used by others.
Globally and within nations there is substantial and
persistent geographic variation in cognitive ability (Lynn
et al. [11]. While intelligence researchers generally attribute
little variance (1-3%) to current climate and geography [12],
geoclimatic models of human behavioral variation have
resurged in interest (for a brief history of these models,
see [15]). For example, 80 authors replied to Van Lange
et al.’s [19] target article on the Climate, Aggression, and
Self-control in Humans model. Moreover, there has been
increased interest in light of possible effects of climate
change (e.g., [16], [18]). Despite this, most research with cog-
nitive ability as the criterion has used global or national
samples where evolutionary history and geography are
strongly confounded. Moreover, these analyses, by fo-
cusing on nations or states as units, limit analytic sample
sizes and the ability to discriminate between predictors.
We addressed these issues by examining U.S. county
differences (N = 3100), which allowed for multiple means
of controlling for demographic confounding. The results
did not provide consistent evidence for any geoclimatic
model. It would be worthwhile repeating this analysis for
other countries for which post-1500 migration waves may
have attenuated associations between evolutionary history
and geography (e.g., Australia, Canada, and Brazil). That
said, geoclimatic variables could still be useful scientic
tools for understanding geographic variation in cogni-
tive ability. If they are not proxies for contemporaneous
environmental factors, they could have had evolutionary
impacts ([28], [43]) which might be evident in countries with
mostly indigenous populations.
5. Conclusion
The present study is limited by several factors. First,
while the sample size is large, the models used here are
only cross-sectional. Although cognitive ability and de-
mographic data have existed at the county level for multi-
ple years (2009-2016), UVR levels do not change quickly
and thus frustrate the use of a xed effects (panel) design.
Second, we reported data from only a single country. It is
possible that relative wealth or some other characteristic
of the U.S. obscure the putative geographical effects of
the variables examined here. Further studies will need to
be done for other countries to support or disconrm this
suggestion. Third, we did not have access to individual
level data. It is possible that our county-level results are
different from results discovered at the individual level
and one cannot draw a definite conclusion that they are
or are not (i.e., the ecological fallacy; [40]) with the present
data or results.
An additional potential limitation is range restriction in
the geoclimatic variables. The range of temperature and
UVR in the U.S. is less than the level of variation across
the globe. That said, the geoclimatic range in the U.S. is
greater than in most other countries, including those for
which associations have been reported. As such, if the
DOI: https://doi.org/10.30564/jgr.v4i1.2765
63
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
range is too restricted for the U.S. then the variables may
be of limited general use when it comes to intra-country
associations. Moreover, the effects found were in direc-
tions inconsistent with typical hypotheses or were not
consistently signicant (e.g., Table 7), evincing no clear
pattern to the geoclimatic results. For this reason we did
not attempt to correct effects for range restriction relative
to global UVR variance. Finally, it should be reiterated
that the analyses conducted here were correlational. That
said, as noted in the introduction, geoclimatic research is
generally limited to correlational designs. Since previous
research, showing an apparently robust relationship be-
tween geoclimatic variables and regional outcomes, has
also been correlational, our conclusion-that there is no
robust association-is relatively uncompromised.
In sum, large, geographically distributed differences
in cognitive ability exist. These differences need to be
accounted for. Several models have been proposed which
have attempted to explain these differences in terms of
contemporaneous geoclimatic ones. However, our present
results agree with the majority opinion of intelligence re-
searchers, that contemporaneous geoclimatic factors are
not major determinants of variation in cognitive ability
[12], at least for regions with geoclimatic variation similar
to that in the U.S. Nonetheless, examining data in regions
other than the U.S. would help to better evaluate these
issues, as it may be that warmer climates or more intense
UVR are needed to trigger the proposed physiological
mechanisms through which these variables might affect
cognitive ability.
References
[1] Lynn, R.,Vanhanen, T. IQ and Global Inequality.
Augusta, GA, US: Washington Summit Publishers,
2006.
[2] Lynn, R., Becker, D. The Intelligence of Nations.
London, GB: Ulster Institute for Social Research,
2019.
[3] Kirkegaard, E. O. W., Pesta, B. J. An S-factor analy-
sis on the provinces of Vietnam: Relationships with
cognitive ability, ethnicity, and latitude. Mankind
Quarterly, 2018, 58(4):562-579.
DOI:10.46469/mq.2018.58.4.2
[4] McDaniel, M. A. Estimating state IQ: Measurement
challenges and preliminary correlates. Intelligence,
2006, 34(6):607-619.
DOI:10.1016/j.intell.2006.08.007
[5] Beaver, K., Wright, J. The association between coun-
ty-level IQ and county-level crime rate. Intelligence,
2011, 39(1):22-26.
DOI:10.1016/j.intell.2010.12.002
[6] Kirkegaard, E. O. W., Fuerst, J. Admixture in Argen-
tina. Mankind Quarterly, 2017, 57(4):542.
DOI:10.46469/mq.2017.57.4.4
[7] Templer, D., Rushton, J. P. IQ, skin color, crime,
HIV/AIDS, and income in 50 U.S. States. Intelli-
gence, 2011, 39, 437-442.
DOI:10.1016/j.intell.2011.08.001
[8] Lynn, R. In Italy, north-south differences in IQ pre-
dict differences in income, education, infant mortal-
ity, stature, and literacy. Intelligence, 2010, 38, 93-
100.
DOI:10.1016/j.intell.2009.07.004
[9] Reeve, C. Expanding the g nexus: Further evidence
regarding the relationship among national IQ, reli-
giosity, and national health outcomes. Intelligence,
2009, 37, 495-505.
DOI:10.1016/j.intell.2009.06.003
[10] Pesta, B. J., McDaniel, M., Bertsch, S. Toward an in-
dex of well-being for the 50 U.S. states. Intelligence,
2010, 38, 160-168.
DOI:10.1016/j.intell.2009.09.006
[11] Lynn, R., Fuerst, J., Kirkegaard, E. O. W. Regional
differences in intelligence in 22 countries and their
economic, social and demographic correlates: A re-
view. Intelligence, 2018, 69, 24-36.
DOI:10.1016/j.intell.2018.04.004
[12] Rindermann, H., Becker, D., Coyle, T. R. Survey of
Expert Opinion on Intelligence: Causes of Interna-
tional Differences in Cognitive Ability Tests. Fron-
tiers in Psychology, 2016, 7.
DOI: 10.3389/fpsyg.2016.00399
[13] Van de Vliert, E., Van Lange, P. A. M. Latitudinal
gradients as scientic tools for psychologists, 2020.
Current Opinion in Psychology, 32, 43-46.
DOI:10.1016/j.copsyc.2019.06.018
[14] Rindermann, H. Cognitive Capitalism: Human Capi-
tal and the Wellbeing of Nations. Cambridge Univer-
sity Press, 2018.
[15] Frost, P. The Original Industrial Revolution. Did
Cold Winters Select for Cognitive Ability?. Psych,
2019, 1(1):166-181.
DOI:10.3390/psych1010012
[16] Sternberg, R. J. (2017). An alternative interpretation
of climate data: Intelligence. Behavioral and Brain
Sciences, 40.
DOI:10.1017/S0140525X16001102
[17] Park, J. Hot Temperature, Human Capital and Adap-
tation to Climate Change. Unpublished Manuscript,
Harvard University Economics Department, 2017.
[18] Randell, H., Gray, C. Climate change and education-
al attainment in the global tropics. Proceedings of the
National Academy of Sciences, 2019, 116(18):8840-
DOI: https://doi.org/10.30564/jgr.v4i1.2765
64
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
8845.
DOI:10.1073/pnas.1817480116
[19] Van Lange, P. A. M., Rinderu, M. I., Bushman, B. J.
The logic of climate and culture: Future avenues for
CLASH. Behavioral and Brain Sciences, 2017, 40,
42-49.
DOI:10.1177/1368430217735579
[20] Van de Vliert, E., Kong, D. T. Cold, heat, wealth, and
culture. In D. Matsumoto & H. C. Hwang (Eds.), Ox-
ford Handbook of Culture and Psychology (2nd ed.).
New York: Oxford University Press, 2018.
[21] León, F. R. Race vis-à-vis latitude: Their influence
on intelligence, infectious diseases, and income.
Mankind Quarterly, 2016, 56(3):411.
DOI:10.46469/mq.2016.56.3.7
[22] Eppig, C., Fincher, C. L., Thornhill, R. Parasite prev-
alence and the distribution of intelligence among the
states of the USA. Intelligence, 2011, 39(2-3):155-
160. doi: 10.1016/j.intell.2011.02.008
[23] Varnum, M. E. Sources of regional variation in social
capital in the United States: Frontiers and pathogens.
Evolutionary Behavioral Sciences, 2014, 8(2):77.
doi: 10.1037/h0098950
[24] León, F. R. The east-to-west decay of math and read-
ing scores in the United States: A prediction from
UVB radiation theory. Personality and Individual
Differences, 2015, 86, 287-290.
DOI:10.1016/j.paid.2015.06.028
[25] León, F. R., Burga-León, A. Why complex cognitive
ability increases with absolute latitude. Intelligence,
2014, 46, 291-299.
DOI:doi.org/10.1016/j.intell.2014.07.011
[26] León, F. R., Burga-León, A. How geography influ-
ences complex cognitive ability. Intelligence, 2015,
50, 221-227.
DOI:10.1016/j.intell.2015.04.011
[27] León, F. R., Burga-León, A. UV Radiation Associ-
ates With State Income Through Complex Cognitive
Ability in the USA. Journal of Individual Differenc-
es, 2018, 39, 18-26.
DOI:10.1027/1614-0001/a000245
[28] Fedderke, J.W., Klitgaard, R.E., MacMurray, J.P.,
Napolioni, V. Diagnosing deep roots of development:
Genetic, disease and environmental factors. Econom-
ic Research Southern Africa, 2014. Retrieved from
http://www.econrsa.org/publications/workingpapers/
diagnosing-deep-roots-development-genetic-dis-
ease-and-environmental
[29] Klitgaard, R., Fedderke, J. W., Napolioni, V. Ge-
ography, Climate, and Genes in Development
Studies. The World Bank Economic Review, 2020,
34(Supplement_1), S46-S51.
DOI:10.1093/wber/lhz034
[30] Meisenberg, G. From Skin to Brain: In Search of
Plausible Messengers. Mankind Quarterly, 2018,
58(4).
DOI:10.46469/mq.2018.58.4.10
[31] Sultan, S., Taimuri, U., Basnan, S. A., Ai-Orabi,
W. K., Awadallah, A., Almowald, F., Hazazi, A.
Low Vitamin D and Its Association with Cognitive
Impairment and Dementia. Journal of Aging Re-
search, 2020.
DOI:10.1155/2020/6097820
[32] León, F. R. Diminished UV radiation enhances
national cognitive ability, wealth, and institutions
through health and education. Personality and Indi-
vidual Differences, 2018a, 120, 52-57.
DOI:10.1016/j.paid.2017.08.025
[33] León, F. R. Boundary conditions and new frontiers
for Cognitive UV Radiation Theory. Mankind Quar-
terly, 2018b, 58(4):673-685.
DOI:10.46469/mq.2018.58.4.11
[34] León, F. R., Hassall, C. UV radiation is associated
with latitudinal trends in cognitive ability of White
children in the USA. Journal of Individual Differenc-
es, 2017, 38(3):155-162.
DOI:10.1027/1614-0001/a000232
[35] León, F. R., Antonelli-Ponti, M. UV radiation theory
and the Lynn (2010): Italian debate. Mankind Quar-
terly, 2018, 58(4):621-649.
DOI:10.46469/mq.2018.58.4.7
[36] León, F. R., Avilés, E. How altitude above sea level
affects intelligence. Intelligence, 2016, 58, 33-41.
DOI:10.1016/j.intell.2016.06.001
[37] Gelade, G. A. The geography of IQ. Intelligence,
2008, 36(6):495-501.
DOI:10.1016/j.intell.2008.01.004
[38] Hassall, C., Sherratt, T. N. Statistical inference and
spatial patterns in correlates of IQ. Intelligence, 2011,
39(5):303-310.
DOI:10.1016/j.intell.2011.05.001
[39] Bromham, L., Hua, X., Cardillo, M., Schneemann, H.,
Greenhill, S. J. Parasites and politics: Why cross-cul-
tural studies must control for relatedness, proximity
and covariation. Royal Society Open Science, 2018,
5(8), 181100.
DOI:10.1098/rsos.181100
[40] Robinson, W. S. Ecological correlations and individ-
ual behavior. American Sociological Review, 1950,
15(195):351-57.
DOI: doi.org/10.1093/ije/dyn357
[41] Kirkegaard, E. O. W., Pesta, B. J. Smart and Poor, or
Rich and Dull? A US County-Level Analysis of the
Relationship between IQ and Presidential-Election
DOI: https://doi.org/10.30564/jgr.v4i1.2765
65
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
Voting Behavior. Mankind Quarterly, 2019, 60, 243-
255.
DOI:10.46469/mq.2019.60.2.7
[42] Pesta, B. J., Poznanski, P. J. Only in America: Cold
Winters Theory, race, IQ and well-being. Intelli-
gence, 2014, 46, 271-274.
DOI:10.1016/j.intell.2014.07.009
[43] Luoto, S. An updated theoretical framework for hu-
man sexual selection: From ecology, genetics, and
life history to extended phenotypes. Adaptive Human
Behavior and Physiology, 2019, 5(1):48-102.
DOI: 10.1007/s40750-018-0103-6
[44] Fischer, D. H. Albion’s Seed: Four British Folkways
in America (Reprint edition). New York, NY: Oxford
Univ Press, 1991.
[45] Woodard, C. American Nations: A History of the
Eleven Rival Regional Cultures of North America
(Reprint edition). New York, N.Y. Penguin Books,
2012.
[46] Becker, D., Rindermann, H., Coyle, T. R. 2013 sur-
vey of expert opinion on intelligence. Presented at
the 14th Annual Meeting of the International Society
for Intelligence Research, Melbourne, Australia,
2013.
[47] Parvane, N. R., Tadepalli, S., Singh, P., Qian, A.,
Joshi, R., Kandala, H., ... Cheriyath, P. Prevalence of
vitamin D deficiency and associated risk factors in
the US population (2011-2012). Cureus, 2018, 10(6).
DOI:10.7759/cureus.2741
[48] Tang, H., Quertermous, T., Rodriguez, B., Kardia,
S. L., Zhu, X., Brown, A., ... Schork, N. J. Genetic
structure, self-identied race/ethnicity, and confound-
ing in case-control association studies. The American
Journal of Human Genetics, 2005, 76(2):268-275.
DOI:10.1086/427888
[49] H. Fang, Q. Hui, Lynch, J., Honerlaw, J., Assimes, T.
L., Huang, J., ... DuVall, S. L. Harmonizing genetic
ancestry and self-identified race/ethnicity in ge-
nome-wide association studies. The American Jour-
nal of Human Genetics, 2019, 105(4):763-772.
DOI:10.1016/j.ajhg.2019.08.012
[50] LeSage, J., Pace, R. K. Introduction to Spatial
Econometrics. Chapman and Hall/CRC, 2009.
[51] Corrado, L., Fingleton, B. Where is the economics in
spatial econometrics?. Journal of Regional Science,
2012, 52(2):210-239.
DOI:10.1111/j.1467-9787.2011.00726.x
[52] Reardon, S. F., Ho, A. D., Shear, B. R., Fahle, E. M.,
Kalogrides, D., DiSalvo, R. Stanford Education Data
Archive (Version 2.1), 2018. Accessed at: https://
edopportunity.org/get-the-data/
[53] Fahle, E. M., Shear, B. R., Kalogrides, D., Reardon,
S. F., Chavez, B., Ho, A. D. Stanford Education Data
Archive Technical Documentation Version 3.0, 2019
[54] Rindermann, H., Thompson, J. Ability rise in NAEP
and narrowing ethnic gaps?. Intelligence, 2013,
41(6):821-831.
DOI:10.1016/j.intell.2013.06.016
[55] McLaughlin, D. H., Bandeira de Mello, V., Blanken-
ship, C., Chaney, K., Esra, P., Hikawa, H. Compar-
ison between NAEP and state reading assessment
results: 2003 (NCES 2008-474). Washington, DC:
US Department of Education, Institute of Education
Sciences. National Center for Education Statistics,
2008a.
[56] McLaughlin, D.H., Bandeira de Mello, V., Blan-
kenship, C., Chaney, K., Esra, P., Hikawa, H., ...
Wolman, M. Comparison between NAEP and State
Mathematics Assessment Results: 2003. Volume 1.
Research and Development Report. NCES 2008-475.
National Center for Education Statistics, 2008b.
[57] Linn, R. L., McLaughlin, D., Thissen, D. Utility and
Validity of NAEP Linking Efforts. American Insti-
tutes for Research, 2009.
[58] Stancavage, F. B., Bohrnstedt, G. W. Examining
the Content and Context of the Common Core State
Standards: A First Look at Implications for the Na-
tional Assessment of Educational Progress. American
Institutes for Research, 2013.
[59] Bandeira de Mello, V., Rahman, T., Fox, M.A., Ji,
C.S. Mapping State Proficiency Standards Onto
NAEP Scales: Results From the 2017 NAEP Reading
and Mathematics Assessments (NCES 2019-040).
U.S. Department of Education, Washington, DC:
Institute of Education Sciences, National Center for
Education Statistics, 2019.
[60] Reardon, S. F., Kalogrides, D., Ho, A. D. Validation
methods for aggregate-level test scale linking: A case
study mapping school district test score distributions
to a common scale. Journal of Educational and Be-
havioral Statistics, 2019, 1076998619874089. https://
cepa.stanford.edu/content/validation-methods-ag-
gregate-level-test-scale-linking-case-study-mapping-
school-district-test-score-distributions-common-scale
[61] National Cancer Institute (N.D.). Geographic Infor-
mation Systems and Science for Cancer Control:
County Level UV Exposure Data for the Continental
United States. Retrieved July 01, 2019, from: https://
gis.cancer.gov/tools/uv-exposure/
[62] Centers for Disease Control and Prevention (N.D.).
North America Land Data Assimilation System
(NLDAS) Daily Air Temperatures and Heat Index
(1979-2011) Request. Retrieved July 01, 2019, from:
https://wonder.cdc.gov/nasa-nldas.html
DOI: https://doi.org/10.30564/jgr.v4i1.2765
66
Journal of Geographical Research | Volume 04 | Issue 01 | January 2021
Distributed under creative commons license 4.0
[63] Cabeza de Baca, T., Figueredo, A. J. The cognitive
ecology of Mexico: Climatic and socio-cultural ef-
fects on life history strategies and cognitive abilities.
Intelligence, 2014, 47(1):63-71.
DOI:10.1016/j.intell.2014.08.007
[64] Van de Vliert, E., Van Lange, P. A. Latitudinal psy-
chology: an ecological perspective on creativity,
aggression, happiness, and beyond. Perspectives on
Psychological Science, 2019, 14(5):860-884.
DOI:10.1177/1745691619858067
[65] Kirkegaard, E. O. W. Inequality across US counties:
An S factor analysis. Open Quantitative Sociology &
Political Science, 2016.
DOI:10.26775/OQSPS.2016.05.23
[66] Parsons, T., Platt, G. M. (1975/2013). The American
University. Harvard University Press.
[67] Drobne, S., Lakner, M. The Influence of the Zona-
tion Effect on a System of Hierarchical Functional
Regions. Business Systems Research Journal, 2018,
9(2):45-54. DOI:10.2478/bsrj-2018-0018
[68] Harrell, F. E. rms: Regression Modeling Strategies
(Version 5.1-3.1), 2019. Retrieved from https://
CRAN.R-project.org/package=rms
[69] Fuerst, J., Kirkegaard, E. O. The genealogy of dif-
ferences in the Americas. Mankind Quarterly, 2016,
56(3):425.
DOI:10.46469/mq.2016.56.3.9
[70] Wong, D. W. The modifiable areal unit problem
(MAUP). In WorldMinds: Geographical Perspectives
on 100 Problems, 2004:571-575). Springer, Dor-
drecht.
DOI:10.1007/978-1-4020-2352-1_93
DOI: https://doi.org/10.30564/jgr.v4i1.2765
... In general, the further north the state, the higher its IQ (and well-being). This replicates previous state-level results (Pesta and Poznanski 2014), though it is inconsistent with county-level results (Pesta et al. 2021). The largest correlation in Table 3, however, is between IQ and FICO credit scores (0.87; see also the 0.85 correlation between well-being and FICO scores). ...
Article
Full-text available
At the level of the 50 U.S. states, an interconnected nexus of well-being variables exists. These variables strongly correlate with estimates of state IQ in interesting ways. However, the state IQ estimates are now more than 16 years old, and the state well-being estimates are over 12 years old. Updated state IQ and well-being estimates are therefore needed. Thus, I first created new state IQ estimates by analyzing scores from both the Program for the International Assessment of Adult Competency (for adults), and the National Assessment of Educational Progress (for fourth and eighth grade children) exams. I also created new global well-being scores by analyzing state variables from the following four well-being subdomains: crime, income, health, and education. When validating the nexus, several interesting correlations existed among the variables. For example, state IQ most strongly predicted FICO credit scores, alcohol consumption (directly), income inequality, and state temperature. Interestingly, state IQ derived here also correlated 0.58 with state IQ estimates from over 100 years ago. Global well-being likewise correlated with many old and new variables in the nexus, including a correlation of 0.80 with IQ. In sum, at the level of the U.S. state, a nexus of important, strongly correlated variables exists. These variables comprise well-being, and state IQ is a central node in this network.
Article
Full-text available
Vitamin D is a neurosteroid hormone that regulates neurotransmitters and neurotrophins. It has anti-inflammatory, antioxidant, and neuroprotective properties. It increases neurotrophic factors such as nerve growth factor which further promotes brain health. Moreover, it is also helpful in the prevention of amyloid accumulation and promotes amyloid clearance. Emerging evidence suggests its role in the reduction of Alzheimer’s disease hallmarks such as amyloid-beta and phosphorylated tau. Many preclinical studies have supported the hypothesis that vitamin D leads to attentional, behavioral problems and cognitive impairment. Cross-sectional studies have consistently found that vitamin D levels are significantly low in individuals with Alzheimer’s disease and cognitive impairment compared to healthy adults. Longitudinal studies and meta-analysis have also exhibited an association of low vitamin D with cognitive impairment and Alzheimer’s disease. Despite such evidence, the causal association cannot be sufficiently answered. In contrast to observational studies, findings from interventional studies have produced mixed results on the role of vitamin D supplementation in the prevention and treatment of cognitive impairment and dementia. The biggest issue of the existing RCTs is their small sample size, lack of consensus over the dose, and age of initiation of vitamin D supplements to prevent cognitive impairment. Therefore, there is a need for large double-blind randomized control trials to assess the benefits of vitamin D supplementation in the prevention and treatment of cognitive impairment.
Article
Full-text available
A small research stream exists which focuses on relationships between political ideology (inferred from voting behavior) and the intelligence of geopolitical subdivisions such as the 50 U.S. states. With U.S. state-level data, IQ scores positively predict votes cast for Democrats, but only when controlling for state racial composition. Here, however, we explore the relationship between IQ and voting behavior at the level of U.S. counties (approx. n = 3,100). We find that county-level IQ weakly predicts more votes cast for Republicans (r's .07 to .13) and less votes cast for Democrats (r's-.10 to-.14). These small relationships are also found in multiple regression analyses, even when demographic data, social status, and population density appear as covariates (3-4% points more votes cast for Republicans per standard deviation of IQ). The effect of general social status was opposite that of intelligence (6-9% points less votes cast for Republicans per standard deviation of social status), which is surprising considering the very strong positive correlation (r = .77) between IQ and social status. Additionally, racial homogeneity by itself predicts voting Republican; however, when other variables are present in the regression model, homogeneity predicts voting Democrat. Results indicate that aggregate-level relationships between intelligence and voting outcomes are more complex than previously thought.
Article
Full-text available
Rushton and Jensen argued that cognitive ability differs between human populations. But why are such differences expectable? Their answer: as modern humans spread out of Africa and into northern Eurasia, they entered colder and more seasonal climates that selected for the ability to plan ahead, in order to store food, make clothes, and build shelters for winter. This cold winter theory is supported by research on Paleolithic humans and recent hunter-gatherers. Tools become more diverse and complex as effective temperature decreases, apparently because food has to be obtained during limited periods and over large areas. There is also more storage of food and fuel and greater use of untended traps and snares. Finally, shelters have to be sturdier, and clothing more cold-resistant. The resulting cognitive demands are met primarily by women because the lack of opportunities for food gathering pushes them into more cognitively demanding tasks, like garment making, needlework, weaving, leatherworking, pottery, and kiln operation. The northern tier of Paleolithic Eurasia thus produced the "Original Industrial Revolution"—an explosion of creativity that preadapted its inhabitants for later developments, i.e., farming, more complex technology and social organization, and an increasingly future-oriented culture. Over time, these humans would spread south, replacing earlier populations that could less easily exploit the possibilities of the new cultural environment. As this environment developed further, it selected for further increases in cognitive ability. Indeed, mean intelligence seems to have risen during recorded history at temperate latitudes in Europe and East Asia. There is thus no unified theory for the evolution of human intelligence. A key stage was adaptation to cold winters during the Paleolithic, but much happened later.
Article
Over the coming decade, much more genetic data will enter into the study of economic development. This paper provides an example and emphasizes the uses and misuses of such information. It has assembled for the first time national frequencies of the ACP1 genetic polymorphism and the Interleukin-6 (IL6-174G) and Interleukin-10 (IL10-1082G) cytokines. These three respond over the centuries to ultraviolet radiation and infectious diseases. The study also looks at a national measure of heterozygotic diversity. In particular, it finds that ACP1 frequencies are significantly related to national outcomes ranging from GDP per capita to type and quality of governance, to measures of national “competitiveness,” to health, to fertility, to measures of satisfaction with life. These associations do not seem explainable by reverse causation nor by the influence of some of the usual variables in studies of long-run development. Nonetheless, these results do not mean that a few genes have a direct causal effect on world development. The ACP1*B variable is surely picking up the influences of many genetic and cultural adaptations over evolutionary time in response to ultraviolet exposure and pathogen burdens. This study's findings thus support other research indicating the importance of disease environments in shaping both genetic and sociocultural adaptations that have influence on development outcomes today. The paper concludes with a discussion of what such strong associations mean and do not mean, in hopes of guiding future studies of genes and other deep roots of economic development.
Article
Linking score scales across different tests is considered speculative and fraught, even at the aggregate level. We introduce and illustrate validation methods for aggregate linkages, using the challenge of linking U.S. school district average test scores across states as a motivating example. We show that aggregate linkages can be validated both directly and indirectly under certain conditions such as when the scores for at least some target units (districts) are available on a common test (e.g., the National Assessment of Educational Progress). We introduce precision-adjusted random effects models to estimate linking error, for populations and for subpopulations, for averages and for progress over time. These models allow us to distinguish linking error from sampling variability and illustrate how linking error plays a larger role in aggregates with smaller sample sizes. Assuming that target districts generalize to the full population of districts, we can show that standard errors for district means are generally less than .2 standard deviation units, leading to reliabilities above .7 for roughly 90% of districts. We also show how sources of imprecision and linking error contribute to both within- and between-state district comparisons within versus between states. This approach is applicable whenever the essential counterfactual question—“what would means/variance/progress for the aggregate units be, had students taken the other test?”—can be answered directly for at least some of the units.
Article
Large-scale multi-ethnic cohorts offer unprecedented opportunities to elucidate the genetic factors influencing complex traits related to health and disease among minority populations. At the same time, the genetic diversity in these cohorts presents new challenges for analysis and interpretation. We consider the utility of race and/or ethnicity categories in genome-wide association studies (GWASs) of multi-ethnic cohorts. We demonstrate that race/ethnicity information enhances the ability to understand population-specific genetic architecture. To address the practical issue that self-identified racial/ethnic information may be incomplete, we propose a machine learning algorithm that produces a surrogate variable, termed HARE. We use height as a model trait to demonstrate the utility of HARE and ethnicity-specific GWASs.
Article
Are there systematic trends around the world in levels of creativity, aggressiveness, life satisfaction, individualism, trust, and suicidality? This article suggests a new field, latitudinal psychology, that delineates differences in such culturally shared features along northern and southern rather than eastern and western locations. In addition to geographical, ecological, and other explanations, we offer three metric foundations of latitudinal variations: replicability (latitudinal gradient repeatability across hemispheres), reversibility (north-south gradient reversal near the equator), and gradient strength (degree of replicability and reversibility). We show that aggressiveness decreases whereas creativity, life satisfaction, and individualism increase as one moves closer to either the North or South Pole. We also discuss the replicability, reversibility, and gradient strength of (a) temperatures and rainfall as remote predictors and (b) pathogen prevalence, national wealth, population density, and income inequality as more proximate predictors of latitudinal gradients in human functioning. Preliminary analyses suggest that cultural and psychological diversity often need to be partially understood in terms of latitudinal variations in integrated exposure to climate-induced demands and wealth-based resources. We conclude with broader implications, emphasizing the importance of north-south replications in samples that are not from Western, educated, industrialized, rich, and democratic (WEIRD) societies.
Article
It is an unmistakable fact of life that animals and plants function differently at lower and higher latitudes with distinct temperatures and rainfall. No less unmistakable are the opposite directions of these latitudinal gradients above and below the equator. Therefore, it would be surprising if there were no opposite north-south gradients in human functioning in the northern and southern hemispheres. And indeed, recent publications and projects have started to validate, integrate, and explain such north-south gradients in cognitive ability, creativity, ingroup-outgroup dynamics, aggressiveness, life satisfaction, and individualism versus collectivism. Our brief review of these contemporary trends cumulates into a latitudinal-tools matrix for further integration and sophistication of the latitude-related ecology of habitual mindsets and practices.