Content uploaded by Emil O. W. Kirkegaard
Author content
All content in this area was uploaded by Emil O. W. Kirkegaard on Jun 19, 2016
Content may be subject to copyright.
MANKIND QUARTERLY 2016 56:4 580-606
580
Inequality in the United States: Ethnicity, Racial Admixture
and Environmental Causes
Emil O. W. Kirkegaard
1
University of Aarhus, Denmark
John Fuerst
2
Independent Researcher, USA
Previously, we looked at the association between overall state-level
biogeographic ancestry (BGA) and overall state-level outcomes. It was
found that European BGA relative to African and Amerindian BGA was
associated with better outcomes. In this paper, the analysis is extended
by looking at the state-level ancestry-outcome associations individually for
Black and Hispanic self-identified race-ethnicity (SIRE) groups. General
socioeconomic factor (S) scores were calculated for US states by SIRE
groups based on three indicators. The S factor loadings were generally
stable across subgroup analyses and the factor scores were stable across
factor analytic extraction methods (for the latter, almost all r's ≈ 1). For
Whites, Blacks and Hispanics, there were strong correlations between
cognitive ability scores and S factor scores across states (r= .55 to .78; N
= 28-50). This pattern also held when all data were analyzed together (r=
.86, N = 115). Furthermore, the size of the Hispanic-White and Black-
White S and cognitive ability gaps strongly correlated across states (r=
.62 to .69; N = 36-37). Lastly, parasite prevalence did not plausibly explain
SIRE gaps in cognitive ability because gaps were smaller in more parasite-
rich states (combined analysis r= -.17, N = 91). We found that climatic
and geospatial variables did not correlate strongly with cognitive ability
and S scores when scores were decomposed by SIRE group, but did so
1
Department of Linguistics, University of Aarhus, Denmark
Email: emil@emilkirkegaard.dk
2
Email: j122177@hotmail.com
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
581
at the total state level, even after statistically controlling for SIRE
composition.
Key words: Inequality, General socioeconomic factor, S factor, USA,
States, Cognitive ability, Intelligence, NAEP, Race, SIRE, Biogeographic
ancestry, Ecology
Fuerst & Kirkegaard (2016a) examined the relationship between estimated
biogeographic ancestry (BGA), cognitive ability and socioeconomic outcomes (S)
across sovereign nations of the Americas as well as across first-level
administrative divisions (states, departments, etc.) within 4 nations (USA, Mexico,
Colombia and Brazil). Generally, strong correlations (r= .5 to .8) between
European BGA, cognitive ability and S were found. One glaring exception to this
pattern was the relationship between European ancestry and socioeconomic
outcomes in the United States. For this country, the zero-order correlation
between European ancestry and S was comparatively low at .39; moreover, when
cold weather and latitude were included in regression and path models, the
relation became moderately to strongly negative (Fuerst & Kirkegaard, 2016a,b).
In Fuerst and Kirkegaard (2016a), we noted that further research was needed for
this country. Two commenters, Leon (2016) and Pesta (2016), presented new US
analyses. In line with previous research, Leon (2016) and Pesta (2016) looked at
total state outcomes, a practice which is problematic given interstate differences
in ethnic composition. In this paper, we attempt to gain analytic leverage by
decomposing state-level outcomes by major self-identified race/ethnicity (SIRE)
groups.
3
3
The term race is a polyseme. In this paper, we are interested in two meanings: one,
self-identified race/ethnicity (SIRE) and, two, biogeographic ancestry (BGA). While
SIRE is a social identity, BGA is a description of relative genetic relatedness. Owing to
centuries of admixture along with social and political whims, the two meanings often do
not or do not neatly correspond. For instance, many members of the SIRE group
Blacks/African Americans have substantial non-(West) African BGA (discussed in more
detail below). In this paper, we will use the acronym SIRE to refer to the social identities.
Sometimes we will use the following abbreviations for SIRE groups: Blacks/African
Americans = B, White Americans = W, Asian Americans = A, Hispanic/Latino
Americans = H and Native Americans = NA.
MANKIND QUARTERLY 2016 56:4
582
1. Data sources
1.1. S-factor scores
Measure of America (http://www.measureofamerica.org/) publishes analyses
as well as socioeconomic datasets for US states, counties and other divisions
(http://www.measureofamerica.org/download-agreement/). As in the case of a
previous study (Kirkegaard, 2015), we downloaded the dataset covering US
states. We used the 2010 dataset as this was the most recent. Because we
wished to analyze outcomes by SIRE groups by state, we were restricted to the
following variables:
1. Human Development Index [American Human Development Index]
2. Life expectancy at birth (years)
3. Less than high school (%)
4. At least high school diploma (%)
5. At least bachelor's degree (%)
6. Graduate degree (%)
7. School enrollment (%)
8. Median earnings (2010 dollars)
9. Health Index
10. Education Index
11. Income Index
Of these, 1 and 9-11 are aggregate variables calculated from the others; as
such, they were of little interest because we wanted to compute our own
aggregate. Variables 2 and 8 were independently useful, but 3-7 were all
education related and strongly overlapped. To avoid possible general factor
contamination/coloring (Jensen, 1998, p. 85; Kirkegaard, in review), we selected
only one of these, number 5. This left us with variables 2, 5 and 8. There are
multiple ways one can extract S factors from the present data. We begin by
analyzing the values for all states (N = 50). We do this because we want to know
how well the S factor scores extracted from these 3 indicators correlate with those
extracted from 81 variables (described in Kirkegaard, 2015). We also analyze the
SIRE state data as units in one analysis (N=154). Results are shown in Figure 1.
As expected, all variables have positive loadings. However, the loading of life
expectancy was somewhat unstable being about .2-.3 lower for the SIRE state
data. The educational variable had a loading of ~1, even when using ranked data.
The correlations between the different total state S factors are shown in Table 1.
While not near unity, the correlation between the 3 and 81 variable S-factor scores
was very strong. The interval and rank order versions were correlated near unity,
so it should not matter much which method is used.
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
583
Figure 1. S factor loadings for total states and SIRE states analyses. Standard
(interval) and rank-order data. TS = total states (all SIREs together), RS = race
states, SIRE x state combinations in one combined analysis.
Table 1. Unweighted correlations between total state socioeconomic (S) factors,
N = 50 states.
Variable
S_3indi
S_3indi_rank
S_81indi
S_3indicators
S_3indicators ranked
0.979
S_81 indicators
0.876
0.871
One can split the dataset by SIRE groups and analyze each subset. Such an
analysis can show if the structure of S is different for the different SIRE groups.
Results are shown in Figures 2 and 3.
With the interval data, life expectancy for Hispanics is a clear outlier with a
loading below 0. However, this anomaly disappears when using ranked data.
Even with ranked data there was some instability, though all loadings were
between .7 and 1. Due to the somewhat unstable factor loadings, it is worth
MANKIND QUARTERLY 2016 56:4
584
examining the cross-method correlations between factor scores by SIRE. These
are shown in Table 2.
Figure 2. S factor loadings for SIRE states by SIRE, using interval data. A, Asian;
B, Black; H, Hispanic; NA, Native American; W, White.
Figure 3. S factor loadings for SIRE states by SIRE, using ranked data. A, Asian;
B, Black; H, Hispanic; NA, Native American; W, White.
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
585
MANKIND QUARTERLY 2016 56:4
586
We are especially concerned with the italicized numbers. Despite the
variation in loadings, the scores were very strongly correlated across methods.
This is expected because of the strong intercorrelations of the variables in
general. Because interval data are generally to be preferred and because the
scores were so highly correlated, we used the interval scores for the remaining
analyses despite the odd loading of life expectancy in the case of Hispanics. This
loading is possibly related to the so-called Hispanic paradox, which is the finding
that Hispanics generally live longer than non-Hispanic Whites, despite being
socioeconomically (and cognitively) worse off (Ruiz, Steffen & Smith, 2013).
As a robustness check, we extracted the S factor scores using all
combinations of extraction methods and scoring methods using the R fa() function
(Revelle, 2015). No method variance across any of the datasets was detected
(i.e., all correlations were near or at unity).
The distributions of the state S scores by SIRE groups are shown in Figure
4.
Figure 4. Empirical density plot of distributions of SIRE states. Socioeconomic
(S factor) scores are plotted on the X axis.
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
587
The three lowest scoring groups are concentrated around -1 to -.5, while
Asians are far ahead, even of Whites. There is substantial overlap between
Blacks and Whites. Descriptive statistics are shown in Table 3. We note that
Whites and Asians have much higher standard deviations than the other groups.
Why is not clear.
Table 3. Descriptive statistics for S by SIRE. N. A., Native Americans.
Statistic / SIRE
Blacks
Asians
Hispanics
N. A.
Whites
mean
-0.624
1.678
-0.924
-1.030
0.266
sd
0.231
0.582
0.230
0.235
0.413
1.2. Cognitive scores
There are presently no published, publicly available US IQ scores
decomposed by state and SIRE groups. As such, we used The National
Assessment of Educational Progress MAIN's (MAIN-NAEP) math and reading
scores as indices of cognitive ability
4
. We used the results from 8th graders for
years 2011 and 2013. The use of achievement scores was justified because it
has been shown that at the aggregate level, scholastic test scores correlate very
strongly with standard IQ test scores (Lynn & Vanhanen, 2012; Rindermann,
2007). One could alternatively use university and/or graduate entrance exam test
scores (e.g., SAT and ACT), but these scores are less representative of state
populations (McDaniel, 2006). The NAEP online explorer provides percentages
of test takers by SIRE group. This is potentially important because, owing to, for
example, migration and population age structure, the SIRE composition of test
takers can markedly differ from the overall SIRE composition of a given state.
Many states had missing data for the percentages of some of the smaller SIRE
groups (e.g., Pacific Islanders in Connecticut). We filled in these cases with 0's.
Figure 5 shows a density plot of state cognitive ability by SIRE groups.
We see a substantial overlap between Black and Hispanic scores, as would
be expected given the magnitude of differences between these groups (e.g., Roth
et al., 2001). There is, however, almost no overlap between these and the white
distribution (the outlier is West Virginia). Table 4 shows the descriptive statistics.
4
Accessed at: http://nces.ed.gov/nationsreportcard/naepdata/
MANKIND QUARTERLY 2016 56:4
588
Figure 5.Empirical density distributions for cognitive ability for SIRE groups at
the state level.
Table 4. Descriptive statistics for cognitive ability by SIRE.
Statistic / SIRE
Blacks
Hispanics
Whites
Mean
255.9
263.1
283.2
SD
5.1
4.8
5.2
1.3. Within SIRE racial admixture data
SIRE racial admixture data was copied from Bryc et al. (2015), who obtained
the results by analyzing customer data from the personal genomics service
23andMe (https://www.23andme.com/). The data are problematic because it is
from a self-selected group. It is unknown how this self-selection influences the
admixture estimates. Because among Blacks and Hispanics, European BGA is
positively correlated with SES and since interest in genomic testing and a
willingness to purchase a test-kit (at $100-200) probably correlates with SES,
black and Hispanic participants were probably, to some degree, indirectly
selected for European ancestry (Fuerst & Kirkegaard, 2015). Additionally, for
Blacks and Hispanics, data are lacking for many states and for other states are
based on small sample sizes. Bryc et al. (2015) showed ternary plots for individual
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
589
admixture (their Figure S4) but they did not show one for the state-level data.
They did, however, report the state admixture percentages by SIRE group in table
form (their Table S2), from where we obtained them. Figure 6 shows a ternary
plot of racial admixture for states by SIRE group.
Figure 6. Ternary plot of the three main sources of genomic admixture in SIRE
states. Plot based on assigned admixture only. Admixture proportions are read
counterclockwise (e.g. about 15-35% European and 65-85% African for self-
identified Blacks).
One can see that Blacks lie on a line between the African and European
clusters with close to no Amerindian admixture. As Whites are almost entirely
European (top corner), owing to the trivial variation in BGA, admixture analysis
would provide meaningless results. Hispanics are mostly European, but show
non-trivial African and Amerindian admixture. Table 5 shows basic descriptive
statistics for each SIRE group and their admixture percentages.
MANKIND QUARTERLY 2016 56:4
590
Table 5. 23andMe based racial admixture estimates by SIRE group. mad =
adjusted median absolute deviation.
SIRE
n
mean
sd
median
mad
min
max
Blacks
African
31
0.74
0.04
0.74
0.03
0.64
0.83
Amerindian
31
0.01
0.00
0.01
0.00
0.00
0.01
European
31
0.23
0.04
0.24
0.03
0.15
0.34
Other
31
0.02
0.00
0.02
0.00
0.01
0.03
Total
31
0.98
0.00
0.98
0.00
0.97
0.99
Hispanics
African
34
0.09
0.05
0.08
0.06
0.01
0.22
Amerindian
34
0.10
0.05
0.09
0.03
0.04
0.21
European
34
0.73
0.07
0.72
0.05
0.57
0.90
Other
34
0.08
0.05
0.07
0.03
0.02
0.34
Total
34
0.92
0.05
0.93
0.03
0.66
0.98
Whites
African
50
0.00
0.00
0.00
0.00
0.00
0.01
Amerindian
50
0.00
0.00
0.00
0.00
0.00
0.00
European
50
0.99
0.00
0.99
0.00
0.98
1.00
Other
50
0.01
0.00
0.01
0.00
0.00
0.02
Total
50
0.99
0.00
0.99
0.00
0.98
1.00
As noted by Bryc et al. (their section: Aggregating Local Ancestry
Information), the estimates do not generally add up to exactly 1. This is because
the researchers adopted a conservative approach which left unassigned some
admixture. We created two additional variables “Other” and “Total”, where the
latter refers to the sum of the assigned admixture and the former refers to the
residual. Inspecting the table makes it clear that Hispanics have substantial
unassigned admixture (mean = 8%). Figure 7 shows a histogram of total assigned
admixture by SIRE group and state.
The far outlier is Hawaii, where presumably a large chunk of the missing
admixture is Pacific Islander/Hawaiian/Asian. The presence of substantial
missing admixture for Hispanics makes interpretations of admixture x outcome
results problematic.
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
591
Figure 7.Histogram of total assigned admixture.
1.4. US states climatic data
Sources are given in our target article (Fuerst & Kirkegaard, 2016a).
2. Analyses
2.1. Cognitive ability and S
We now turn to the relationship between cognitive ability and socioeconomic
outcomes (S). Generally, this relationship has been found to be strongly positive
(r's .5 to .9) across many aggregated datasets (Fuerst & Kirkegaard, 2016a;
Kirkegaard, 2014, 2015), but no meta-analysis has yet summarized the results.
In our target article, we reported a weighted correlation of .70 for cognitive ability
and S for the states of the US. Figure 8 depicts the cognitive ability and S relation
for each SIRE group and for all three groups aggregated together. All datasets
show positive relationships as expected, though the slope is somewhat steeper
for Whites. Blacks have somewhat higher S-factor scores than Hispanics despite
having lower cognitive ability scores, thus giving rise to a slight Simpson's
paradox (for a discussion of Simpson's paradox, see: Kievit et al., 2013). The
relatively low socioeconomic status of Hispanics is probably due, in part, to the
fact that many are or are the children of recent low-skill immigrants. Table 6 shows
correlations and sample sizes.
MANKIND QUARTERLY 2016 56:4
592
Figure 8. Scatter plot of cognitive ability (NAEP score) and socioeconomic (S-
factor) score by SIRE by state. SIRE groups are marked by colors. The black line
is the regression based on all combined datasets. B = Blacks, H = Hispanics, W
= Whites.
Table 6. Cognitive ability x S correlations for total states ("All"), each SIRE state
and all SIRE states combined in one analysis.
Statistic/SIRE
All
Black
Hispanic
White
Combined
r
0.67
0.67
0.55
0.78
0.86
N
50
37
28
50
115
All correlations are strong and positive. The combined dataset exhibits the
strongest correlation, which is expected because the variance is increased as can
be seen in the scatter plot of Figure 8. A further finding is that there is almost a
positive manifold (i.e., all intercorrelations are positive) for NAEP and S scores
across SIRE groups. For instance, in states where Whites are socioeconomically
better off, Blacks are smarter (r=.566; N=37), and vice versa (r=.583). The
correlation matrix is given in the appendix (see supplementary material,
Appendix: NAEP_S_cors).
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
593
2.2. State SIRE composition and cognitive ability
In our target article, we attempted to predict the cognitive ability of the states
of the US from their estimated genomic admixture. To create state admixture
estimates, we used SIRE admixture estimates in conjunction with 2010 census
SIRE state percentages. Since we were interested in the relative effect of
European, African and Amerindian BGA on outcomes we excluded Asian and
Pacific Islander SIRE groups from our computations. Based on multiple
regressions, we found that using African or Amerindian ancestry in conjunction
with European ancestry offered no incremental validity over using just European
ancestry.
Here, we rerun the analysis using NAEP test taker percentages for all SIRE
groups. We include the state of Hawaii in this analysis, so N = 50. We carried out
multiple regression analyses, testing all possible models (Nmodels=127), with SIRE
group percentages as predictors. A number of the models generated senseless
results (e.g., all predictors had excessively large negative betas); this problem
seems to result when too many SIRE composition predictors are used, probably
due to multicollinearity and a lack of free variance. Disregarding them, the top 10
models according to adjusted r2are shown in Table 7.
Table 7. Model statistics for predicting state NAEP scores. Beta coefficients are
shown for each SIRE group. Top 10 models by adj. R2. N. A., Native American;
P. I., Pacific Islander.
Model #
White
Black
Hispanic
Asian
N. A.
P. I.
Mixed
adj. R2
117
-0.600
-0.370
-0.326
-0.426
0.197
0.322
126
-0.601
-0.430
0.227
-0.311
-0.638
0.199
0.318
87
-0.603
-0.384
-0.302
-0.336
0.307
111
0.886
0.216
0.518
-0.433
0.274
0.305
124
0.564
-0.228
0.379
-0.175
-0.480
0.255
0.303
114
-0.604
-0.443
0.221
-0.287
-0.541
0.302
107
0.612
-0.163
0.466
-0.524
0.241
0.301
82
0.701
0.591
-0.576
0.250
0.297
125
0.882
0.220
0.497
-0.091
-0.418
0.284
0.295
108
0.415
-0.315
-0.231
-0.204
0.240
0.294
We see some familiar patterns: White and Asian percentages are
consistently associated with positive betas, while Black, Native American and
Pacific Islander percentages are associated with negative ones. The betas for
MANKIND QUARTERLY 2016 56:4
594
percentage of Hispanics varied depending on the other predictors in the model.
The correlation matrix is given in supplementary material, Appendix
(NAEP_beta_cors).
Another summary of the results is shown below in Table 8. Here, we show
the median and adjusted median absolute deviation (mad) beta for each SIRE
predictor. These represent robust (that is, to the influence of the invalid models'
betas) alternatives to means and standard deviations. When analyzed this way,
the usual patterns emerge, except that the effect of White% is greater than that
of Asian%. The adjusted multiple R of the best fitting model is .57 (the square root
of .322 from model 117), vs. .64 in the target article. Again, we fail to find
substantial evidence of incremental validity for predictors beyond an index of
European ancestry.
Table 8. Summary statistics of beta coefficients for model predictors of state
NAEP scores. N. A., Native American; P. I., Pacific Islander.
Measure
White
Black
Hispanic
Asian
N. A.
P. I.
Mixed
median
0.632
-0.456
-0.224
0.206
-0.068
-0.338
0.136
mad
0.230
0.180
0.266
0.520
0.089
0.451
0.070
2.3. Within SIRE racial admixture, cognitive ability and S
A prediction from the racial model tested by Fuerst & Kirkegaard (2016a) is
that the scores of black and Hispanic individuals should positively correlate with
the individuals' % of European admixture. The underlying logic applies to state
scores too – though, on the state level, there are more confounds. As mentioned
in Section 1.3, the admixture estimates available are questionable, especially for
Hispanics, in which case a non-trivial portion of admixture was unassigned. Still,
it is worth looking at the associations. Tables 9 and 10 show the correlation
matrices for Blacks and Hispanics, respectively.
Table 9. African American BGA by state and outcome variables. Unweighted
correlations. N's 31 for admixture variables and 37 for S x CA.
African%
Amerindian%
European%
Other%
CA
Amerindian%
0.077
European%
-0.993
-0.147
Other%
-0.416
0.245
0.314
CA
-0.330
0.185
0.298
0.338
S
-0.295
0.356
0.253
0.350
0.671
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
595
Table 10. Hispanic racial admixture by state and outcome variables. Unweighted
correlations. N's 34, 24, 34 for admixture x CA, admixture x S, and CA x S,
respectively.
African
%
Amerindian
%
European
%
Other
%
CA
Amerindian%
-0.424
European%
-0.206
-0.497
Other%
-0.312
0.161
-0.655
CA
0.276
-0.237
0.137
-0.229
S
0.450
-0.559
0.132
-0.010
0.549
For the African American SIRE group, we see the expected findings, namely
that more European admixture correlates with better outcomes and that more
African admixture correlates with worse. Amerindian ancestry is also positively
correlated with outcomes, but, as shown in Table 5, the variation in Amerindian
ancestry among African Americans is trivial (M = 0.006, SD = 0.002), assuring
that any association is likely spurious. For Hispanics, as expected, Amerindian
ancestry is negatively associated with outcomes. However, unexpectedly, African
ancestry is positively so associated. As shown in Table 5, the standard deviation
of African admixture is about the same as that for Amerindian admixture, so the
results are not dismissible on the grounds of low variation. It is possible that the
results are biased due to the influence of other variables not included in the
analysis.
Thus, we used hierarchical regression with plausible predictors to see how
the results are affected. We include White cognitive ability as an index of the
quality of schools. Table 11 shows the models. We start with a baseline model
which shows the zero-order correlation between state-level African admixture and
state-level African American cognitive ability. We then add our parasite load
variable. This seems to acquire some of the validity of African%, though the
relation stays negative. We then add White cognitive ability, which turns out to be
a strong positive predictor. In models 4a and 4b we try adding a second
admixture variable, but they have no incremental validity. We could not add
European% because that would create multicollinearity (rAfrican% x European% = -
0.993).
Next we repeat the analysis for Hispanics. Table 12 shows the results,
presented using two different bases: European% and African%. This is because
the racial model being tested makes relatively clear predictions concerning the
expected associations (positive for European and negative for African). We add
variables as before. It is worth noting that parasite prevalence has a strong
MANKIND QUARTERLY 2016 56:4
596
positive beta, which is not as expected if parasite burden is lowering the cognitive
ability of Hispanics. As before, White cognitive ability is a moderate to strong
positive predictor across all models. Of special interest is that European% begins
as weakly positive but becomes more strongly positive in the best models, while
African% begins as weakly positive but becomes more negative in the better
models. The full set of models are given in the appendix (see supplementary
material, Appendix: NAEP_A_models, NAEP_H_models).
Table 11.Hierarchical regression models for African American cognitive ability.
Model
White CA
Parasites
African
%
Amerindian
%
Other
%
adj. R²
1
-0.349
0.078
2
-0.388
-0.101
0.109
3
0.555
-0.112
-0.221
0.506
4a
0.551
-0.116
-0.220
0.013
0.487
4b
0.543
-0.109
-0.197
0.066
0.491
Table 12. Hierarchical regression models for Hispanic cognitive ability.
Model
White
CA
Parasites
African
%
European
%
Amerinidian
%
Other
%
adj. R²
European% as base
1
0.121
-0.012
2
0.536
0.168
0.175
3
0.457
0.716
0.321
0.371
4a
0.461
0.687
0.274
0.358
4b
0.505
0.809
0.469
0.191
0.372
4c
0.486
0.753
0.318
0.351
African% as base
1
0.245
0.047
2
0.480
0.040
0.139
3
0.355
0.687
-0.075
0.252
4a
0.482
0.791
-0.270
-0.333
0.338
4b
0.374
0.686
-0.127
-0.164
0.261
4c
0.468
0.753
-0.046
0.318
0.351
We now present the regression results for S. These are analogous to those
for cognitive ability except that parasite prevalence is not used as a predictor.
Tables 13 and 14 show the results for African Americans and Hispanics,
respectively. The results are similar to the cognitive ability ones. For African
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
597
Americans, White S is a strong positive predictor and African% is a moderate
negative predictor.
Table 13. Hierarchical regression models for African American socioeconomic
outcomes (S factor scores).
Model
White S
African
%
Amerindian
%
Other
%
adj. R²
1
-0.291
0.055
2
0.692
-0.257
0.714
3
0.671
-0.262
0.059
0.706
3a
0.696
-0.264
-0.018
0.703
4
0.677
-0.276
0.065
-0.032
0.696
Table 14. Hierarchical regression models for Hispanic socioeconomic outcomes
(S factor scores).
Model
White S
African
%
European
%
Amerindian
%
Other
%
adj. R²
European% as base
1
0.150
-0.027
2
0.374
0.369
0.039
3a
0.370
0.065
-0.596
0.344
3b
0.364
0.434
0.083
-0.004
4
0.412
-0.281
-0.744
-0.343
-0.380
African% as base
1
0.520
0.166
2
0.265
0.555
0.203
3a
0.349
0.287
-0.503
0.397
3b
0.234
0.586
0.071
0.169
4
0.412
0.207
-0.552
-0.126
0.380
Both
2
0.518
0.142
0.144
3
0.446
0.573
0.401
0.269
4a
0.412
0.327
0.163
-0.440
0.380
4b
0.412
0.801
0.805
0.498
0.380
For Hispanics, as before, White S is a moderate to strong predictor across
all models. Unexpectedly, again, African% is a positive predictor even in the
company of other predictors. Moreover, the size of the beta is larger than that for
MANKIND QUARTERLY 2016 56:4
598
European%. The full set of models are given in the appendix (see supplementary
material, Appendix: S_A_models, S_H_models). These results are curious since
they suggest that in the Hispanic population African ancestry is more positively
associated with outcomes than is European ancestry and, yet, on the national
level, self-identifying black Hispanics perform worse than do self-identifying white
Hispanics. For an illustration of the latter point, 2015 grade 8 NAEP math and
reading results are shown in Table 15 for all Hispanics, Mexicans, Cubans, and
Puerto Ricans. Nothing sticks out in these which would account for the lack of a
negative association between African ancestry and cognitive ability on the state
level. Self-reported “Black” identity is consistently associated with lower scores
across all major Hispanic groups. It is possible that SIRE does not well track
ancestry for Hispanics. Alternatively, perhaps the African ancestry estimates were
off owing to the unassigned admixture. This is a topic which might need to be
explored more in the future.
Table 15. 8th-grade NAEP scores for various Hispanic subgroups in 2015.
NAEP 2015
Math
Reading
White Hispanic
276
259
White and Black Hispanic
269
254
Black Hispanic
260
242
Mexican (all)
270
253
White Mexican
275
257
White and Black Mexican
268
249
Black Mexican
261
242
Cuban (all)
272
256
White Cuban
280
264
White and Black Cuban
265
244
Black Cuban
256
238
Puerto Rican (all)
266
254
White Puerto Rican
272
262
White and Black Puerto Rican
265
252
Black Puerto Rican
260
244
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
599
2.4. Environmental causes of cognitive ability and S across states
A number of environmental factors have been proposed to explain
differences in cognitive ability and socioeconomic development between
countries and administrative units therein (Eppig, Fincher & Thornhill, 2011;
Kanazawa, 2008; León, 2016; León & Burga León, 2014). One study examined
cognitive ability, socioeconomic outcomes and temperature for US states and
statistically controlled for SIRE composition (Pesta & Poznanski, 2014). We argue
that a preferable method is to disaggregate scores by major SIRE groups and to
then correlate those scores with the variables of interest. As we have
disaggregated cognitive ability and S-factor scores we can conduct this analysis.
We also include the total state scores for comparison. Table 16 shows part of the
correlation matrix.
Table 16. Correlations of cognitive ability (CA) and socioeconomic development
(S) of Blacks (B), Hispanics (H) and Whites (W) with environmental variables.
CA All
CA W
CA B
CA H
S All
S W
S B
S H
Parasites
-0.58
-0.21
-0.36
0.23
-0.54
-0.08
-0.33
0.27
Longitude
0.27
0.08
-0.13
0.13
0.16
-0.10
-0.21
0.44
Latitude
0.55
0.28
0.15
0.03
0.46
0.04
0.34
-0.23
Temperature
-0.59
-0.29
-0.16
0.01
-0.50
-0.06
-0.20
0.24
Rain
-0.23
-0.20
-0.11
0.12
-0.12
-0.08
-0.33
0.68
Humid morning
-0.12
-0.25
-0.26
0.20
-0.14
-031
-0.36
0.37
Humid afternoon
0.09
-0.05
0.01
0.11
0.25
-0.004
-0.03
0.43
Sunshine
-0.28
-0.08
-0.05
-0.18
-0.30
0.02
0.05
-0.38
In general, there are often sizable correlations between the outcomes –
cognitive and S scores – and environmental variables when looking at the total
state scores (e.g., temperature x CA = -.59). However, the correlations are much
smaller and the directionality is inconsistent when looking at the disaggregated
scores. Results for Hispanics, in Table 16, are often at odds with those for the
other two SIRE groups. This perhaps owes to migration-related effects (e.g.,
language bias). The true (statistical) effect of the environmental variables is
probably best indexed by the average of these variables' correlation with White
and Black outcomes. Though, an alternative possibility is that the association in
the case of Whites and Blacks is spurious, resulting from patterns of historic
selective migration from the southeast to the north, and that the true
environmental effect is indexed by the effect on Hispanic scores.
MANKIND QUARTERLY 2016 56:4
600
Parasite prevalence, latitude, and temperature are generally associated with
outcomes in the directions noted by Leon (2016) and Pesta (2016), yet the
magnitudes of the associations with SIRE-decomposed state scores are notably
smaller than those based on total state scores. To facilitate comparison with
Pesta & Poznanski's (2014) results, we ran the partial correlations for total state
scores and the environmental variables controlling for the percentage of Whites
(in line with the previously mentioned authors' method). This method of controlling
left substantially larger correlations between the environmental variables and the
outcome variables (e.g., CA x Temp = -.41). This illustrates, to our minds, that
statistically controlling for SIRE percentage does not completely remove the effect
of SIRE related differences. The full partial correlation matrix is given in the
appendix (see supplementary material, Appendix: Partial_environ_vars).
2.5. Cognitive ability and S gaps across states
2.5.1. Cognitive ability gaps as a cause of S gaps
Across states there are substantial differences in the magnitude of the gaps
between the majority group (Whites) and, respectively, Blacks and Hispanics.
Such differences call out for an explanation. If cognitive ability differences cause
S differences (the central thesis of cognitive sociology), then S and cognitive
ability gaps should be correlated across states. Figure 9 shows the scatter plot.
The correlations are strong at .62 (N=37) and .69 (N=36) for Blacks and
Hispanics, respectively. West Virginia (WV) is a clear outlier in terms of the
magnitude of the B/W gaps, but it nonetheless falls neatly on the regression line
for S-factor and cognitive gaps, as, in this case, both are near zero.
Figure 9. Scatter plot of cognitive ability and socioeconomic (S) gaps to the
White population for Blacks (B) and Hispanics (H).
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
601
2.5.2. Differences in European ancestry, and gaps in cognitive ability and S
across states
In Section 2.3, we examined SIRE BGA and outcomes. We found, for Blacks,
that European ancestry was associated with positive outcomes and that African
ancestry was associated with negative ones. However, for Hispanics, African
ancestry was associated with positive outcomes and was so more strongly than
was European ancestry. This is a puzzling finding, one inconsistent with a broad
collection of evidence from other sources. There is an alternative way to examine
the dataset, namely to correlate the magnitudes of the within-state differences in
ancestry with the magnitude of the within-state differences in outcomes for Whites
and Blacks/Hispanics across states. This analytic strategy has the advantage of
side-stepping factors (such as temperature) which vary between states and which
have more or less uniform effects on members within states. The R~CA-S model
would predict a positive relation between differences in European ancestry and
differences in cognitive ability and S. Table 17 shows the correlation matrix.
Table 17. Differences in European ancestry (ΔEuro), cognitive ability (ΔCA) and
socioeconomic outcomes (ΔS) between majority and minority SIRE groups
across states. Unweighted correlations. N's for Blacks = 31-45, N's for Hispanics
= 24-46.
SIRE
Variable
ΔEuro%
ΔCA
Black/White
ΔCA
0.28
ΔS
0.16
0.62
Hispanic/White
ΔCA
0.36
ΔS
0.55
0.69
We see that differences in European ancestry are positively associated with
differences in outcomes for both comparisons. Oddly again, for Hispanics, African
ancestry positively correlates with both cognitive ability and S at, respectively, .12
and .35. While these are less than the correlations between European ancestry
and outcomes, the associations are in the unpredicted direction.
2.5.3. Parasites as a cause of cognitive ability gaps
As discussed in Section 2.4, it has been proposed that parasite prevalence
is an environmental cause of US state-level and national differences in cognitive
ability (Eppig, Fincher & Thornhill, 2010, 2011). If one accepts this, one might also
propose that group differences in cognitive ability within countries could be
MANKIND QUARTERLY 2016 56:4
602
explained by differential parasite exposure. In this model, the lower scoring
groups are more exposed to parasites than the higher scoring groups, thus
lowering their cognitive ability. This leads to a prediction, namely that the group
gaps should be larger in the states with higher parasite prevalence rates. We
tested this by correlating human parasite prevalence rates with the size of the
gaps (as explained earlier). Figures 10 and 11 show the scatterplots.
Figure 10. Prevalence of human parasites and White-Black cognitive ability gap
by state.
Figure 11. Prevalence of human parasites and White-Hispanic cognitive ability
gap by state.
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
603
The correlation for the White-Black gaps is slightly positive (with the
confidence interval widely overlapping 0) while that for Hispanics is strongly
negative. If one combines the analyses, the correlation is -.17 [CI95: -.36 to .04].
3. Discussion and conclusion
We found that the state-level cognitive ability x S-factor relation replicated for
Blacks, Whites and Hispanics in the US (Section 2.1). In line with previous results,
it was found that adding multiple SIRE groups did not substantially increase
predictive validity relative to using European ancestry alone (Section 2.2).
Additionally, associations of environmental factors with outcomes were weaker
than previously reported when scores were decomposed by SIRE groups as
compared to when the statistical effect of SIRE percentages was controlled for
(Section 2.4).
Regarding admixture, it was found (Section 2.3) that for Blacks, state-level
European admixture was positively associated with better outcomes (mean r=
.28), while African admixture was associated with worse outcomes (mean r= -31,
N = 31). For Hispanics, partially contrary to prediction, both African and European
admixture were related to better outcomes (mean r's = .36 and .13), while
Amerindian admixture was associated with worse outcomes (mean r= -.40; N =
24-34). Both sets of results were robust to controls for certain external factors.
The same pattern held when we alternatively correlated the magnitude of the B/W
and H/W European ancestry differences with differences between SIRE groups
in cognitive ability and socioeconomic outcomes (Section 2.5). The reason why
state-level African ancestry among Hispanics positively correlates with state-level
Hispanic outcomes is unclear, especially in light of the relatively poor
performance of self-reported black Hispanics.
Taken together, the results illustrate how using SIRE decomposed scores
can provide analytic leverage when conducting cross-state analyses. It would be
interesting to examine the S factor at lower levels of aggregation for the US.
Several datasets suitable for this purpose are known to exist: 1) Counties
(N=3143), 2) Congressional districts (N=436), 3) Louisiana parishes (N=36), and
4) Metropolitan Areas (N=25), all from Measure of America. SIRE composition
data are given in the datasets as well, making it possible to conduct crude
admixture analyses. While cognitive data are not found in these datasets,
cognitive data are known to exist for counties (Barnes, Beaver & Boutwell, 2013;
Boutwell et al., 2013). Furthermore, because some of these datasets are nested
inside each other, it is possible to conduct multi-level analyses as well.
MANKIND QUARTERLY 2016 56:4
604
Supplementary material and acknowledgments
R source code, data and the appendix (appendix.xslx) can be found at the
project repository at Open Science Framework https://osf.io/dzqen/files/.
References
Barnes, J.C., Beaver, K.M. & Boutwell, B.B. (2013). Average county-level IQ predicts
county-level disadvantage and several county-level mortality risk rates. Intelligence 41:
59–66. http://doi.org/10.1016/j.intell.2012.10.007
Boutwell, B.B., Franklin, T.W., Barnes, J.C., Beaver, K.M., Deaton, R., Lewis, R.H.,
Petkovsek, M.A. (2013). County-level IQ and fertility rates: A partial test of differential-K
theory. Personality and Individual Differences 55: 547–552. http://doi.org/10.1016/
j.paid.2013.04.018
Bryc, K., Durand, E.Y., Macpherson, J.M., Reich, D. & Mountain, J.L. (2015). The genetic
ancestry of African Americans, Latinos, and European Americans across the United
States. American Journal of Human Genetics 96: 37–53. http://doi.org/10.1016/
j.ajhg.2014.11.010
Eppig, C., Fincher, C.L. & Thornhill, R. (2010). Parasite prevalence and the worldwide
distribution of cognitive ability. Proceedings of the Royal Society of London B: Biological
Sciences, rspb20100973. http://doi.org/ 10.1098/rspb.2010. 0973
Eppig, C., Fincher, C.L. & Thornhill, R. (2011). Parasite prevalence and the distribution of
intelligence among the states of the USA. Intelligence 39: 155–160.
Fuerst, J. (2015). Nature of race: the genealogy of the concept and the biological
construct’s contemporaneous utility. Open Behavior Genetics.
Fuerst, J. & Kirkegaard, E.O.W. (2015). Admixture in the Americas. Presented at the 2015
London Conference on Intelligence, London, UK. Retrieved from
https://docs.google.com/presentation/d/1hjhOiitk0MnqMHgVthyj8j7qa4qcAqPUDNaTT8r
petg/e dit?pli=1#slide=id.p
Fuerst, J. & Kirkegaard, E.O.W. (2016a). Admixture in the Americas: Regional and
national differences. Mankind Quarterly 56: 256-374.
Fuerst, J. & Kirkegaard, E.O.W. (2016b). The genealogy of differences in the Americas.
Mankind Quarterly 56: 426-482.
KIRKEGAARD, E.O.W & FUERST, J. INEQUALITY IN THE UNITED STATES
605
Jensen, A.R. (1998). The g Factor: The Science of Mental Ability. Westport, Conn.:
Praeger.
Kanazawa, S. (2008). Temperature and evolutionary novelty as forces behind the
evolution of general intelligence. Intelligence 36: 99–108.
Kievit, R., Frankenhuis, W.E., Waldorp, L. & Borsboom, D. (2013). Simpson’s paradox in
psychological science: A practical guide. Quantitative Psychology and Measurement 4:
513. http://doi.org/10.3389/fpsyg.2013.00513
Kirkegaard, E.O.W. (In review). Some new methods for exploratory factor analysis of
socioeconomic data. Open Quantitative Sociology & Political Science.
Kirkegaard, E.O.W. (2014). The international general socioeconomic factor: Factor
analyzing international rankings. Open Differential Psychology. Retrieved from
http://openpsych.net/ODP/2014/09/the-international-general-socioeconomic-factor-
factor-analyzing- international-rankings/
Kirkegaard, E.O.W. (2015). A replication of the S factor among US states using a new
and larger dataset. The Winnower. Retrieved from https://thewinnower.com/papers/1418-
a-replication-of-the-s-factor-among-us-states-using-a-new-and-larger-dataset
León, F.R. (2016). Race vis-à-vis latitude: Their influence on intelligence, infectious
diseases, and income. Mankind Quarterly 56: 412-420.
León, F.R. & Burga-León, A. (2014). Why complex cognitive ability increases with
absolute latitude. Intelligence 46: 291–299. http://doi.org/10.1016/j.intell.2014.07. 011
Lynn, R. & Vanhanen, T. (2012). Intelligence: A Unifying Construct for the Social
Sciences. London: Ulster Institute for Social Research.
McDaniel, M.A. (2006). State preferences for the ACT versus SAT complicate inferences
about SAT-derived state IQ estimates: A comment on Kanazawa (2006). Intelligence 34:
601–606. http://doi.org/10.1016/j.intell.2006.07.005
Pesta, B.J. (2016). Does IQ cause race differences in well-being? Mankind Quarterly 56:
421-425.
Pesta, B.J. & Poznanski, P.J. (2014). Only in America: Cold Winters Theory, race, IQ and
well-being. Intelligence 46: 271–274. http://doi.org/10.1016/ j.intell.2014.07.009
MANKIND QUARTERLY 2016 56:4
606
Revelle, W. (2015). psych: Procedures for Psychological, Psychometric, and Personality
Research (Version 1.5.4). Retrieved from http://cran.r-project.org/web/packages
/psych/index.html
Rindermann, H. (2007). The g-factor of international cognitive ability comparisons: The
homogeneity of results in PISA, TIMSS, PIRLS and IQ-tests across nations. European
Journal of Personality 21: 667–706. http://doi.org/10.1002/per.634
Roth, P.L., Bevier, C.A., Bobko, P., Switzer, F.S. & Tyler, P. (2001). Ethnic group
differences in cognitive ability in employment and educational settings: A meta-analysis.
Personnel Psychology 54: 297–330. http://doi.org/10.1111/j.1744-6570.2001.tb00094.x
Ruiz, J.M., Steffen, P. & Smith, T.B. (2013). Hispanic mortality paradox: A systematic
review and meta-analysis of the longitudinal literature. American Journal of Public Health
103(3): e52–e60. http://doi.org/10.2105/AJPH.2012. 301103
