ArticlePDF Available

Is National Mental Sport Ability a Sign of Intelligence? An Analysis of the Top Players of 12 Mental Sports

  • Ulster Institute for Social Research

Abstract and Figures

Research at the individual level shows strong positive relationships between performance in video games and on intelligence tests. Together with evidence of above average IQs of players of traditional mental sports such as chess, this suggests that national IQs should be strongly related to national performance on mental sports. To investigate this, lists of top players for 12 different electronic sports (e-sports) and traditional mental sports were collected from a variety of sources (total n = 36k). Using a log count approach to control for population size, national cognitive ability/IQ was found to be a predictor (p<.05) of the relative representation of countries among the top players for every game except Go. When an overall mental sports score was calculated using a factor analytic approach, the factor scores correlated r = .79 with Lynn and Vanhanen's (2012) published national IQs. The pattern was somewhat nonlinear such that national IQs below 85 seemed to have no relationship. The games that related most strongly with the general factor of mental sport ability also correlated more strongly with national IQs (r = .94). The relationship was fairly robust to controls for geographical region (coefficient 74% of the original in chosen model specification).
Content may be subject to copyright.
Is National Mental Sport Ability a Sign of Intelligence?
An Analysis of the Top Players of 12 Mental Sports
Emil O. W. Kirkegaard*
Ulster Institute for Social Research, London
*Address for correspondence:
Research at the individual level shows strong positive
relationships between performance in video games and on
intelligence tests. Together with evidence of above average IQs of
players of traditional mental sports such as chess, this suggests that
national IQs should be strongly related to national performance on
mental sports.
To investigate this, lists of top players for 12 different electronic
sports (e-sports) and traditional mental sports were collected from a
variety of sources (total n = 36k). Using a log count approach to
control for population size, national cognitive ability/IQ was found to
be a predictor (p<.05) of the relative representation of countries
among the top players for every game except Go. When an overall
mental sports score was calculated using a factor analytic approach,
the factor scores correlated r = .79 with Lynn and Vanhanens (2012)
published national IQs.
The pattern was somewhat nonlinear such that national IQs
below 85 seemed to have no relationship. The games that related
most strongly with the general factor of mental sport ability also
correlated more strongly with national IQs (r = .94). The relationship
was fairly robust to controls for geographical region (coefficient 74%
of the original in chosen model specification).
Key words: Intelligence, IQ, Cognitive ability, Mental sport, e-sport,
Chess, Go, Poker, Scrabble, Starcraft 2, DOTA 2, League of
Legends, LOL, Counter-Strike, Hearthstone, Overwatch, Super
Smash Bros
Enormous amounts of human time are spent playing a variety of mental
sports/games1, either involving a computer or not. A large amount of this time is
spent playing competitive games, that is, against other players (PvP; player
versus player) instead of against the computer (PvE; player versus environment).
Major games with PvP gameplay usually have an in-game ranking list as well as
large tournaments offering prizes of thousands or millions of US dollars. This
naturally creates a focus on player ability, and its causes and correlates. Aside
from time spent playing the games and thereby improving through practice, an
obvious candidate for explaining differences in player skill is (general) cognitive
ability/general intelligence2.
Cognitive ability is known to be strongly related to a wide variety of
vocational, educational/training, social and health outcomes and has been
described (in part) as a very general mental capability that, among other things,
involves the ability to reason, plan, solve problems, think abstractly, comprehend
complex ideas, learn quickly and learn from experience.(Gottfredson, 1997, p.
13; see also Hunt, 2010; Jensen, 1998; Strenze, 2015). Since mastering a game,
in particular a complex game, requires understanding of game mechanics,
planning tactics/strategy and learning from experience, cognitive ability is a
plausible cause of game skill.
A number of studies have examined the relationship between measures of
cognitive ability and skill in a variety of games, both computerized (esports) and
not (usually board games). For instance, a meta-analysis of studies of chess
players and non-chess players found a mean group difference of .50d on
intelligence tests (Sala et al., 2017). In another study, Ángeles Quiroga et al.
(2015) (see also Foroughi et al., 2016) administered 12 video games to 188
students along with traditional standardized cognitive tests, and found that they
measured nearly identical constructs (rs .93 to .96).
Given the positive relationships between game performance and cognitive
ability seen between individuals, we might wonder whether national skills at
mental sports also show relationships to national intelligence. Indeed, in a recent
1 Similar to Jensen (1998, p. 49ff), we define mental sport as sports where player
performance is not strongly linked to physical features such as height, grip strength or
maximum running speed.
2 We will use cognitive ability/intelligence in lieu of the more technical general cognitive
ability, general mental ability, or general intelligence and in lieu of the less readable
gca / gma / g. However, by this usage we refer to the general cognitive ability captured
by the g factor (Jensen, 1998).
series of blogposts, Zambian biochemist Chanda Chisala (2016 etc.) argued that
Nigerian scrabble performance was much higher than one would expect based
on its national IQ estimate (given as 71.2 by Lynn & Vanhanen, 2012). His
interpretation is that the tests are missing real ability present in this and other
African populations. His study, however, was not systematic and did not include
many other mental sports. The purpose of the present study is to systematically
examine the relationship between national skill at mental sports and national
cognitive ability.
1. Data
Table 1. Sample sizes by game
Top players
N countries
We chose to rely on counts of top players for games since ranking lists are
widely available and they contain the necessary nationality data. For games, we
included every esport listed at with at
least 1,000 top players. This resulted in the inclusion of 8 games: DOTA 2 (dota2),
League of Legends (lol), Counter-Strike: Global Offensive (csgo), Starcraft 2
(sc2), Counter-Strike (cs), Hearthstone (hs), Overwatch (ow), and Super Smash
Bros. Melee (ssbm). Furthermore, we included major non-esports: Chess, Go,
Poker and Scrabble. In each case, we found a major ranking website3 and
scraped4 the contents. The number of top players by game is shown in Table 1.
For predictor variables, we used the national IQs from Lynn and Vanhanen
(2012). Population sizes, country, continents and regions are from the U.N. All
statistical analyses were done in R. The player lists were scraped using Rs rvest
package (Wickham & RStudio, 2016), and in one case UNIXs curl. See analysis
code for details.
2. Analyses
2.1. Counting players by country
All datasets from were already given in
country-level form. However, sometimes a few subnational units were used on
the site which would have caused issues due to inconsistencies across datasets5
and missing covariate data. This was seen for subdivisions of United Kingdom
(England, Scotland, Wales, N. Ireland), China (Hong Kong, Macao), Denmark
(Greenland, Faeroe Islands) and the Netherlands (Aruba etc.). In every case, the
subdivisions were merged with their mother countries using the motherlands
For the other datasets, they were given by player. Every dataset included
some form of a national identifier, either by stating the country or by having a flag.
We converted the flags to their countries and aggregated to the country level.
2.2. Modeling count data with rare persons
The most straightforward approach to modeling count data is to convert it to
the more common per capita form, which facilitates easy intuitive understanding
and interpretation. Unfortunately, this transformation a form of direct control for
population size results in a severe increase in the standard error of the estimate.
To see this, consider a small country like Iceland (population 330k). Suppose we
are interested in an event (person, in this case) that is very rare, with an average
frequency of 1 in a million. This means we would expect 0.33 of them in Iceland
3 These were:,,,
4 Automatically downloaded using a computer script.
5 For instance, if a given small unit is only available in half of the datasets, it would seem
to have particularly low performance in the others, insofar as one uses imputed values.
on average. How would the sampling distribution look for Iceland? We sampled
this scenario 106 times. The results are given in Table 2.
Table 2. Iceland (330k population) sampling of rare person (1 in 106). 106
Event count
In 72% of the cases, no person was observed, yielding a per capita rate of
zero. On the other hand, 24% of samples yielded a single person, yielding a per
capita rate of 1 in 330k about three times the true value. A further ~4% of
samples yielded higher counts and even more extreme per capita rates. Note that
every sampling outcome would yield a quite inaccurate estimate of the true per
capita rate (1 in 106).
An alternative, but less intuitive, way of approaching the problem is to include
population size in the regression model along with whatever predictors one is
interested in. To avoid non-normality in residuals, usually log values of population
size are employed. This approach avoids the extreme standard error for countries
because the counts themselves are used, not the per capita rates. If we have a
small country like Iceland, a small count is expected (~0.3 in the above), and
small deviations around it (0-5) do not constitute major outliers in the modeling.
To illustrate this, we simulated fictive world-wide data (200 countries) where
countries varied strongly in population size (sampled from the true country sizes
with repeats) and in their per capita rate of top persons (from 1 in 106 to 1 in 109,
normally distributed). Moreover, the per capita rate of the country was strongly
caused (r = .70) by a single predictor (also normally distributed). The data were
then modeled using three approaches:
1) Per capita: per capita rate ~ cause
2) Straight count: count ~ population size + cause
3) Log-count: log10(count + 1) ~ log10(population size) + cause6
The model statistics were saved and analyzed. The results showed that the
log count approach was better than the per capita method in two ways. First, it
produced much smaller p values and the most under the conventional thresholds,
as shown in Figure 1; though it should be said that the straight count approach
was slightly superior.
Second, the log count method produced the smallest variation in the beta
estimates in terms of coefficient of variation (standard deviation divided by mean).
Table 3 shows the summary of key model statistics. Note that it is not sensible to
compare R2 (or adj. R2) for the modeling choices because the count approaches
include population size as a predictor. Because this explains most of the variance,
these models almost always have quite large R2s even when the prediction is
Figure 1. Distribution of p values for the causal variable in n = 104 simulation
runs. The vertical lines mark the traditional cutoffs of .05, .01, .005, and .001.
6 “~” (tilde) means is predicted byor regressed unto”, i.e. left-side of tilde is the
dependent variable and the right side terms are the predictors.
Table 3. Summary statistics of model parameters from world simulations. SD =
standard deviation, CV = coefficient of variation, MAD = median absolute
Mean SD Median MAD Min Max Skew Kurtosis CV
0.08 0.03 0.08 0.02
0.16 0.34
log count/p
0.13 0.36 0.12 0.06
7.75 3.63 269.84 2.84
per capita/p
2.3. Predicting performance of individual mental sports
There are a number of ways the data can be modeled. One approach is to
model each dataset independently, and then meta-analyze the results. This
approach will produce unbiased results insofar that sampling error is not
systematic (e.g. biased towards zero). In every dataset, many countries had not
a single top player. For these countries, 0’s were imputed. After this, the following
model was fit to each dataset:
log10(player count + 1) ~ log10(population) + IQ + UN macroregion
Two aggregate estimates were then computed based on the model beta
estimates. The first was the arithmetic mean, which was 0.0235. The second was
a random effects meta-analytic estimate which was .0226 [.0162 to .0289, p
< .0001]7. The between sample heterogeneity was 70% (p < .0001), indicating
substantial between sample variation unrelated to sampling error itself. This is the
expected result if culture strongly affects to which games populations decide to
put their national talent pool. Figure 2 shows the Forest plot.
Visual inspection of the Forest plot indicated that Go was an outlier, which is
not surprising because it is played almost exclusively in East Asian countries (no
non-East Asian country had 5 players in the dataset of 946 players), producing
a much larger sampling error than the analytic standard error would indicate.8
Because of this, the meta-analysis was re-run without Go. This increased the
7 95% analytic confidence interval. The analysis was done using metafor (Viechtbauer,
8 The analytic standard error is based on the sample size and standard deviation.
However, there were only 14 country observations for this dataset, the remaining
values were imputed as 0’s.
arithmetic mean effect size to .0253 and the random effects estimate to .0246
[.0195 to .0297, p < .0001].
Figure 2. Forest plot for random effects meta-analysis of effect size of national
To check robustness to analytic choices, two further modeling variants were
run. First, without imputed 0’s for countries with no players. Second, without the
regional control. Both approaches produced higher estimates for the IQ beta
at .0309 [.0229 to .0389] and .0421 [.0339 to .0504], respectively, using the
random effects model.
2.4. Aggregate mental sport performance
An alternative approach is to avoid modeling IQ at the level of the individual
game, but to aggregate game performance to a single overall metric, which is
then modeled as a function of IQ and other predictors. This was done by fitting a
simple population size model to each game:
log10(player count + 1) ~ log10(population)
and saving the residuals.9 The residuals were then factor analyzed to see if it was
9 For this purpose, the zero-imputed datasets were used.
sensible to interpret an aggregate metric as a measure of a latent general gaming
ability. The mean inter-correlation was .57, and every game loaded positively on
the general factor. Figure 3 shows the scatterplot between national IQ and
general gaming ability score.
Figure 3. General gaming ability score and national IQ. Orange line = linear fit
(top left), blue line = local regression fit (span = 1.00). Weighted by square root of
population size.
The scatterplot revealed a somewhat nonlinear pattern where a large
proportion of countries had similar near zero performance, roughly those below
IQ 80, while there was a linear or perhaps upwardly curving trend above IQ 80.
The strongest negative outlier was North Korea, presumably due to very limited
internet access despite presumed high IQ.10 There were three notable positive
10 No IQ test data is available for North Korea (Lynn & Vanhanen, 2012). However, the
IQ has been estimated based on the neighboring countries (China and South Korea).
outliers: South Africa, Brazil and United States. All of these feature a dominant
European-derived population, though its fraction of the population varies and is
probably best interpreted as a smart fraction effect (La Griffe du Lion, 2002;
Rindermann, Sailer & Thompson, 2009). Stated differently, there is much
cognitive stratification in these countries, with subpopulations that differ widely in
cognitive ability.
As a robustness test, the mean and median game performance were also
calculated from the residuals as an alternative to the factor scores. However, both
of these correlated .99 with the factor scores, so this had little effect.
A series of regression models were fitted to the final dataset. First, a simple
linear regression model with national IQ as the predictor. Second, a nonlinear
model (restricted cubic spline with 3 knots; Harrell, 2015) with national IQ as
predictor to determine the gain in validity from modeling the nonlinear pattern
seen in the scatterplot. The R2 adj. of this model was .693 compared to .653 of
the linear model, i.e. there was a slight increase as expected (p < 5.3 * 10-7 for
model comparison).
Third, three-variant models with regional controls to adjust for any culture-
related confounders in mental sports interest (or interest related to the sample of
included games). These were based on UN classifications of countries (5
continents and 23 regions), as well as a recoding of region into macro-regions to
decrease the number of areas from 23 to 14 (details found in the supplementary
materials). The three regional control models increased the validity beyond IQ
alone and were about equivalent in terms of adj. R2 (.746, .790, .760,
respectively). We used the macro-region variant for further modeling as this gave
the best trade-off between likely overfitting and simplicity.
Fourth, the nonlinear terms for IQ were dropped to see if the regional controls
would be sufficient to account for the nonlinearity. This model had an adj. R2
of .742, only slightly worse than the nonlinear model, suggesting that cultural
effects could account for most of the nonlinear IQ relationship. Still, the nonlinear
model fit better and this was probably not a coincidence (p = .0001), and thus it
was kept.
Fifth, for comparison, three models with just the regional controls were fitted
to see how well these could proxy for cognitive ability. While continents alone
were clearly insufficient, using either regions or macro-regions produced nearly
the same levels of R2 adj. (.55, .89 and .90 for continents, macro-regions, and
regions, respectively). In case of the latter two, this is not surprising because
sorting countries into 23 or 15 classes allows one to very closely approximate
national IQs.
Sixth, at the request of a reviewer, an exploratory analysis was run by adding
percentage of internet users as a covariate (Fig. 4). Models were then refit both
with and without the spline for IQ. Likelihood tests showed that this improved
model fit in both cases (p for linear models = .002, p for spline models = .01).
However, the increase in explained variance was generally small, about 2% (.742
adj. R2 to .767 for linear, and .760 to .781 for spline). The beta for IQ in the linear
model changed from .070 to .046 (-34.8%), and from .098 to .086 (-12.5%) for the
spline model (second slope value). In all cases, the p value for IQ remained very
small (p < .002 in all cases). Furthermore, percentage internet users showed the
strongest relationships with games that loaded the highest on the general mental
sports factor (r = .92, plot can be found in supplementary materials), not with
games that require internet access as one would expect if it was a strong
confounder. This is the pattern one would expect if percentage internet users
were a proxy for other variables related to country well-being. Table 4 shows the
final model fit (model 2b), and Table 5 shows summary statistics for the models
Figure 4. Map of mobile broadband internet penetration.
Table 4. Full model fit; n = 195 countries, R2 adj. = .760. Weighted by square root
of population size. CI = 95% confidence interval. IQ was modeled as a restricted
cubic spline following Harrell 2015's rms package (
CI lower
CI upper
Intercept -0.398 -3.841 3.045 0.820
IQ 0.006 -0.037 0.049 0.798
IQ’ 0.098 0.047 0.149 <0.001
UN_macroregion=N&W Europe + offshoots
UN_macroregion=Latin America
UN_macroregion=Central Asia
UN_macroregion=Eastern Asia
UN_macroregion=Eastern Europe
UN_macroregion=Melanesia -1.005 -2.299 0.290 0.128
UN_macroregion=Micronesia -0.390 -2.973 2.193 0.766
UN_macroregion=MENA -0.999 -1.679 -0.319 0.004
UN_macroregion=South-Eastern Asia -0.921 -1.510 -0.332 0.002
UN_macroregion=Southern Asia
UN_macroregion=Southern Europe
Table 5. Summary statistics for fitted models. Weighted by square root of
population size.
Regional controls
Adj. R2
It is noteworthy that every regional dummy has a negative beta estimate
compared to the north and west European reference group. Interpretation of this
is unclear. It might reflect bias in data collection where games that non-Europeans
play were missed. This seems unlikely given the inclusiveness of the dataset and
the systematic sampling approach (i.e. include all games with 1,000 top
players). Alternatively, it might reflect a genuine cultural effect where N&W
European cultures place more emphasis on performance in mental sports,
perhaps mediated through individualism.
The effect of national IQ remained quite substantial in all models, and only
partially shrunk when regional controls were added. If we compare the linear
models 1a and 2d, we find that the betas for IQ were 0.095 and 0.074, meaning
that IQ retained 78% of its validity.
Finally, because we have 12 indicators of general mental sport skill, it is
possible to conduct a moderator analysis using Jensens method of correlated
vectors, shown in Figure 5.
A strong pattern is seen such that the games that more strongly reflect
general gaming ability also show stronger correlations to national IQ, as predicted
from a latent factor model. It also offers an explanation for the strong performance
of Nigerians on scrabble, as scrabble was the worst indicator of general mental
sports ability (factor loading = .27).
Figure 5. Jensen's method applied to the general mental sports factor. The X-
axis shows the loading of each game on the general mental sports factor.
3. Discussion and conclusion
The present study investigated national mental gaming skill. Across 12
diverse games, it was found that skill at one game relates positively to skill at
another game, supporting a general mental sports ability interpretation. Scoring
the countries on this general ability results in a correlation of .79 with Lynn and
Vanhanens national IQs (weighted by square root of population size).
The strong relationship with national IQ mostly held when regional controls
were included. The beta coefficient for IQ decreased by about 26%. This might
reflect some cultural confounding or measurement error in the national IQs
(Westfall & Yarkoni, 2016). Nevertheless, the strong relationship fits well with both
individual level results obtained in multiple studies (Ángeles Quiroga et al., 2015;
Foroughi et al., 2016; Kokkinakis et al., 2017; Spitz, 1978), as well as large
between country differences in cognitive ability reported by a number of authors
(Lynn & Vanhanen, 2012; Rindermann, 2007).
At the request of a reviewer, internet usage was entered as another covariate
based on the obvious idea that if one has no internet access, it is hard to play
computer games that require internet access. Doing this improved model fit
slightly (by about 2% adj. R2 across comparisons), but IQ remained highly
predictive in all cases though its slope was reduced (-24% on average).
Percentage internet users showed the strongest correlations to game
performance in those that loaded most strongly on the general mental gaming
factor (r = .92), not the games that require internet access, indicating that its
function in the models was probably to act as a proxy for other variables, making
its interpretation difficult.
With regards to the arguments by Chisala concerning Nigeria and Scrabble,
we find no support for mismeasured Nigerian intelligence. It is true, as he noted,
that Nigerians are much better at Scrabble than one would expect (standardized
residual for Scrabble is 1.55, rank 3). However, Nigerians underperform on the
remaining 11 sports (all residuals are negative), and the overall factor score of
Nigeria (-1.39) is about what one would expect based on the current estimates of
the countrys mean cognitive ability and the relation to national IQ. Indeed, based
on a linear regression with just IQ, the standardized residual for Nigeria is only -
0.16, meaning that the country performs very slightly worse at mental sports in
general than one would expect based on its mean national intelligence. The
predicted IQ of Nigeria was 73.1 based on a nonlinear model with just game
performance as the predictor, Lynn and Vanhanens (2012) estimate was 71.2,
and a recent large-scale study (n 11k) using Ravens Standard Progressive
Matrices found a mean IQ of 65.5 (Hur, Nijenhuis & Jeong, 2017).11 The
anomalously high Scrabble performance is not plausibly interpreted as hidden
ability, but rather as a culture-specific preference for a specific sport.
Supplementary material and acknowledgments
Supplementary materials including code, high quality figures and data can
be found at
Ángeles Quiroga, M., Escorial, S., Román, F.J., Morillo, D., Jarabo, A., Privado, J., … &
Colom, R. (2015). Can we reliably measure the general factor of intelligence (g) through
commercial video games? Yes, we can! Intelligence 53 (supplement C): 1-7.
Becker, D. (2017, November 30). THE NIQ-DATASET. Retrieved from
Chisala, C. (2016, October 31). Scrabble spells doom for the racial hypothesis of
intelligence. Retrieved from
Foroughi, C.K., Serraino, C., Parasuraman, R. & Boehm-Davis, D.A. (2016). Can we
create a measure of fluid intelligence using Puzzle Creator within Portal 2? Intelligence
56 (supplement C): 58-64.
Gottfredson, L.S. (1997). Mainstream science on intelligence: An editorial with 52
signatories, history, and bibliography. Intelligence 24: 13-23.
Harrell, F.E. (2015). Regression Modeling Strategies: With Applications to Linear Models,
Logistic and Ordinal Regression, and Survival Analysis, 2nd edition. New York: Springer.
Hunt, E. (2010). Human Intelligence. Cambridge University Press.
Hu r, Y. -M., te Nijenhuis, J. & Jeong, H.-U. (2017). Testing Lynns theory of sex differences
in intelligence in a large sample of Nigerian school-aged children and adolescents
(N>11,000) using Ravens Standard Progressive Matrices Plus. Mankind Quarterly 57:
Jensen, A.R. (1998). The g Factor: The Science of Mental Ability. Westport CT: Praeger.
11 This study did not calculate a norm referenced IQ score. However, the conversion is
provided by David Beckers National IQ dataset, v1.2 (Becker, 2017).
Kokkinakis, A.V., Cowling, P.I., Drachen, A. & Wade, A.R. (2017). Exploring the
relationship between video game expertise and fluid intelligence. PLoS ONE 12(11):
La Griffe du Lion (2002, March 1). The smart fraction theory of IQ and the wealth of
nations. Retrieved from
Lynn, R. & Vanhanen, T. (2012). Intelligence: A Unifying Construct for the Social Sciences.
London: Ulster Institute for Social Research.
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.
Rindermann, H., Sailer, M. & Thompson, J. (2009). The impact of smart fractions,
cognitive ability of politicians and average competence of peoples on social development.
Talent Development & Excellence 1: 3-25.
Sala, G., Burgoyne, A.P., Macnamara, B.N., Hambrick, D.Z., Campitelli, G. & Gobet, F.
(2017). Checking the Academic Selectionargument. Chess players outperform non-
chess players in cognitive skills related to intelligence: A meta-analysis. Intelligence 61:
Spitz, H.H. (1978). The universal nature of human intelligence: Evidence from games.
Intelligence 2: 371-379.
Strenze, T. (2015). Intelligence and success. In: S. Goldstein, D. Princiotta & J.A. Naglieri
(eds.), Handbook of Intelligence, pp. 405-413. New York, NY: Springer.
Viechtbauer, W. (2015). metafor: Meta-Analysis Package for R (Version 1.9-8). Retrieved
Westfall, J. & Yarkoni, T. (2016). Statistically controlling for confounding constructs is
harder than you think. PLoS ONE 11(3): e0152719.
Wickham, H. & RStudio (2016). rvest: Easily Harvest (Scrape) Web Pages (Version 0.3.2).
Retrieved from
... We included regional dummies (labeled macroregions) based on the United Nations' classification, which we modified to reduce the number of dummies to a more manageable level (n = 14). This level of dummies is intermediate between continents (n = 5) and UN regions (n = 23) and has been used in prior studies (Kirkegaard, 2019a). The supplementary materials show maps of the classifications used ( Figure S1-S3). ...
... While the situation is not perfect for the cognitive data, there is a diverse set of cognitive indicators all of which are very strongly related, sometimes estimated to be correlated at 1.0 when adjusted for measurement error (Lynn & Becker, 2019;Lynn & Meisenberg, 2010;Lynn & Mikk, 2007;Rindermann, 2007Rindermann, , 2018. Kirkegaard (2019a) even showed that one can extract a general mental gaming ability score from a diverse set of 12 mental games, and this correlates .87 with Lynn's national IQs. Furthermore, there are many within-country studies of measurement invariance which generally support this for major population groups, with minimal bias in reported effect sizes (Beaujean & McGlaughlin, 2014;Frisby & Beaujean, 2015;Lasker et al., 2019). ...
Full-text available
Patient people fare better in life than impatient people. Based on this and on economic models, many economists have claimed that more patient countries should fare better than less patient countries. We utilize cross-national data in non-cognitive traits measured in the Global Preference Survey (GPS). This survey measured six non-cognitive traits — risk and time preferences, positive and negative reciprocity, altruism, and trust — across 76 countries in about 80,000 persons. As such, it provides the best current database of economics-focused non-cognitive traits. We combine this database with existing estimates of national intelligence (national IQs) and model country outcomes as a function of these predictors. For outcomes, we used the 51 national well-being indicators from the Social Progress Index (SPI) as well as the composite extracted from this, the general socioeconomic factor. We find that non-cognitive variables, time preference included, are only weakly predictive of national well-being outcomes when national IQs are also in the model. The median β across the indicators was 0.11 for time preference but 0.39 for national IQ. We replicated these results using six economic indicators, again with similar results: median βs of 0.15 and 0.52 for time preference and national IQ, respectively. Across all our results, we found that national IQ has 2-4 times the predictive validity of time preference. These results are fairly robust to inclusion of a spatial autocorrelation control, alternative measures of national IQ and time preference, or no controls. Our results suggest that the importance of national non-cognitive traits, including time preference, is overestimated or that these traits are mismeasured.
... Third, video games would be a useful way for estimating the average IQ level of populations (Kirkegaard, 2018). They might contribute to assessing inaccessible groups (video games can be implemented in cell phones) and also to systematically analyze the link between intelligence and health in real time, as suggested by Kokkinakis and colleagues (2017). ...
Video games can be useful tools for assessing intelligence and cognitive differences. First, available video games are grouped into thirteen genres defining their basic features. Second, empirical studies relating intelligence and video games are reviewed. Results show very strong correlations between intelligence and video game performance at the latent level, which suggests that the latter taps core facets of the intelligence concept. Third, regarding cognitive processes, studies have focused on “action video games.” Results have shown that video game experience (hours per week) correlates mainly with visuospatial cognition, perception, and attention. Fourth, key neural correlates of video game performance are also discussed. The final section enumerates required features for a video game to properly measure intelligence differences using a video game elaborated for research purposes (Forgotten Depths).
... Other Outcomes We also decided to investigate specific claims from the literature. Kirkegaard (2019a) suggested that intellectual class ability could help to explain certain outliers in the relationship between national IQ and performance in international mental sports. Rindermann (2011) made a similar argument concerning winners of the International Mathematical Olympiad (IMO). ...
Full-text available
Previous studies have found that the ability of a country’s cognitive elite is generally more predictive than the average ability. However, these studies have relied on sub-optimal methods. Here, the authors tested smart fraction theory, as it is known, using a pre-residualization approach, which obviates the problem of collinearity. For outcome variables, they utilised the 51 indicators of the Social Progress Index, as well as 6 economic variables. Like in previous research, the authors operationalized the ability of the intellectual class as the 95th percentile score. Consistent with smart fraction theory, they found evidence that the ability of the intellectual class does influence outcomes over and above that of the average ability. For the 93 countries with available data, average ability explained 57% of the variation in country performance. Adding the 95th percentile score increased this to 66%. The authors discuss their findings in light of the existing literature.
... We included regional dummies, one based on continents and one based on macroregions. The macroregions were copied from a previous study and based on UN classifications (Kirkegaard, 2019a). Maps with the classification schemes are given in the appendix. ...
Full-text available
We estimated crime rates among 70 origin-based immigrant groups in the Netherlands for the years 2005-2018. Results indicated that crime rates have overall been falling for each group in the period of study, and in the country as a whole, with about a 50% decline since 2005. Immigrant groups varied widely in crime rates, with East Asian countries being lower and Muslim countries, as well as Dutch (ex-)colonial possessions in the Caribbean, being higher than the natives. We found that national IQ and Muslim percentage of population of the origin countries predicted relative differences in crime rates, r’s = .64 and .45, respectively, in line with previous research both in the Netherlands and in other European countries. Furthermore, we carried out a survey of 200 persons living in the Netherlands to measure their preferences for immigrants for each origin country in terms of getting more or fewer persons from each country. Following Carl (2016), we computed a mean opposition metric for each country. This correlated strongly with actual crime rates we found, r’s = .46 and .57, for population weighted and unweighted results, respectively. The main outliers in the regression were the Dutch (ex-)colonial possessions, and if these are excluded, the correlations increase to .68 and .66, respectively. Regressions with plausible confounders (Muslim percentage, geographical fixed effects) showed that crime rates continued to be a useful predictor of opposition to specific countries. The results were interpreted as being in line with a rational voter preference for less crime-prone immigrants.
... Third, video games would be a useful way for estimating the average IQ level of populations (Kirkegaard, 2018). They might contribute to assessing inaccessible groups (video games can be implemented in cell phones) and also to systematically analyze the link between intelligence and health in real time, as suggested by Kokkinakis and colleagues (2017). ...
Full-text available
The Cambridge Handbook of Intelligence
Full-text available
One commonly studied aspect of the importance of IQ is its validity in predicting job performance. Previous research on this subject has yielded impressive results, regularly finding operational validities for general mental ability exceeding 0.50. In 2015, Ken Richardson and Sarah Norgate criticized the research on the relationship between IQ and job performance, reducing it to virtually nothing. Their assessment of this topic has enjoyed little criticism since its publication despite the crux of their arguments being undermined by readily available empirical evidence and thirty years of replication of the contrary. This article replies to their main criticisms, including the construct validity of IQ tests and supervisory ratings, the validity of the Hunter-Schmidt meta-analytic methods, and possible psychological confounders.
Full-text available
Hundreds of millions of people play intellectually-demanding video games every day. What does individual performance on these games tell us about cognition? Here, we describe two studies that examine the potential link between intelligence and performance in one of the most popular video games genres in the world (Multiplayer Online Battle Arenas: MOBAs). In the first study, we show that performance in the popular MOBA League of Legends’ correlates with fluid intelligence as measured under controlled laboratory conditions. In the second study, we also show that the age profile of performance in the two most widely-played MOBAs (League of Legends and DOTA II) matches that of raw fluid intelligence. We discuss and extend previous videogame literature on intelligence and videogames and suggest that commercial video games can be useful as 'proxy' tests of cognitive performance at a global population level.
Full-text available
Sex differences in intelligence have been much disputed for many decades. The present study examined the issues of whether sex differences in intelligence change during development. In total, 11,164 children (mean age = 13.5 years; SD = 2.6 years) completed the Standard Progressive Matrices Plus (SPM+). From age 8 to 19 years, sex differences in the total score of the SPM+ increased from-0.06d (favoring females) to 0.46d (favoring males), with an average of 0.23d. Our findings support Lynn's developmental theory of sex differences in cognitive abilities.
Full-text available
Substantial research in the psychology of expertise has shown that experts in several fields (e.g., science, mathematics) perform better than non-experts on standardized tests of intelligence. This evidence suggests that intelligence plays an important role in the acquisition of expertise. However, a counter argument is that the difference between experts and non-experts is not due to individuals' traits but to academic selection processes. For instance, in science, high scores on standardized tests (e.g., SAT and then GRE) are needed to be admitted to a university program for training. Thus, the “academic selection process” hypothesis is that expert vs. non-expert differences in cognitive ability reflect ability-related differences in access to training opportunities. To test this hypothesis, we focused on a domain in which there are no selection processes based on test scores: chess. This meta-analysis revealed that chess players outperformed non-chess players in intelligence-related skills ( = 0.49). Therefore, this outcome does not corroborate the academic selection process argument, and consequently, supports the idea that access to training alone cannot explain expert performance.
Full-text available
Social scientists often seek to demonstrate that a construct has incremental validity over and above other related constructs. However, these claims are typically supported by measurement-level models that fail to consider the effects of measurement (un)reliability. We use intuitive examples, Monte Carlo simulations, and a novel analytical framework to demonstrate that common strategies for establishing incremental construct validity using multiple regression analysis exhibit extremely high Type I error rates under parameter regimes common in many psychological domains. Counterintuitively, we find that error rates are highest-in some cases approaching 100%-when sample sizes are large and reliability is moderate. Our findings suggest that a potentially large proportion of incremental validity claims made in the literature are spurious. We present a web application ( that readers can use to explore the statistical properties of these and other incremental validity arguments. We conclude by reviewing SEM-based statistical approaches that appropriately control the Type I error rate when attempting to establish incremental validity.
Researchers have suggested that video games can be used to measure cognitive abilities (e.g., Boot, 2015). To that end, we wanted to test the feasibility of developing a measure of fluid intelligence within a commercially-available video game. We created a 15 chamber (i.e., problems to be solved) test called the “Portal 2 Test Battery” using the Puzzle Creator within the popular problem-solving video game Portal 2. In study 1, 35 video game players (VGPs) who had experience playing Portal 2 completed this test and Ravens Advanced Progressive Matrices (RAPM), a validated measure of fluid intelligence, revealing a strong relationship (r = .65) between the two. In study 2, 100 participants, both VGPs and non-video game players (nVGPs) completed the Portal 2 Test Battery, RAPM, and the Bochumer Matrices Test (BOMAT), another validated measure of fluid intelligence. Structure equation modeling revealed a strong relationship between the latent variables fluid intelligence and portal ability (r = .78). Additionally, the reliability of our test was acceptable in both studies (α = .80 and α = .76). These results provide the first evidence that it may be feasible to create a measure of fluid intelligence using the Puzzle Creator within Portal 2.
The chapter reviews the research on the relationship between intelligence and various forms of success. The chapter also discusses how this relationship is influenced by genes and society.