Personality and Individual Differences 206 (2023) 112110

Contents lists available at ScienceDirect

Personality and Individual Differences

journal homepage: [Link]/locate/paid

The future of intelligence: A prediction of the FLynn effect based on past

student assessment studies until the year 2100
Heiner Rindermann *, David Becker
Department of Psychology, Chemnitz University of Technology, Wilhelm-Raabe-Str. 43, D-09107 Chemnitz, Germany

A R T I C L E I N F O A B S T R A C T

Keywords: In the 20th century, a strong increase in IQ test values was observed. However, the results of the last few decades
Flynn effect point to an end to the FLynn effect or even its reversal. Scientists came to skeptical assumptions for the 21st
Future of intelligence century. There is hardly any research on the possible future development of intelligence. Here we present a
IQ prediction
statistical approach: Results from 24 student achievement surveys (TIMSS, PISA, PIRLS) from 1995 to 2019 serve
Intelligence prediction
Linear and non-linear regression
as the data basis. In order to be included in the analysis, at least five measurements per country had to be
available (N = 79 countries). For statistical estimation, trends (calculated via subtractions between measure­
ments within one student survey approach), linear regressions and nonlinear quadratic regressions (both using
any given student achievement data) were applied. A correction for outliers smoothed the results. For the sample
of 79 countries, IQs would increase by about 10 IQ points by 2100 (international mean IQ 101). We compare the
results with results from theoretical models. The theoretical models came to less optimistic conclusions. The
results depend on the selected statistical (e.g., linear or nonlinear, outlier correction or not) and theoretical
assumptions. We recommend that also other authors address this research question.

1. The FLynn effect in the 20th century education, health and brain size) and in the indicators of cognitive
abilities (e.g., in share of employees in professional occupations,
In the 20th century, there was a steep increase in IQ test scores of educational degrees acquired and complexity of everyday life), most
about dec = 2.31 IQ points (dec: per decade) (Trahan et al., 2014) or dec researchers now assume a real increase in cognitive ability (intelligence
= 2.83 IQ points (Pietschnig & Voracek, 2015). The increase in IQ test and knowledge). Some even assume that the rise in intelligence has been
scores from generation to generation was first observed in soldiers in the happening for centuries (Oesterdiekhoff, 2014).
United States before World War II (Rundquist, 1936). It was later seen Another issue discussed was whether the gains were in g (general
throughout the developed world, followed by IQ increases in developing factor of intelligence) or in several specific abilities (te Nijenhuis & van
countries (Flynn, 2012). The American-New Zealand political scientist der Flier, 2013). Despite the observable rise in intelligence test scores,
James Flynn's seminal 1984 review “The mean IQ of Americans: Massive other researchers assume that genotypic intelligence has been declining
gains 1932 to 1978” popularized the secular rise of IQ test performance for some time.1 The “co-occurrence model” assumes two simultaneous
in academia and the media. The effect was later labeled the “Flynn ef­ processes: an increase in IQ test scores that has nothing to do with the g-
fect”. In the 1980s, Richard Lynn (1982) described the same phenome­ factor of intelligence and a decrease in genetic intelligence that would be
non for Japan. So two researchers, Lynn and Flynn, or Flynn and Lynn, on g (Egeland, 2022; Woodley of Menie & Fernandes, 2015). Indicators
rediscovered the secular IQ rise, which is why we call it the “FLynn for a more g-related decrease are for example a decline in mental speed
effect”. and other basic cognitive processes or a lessening of important in­
At the outset of research on the FLynn effect, it was doubted whether novations and extremely intelligent people (Dutton & Woodley of
the observed IQ test scores represent true increases in intelligence or just Menie, 2018; Woodley, 2012).
IQ inflation (e.g., Flynn, 1987). However, since there are similar de­ Nevertheless, results of the last decades point to an end of the FLynn
velopments in the supporting conditions of intelligence (e.g., in effect or even to its reversal in the developed world. In countries with

* Corresponding author.
E-mail addresses: [Link]@[Link] (H. Rindermann), [Link]@[Link] (D. Becker).
1
It is also called the Woodley effect after Michael Woodley of Menie.

[Link]
Received 25 November 2022; Received in revised form 27 January 2023; Accepted 28 January 2023
Available online 6 February 2023
0191-8869/© 2023 Elsevier Ltd. All rights reserved.
H. Rindermann and D. Becker Personality and Individual Differences 206 (2023) 112110

negative trends, the average decrease was about dec = − 2.44 IQ points of education in 21st century (e.g., Lutz et al., 2014). Education is one of
(Dutton et al., 2016) or dec = − 1.66 IQ points (Woodley of Menie et al., the most important determinants of cognitive ability (e.g., Ritchie &
2018). The FLynn effect seems to be petering out, at least in western Tucker-Drob, 2018, 3.39 IQ points per school year).
countries. This skepticism was also expressed in a survey of scientists in Another important factor is migration and the demographic change
the field of FLynn effect research (Rindermann et al., 2017, Table 3). In that it causes. One example from the USA: In the US student achieve­
addition, the quality of the framework conditions for intelligence ment study NAEP (National Assessment of Educational Progress), which
development can hardly be improved or further improvements (e.g., in measures reading ability and mathematics, the results in the different
education and health) are accompanied by diminishing returns. Finally, ethnic groups have increased over the years. However, as the percent­
low birth rates among the educated, especially among educated women, ages of the relatively higher test-scoring ethnic group (Whites) decrease
and immigration from countries with lower levels of education and and the relatively lower-testing ethnic group (Hispanics) increase, the
cognitive ability weaken the demographic basis for further FLynn ef­ overall trend has remained flat since the 1970s (Rindermann &
fects. This is particularly problematic because a further increase in Thompson, 2013). This means that predictions hardly show any growth
cognitive demands in the world of work and life is to be expected (Hunt, – but at least no decline either (Rindermann & Pichelmann, 2015). The
1995; OECD, 2013). fear of a similar development for Germany was expressed in 2011 by the
While developing countries are cognitively catching up (Meisenberg German “PISA pope” Jürgen Baumert (2011): There would be gains in
& Lynn, 2023; Meisenberg & Woodley, 2013), gains have reversed, PISA, presumably through teaching and school reforms, but these would
particularly in German-speaking countries (see Figs. 1 and 2): Gains in be “cancelled out” by socio-structural change.
cognitive abilities in the past have turned into losses, which nonlinear Previous prediction models considered several factors (Rindermann,
regression models show particularly well. 2018; Rindermann & Thompson, 2011): the effects of asymmetric fertility
In an overall view of the development of student achievement studies (asymmetric birth rates and generation spans), migration (increasing
in Germany, the development looks just as critical: According to shares of immigrants and different ability levels of immigrants), further
Woessmann (2021), who evaluated 43 surveys from PISA, TIMSS, IGLU environmental improvements (e.g., in health and education), and different
and IQB over the last 20 years, ability levels are falling (Fig. 2). Between limits to further improvements (easier lower-level environmental im­
2010 and 2019, there was a decrease of − 14 points on the student provements, easier lower-level cognitive improvements, and different
assessment scale (SASQ, M = 500, SD = 100) equivalent to − 2.10 IQ in finishing lines or target points). These factors also interact, for example,
the intelligence test scale (IQ, M = 100, SD = 15). If we extrapolate that further environmental improvement and easier cognitive growth at
for another 80 years, Germany would be at an IQ of 82 in the year 2100. lower levels lead to reduced differences between natives and immi­
The consequences could be dramatic. They may not be as big as por­ grants, within societies across generations, and between societies as a
trayed in the 2006 science fiction comedy “Idiocracy”, but given the difference between old residents and new immigrants.
theoretically and empirically well-supported premises, the implications However, any theory-based prediction is difficult and delicate for
are bound to be negative. two reasons: Initially, we have limited knowledge and all models will
Is this just a temporary decline? Or the tipping point from which only take into account a limited set of factors. For example, the aging of
things go irreversibly downhill (Dutton & Woodley of Menie, 2018)? societies has not been taken into account so far, partly because tests are
How can we predict future trends? Other researchers such as Lynn usually administered only to younger samples. However, the test per­
(2011) or Nyborg (2012) have given general overviews or presented formance of middle-aged and older individuals is also relevant for so­
individual estimates for individual countries, we try to present results cietal development (e.g., Hanushek et al., 2015). In particular, there are
for a wide range of countries from developed to developing countries. As also political-ideological problems: If asymmetric child rates and
can already be seen in Fig. 1, two statistical models strive with each immigration are relevant conditional factors, but both are seen as
other: the assumption of linear progress (the straight line) and the politically incorrect (“dysgenics”, “eugenics”, “racism”, “xenophobia”)
assumption of a curvilinear relationship (the curve). and therefore not addressed, then models become fundamentally wrong.
More fundamentally, two competing prediction approaches can be (2) Second, an omission of theoretical considerations and a pure use
distinguished: of statistical methods of prediction based on an extrapolation of past devel­
(1) First, the use of theoretical models based on causal theories and the opment. The prerequisite for this is the existence of multiple measure­
use of knowledge (or assumptions) about the development of these de­ ment time points, the more the better. This requirement is also a
terminants. For example, there are assumptions about the development disadvantage because for many countries, including very important ones
(China, India, nearly entire sub-Saharan Africa), such measurements are
not available in sufficient numbers. In addition, one cannot completely
dispense with assumptions. First, there is the basic assumption that past
development can serve as a benchmark for future development. Then
one also has to make decisions regarding the kind of development, such
as linear or nonlinear. Finally, plausibility considerations must not be
ignored. For example, an IQ of 200 or even a negative IQ is to be
excluded.
We now attempt to set up such an alternative, theory-reduced model
of prediction. To do this, we use three different statistical methods:
Calculating average trends between 1995 and 2018, using linear
regression (independently done by the first and second author), and
using nonlinear quadratic regression. These values are adjusted with an
outlier correction. At the end, the mean corrected statistical prediction
values are compared with predictions based on theoretical models.

Fig. 1. Intelligence development based on spatial perception between 1976

and 2015 in Austria (Pietschnig & Gittler, 2015; reprinted with permission by
Elsevier): ability went up until 1997, since then down.

2
H. Rindermann and D. Becker Personality and Individual Differences 206 (2023) 112110

Fig. 2. Development of the results in student achievement studies in Germany between 2000 and 2019 (Woessmann, 2021; reprinted with permission of the author).

2. Method attendance rates), TIMSS and PIRLS scores were similarly corrected for
youth not in the school system and for age (whether students in grade 4
2.1. Large scale student assessment studies as data source for cognitive or 8 compared to other countries were relatively old or young). For
ability several countries for which, for example, only regional data were
available (such as China or Venezuela) or obviously erroneous data
Student achievement tests are a good indicator of cognitive ability, (Kazakhstan), further corrections were made (for a detailed description,
additionally, causes and consequences of student achievement and in­ see Rindermann, 2018, Appendix). However, since several measurement
telligence are similar (Kaufman et al., 2012; Rindermann & Baumeister, points were necessary to be included in the analyses and problematic
2015). Similar to intelligence tests, PISA scales show a strong g factor data were mostly available from those countries that participated rather
(about 70 % of variance; Pokropek et al., 2021). PISA scales are also rarely, the remaining country sample is characterized by relatively
highly correlated with intelligence test scales (the German CogAT, at the fewer corrections (China and Venezuela, for example, have too few
observed level ro = .47, at the latent level rl = .86; Brunner, 2008, p. measurement points).
158). By calculating the mean (across the dimensions of reading, To be included in the development measurements along time
mathematics, and science) specific factors are averaged out. We drew on (regardless of the specific methods trend, linear or nonlinear regression),
the following sources for abilities: at least five measurements had to be given per country. The total sample
TIMSS (Trends in International Mathematics and Science Study) of 104 countries with results in the aforementioned student achievement
measures competence in mathematics and science for mostly fourth and studies thus dropped to 79 countries.
eighth graders and, depending on school enrolment age, for third and The mean score in the PISA 2018, TIMSS 2019, and PIRLS 2016
seventh graders in some countries. TIMSS focuses on core aspects of student achievement studies was taken as the starting value.
curricula in different countries, with greater emphasis on the curricula
of developed countries. Data from 1995 to 2019 were used (1995, 1999,
2003, 2007, 2011, 2015 and 2019). Data for 4th and 8th grades were 2.2. Statistical analysis
treated separately. Sources were international reports (e.g., Mullis et al.,
2020). In 1999, only 8th grade students were tested. In studies with different dimensions (TIMSS, PISA), the results of the
PISA (Programme for International Student Assessment) from 2000, different dimensions were first combined into an overall result.
2003, 2006, 2009, 2012, 2015 and 2018, 15-year-old students in
reading, mathematics and science, 2003 also problem solving, di­ 2.2.1. Trend analysis
mensions (tasks) are not closely related to the content taught in school, “Trend analysis” means that differences between two measurement
but represent a more general “literacy”. All student achievement tests points were calculated by subtraction. One example: Trend PISA
use scales with a mean of 500 and a standard deviation of 100. Sources 2018–2015 is the PISA average 2018 minus the PISA average 2015.
were international PISA reports (e.g., OECD, 2019). Because seven measurement time points are available from PISA, 21
PIRLS (Progress in International Reading Literacy Study) from 2001, differences could be calculated (e.g., 2015 minus 2012 and 2009 minus
2006, 2011 and 2016 (fourth graders in reading, not close to school 2000). Each difference was converted to a decadal (dec) 10-year value
curricula, “literacy”; e.g., Mullis et al., 2017). (e.g. ((2009 − 2003) / 6) × 10). At the end, the differences were
All student achievement test data were used as corrected, i.e., PISA arithmetically averaged.
scores were corrected for youth not in the school system (school Mean differences (unweighted trends) were calculated separately for
each of the PISA, TIMSS 4th grade, TIMSS 8th grade, and PIRLS studies.

3
H. Rindermann and D. Becker Personality and Individual Differences 206 (2023) 112110

The four trends were then (unweighted) averaged. A maximum of 63 the estimates from the first available measurement. To arrive at
different trends can be calculated (21 in TIMSS 8, 15 in TIMSS 4, 21 in consistent values here, for the second author's linear (and nonlinear)
PISA and 6 in PIRLS). In a second variant, an outlier correction was estimates, the increases from 2020 onward were summed to the speci­
applied (see below) within each study (TIMSS 4, TIMSS 8, PISA, PIRLS). fied starting value in 2020 (the mean of all school performance studies).
Values in the student achievement scale (M = 500, SD = 100) were The very high correlations for the years 2030, 2050 and 2100 have even
arithmetically converted to IQ scores (minus 500, divided by 100, increased slightly (r = .999, .998 and .998). The average mean differ­
multiplied by 15, plus 100) resulting in the usual IQ scale (M = 100, SD ence for the years 2030, 2050 and 2100 was now only 0.20 IQ points.
= 15). As a common benchmark, results were compared with the
average of British natives that was set to an IQ of 100 (“Greenwich-IQ”). 2.2.3. Nonlinear regression
Values for years 2030, 2050, and 2100 were obtained simply by The second author additionally set up a nonlinear quadratic regres­
adding the trends (per decade) from the baseline (set at 2020). This sion (y = a + bx + cx2) for the same data set described above. The
analysis was done by the first author. different formulas here for each country, respectively their coefficients,
were used to predict the mean IQs for the years 2030, 2050, and 2100. In
2.2.2. Linear regression a second variant (suggested by a reviewer), we used a hierarchical
The first step was to bring the 24 measurement time points (6 from method: We started with a linear regression for all countries. Then we
TIMSS-4, 7 from TIMSS-8, 7 from PISA, 4 from PIRLS) onto a common added a nonlinear quadratic regression, but its values were used only for
scale. No adjustments were made within the studies (TIMSS 4, TIMSS 8, those countries when the explained variance increased by 5 % (Δ R2 >
PISA, PIRLS), that is, we assumed constant, comparable scales. How­ .05). This was the case in 45 % of the countries (nonlinear quadratic).
ever, adjustments were made between studies (TIMSS 4, TIMSS 8, PISA, Because we later computed an average from the various calculation
PIRLS). As far as possible, the same years were taken for finding the methods, including the linear and nonlinear methods, we did not pursue
adjustments. The adjustment formulas were calculated in common this conditional nonlinear quadratic regression approach any further.
country samples.
PIRLS 2001 was adjusted in mean and standard deviation to PISA 2.2.4. Outlier correction
2000 (PISA more countries, older students, more dimensions). The value Due to plausibility considerations, an outlier correction was made:
by which PIRLS 2001 was adjusted was adopted for all other PIRLS Trend analysis method: For changes within 10 years equal or greater
measurement points. than ±40 SASQ (or d = ±0.40), an outlier correction was applied.
TIMSS grade 4 2003 was adjusted in mean and standard deviation to “SASQ” means student achievement study quotient with a mean of 500
PISA 2003 (PISA more countries, older students, more dimensions). The and a standard deviation of 100. Changes equal to or >40 SASQ but <50
value by which TIMSS grade 4 2003 was adjusted was adopted for all SASQ were multiplied by a factor of 0.6. So reduction by 40 %. Changes
other TIMSS grade 4 measurement points. equal to or >50 SASQ but <60 SASQ were multiplied by a factor of 0.3.
The same was done for TIMSS grade 8: TIMSS grade 8 2003 was So reduction by 70 %. Changes equal to or >60 SASQ were multiplied by
adjusted in mean and standard deviation to PISA 2003 (PISA more a factor of 0.15. This results in a reduction of 85 % (of values above 60
countries, more dimensions). The value by which TIMSS grade 8 2003 SASQ). At the end, all SASQ scores were converted to IQ scores.
was adjusted was adopted for all other TIMSS grade 8 measurement Regression analysis method: Here we applied a similar outlier correc­
points. tion (adapted to the IQ scale): Changes equal to or >6 IQ in 10 years but
When multiple measurements were available for a year (e.g., from <7.5 IQ were multiplied by a factor of 0.6 (reduction by 40 %). Changes
PISA and TIMSS 4 in 2003), these values were arithmetically averaged equal to or >7.5 IQ but <9 IQ were multiplied by a factor of 0.3
(after common scaling). At the end, measurements were available for 14 (reduction by 70 %). Changes equal to or >9 IQ were multiplied by a
years (1995, 1999, 2000, 2001, 2003, 2006, 2007, 2009, 2011, 2012, factor of 0.15 (reduction of 85 % of values above 9 IQ).
2015, 2016, 2018, 2019). However, values were available for different
countries at different measurement times. The most measurements were 2.2.5. Regions
available for Bulgaria, Canada, the Czech Republic, Great Britain, Hong Countries were assigned to regions based on geographical and cul­
Kong, Hungary, New Zealand, Russia and the USA (14 each). The fewest tural criteria. For example, Singapore was assigned to East Asia. For each
measurements came from Belarus, Belize, Brunei, El Salvador, region, a sample country was selected and values were reported for that
Honduras, Mauritius, Mongolia, Pakistan and Venezuela (one mea­ country in the results tables. Averages were not weighted by country
surement each). These were all excluded, as were large and important population size, nor was the global mean (at the end of each table). The
countries such as China (4 measurement time points) and India (2 number of countries per region is given in Tables 1 to 3. The most
measurement time points). 79 countries had sufficient data. countries are found for North Africa and the Middle East (15 countries,
We independently calculated two linear regressions (y = a + bx), the e.g. Egypt and Iran), then for Southeast Europe (11 countries, e.g.
first author using SPSS, the second author using R. A separate regression Greece, Romania and Israel). Israel and Singapore are examples of
formula was created for each country and the coefficients found here categorization based not only on geographical but also on cultural
were used to forecast the mean cognitive abilities of societies on the IQ criteria. The fewest countries are found for Central-South Asia (only
scale for the years 2030, 2050 and 2100. In order to increase the number Kazakhstan, e.g., India has not enough measurements), sub-Sahara Af­
of data points per country and thus improve the accuracy of the linear rica (2 countries, Botswana and South Africa), North-English America (2
and polynomial regression formulas formed (see next section), the sec­ countries, Canada and USA) and Australia-New Zealand (both 2
ond author included for his analysis additional annual results for some countries).
countries: Whenever a country temporarily dropped out of a study if
there was only one wave, the mean value was calculated from the pre­ 3. Results
ceding and following waves, e.g. from PISA 2012 and 2018, if PISA 2015
was missing. Both methods came to very similar results. Estimates for Table 1 shows the results for the 2020 baseline and 2100 estimates
2030, 2050, and 2100 were correlated with r = 0.982, 0.989, and 0.995, based on trend analysis, first-author linear regression analysis, second-
respectively (79-country sample). However, the average difference be­ author linear regression analysis, nonlinear regression analysis and the
tween the two estimates was about 3.35 IQ points (first author's values conditional linear-nonlinear analysis. The two estimates based on linear
higher), even for the year 2030. The reason was that the first author ran regression correlate with r = .998. The trend based estimate correlates at
his estimate from 2020 (2020 was set as the starting value, the general r = 0.840 with the mean linear estimate. However, the nonlinear 2100
student achievement mean 2016 to 2019), while the second author ran estimate does not correlate with the trend 2100 estimate (r = − 0.095)

4
H. Rindermann and D. Becker Personality and Individual Differences 206 (2023) 112110

Table 1
Comparison of models (trends-subtractions, linear regressions, nonlinear-quadratic regression, conditional linear-nonlinear), uncorrected values for the year 2100.
CA CA CA CA CA CA
2020 2100 2100 2100 2100 2100
Trend Linear A1 Linear A2 Nonlinear lin/nonlin

Africa (sub-Sahara, 2) 68.75 74.91 99.23 99.31 1177.55 1177.55

South Africa 67.07 70.44 122.89 123.07 1052.67 1052.67
N-Africa M-East (15) 82.38 114.38 115.89 115.60 421.20 334.80
Egypt 74.77 67.73 23.48 24.37 802.77 802.77
America (North-E, 2) 99.17 99.53 99.27 98.13 129.25 98.13
USA 98.32 101.53 101.68 99.11 199.92 99.11
America (Latin-CS, 8) 84.74 120.60 131.29 131.84 189.14 261.54
Mexico 82.46 111.80 111.40 111.26 13.66 111.26
Asia (Central-S, 1) 89.19 114.88 138.03 136.39 − 1398.81 − 1398.81
Kazakhstan 89.19 114.88 138.03 136.39 − 1398.81 − 1398.81
East Asia (6) 103.26 113.89 107.18 107.29 332.41 333.74
Japan 103.61 101.87 96.25 96.97 274.01 274.01
Southeast Asia (3) 83.12 76.72 67.76 67.38 508.72 508.72
Indonesia 79.06 89.12 95.35 95.06 461.46 461.46
Australia-NZ (2) 97.66 93.80 90.43 89.46 87.54 89.46
New Zealand 96.84 85.27 89.02 87.24 11.80 87.24
Western Europe (5) 97.79 94.43 93.60 93.51 51.95 34.51
United Kingdom 99.57 105.09 102.04 101.25 193.97 101.25
Scandinavia (5) 98.25 97.44 96.71 96.50 74.41 96.50
Norway 98.21 115.38 109.09 108.61 318.21 318.21
Central Europe (5) 97.91 106.46 108.17 105.49 − 171.53 − 36.01
Germany 97.89 99.24 113.92 99.49 − 170.91 − 170.91
Eastnorth Europe (9) 98.13 107.76 106.04 105.81 169.10 181.06
Russia 98.94 118.28 110.18 109.34 385.34 385.34
Southwest Europe (5) 94.32 99.81 106.65 106.51 180.62 248.91
Italy 95.18 94.39 97.03 96.94 − 91.22 96.94
Southeast Europe (11) 89.53 96.84 103.46 102.75 292.01 235.82
Greece 92.37 89.25 79.83 79.57 98.29 79.57
Average (79) 91.29 104.70 107.20 106.80 229.78 226.80

Notes. CA: cognitive ability in an IQ-scale (Greenwich-IQ); 79 countries, in parentheses number of countries per region, (e.g., for sub-Saharan Africa the 2 countries
Botswana and South Africa). The division into regions is based on geographical and cultural criteria. An example country is presented for each region. N-Africa M-East:
North Africa, Middle East; North-E: North, English; CS, Central-S: Central-South, Australia-NZ: Australia-New Zealand, also called Trans-Tasman; Trend: estimation for
2100 based on trends (subtraction) between measurement points; Linear A1: estimation for 2100 based on linear regression of student assessment results between 1995
and 2019, done by the first author; Linear A2: the same, done by the second author; Nonlinear: estimation for 2100 based on nonlinear quadratic regression of student
assessment results between 1995 and 2019; lin/nonlin: hierarchical linear/nonlinear quadratic regression (if additional R2 > .05).

and not with the mean linear 2100 estimate (r = − 0.080). The results of 107.20 and 106.80). The result of the estimation based on the nonlinear
the conditional method (nonlinear if additional R2 > 0.05) correlates quadratic regression model deviates greatly from this (M = 229.78,
only (and highly) with the nonlinear 2100 estimate (r = 0.914, similarly the conditional). Results depend strongly on whether a linear
Spearman r = 0.780), not with the linear (r = − 0.042, Spearman r = or nonlinear development is assumed. Likewise, the direction of devel­
0.147). If for outliers corrected values are used the correlations are opment (whether positive or negative) often differs from each other.
similar: trend with linear r = 0.799, trend with nonlinear r = 0.107, There is agreement on the direction of development for the following
linear with nonlinear r = − 0.078, nonlinear with conditional r = 0.838. regions: increases for sub-Saharan Africa, North Africa and the Middle
The uncorrected values correlate with the corrected values at r = 0.975 East, for Latin America, East Asia, Eastnorth Europe, Southwest Europe
(trend), r = 0.987 (linear), r = 0.717 (nonlinear) and r = 0.689 (linear/ and Southeast Europe; and decreases for Trans-Tasman, Western Europe
nonlinear). and Scandinavia. A positive development is predicted for Central Europe
Estimates based on the nonlinear formula led to much more extreme using a linear regression, but a negative one using a non-linear
results (see Table 1). For example, an IQ of 1053 was predicted for the regression.
year 2100 in South Africa and − 1399 for Kazakhstan (see Discussion for Table 2 shows the estimates for the years 2030, 2050 and 2100 based
plausibility considerations). We calculated the mean absolute gains on the arithmetic mean of the three estimation methods trend, linear
(estimate 2100 minus cognitive ability measure of 2020) for each regression (mean value of the two models of the first and second author)
method as an indicator of statistical extremity. The values were 19.85 IQ and nonlinear regression (not conditional) as well as for the uncorrected
points (trend), 24.54 (linear 1), 24.05 (linear 2), 24.29 (linear mean), vs. corrected estimates. In the last columns on the left and right, the
but 309.91 IQ points for the estimation based on the nonlinear model. mean increases per decade (dec in IQ) between 2020 and 2100 are
The more data are given for a country in TIMSS, PISA and PIRLS, the less shown.
extreme were the estimates, correlation between number of measure­ The outlier correction we applied is very effective. While the un­
ment points and uncorrected mean gains (mean of the three methods) corrected absolute mean IQ gains per decade ('20–100) are 13.10 IQ
was r = − .38 (N = 79 countries). If one takes the number of possible points (left middle column of Table 2, there an arithmetic mean), the
comparisons (subtractions in the trend analysis, e.g. comparison be­ corrected ones are much smaller at 1.78 IQ points (last left column of
tween PISA 2012 and 2009 and between 2012 and 2006) instead of the Table 2). The average gains (positive and negative values as they are)
number of measurement points, the correlation is somewhat higher (r = are 6.98 IQ points and 1.25 IQ points (last row of Table 2). The outlier
− .44, N = 79). From this it can be concluded that more data leads to correction leads to significantly less extreme results with the nonlinear
higher reliability of the estimates. model (uncorrected SD = 229.78, corrected SD = 95.43, compared to
Looking at the international means, the results of the trend analysis the linear model: uncorrected SD = 106.80, corrected SD = 105.73).
and the two linear regression models are quite similar (M = 104.70, Again, the more data are given (as measurement points or as

5
H. Rindermann and D. Becker Personality and Individual Differences 206 (2023) 112110

Table 2
Comparison of uncorrected and corrected models' means.
Uncorrected models' mean Corrected models' mean

CA CA CA CA IQ gain decade CA CA CA IQ gain decade

2020 2030 2050 2100 ('20–100) 2030 2050 2100 ('20–100)

Africa (sub-Sah, 2) 68.75 86.18 146.96 450.58 47.73 75.14 79.30 89.68 2.62
South Africa 67.07 84.96 142.69 415.36 43.54 74.20 79.82 93.88 3.35
N-Africa M-East (15) 82.38 90.75 114.50 217.11 16.84 87.02 91.69 103.43 2.63
Egypt 74.77 79.32 108.51 298.14 27.92 75.95 72.21 62.84 − 1.49
America (North, 2) 99.17 99.56 101.07 109.16 1.25 99.56 99.75 99.86 0.09
USA 98.32 99.99 105.69 133.95 4.45 99.99 101.13 103.26 0.62
America (Latin, 8) 84.74 92.67 108.14 147.10 7.80 88.89 94.61 110.16 3.18
Mexico 82.46 84.98 87.25 78.93 − 0.44 84.89 88.75 98.65 2.02
Asia (Central, 1) 89.19 79.40 17.50 − 382.24 − 58.93 87.41 90.56 98.43 1.16
Kazakhstan 89.19 79.40 17.50 − 382.24 − 58.93 87.41 90.56 98.43 1.16
East Asia (6) 103.26 107.01 119.91 184.51 10.16 106.12 107.79 111.65 1.05
Japan 103.61 105.44 113.31 157.50 6.74 105.37 105.63 104.37 0.10
Southeast Asia (3) 83.12 87.46 106.88 217.67 16.82 85.61 84.49 81.69 − 0.18
Indonesia 79.06 85.87 108.17 215.26 17.03 83.64 86.30 92.97 1.74
Australia-NZ (2) 97.66 96.79 95.06 90.43 − 0.90 96.79 95.83 93.56 − 0.51
New Zealand 96.84 94.49 88.13 61.73 − 4.39 94.49 92.24 87.01 − 1.23
Western Europe (5) 97.79 96.72 93.61 79.98 − 2.23 96.78 96.02 94.54 − 0.41
United Kingdom 99.57 101.28 106.84 133.57 4.25 101.28 102.60 105.24 0.71
Scandinavia (5) 98.25 97.80 96.36 89.48 − 1.10 97.90 97.53 96.68 − 0.20
Norway 98.21 102.80 116.75 180.82 10.33 102.45 104.88 110.96 1.59
Central Europe (5) 97.91 95.15 82.94 13.92 − 10.50 96.45 97.37 99.37 0.18
Germany 97.89 95.05 82.56 11.68 − 10.78 95.28 95.78 97.05 − 0.10
Eastnorth Europe (9) 98.13 100.12 105.48 127.59 3.68 99.95 101.46 104.92 0.85
Russia 98.94 104.53 122.08 204.46 13.19 103.47 106.26 113.24 1.79
Southwest Europe (5) 94.32 96.23 102.08 129.00 4.34 94.93 96.11 99.67 0.67
Italy 95.18 92.82 83.51 33.39 − 7.72 92.98 92.15 91.77 − 0.43
Southeast Europe (11) 89.53 93.48 105.93 163.98 9.31 92.47 94.39 99.20 1.21
Greece 92.37 91.43 90.02 89.08 − 0.41 91.43 90.26 87.35 − 0.63
Average 91.29 95.07 105.26 147.16 6.98 93.51 95.69 101.25 1.25

Table 3
Comparison of two former theoretical models with the statistical model.
Theoretical model Theoretical model Statistical model
(2011) (2018) (Corrected mean)

CA CA IQ gain decade CA IQ gain decade CA IQ gain decade

2020 2100 ('10–100) 2100 ('10–100) 2100 ('20–100)

Africa (sub-Sahara, 6/2) 68.75 78.14 0.01 85.35 1.58 89.68 2.62
South Africa 67.07 70.30 0.51 87.84 1.97 93.88 3.35
N-Africa M-East (19/15) 82.38 83.49 0.24 92.01 0.99 103.43 2.63
Egypt 74.77 81.71 − 0.11 86.96 0.36 62.84 − 1.49
America (North-E, 2/2) 99.17 98.63 − 0.20 97.07 − 0.26 99.86 0.09
USA 98.32 97.26 − 0.19 95.77 − 0.28 103.26 0.62
America (Latin-CS, 11/8) 84.74 83.17 0.29 93.33 1.06 110.16 3.18
Mexico 82.46 82.95 − 0.24 93.57 0.86 98.65 2.02
Asia (Central-S, 3/1) 89.19 83.67 0.33 91.69 1.14 98.43 1.16
Kazakhstan 89.19 85.03 0.19 92.54 0.33 98.43 1.16
East Asia (8/6) 103.26 100.70 − 0.23 100.08 − 0.12 111.65 1.05
Japan 103.61 102.50 − 0.31 100.42 − 0.36 104.37 0.10
Southeast Asia (4/3) 83.12 86.70 0.03 91.44 0.62 81.69 − 0.18
Indonesia 79.06 83.84 0.21 94.09 1.09 92.97 1.74
Australia-NZ (2/2) 97.66 99.84 − 0.14 98.94 − 0.01 93.56 − 0.51
New Zealand 96.84 99.24 − 0.14 98.65 − 0.06 87.01 − 1.23
Western Europe (5/5) 97.79 98.34 − 0.24 94.33 − 0.50 94.54 − 0.41
United Kingdom 99.57 98.69 − 0.15 95.67 − 0.44 105.24 0.71
Scandinavia (5/5) 98.25 97.78 − 0.20 96.01 − 0.32 96.68 − 0.20
Norway 98.21 96.74 − 0.02 97.49 − 0.05 110.96 1.59
Central Europe (5/5) 97.91 97.45 − 0.31 91.71 − 0.79 99.37 0.18
Germany 97.89 97.22 − 0.29 93.35 − 0.61 97.05 − 0.10
Eastnorth Europe (10/9) 98.13 97.17 − 0.18 96.66 − 0.05 104.92 0.85
Russia 98.94 96.39 − 0.20 95.88 − 0.15 113.24 1.79
Southwest Europe (5/5) 94.32 92.58 − 0.30 94.65 − 0.16 99.67 0.67
Italy 95.18 93.01 − 0.44 94.10 − 0.38 91.77 − 0.43
Southeast Europe (11) 89.53 89.78 − 0.15 94.70 0.41 99.20 1.21
Greece 92.37 91.49 − 0.35 91.00 − 0.44 87.35 − 0.63
Average (97/79) 91.29 90.21 − 0.03 93.88 0.41 101.25 1.25
(Standard deviation) (8.61) (7.95) (0.38) (3.85) (0.82) (15.72) (2.01)

Notes. 97 (two former models) 79 (present model) countries, in parentheses number of countries per region, (e.g., for sub-Saharan Africa 6 or 2). The former estimations
did not require multiple measurements from different years, so the number of countries is higher.

6
H. Rindermann and D. Becker Personality and Individual Differences 206 (2023) 112110

comparisons between them) the less extreme are the gains (on average r averages at the beginning of the 21st century (rth2011 = .95, rth2018 = .66,
= − .42). rst2022 = .23; N = 79). The order of regions in mean cognitive levels
Based on past development between 1995 and 2019 for the following hardly changes according to the theoretical models, but it does some­
regions a positive cognitive national development is expected in 21st what according to the statistical model: According to the 2011 theo­
century: sub-Saharan Africa, North Africa and the Middle East, North retical model, East Asia, Australia-New Zealand, and North America are
America, Latin America, East Asia, Eastnorth Europe, Southwest Europe leading in 2100 (with IQs around 100), similar as in the 2018 theoretical
and Southeast Europe. A negative development is expected for: Trans- model. However, according to the statistical model, East Asia is followed
Tasman, Western Europe and Scandinavia. The development for Cen­ by Latin America, East-North Europe, and the Middle East in 2100.
tral Asia is unclear (we only have data from one country, Kazakhstan), Table 4 shows the correlations among the three starting values (IQ
for Southeast Asia and Central Europe. In all of these regions, outlier values based on tests, models 2011, 2018 and 2020), the three estimates
correction leads to much smaller (absolute) values for the nonlinear for the year 2100 and the three decadal increases. The calculated
estimates. Thus, the final uncorrected and corrected estimates differ average IQs for the beginning of the 21st century are quite similar
greatly. For Southeast Asia, the (across three methods averaged) (correlations around r = .94), but the 2100 statistical estimates do not
development is nonlinear (it is at least somewhat nonlinear everywhere, correlate with theoretical 2100 estimates (r = .02 and .24). After all, the
but is particularly evident here): The corrected estimate for 2030 shows decadal increases correlate positively (r = .59 and .63). The 2100 esti­
an increase (from 2020 83.12 IQ points to 85.61) but then a decrease mates and decadal gains of the two theoretical models correlate well (r
(2050: 84.49 and 2100 81.69 IQ points). = .66 and .71). The means show for the statistical model a much more
Following the corrected statistical estimates the regions with the positive cognitive societal development in 21st century (in the same 79
highest IQs worldwide in 2100 would be East Asia (111.65 IQ points), country sample estimates for 2100: IQs 91.67, 94.48 vs. 101.25). Simi­
Latin America (110.16) and Eastnorth Europe (IQ 104.92). The regions larly, the mean decadal gains differ: − .08 and .23 vs. 1.25. According to
with the lowest were Southeast Asia (IQ 81.69), sub-Saharan Africa (IQ the theoretical models the countries become more similar, but not ac­
89.68) and Trans-Tasman (IQ 93.56). cording to the statistical model. In all models, gains are higher for
The worldwide development seems to be positive according to the countries at lower ability levels (correlation between starting values and
statistical model. According to the corrected estimate, global IQ in­ gains: rth2011 = − .76, rth2018 = − .92, rst2022 = − .31; N = 79).
creases from 91.29 in 2020 to 93.51 in 2030, then reaches 95.69 in
2050, and finally 101.25 IQ points in 2100. 4. Discussion
Table 3 compares the results of the two former theoretical models
(presented in 2011 and 2018) based on assumptions (asymmetric In this study, we present the results of a statistical prediction of the
fertility, migration, environmental improvements, different limits to development of intelligence on a societal level in the 21st century.
further improvements) and of the current statistical model based on past “Statistical” means that the past development between 1995 and 2019 is
development (between 1995 and 2019, average of the three methods used as the basis for the prediction. No explicit theoretical assumptions
trend, linear and nonlinear, corrected). While a slight decrease or only a were formulated. Three methods were used: Calculation of average
small increase is expected in the theoretical models (from 91.29 in 2020 gains (trends) by subtracting all possible test time points within a study
to IQ 90.21 or 93.88 in 2100), according to the statistical model an in­ (within TIMSS grade 4, TIMSS grade 8, PISA and PIRLS; a maximum of
crease of about 10 IQ points is expected worldwide in the next 80 years 24 measurement points) and then averaging them. The result is the
(IQ 101.25). Note that in Table 3 the country samples differ slightly (97 average gain over 10 years (dec) converted to an IQ point scale. Second,
and 79 countries; for equal country samples, see Table 4). all student assessment scores were converted to a common scale and the
The models based on theoretical assumptions also lead in their es­ (maximum) 14 measurement years between 1995 and 2019 were used
timates to a convergence of countries. The standard deviations are to calculate a linear regression formula for each country (y = a + bx).
declining (from 8.61 in 2020 to 7.95 or 3.85 in 2100). On the other Third, for the same student assessment data a nonlinear regression for­
hand, the statistical model leads to an increase in the differences be­ mula (y = a + bx + cx2) was calculated. (In a variant, we conditionally
tween countries (SD = 15.72). On closer inspection, this is because, combined the linear with the nonlinear method.) The results were cor­
according to the theoretical models, developing countries are catching rected for outliers and averaged.
up and developed countries can no longer continue to make cognitive According to the average (corrected for outliers) of the three
gains or are even experiencing declines: For example, sub-Saharan Af­ methods, worldwide IQ will increase by about 10 IQ points from 2020
rica, the Middle East, Latin America, and Southeast Asia are catching up, (IQ 91) until 2100 and then reach 101 IQ points (dec = 1.25 IQ points).
while Europe, North America, Trans-Tasman, and East Asia remain However, the results heavily depend on chosen prediction methods,
stable or experience a cognitive decline (particularly Central Europe in outlier correction and region: Estimates based on the nonlinear formula
the 2018 theoretical model, from an IQ of 97.91 in 2020 to 91.71 in lead to much more extreme results (e.g., IQ 1178 for sub-Saharan Africa
2100). According to the statistical model, developing countries are also in 2100 or IQ − 1399 for Central Asia). Such a difference depending on
catching up here, but some European countries, North America and the linear or nonlinear equation could already be seen in the study with
especially East Asia are also showing positive development. The sharp data from German-speaking countries (Pietschnig & Gittler, 2015).
increase in Latin American scores (from IQ 84.74 in 2020 to 110.16 in Trend and linear models produce results that are much more similar to
2100) further increases country differences. expert opinions (Rindermann et al., 2017). As intended, the outlier
Estimates based on theoretical models correlate more closely with IQ correction leads to significantly less extreme results, especially for the

Table 4
Correlations between starting values, 2100 estimates and decadal increases, additionally means and standard deviations.
Starting values 2100 estimates Decadal increases

2011 (theo.) 2018 (theo.) 2022 (stat.) 2011 (theo.) 2018 (theo.) 2022 (stat.) 2011 (theo.) 2018 (theo.) 2022 (stat.)

2011 (theoretical) 1 .96 .92 1 .66 .02 1 .71 .59

2018 (theoretical) 1 .95 1 .24 1 .63
Means 92.38 92.44 91.29 91.67 94.48 101.25 − 0.08 0.23 1.25
(and SD) (9.36) (7.65) (8.61) (7.45) (3.17) (15.72) (0.31) (0.68) (2.01)

Notes. Number of countries reduced to 79 (all correlations, means and standard deviations are given for the same country sample).

7
H. Rindermann and D. Becker Personality and Individual Differences 206 (2023) 112110

nonlinear model. In all models, East Asia still leads in cognitive ability in one-quarter of the teachers cannot subtract double-digit numbers [in
2100, while in sub-Saharan Africa the results are relatively weak. In Togo] and one-third of the teachers cannot multiply double-digit
addition, all methods project that increases in the 21st century will be numbers [in various African countries].” (Bold et al., 2017, pp. 192f.)
larger for countries with currently lower levels of test results. As “In Kenya, Tanzania, and Uganda, when grade 3 students were asked
described for past empirical data (Meisenberg & Woodley, 2013) lower recently to read a sentence such as ‘The name of the dog is Puppy’, three-
achieving countries may catch up. There appears to be a specific his­ quarters did not understand what it said.” (World Bank, 2018, p. 3)
torical time frame for the FLynn effect: The slowdown of the FLynn ef­ Examples can also be found in the geographical centers of anywheres, at
fect in developed countries, but its persistence in developing countries, international airports, e.g. at Frankfurt Airport:
may lead to a reduction in international disparities.
“A recorded female voice advised us to board the plane from the front
Several results of the predictions contradict expert opinions and
if we have a ‘low seat number’, and from the back, in case we had a
plausibility considerations, e.g. IQs above 200 or negative values. The
‘high seat number’. I chuckled because this is of course idiotic. No, of
outlier corrections – based on plausibility considerations – reduced such
course it is not idiotic to board the plane that way. However, if you
implausible results. But what is plausibility? Plausibility considerations
do not know how many rows the plane you are about to board has, it
here are assumptions about the magnitude of the changes and about the
is quite unclear if your seat number is indeed high. ... Just consider
pattern of differences between regions. Since there are no individuals (or
how much has to go wrong to make all of this happen. The Frankfurt
even tests that can measure this) with scores above 200 or below 0, such
Airport is run by a large organization. ... Yet, nobody of the parties
results are also not realistic at the national level. It is therefore simply
involved thought it prudent to point out that a key bit of information
impossible that a linear or nonlinear increase or decrease found between
is missing and that their work is not particularly useful. ... There were
1995 and 2019 will continue for decades. The other aspect concerns the
probably at least five people involved, the guy having the idea, some
order of regions and differences between them. Although the absolute
PR people, the person recording the voice track, some empty suits in
values are not stable over time (FLynn effect), the patterns are to a
meetings who discuss this project repeatedly. Yet, all of them
certain extent stable: Past technological and intellectual development,
thought that they are about to produce a fine piece of work.”
even up to two thousand years back, are positively correlated to per­
(Elias, 2019)
formance on student achievement and intelligence tests today (Comin
et al., 2010; Mokyr, 2005; Murray, 2003; Rindermann, 2018, p. 122). We do not have to travel that far, we do not have to go back into the
Qualitative analyses of cognitive activities and achievements in the past deep past, we do not have to visit other continents by plane, we just have
and today point in a similar direction (Rindermann, 2018, chapter to step outside the door of our Psychological Institute and go to the next
4.4.3). building, a supermarket. This week, we saw these price tags (Fig. 3):
However, doubts based on plausibility considerations do not only There is an advertisement for chocolate. On the left price tag, the
relate to future or predicted increases in intelligence, but also to past price is reduced by 33 % from 1.19 euros to 0.79 euros, and on the right
ones. Can one really assume an increase in intelligence of around 15 to from 1.15 euros by 52 % to the same 0.79 euros. First of all, it is a bit silly
30 IQ points in the 20th century, or have only the way of thinking and to put up two different price tags for the same merchandise. Then a
the ability to take tests changed? Further doubts are expressed about the glaring calculation error is found on the right. Finally, by hanging both
extent of international differences and in particular about the low in­ signs next to each other, everyone must notice this obvious mistake.
telligence and student test scores in many developing countries. What However, this is not the case. As with the airport example above,
we need here is a collection and analysis of people's everyday thinking, cognitive errors come together at different levels, in this case from
especially about people outside the university context. We as scientists management to sales staff.
often only move in a “super-WEIRD” social context with little or no Overall, this means that there are enormous cognitive differences
contact to thinking outside of this “bubble”. Super-WEIRD does not only between the past and the present, between different regions of the
mean “western, educated, industrialized, rich and democratic” but also world; and even in countries that are at the tail end of the FLynn effect of
highly intelligent – the colleagues of scientists, their university students, one or more centuries, there are striking examples of low cognitive
partners, friends and their children (by age comparison), almost ability in everyday life. At the societal level, there is still much room for
everyone with whom they spend more than a few seconds and exchange further upward cognitive development, even more in developing coun­
two words. In such a lifeworld, thinking and living at a low level of tries. 10 to 20 IQ points would certainly be conceivable. We must not be
intelligence becomes simply unimaginable, completely implausible. too cautious with our plausibility considerations.
Some examples: There were trials against animals in Europe up until the However, predictions based solely on a statistical model, i.e. pre­
16th century (Dinzelbacher, 2006): dictions based only on past developments of a few decades and extrap­
olating them into the more distant future, are very likely to be wrong. In
“Caterpillars devastated some fields in the region of Lausanne,
addition, purely statistical predictions are not possible, because one
Switzerland, in 1519. An official messenger went to them and or­
must always decide on a statistical variant, such as additive continuation
dered [them] to appear in court to a scheduled date. The judge held
(trend), linear or nonlinear development. Nevertheless, purely statistical
some of these animals in his hands and ordered them to leave the
forecasts are not unreasonable – if they are only used for a short period
region within three days.”
of time and are based on as many different models as possible (e.g.
(Oesterdiekhoff, 2009, p. 347f.)
trends, linear, nonlinear).
While that was a glimpse into the past, a look at other regions with A combination of both approaches, the statistical one, which con­
significantly weaker test results is just as revealing. In the late 1990s, the tinues the past development, and the theoretical one, which formulates
first author asked about 20 adult people in Cuba how many times a day a explicit assumptions, seems to be optimal: For the first 10 to 20 years to
broken analog watch with a dial and hands that stopped working predict, one uses a statistical model, one assumes a continuation of the
showed the correct time. This question relates to daily life in Cuba, as previous development. However, the more time progresses, the more
many clocks did not work. About 75 % said, “never shows the correct should be chosen sophisticated theoretical models for a prediction. Over
time”, 15 % “once” and 10 % “twice” (Rindermann, 2018, p. 115). In the years, the influence of statistical formulas could be tapered off and
Nigeria, about 82.5 % of surveyed urban middle-class people believed that of theoretical models gradually increased. For short-term statistical
that God can put money in one's purse “if that person really needs money forecasting, a curvilinear model should only be used if there is clear
at that moment” (73.8 % “highly agreed”; study by Luisa Falkenhayn; theoretical support for it, such as a change in population composition.
Rindermann et al., 2014). According to studies by international research The trend model that arrives at (positive or negative) mean gains via
teams, teachers and students in the same region lack basic skills: “Almost subtractions between as many different measurement points as possible

8
H. Rindermann and D. Becker Personality and Individual Differences 206 (2023) 112110

Fig. 3. Chocolate advertising (Lidl store in Chemnitz, May 11, 2022).

seems to us to be the most convincing: simple, comprehensible, 8. Countries learn from each other, and countries with culturally and
controllable. Outliers can be corrected. genetically similar populations will continue to be more similar in
The longer the prediction period, the more theoretical models must cognitive abilities.
be used. Such models must take into account: 9. The predictions can be different for different levels of ability – the
level of ability of the intellectual classes, which are particularly
1. Further opportunities to improve the environmental conditions important for the development of society (Coyle et al., 2018; Kirke­
relevant for intelligence of societies, especially at medium and low gaard & Carl, 2022), could develop less well (Dutton & Woodley of
IQ levels (within societies and between societies). Menie, 2018).
2. Differential birth rates (better-educated and more intelligent adults
have fewer children, especially women; asymmetric fertility) show Previous, less complex models need to be refined and improved; in
environmental (via family, neighborhood, schools etc.) and genetic particular, the aspect of age-related changes in intelligence should be
consequences for future generations. added. All reasonable predictions are based on observations of past
3. Different possible target points of IQ development. It is plausible to development (here at the societal level), on empirically well-supported
assume that in societies with a tradition of intellectual achievement, evidence and argumentatively supported theories for the causes of
such achievements will also be attainable in the future, and that development, and on equally well-supported assumptions for future
where results are currently rather low, but past achievements have development. Since different researchers may have different assump­
been high, particular progress can be expected in the future. tions about this, ranging from very optimistic (“The improving state of
4. Countries differ in the ability levels of their immigrants and emi­ the world”, Goklany, 2007; “It's getting better all the time”; Moore &
grants. Europe, in particular, has on average a larger share of Simon, 2000) to very pessimistic ones (“collapsing West”, Dutton &
immigration of less skilled people. Predictions of immigration and Rayner-Hilles, 2022, Chapter 1; “decay of Western civilization”, Nyborg,
emigration, as well as where people will come from and where they 2012; “eventual collapse”, Weiss, 2007), it would be good if this
will go, must be taken into account. research question were pursued by more researchers.
5. There are opportunities for catch-up development for immigrants (e. On a more technical-statistical level, the following adaptations can
g., te Nijenhuis et al., 2004), but depending on the level of immi­ be made: Trends and regressions could be calculated (and then aver­
gration, on culture and on environmental quality. aged) not for study means, but for the dimensions literacy, mathematics
6. Intelligence will not only increase or decrease in general, but also and natural sciences. The regressions themselves can also be calculated
change due to the increasing average age of the population (aging within studies (e.g., first in TIMSS-4, TIMSS-8, PISA, and PIRLS and then
societies). averaging the weights) or without averaging across studies if results are
7. Older age of parents can also lead to more mutations and better available from multiple studies in one year (e.g., 2007 from TIMSS 4 and
health systems to an accumulation of mutations over generations TIMSS 8; or 2011 from PIRLS, TIMSS 4, and TIMSS 8; or 2015 from PISA,
(Woodley of Menie & Fernandes, 2016). TIMSS 4, and TIMSS 8). The advantage could be that less extreme pre­
dictions are made thanks to the more measurement time points avail­
able. However, the goal is not so much to make dimension- or study-

9
H. Rindermann and D. Becker Personality and Individual Differences 206 (2023) 112110

specific predictions, but rather for general cognitive human capital, as Lynn, R., & Vanhanen, T. (2012). National IQs: A review of their educational, cognitive,
economic, political, demographic, sociological, epidemiological, geographic and
this general cognitive ability is most important for the development of
climatic correlates. Intelligence, 40, 226–234.
societies (e.g., Jones, 2016; Lynn & Vanhanen, 2012; Meisenberg, Meisenberg, G. (2014). Cognitive human capital and economic growth in the 21st
2014). In addition, dimension averages and study averages lead to an century. In T. Abrahams (Ed.), Economic growth in the 21st century (pp. 49–106). New
increase in the reliability and validity of the measurements and York: Nova Publishers.
Meisenberg, G., & Lynn, R. (2023). Ongoing trends of human intelligence. Intelligence, 96,
predictions. Article 101708.
Meisenberg, G., & Woodley, M. A. (2013). Are cognitive differences between countries
CRediT authorship contribution statement diminishing? Evidence from TIMSS and PISA. Intelligence, 41, 808–816.
Mokyr, J. (2005). The intellectual origins of modern economic growth. Journal of
Economic History, 65, 285–351.
Heiner Rindermann developed the idea. Heiner Rindermann and Moore, S., & Simon, J. L. (2000). It's getting better all the time: 100 greatest trends of the last
David Becker performed the statistical analyses. Heiner Rindermann 100 years. Washington, DC: Cato.
Mullis, I. V. S., Martin, M. O., Foy, P., & Hooper, M. (2017). PIRLS 2016. International
drafted the manuscript. Heiner Rindermann and David Becker revised results in reading. Chestnut Hill, Mass: TIMSS & PIRLS International Study Center.
the entire manuscript. Both authors discussed the results and contrib­ Mullis, I. V. S., Martin, M. O., Foy, P., Kelly, D. L., & Fishbein, B. (2020). TIMSS 2019
uted to the final manuscript. international results in mathematics and science. Chestnut Hill, Mass: TIMSS & PIRLS
International Study Center.
Murray, C. (2003). Human accomplishment: The pursuit of excellence in the arts and sciences,
Declarations of competing interest 800 B.C. to 1950. New York: Harper-Collins.
Nyborg, H. (2012). The decay of Western civilization: Double relaxed Darwinian
selection. Personality and Individual Differences, 53, 118–125.
There are no conflicts of interest.
OECD. (2013). OECD skills outlook 2013. Paris: OECD.
OECD. (2019). PISA 2018 results (Volume I): What students know and can do. Paris: OECD.
Data availability Oesterdiekhoff, G. W. (2009). Trials against animals: A contribution to the
developmental theory of mind and rationality. Mankind Quarterly, 49, 346–380.
Oesterdiekhoff, G. W. (2014). The rise of modern, industrial society. The cognitive-
Data will be made available on request. developmental approach as a new key to solve the most fascinating riddle in world
history. Mankind Quarterly, LIV(3&4), 262–312.
References Pietschnig, J., & Gittler, G. (2015). A reversal of the Flynn effect for spatial perception in
german-speaking countries: Evidence from a cross-temporal IRT-based meta-analysis
(1977–2014). Intelligence, 53, 145–153.
Baumert, J. (2011, April 20). Sinkende Schülerzahlen, mehr Einwandererkinder – der Pietschnig, J., & Voracek, M. (2015). One century of global IQ gains: A formal meta-
PISA-Forscher Jürgen Baumert warnt vor einem Bildungsabstieg. [Falling numbers analysis of the Flynn effect (1909–2013). Perspectives on Psychological Science, 10(3),
of schoolchildren, more immigrant children – PISA researcher Jürgen Baumert warns 282–306.
of an educational decline.] Die Zeit. Retrieved from [Link]/2011/17/C-Inte Pokropek, A., Marks, G. N., & Borgonovi, F. (2021). How much do students' scores in
rview-Baumert. PISA reflect general intelligence and how much do they reflect specific abilities?
Bold, T., Filmer, D., Martin, G., Molina, E., Stacy, B., Rockmore, [Link], W., … (2017). Journal of Educational Psychology. [Link]
Enrollment without learning: Teacher effort, knowledge, and skill in primary schools Rindermann, H. (2018). Cognitive capitalism: Human capital and the wellbeing of nations.
in Africa. Journal of Economic Perspectives, 31, 185–204. Cambridge: Cambridge University Press.
Brunner, M. (2008). No g in education? Learning and Individual Differences, 18(2), Rindermann, H., & Baumeister, A. E. E. (2015). Validating the interpretations of PISA
152–165. and TIMSS tasks: A rating study. International Journal of Testing, 15(1), 1–22.
Comin, D. A., Easterly, W., & Gong, E. (2010). Was the wealth of nations determined in Rindermann, H., Becker, D., & Coyle, T. R. (2017). Survey of expert opinion on
1000 BC? American Economic Journal: Macroeconomics, 2, 65–97. intelligence: The FLynn effect and the future of intelligence. Personality and
Coyle, T. R., Rindermann, H., Hancock, D. G., & Freeman, J. (2018). Nonlinear effects of Individual Differences, 106, 242–247.
cognitive ability on economic productivity. Journal of Individual Differences, 39(1), Rindermann, H., Falkenhayn, L., & Baumeister, A. E. E. (2014). Cognitive ability and
39–47. epistemic rationality: A study in Nigeria and Germany. Intelligence, 47, 23–33.
Dinzelbacher, P. (2006). Das fremde Mittelalter. Gottesurteil und Tierprozess. In Foreign Rindermann, H., & Pichelmann, S. (2015). Future cognitive ability: US IQ prediction
middle ages. Ordeal and trail against animals. Paderborn: Schöningh. until 2060 based on NAEP. PLoS ONE, 10(10), Article e0138412.
Dutton, E., & Rayner-Hilles, J. O. A. (2022). The past is a future country: The coming Rindermann, H., & Thompson, J. (2011). Intelligence of the future and economic
conservative demographic revolution. Exeter: Imprint Academic. development. Talk at 10. In December 2011 at the 12th Conference of the International
Dutton, E., van der Linden, D., & Lynn, R. (2016). The negative Flynn effect: A systematic Society for Intelligence Research (ISIR) in Limassol, Cyprus.
literature review. Intelligence, 59, 163–169. Rindermann, H., & Thompson, J. (2013). Ability rise in NAEP and narrowing ethnic
Dutton, E., & Woodley of Menie, M. A. (2018). At our wits‘ end: Why we’re becoming less gaps? Intelligence, 41(6), 821–831.
intelligent and what it means for the future. Exeter: Imprint Academic. Ritchie, S. J., & Tucker-Drob, E. M. (2018). How much does education improve
Egeland, J. (2022). The ups and downs of intelligence: The co-occurrence model and its intelligence? A meta-analysis. Psychological Science, 29, 1358–1369.
associated research program. Intelligence, 92, Article 101643. Rundquist, E. A. (1936). Intelligence test scores and school marks of high school seniors
Elias, A. S. (2019, January 3). The decline of collective intelligence in Germany: in 1929 and 1933. School & Society, 43, 301–304.
Frankfurt airport. Commentary for Rational Men. Retrieved from: [Link]. te Nijenhuis, J., de Jong, M.-J., Evers, A., & van der Flier, H. (2004). Are cognitive
com/2019/01/03/the-decline-of-collective-intelligence-in-germany-frankfurt-airpor differences between immigrant and majority groups diminishing? European Journal
t. of Personality, 18, 405–434.
Flynn, J. R. (1984). The mean IQ of Americans: Massive gains 1932 to 1978. Psychological te Nijenhuis, J., & van der Flier, H. (2013). Is the Flynn effect on g?A meta-analysis.
Bulletin, 95, 29–51. Intelligence, 41(6), 802–807.
Flynn, J. R. (1987). Massive IQ gains in 14 nations: What IQ tests really measure. Trahan, L. H., Stuebing, K. K., Fletcher, J. M., & Hiscock, M. (2014). The Flynn effect: A
Psychological Bulletin, 101, 171–191. meta-analysis. Psychological Bulletin, 140, 1332–1360.
Flynn, J. R. (2012). Are we getting smarter? Rising IQ in the twenty-first century. Cambridge: Weiss, V. (2007). The population cycle drives human history – From a eugenic phase into
Cambridge University Press. a dysgenic phase and eventual collapse. Journal of Social, Political and Economic
Goklany, I. M. (2007). The improving state of the world. Washington, DC: CATO Institute. Studies, 32(3), 327–358.
Hanushek, E. A., Schwerdt, G., Wiederhold, S., & Woessmann, L. (2015). Returns to skills Woessmann, L. (2021). Testleistungen deutscher Schüler:innen über die Zeit, Methodik der
around the world: Evidence from PIAAC. European Economic Review, 73, 103–130. Darstellung. [Test performance of German students over time.]. München: IFO.
Hunt, E. (1995). Will we be smart enough? A cognitive analysis of the coming workforce. New Woodley, M. A. (2012). The social and scientific temporal correlates of genotypic
York: Russell Sage Foundation. intelligence and the Flynn effect. Intelligence, 40, 189–204.
Jones, G. (2016). Hive mind: How your nation's IQ matters so much more than your own. Woodley of Menie, M. A., & Fernandes, H. B. F. (2015). Do opposing secular trends on
Stanford University Press. backwards and forwards digit span evidence the co-occurrence model? Intelligence,
Kaufman, S. B., Reynolds, M. R., Liu, X., Kaufman, A. S., & McGrew, K. S. (2012). Are 50, 125–130.
cognitive g and academic achievement g one and the same g? An exploration on the Woodley of Menie, M. A., & Fernandes, H. B. F. (2016). The secular decline in general
woodcock-Johnson and Kaufman tests. Intelligence, 40, 123–138. intelligence from decreasing developmental stability: Theoretical and empirical
Kirkegaard, E. O. W., & Carl, N. (2022). Smart fraction theory: A comprehensive re- considerations. Personality and Individual Differences, 92, 194–199.
evaluation. Comparative Sociology, 21(6), 677–699. Woodley of Menie, M. A., Peñaherrera-Aguirre, M., Fernandes, H. B. F., & Figueredo, A.-
Lutz, W., Butz, W. P., & Kc, S. (Eds.). (2014). World population and human capital in the J. (2018). What causes the anti-Flynn effect? A data synthesis and analysis of
twenty-first century. Oxford: Oxford University Press. predictors. Evolutionary Behavioral Sciences, 12(4), 276–295.
Lynn, R. (1982). IQ in Japan and the United States shows a growing disparity. Nature, World Bank. (2018). Learning to realize education's promise. Washington, DC, USA: World
297, 222–223. Bank Group.
Lynn, R. (2011). Dysgenics. Genetic deterioration in modern populations. London: Ulster
Institute for Social Research.

10