Abstract
Research about the Flynn effect, the secular rise in IQ, is heavily based on conscript data from successive male birth cohorts. This inevitably means that two distinct phenomena are mixed: fertility differences by IQ group (‘compositional Flynn effect’), and any difference between parents and children (‘within-family Flynn effect’). Both will influence trends in cognitive ability. We focused on the latter phenomenon, exploring changes in cognitive abilities during adolescence within one generation, and between two successive generations within the same family. We identified determinants and outcomes in three linked generations in the Stockholm Multigenerational Study. School and conscript data covered logical/numerical and verbal scores for mothers at age 13, fathers at 13 and 18, and their sons at 18. Raw scores, and change in raw scores, were used as outcomes in linear regressions. Both parents’ abilities at 13 were equally important for sons’ abilities at 18. Boys from disadvantaged backgrounds caught up with other boys during adolescence. Comparing fathers with sons, there appeared to be a positive Flynn effect in logical/numeric and verbal abilities. This was larger if the father had a working-class background or many siblings. A Flynn effect was only visible in families where the father had low general cognitive ability at 18. We conclude that there is a general improvement in logical/numeric and verbal skills from one generation to the next, primarily based on improvement in disadvantaged families. The Flynn effect in Sweden during the later 20th century appears to represent a narrowing between social categories.
Introduction
IQ tests are usually standardised to have a mean value of 100 and a standard deviation of 15. They evaluate the relative ranking of the test-takers compared to a representative sample of peers of the same population with the same age at the same period of time. This standardisation conceals any differences between countries or regions as well as changes over time.
Improvement of cognitive abilities across birth cohorts has nevertheless been documented in many studies. A meta-study covering 271 empirical samples estimated the yearly increase in fluid intelligence to be 0.41 IQ points during the 20th century (Pietschnig and Voracek, 2015). Three other IQ measures grew in a similar fashion. The secular rise in cognitive abilities was observed in 1968 by Schaie and Strother, but it was the compilation of unstandardised, raw IQ data from many countries by sociologist James Flynn (1987; 2009) that led to this phenomenon being described as the Flynn effect.
Flynn himself did not coin the term but commented: ‘The “Flynn effect” is the name that has become attached to an exciting development, namely, that the twentieth century saw massive IQ gains from one generation to another’ (Flynn, 2009: 1–2). A positive Flynn effect could therefore be defined as ‘IQ gains from one generation to the next’. A reversal of this effect, a negative Flynn effect, would instead refer to falling IQ from one generation to the next.
Two distinct phenomena are involved. First, any difference between parents and their own children will influence secular change in cognitive ability. This phenomenon, which is rarely studied, could be referred to as a ‘within-family Flynn effect’ or a ‘parents-to-children Flynn effect’. This is also the focus of our paper. Second, insofar as IQ is inherited, any fertility differences by IQ level will influence trends in average cognitive ability. This could be referred to as a ‘compositional Flynn effect’. Any compositional effect would mix with the family Flynn effect in analyses of successive birth cohorts.
Kolk and Barclay (2019), analysing Swedish data, found ‘an unambiguously positive relationship between cognitive ability and fertility’. This was the case ‘even after accounting for socioeconomic status in the family of origin’. This means that comparisons of successive Swedish birth cohorts will mix a compositional Flynn effect with a family Flynn effect. This provides us with a strong rationale for looking specifically at IQ change from parents to their children, something Thomson called for as far back as 1946.
The positive Flynn effect, as reviewed by Pietschnig and Voracek (2015), stands in contrast to the previously strong conviction that the high fertility of low social classes would result in falling cognitive abilities over time, the so-called dysgenic hypothesis (Lynn, 1996). Raymond Cattell estimated that ‘national intelligence is declining at the rate of one point of IQ for every ten years’ (Cattell, 1936: 193). Godfrey Thomson, ten years later, estimated the fall in intelligence from one generation to the next to be between 2 and 3 IQ points (Thomson, 1946). The conclusion was drawn that a genetically based deterioration in national intelligence was already under way.
The dysgenic hypothesis, as formulated then, rejected environmental determinants of IQ. Cattell wrote that IQ is ‘largely inborn and constitutional like the colour of our eyes’ and ‘unaffected by normal variation of the environment’ (Cattell, 1936: 189, 190). This simplistic idea about a direct one-to-one relation between a person’s genetic set-up and their results on IQ tests has been contested (Neisser et al, 1996). The dysgenic hypothesis has nevertheless been revitalised by some authors, following new evidence that the Flynn effect might be exhausted in some populations (Dutton and Lynn, 2013). Bratsberg and Rogeberg (2018), however, drew the conclusion that both the Flynn effect and its reversal were environmentally driven rather than caused by differential fertility.
Why do cognitive abilities, as measured by raw scores in IQ tests, increase over historical time in most modern populations? Flynn himself gave no clear answer. In his 2009 book he avoids any simple explanation. The change in abilities across generations is likely to be related to long-term cultural and social changes, and perhaps to interaction between such changes and our genetic characteristics, which should imply that these changes modify gene expression during development. He focuses on the changing environment rather than IQ-related patterns of fertility.
Pietschnig and Voracek (2015) sum up the main suggested explanations of the Flynn effect. These range from improved nutrition in utero and in childhood (Barker and Edwards, 1967; Lynn, 1990; Lucas et al, 1992; Nisbett et al, 2012; Flensborg-Madsen and Mortensen, 2017) via reduced childhood disease (Eppig et al, 2010) to the expansion of a school system that promotes abstract problem solving (Husén and Tuijnman, 1991; Teasdale and Owen, 2005). Parental education and the changing nature of work promoted a richer cognitive environment in the home by the end of the 20th century than at its beginning, we assume. Smaller families also enable parents to spend more time with each child.
Apart from these explanations, Pietschnig and Voracek (2015) also point to ‘hybrid’ (social/biological) factors, such as heterosis (or ‘outbreeding’; Mingroni, 2007) and environmentally induced genomic imprinting. Outbreeding means that a larger pool of genes are potential contributors to the new individual. This has been linked to ‘hybrid vigour’. A recent study (Zhu et al, 2018) suggests that outbreeding promotes educational achievement, especially in male offspring. Genomic imprinting along the paternal line was suggested already in 1990 by Miles Storfer. He suggested that ‘each new generation … receives some biological benefit from the cognitively enriching environmental experiences of its forefathers’ (Storfer, 1990). The imprinting hypothesis lacks hard evidence but is in accordance with new thinking in epigenetics (Benito et al, 2018).
Virtually all explanations involve some aspect of the family environment. We will be able to address a limited number of these. Intelligence is inherited from both parents and develops early in life under the influence of family, school and society (LeWinn et al, 2020). Parents and family provide a large part of the environment in which the new individual grows and develops. Parental education and household social class as well as family size are obvious examples. Improvements in these characteristics from parents to children are potential contributors to the Flynn effect.
The relative contribution of the father and mother (beyond that of their genes) to their children’s IQ gives an indication of which environmental factors may be of importance. For instance, there is a large literature about the influence of the maternal and foetal environments on cognitive abilities. If foetal nutrition was all-important for offspring IQ, focusing on the maternal environment would be a priority in order to understand the Flynn effect. If maternal and paternal factors are equally important, certain paternal factors must exist that counterbalance this maternal influence.
Environmental influences modify cognitive abilities throughout childhood, probably more so the earlier they are present (Bloom, 1964). The idea of sensitive or critical periods in childhood should also be considered (Kuh et al, 2003). Different age periods in an individual’s development might be more or less sensitive to the environmental ‘programming’ of IQ.
Bocéréan et al (2003) documented a Flynn effect as early as among 3–5-year olds. Jonsson (2004) observed that number of years in preschool, net of parents’ class and education, predicted the choice of further theoretical education at age 16. Husén and Tuijnman (1991) found that a child’s schooling has a direct effect on adult IQ, when variations in home background and child IQ were controlled for. Härnqvist (1968a; 1968b) similarly found changes between ages 13 and 18 in general IQ among boys to be substantial and to differ according to their educational experience.
If early schooling, including preschool, is important, we would expect the increasing availability of schools and preschools to improve IQ over generations as well as IQ growth during childhood, suggesting a link between childhood growth in IQ and parent to child improvement. At the same time, any nutritionally or school-driven positive Flynn effect could be exhausting itself, approaching a ceiling. Meisenberg et al (2005), referring to the most advanced countries, suggested that ‘intelligence in these countries is reaching a biological limit’.
Social inequalities in access to higher education have been reduced in Sweden over time, according to the study by Breen et al (2009). If schooling contributes to the Flynn effect in Sweden we would therefore expect it to also contribute to a narrowing of social class–based IQ differences over the period under study.
Aims
We aimed to examine how the change in IQ from parents to children depends on parental and family influences, such as parental education, household social class, family size and crowdedness at home. What kind of intellectual and physical environment in childhood and adolescence, provided by school and family, contributes to the Flynn effect?
We explored changes in cognitive abilities both during adolescence within one generation and between two successive generations within the same family. We used information on raw scores of two cognitive measures, logical/numerical and verbal, for individuals embedded in a rich multigenerational database. Four distinct research questions were asked (see ‘Four research questions’ in next section).
Methods
Study population
Stockholm Multigenerational Study (SBC Multigen) is built around 14,608 members of the original Stockholm Birth Cohort (SBC) born 1953 (Generation 1, G1). It includes their siblings (also G1); parents (G0); children (G2) and grandchildren (G3); and the other parent of individuals in G2 and G3 (belonging to G1 and G2, respectively). In total, the population consists of 285,340 unique individuals, some being represented in more than one category. A detailed description can be found elsewhere (Almquist et al, 2020).
Data selections
We constructed four different, partly overlapping data selections, based on the availability of key variables. We denote the data selections by the letters A–D.
Data selection A provides the richest information. It consists of men from SBC (G1) who undertook cognitive tests both in school at age 13 and at military conscription around age 18, and their sons (G2) who took the cognitive tests at conscription around age 18. Information about mothers and fathers of G1 – G0 – is also available.
Data selection B consists of female SBC members (G1) who undertook cognitive tests in school, and their sons (G2) who took the cognitive tests at military conscription.
Data selection C includes SBC men who undertook cognitive tests both in school and at conscription. This data selection does not include the sons (G2).
Data selection D includes all available father–son pairs (G1 and G2) in the SBC Multigen, where both the father and the son undertook the cognitive tests at conscription. The fathers do not need to belong to the original SCB who took the cognitive tests at age 13.
Characteristics of the data selections are shown in Table 1.
Description of the data selections drawn from Stockholm Multigenerational Study
| Data selection A | Data selection B | Data selection C | Data selection D | |
|---|---|---|---|---|
| Who is in the data selection |
|
|
G1: SBC boys with test scores available from age 13 and 18 years | G1 and G2: All available father–son pairs in SBC Multigen database, where both have cognitive test scores available from age 18 |
| Study questions that can be answered within the data selection | Q1, Q2, Q3, Q4 | Q1 | Q2 | Q3 |
| Sample size | 279 father–son pairs, | 855 mother–son pairs, | 5,469 fathers (sons not included in this data selection) | 7,930 father–son pairs, |
| 264 unique fathers | 782 unique mothers | 7,394 unique fathers | ||
| Birth years | ||||
|
|
|
|
|
| Age at son’s birth | ||||
| G1, Fathers | 17–23 | not included | sons not included | 14–26 |
| G1, Mothers | not included | 15–23 | sons not included | not included |
| Year of conscription | ||||
|
|
|
|
|
| Age at conscription of G1 | ||||
|
|
not applicable |
|
|
| Age at conscription of G2 | ||||
|
|
|
sons not included |
|
| Conscription office of G1 | ||||
| Kristianstad | 0 (0%) | not applicable | 62 (1%) | 2,133 (27%) |
| Gothenburg | 1 (0%) | 36 (1%) | 1,519 (19%) | |
| Solna (Stockholm area) | 271 (97%) | 5,318 (97%) | 1,869 (24%) | |
| Karlstad | 5 (2%) | 34 (1%) | 1,165 (15%) | |
| Östersund (Southern Norrland) | 2 (1%) | 15 (0%) | 642 (8%) | |
| Östersund (Northern Norrland) | 0 (0%) | 4 (0%) | 602 (8%) | |
| Conscription office of G2 | ||||
| Kristianstad | 8 (3%) | 46 (5%) | sons not included | 2,167 (27%) |
| Gothenburg | 1 (0%) | 19 (2%) | 1,238 (16%) | |
| Solna (Stockholm area) | 234 (84%) | 680 (80%) | 1,553 (20%) | |
| Karlstad | 20 (7%) | 59 (7%) | 1,377 (17%) | |
| Östersund (Southern Norrland) | 9 (3%) | 42 (5%) | 881 (11%) | |
| Östersund (Northern Norrland) | 7 (3%) | 9 (1%) | 714 (9%) | |
| Test score of G1 at age 13 | ||||
| Numerical |
|
|
|
not measured |
| Verbal |
|
|
|
|
| Test score of G1 at age 18 | ||||
| Logical |
|
not applicable |
|
|
| Verbal |
|
|
|
|
| Low ability of G1 at age 18 | ||||
| NO | 179 (64%) | not applicable | 4,052 (74%) | 4,255 (54%) |
| Yes | 100 (36%) | 1,417 (26%) | 3,675 (46%) | |
| Test score of G2 at age 18 | ||||
| Logical |
|
|
sons not included |
|
| Verbal |
|
|
|
|
| Social group of G0 father in 1960 | ||||
| Non-manual worker | 94 (34%) | 294 (34%) | 2,780 (51%) | 1,369 (17%) |
| Manager or self-employed | 24 (9%) | 97 (11%) | 646 (12%) | 1,654 (21%) |
| Manual worker | 151 (54%) | 420 (49%) | 1,871 (34%) | 4,472 (56%) |
| Not working/missing | 10 (4%) | 44 (5%) | 172 (3%) | 435 (5%) |
| Overcrowded home 1960 | ||||
| No | 200 (72%) | 598 (70%) | 4,492 (82%) | 5,584 (70%) |
| Yes | 75 (27%) | 238 (28%) | 925 (17%) | 2,212 (28%) |
| Missing | 4 (1%) | 19 (2%) | 52 (1%) | 134 (2%) |
| Education of G0 parents in 1960 | ||||
| At least one university | 11 (4%) | 17 (2%) | 496 (9%) | not available |
| At least one upper secondary | 20 (7%) | 78 (9%) | 903 (17%) | |
| Both less than upper secondary | 240 (86%) | 700 (82%) | 3,806 (70%) | |
| Missing | 8 (3%) | 60 (7%) | 264 (5%) | |
| Years at school of G1 by age 17 | ||||
| >9 years | 136 (49%) | 400 (47%) | 3,998 (73%) | 2,914 (37%) |
| 9 years | 132 (47%) | 421 (49%) | 1,334 (24%) | 2,966 (37%) |
| <9 years | 10 (4%) | 19 (2%) | 95 (2%) | 1,355 (17%) |
| 1 (0%) | 15 (2%) | 42 (1%) | 695 (9%) | |
| Sibship size of G1 | ||||
| Only child | 33 (12%) | 78 (9%) | 713 (13%) | 301 (4%) |
| 2 children | 79 (28%) | 280 (33%) | 2,140 (39%) | 1,006 (13%) |
| 165 (59%) | 490 (57%) | 2,583 (47%) | 3,317 (42%) | |
| 2 (1%) | 7 (1%) | 33 (1%) | 3,306 (42%) | |
| Sibship size of G2 | ||||
| Only child | 68 (24%) | 107 (13%) | sons not included | 1,332 (17%) |
| 2 Children | 115 (41%) | 358 (42%) | 3,165 (40%) | |
| 3 or more children | 96 (34%) | 390 (46%) | 3,433 (43%) | |
Notes: G = generation; m = mean; sd = standard deviation; Q = question.
Measurements of cognitive ability
Raw scores for cognitive tests were available from two sources:
SBC boys and girls performed a test at school in 1966 at age 13. It consisted of three subtests: numerical (‘number series’), verbal (‘verbal opposites’) and spatial, each giving raw scores from 0 to 40. The first two tests were included in the current study.
The Swedish Enlistment Battery was first used in 1944, modified several times between 1947 and 1967, and then again in 1980 and in 1994 (Carlstedt, 2000). Members of G1 were tested between 1969 and 1977, when the 1967 version of the battery was in use. G2 enlisted between 1985 and October 1994, using the 1980 version of the same tests before the newest version was adopted (Dalen et al, 2008). We excluded the 2% who enlisted later or were under 17 or over 19 years old at conscription.
Both enlistment batteries contain four subtests covering logical, verbal, spatial and technical abilities. Logical and verbal tests consist of 40 items with a maximum score of 40. Further details of the tests (Ståhlberg-Carlstedt and Sköld, 1981) are classified by Krigsarkivet [the Military Archives]. Compared to the 1967 test, changes in the Enlistment Battery of 1980 were substantial for spatial and technical tests and we thus found it necessary to restrict the study to logical and verbal abilities. In the logical abilities test (‘Instructions’) the task is to choose the answer that fulfils the conditions given by the instructions; the test measures problem solving, induction capacity and numerical ability (Carlstedt and Mårdberg, 1993). The verbal abilities test is based on concept discrimination in the 1967 version and on synonyms in the 1980 version: a correct synonym for a target word should be chosen from four alternatives. Both verbal tests aim to measure the same underlying dimension, but the precision of the latter test was improved and found to be higher (Carlstedt and Mårdberg, 1993).
For use within the military system, these scores as well as the sum score of all four subtests were normalised on a nine-point (stanine) scale with a mean of 5 and a standard deviation of 2. The standardisation was performed for each year (Kolk and Barclay, 2019), so the stanine measure was always relative within a year and there could be no increase or decline in these scores over time. Raw and standardised test scores, but not the norms for the four subtests from the conscription examinations, were available to us. We used stanine scores only when categorising the general ability of fathers (G1) into high versus low. Otherwise, the raw scores were used in the analyses.
Determinants and covariates
We used social class, overcrowding (more than two persons per room in the household, excluding kitchen) and parental education as recorded in the 1960 census. Using the 1970 census we classified G1 schooling into three broad classes: <9 years (completed 7 or 8 years of schooling); 9 years (completed 9-year compulsory school or similar, and not currently continuing studies); >9 years (completed 9 or more years of education and/or currently studying).
Low general ability in G1 was defined as having a total stanine score of 4 or lower at conscription.
Using demographic data and family links, we detected each participant’s birth order and number of siblings (based on link to biological mother).
Enlistment office, age (in years, calculated as year of enlistment minus year of birth) at enlistment, and father’s or mother’s (G1) age at son’s birth were used as control variables.
Missing values were included as a separate category in the analyses for variables where missingness exceeded 1–2%. For remaining variables, those with missing values were excluded from the modelling. Categories and distribution of all variables are shown in Table 1.
Four research questions
- Q1Do mother’s and father’s (G1) cognitive abilities measured at age 13 predict their son’s (G2) cognitive abilities at age 18 equally well?
- Q2Do parental and family factors (G0) predict changes in cognitive abilities between the ages of 13 and 18 among boys (G1)? What is the effect of own (G1) education between ages 13 and 18 on this change?
- Q3Do parental and family factors (G0) predict changes in cognitive abilities between fathers (G1) and sons (G2) at age 18? Is the change different in fathers with different general ability at baseline?
- Q4How do the changes in cognitive abilities from age 13 to age 18 in G1 men relate to changes from G1 to G2 (fathers to sons)?
Data selection A can be used to answer all the questions Q1–Q4, but does not provide optimal statistical power. We thus reran each analysis within one of the other, larger subsamples with comparable information available (see Table 1) and compared the results in order to draw conclusions about the size and direction of associations.
Statistical analysis
The raw scores (for research question Q1) or change in raw scores (Q2–Q4) were used as outcome in linear regression models. We calculated change scores by subtracting the score of the earlier test from the score of the later test (score at age 18 minus score at age 13 for changes within G1; score of G2 at conscription minus score of G1 for intergenerational change). The tests, each consisting of 40 questions, aim to measure the same dimensions, but are not identical. Thus, the absolute values of the scores and the absolute differences are not readily comparable. A higher score value does, however, indicate higher abilities, and a larger change score indicates a greater positive change, relative to others.
The models included the determinants of interest and were adjusted for conscription office and age at conscription but not for the baseline value of cognitive score (Glymour et al, 2005). Robust confidence intervals (95% CI) were retrieved (G2 clustered for G1). The results are reported as differences between categories or adjusted predictions (Stata command margins). The analyses were performed with StataSE 16.0 (Stata Corp LP, College Station, Texas, US).
Ethical permissions
The study is part of the RELINK research programme. Stockholm Regional Ethical Board gave ethical permission (dnr 2017: 684-32).
Data availability
The data that support these findings are available on reasonable request to the corresponding author. The availability of social and conscript data is subject to restrictions imposed by the National Board of Health and Welfare (Socialstyrelsen), Statistics Sweden (SCB) and Swedish Armed Forces (Rekryteringsmyndigheten), in accordance with Swedish and European legislation on privacy protection. This means that data can only be accessed and analysed at a specified venue in Stockholm.
Results
The four data selections are described in Table 1. Selection D included a broader range of birth years of G1 and G2, data from all over Sweden, and information only from registers. The distribution of covariates and patterns of missingness were thus different from those of the other data selections. In order to account for these differences in the data selections, all the analyses were adjusted for the covariates.
Figure 1 presents average values of logical/numerical test score by covariate categories. The general tendency is that the levels were higher in the more advantaged groups (non-manual social class and non-crowded homes). The differences were largest at age 13 in G1. In G1 at age 18 this pattern was intact, although the advantage of those in a privileged position was reduced in every instance. This pattern was also dominant in G2 at age 18. The overall variability of test scores was smallest at age 18 in G1. This could explain some of the observed convergence, but not all, because the variability was larger again at age 18 in G2. The patterns were broadly similar for verbal scores (not shown).
Average values of logical/numerical test scores across categories of covariates, adjusted for conscription office and age at conscription
Citation: Longitudinal and Life Course Studies 2023; 10.1332/175795921X16708793393107
Average values of logical/numerical test scores across categories of covariates, adjusted for conscription office and age at conscription
Citation: Longitudinal and Life Course Studies 2023; 10.1332/175795921X16708793393107
Average values of logical/numerical test scores across categories of covariates, adjusted for conscription office and age at conscription
Citation: Longitudinal and Life Course Studies 2023; 10.1332/175795921X16708793393107
Research question Q1 – intergenerational continuity
Logical/numerical and verbal test scores of G1 men and women at age 13 were linearly and in a nearly identical way associated with the corresponding scores in their sons (G2) at age 18 (Figure 2). Quadratic terms, allowing for more flexible shape of associations, as well as interaction terms with parent’s sex, were virtually zero and did not improve the model (p-values for the quadratic terms were between 0.348 and 0.756; p-values for the interaction terms with parent’s sex were 0.628 for logical/numerical and 0.800 for verbal score). Slope coefficients for logical/numerical and verbal test scores were 0.32 (0.21, 0.42) and 0.33 (0.21, 0.44) (men) and 0.28 (0.22, 0.34) and 0.33 (0.27, 0.39) (women), respectively.
The relationship between cognitive abilities in parents (G1) and sons (G2), adjusted predictions together with 95% confidence limits; adjusted for conscription office and age at conscription
Citation: Longitudinal and Life Course Studies 2023; 10.1332/175795921X16708793393107
The relationship between cognitive abilities in parents (G1) and sons (G2), adjusted predictions together with 95% confidence limits; adjusted for conscription office and age at conscription
Citation: Longitudinal and Life Course Studies 2023; 10.1332/175795921X16708793393107
The relationship between cognitive abilities in parents (G1) and sons (G2), adjusted predictions together with 95% confidence limits; adjusted for conscription office and age at conscription
Citation: Longitudinal and Life Course Studies 2023; 10.1332/175795921X16708793393107
Research question Q2 – adolescent change in G1 men
Family factors
Logical/numerical ability of G1 men living in crowded accommodation at age 7 and with parents (G0) of low education improved more from age 13 to age 18 than it did for other men (Table 2).
Change in test scores (95% confidence intervals) between ages 13 and 18 (study question Q2), or between two generations (study question Q3), predicted by parental and familial factors. Estimated with linear regression
| Study question Q2 | Study question Q3 | |||
|---|---|---|---|---|
| Data selection A n = 259 | Data selection C n = 5,385 | Data selection A n = 274 | Data selection D n = 7,796 | |
| Change in logical/numerical test score | ||||
| Sibship size of G1 | ||||
| Only child | ref | ref | ref | ref |
| 2 children | −1.8 (−4.4, 0.9) | −0.4 (−0.9, 0.2) | 2.3 (−0.9, 5.4) | 0.8 (−0.1, 1.8) |
| 3 or more children | −1.5 (−4.0, 0.9) | −0.4 (−0.9, 0.2) | 3.2 (0.4, 6.1) | 1.3 (0.4, 2.2) |
| Missing | excl | excl | excl | 1.3 (0.4, 2.2) |
| Wald test | p = 0.3994 | p = 0.3414 | p = 0.0756 | p = 0.0157 |
| Social group of G0 father in 1960 | ||||
| Non-manual worker | 0.1 (−1.9, 2.1) | −0.2 (−0.6, 0.2) | 0.5 (−1.4, 2.5) | −0.7 (−1.1, −0.2) |
| Manager or self-employed | −1.2 (−4.1, 1.7) | −0.3 (−0.8, 0.3) | −0.6 (−3.8, 2.6) | 0.2 (−0.3, 0.6) |
| Manual worker | ref | ref | ref | ref |
| Not working/missing | 2.8 (−2.0, 7.7) | −0.8 (−1.9, 0.3) | 2.2 (−3.7, 8.0) | −0.3 (−1.1, 0.6) |
| Wald test | p = 0.5274 | p = 0.4402 | p = 0.8108 | p = 0.0122 |
| Overcrowded home 1960 | ||||
| No | ref | ref | ref | ref |
| Yes | 0.2 (−1.7, 2.1) | 0.9 (0.4, 1.4) | 0.4 (−1.6, 2.5) | −0.0 (−0.4, 0.4) |
| Wald test | p = 0.8539 | p = 0.0005 | p = 0.6770 | p = 0.9141 |
| Education of G0 parents in 1960 | ||||
| Both less than upper secondary | ref | ref | Variable not included | Variable not available |
| At least one upper secondary | −3.6 (−6.8, −0.4) | −0.9 (−1.4, −0.4) | ||
| At least one university | −1.1 (−5.5, 3.2) | −1.3 (−2.0, −0.7) | ||
| Missing | −1.6 (−6.9, 3.6) | 0.1 (−0.8, 1.0) | ||
| Wald test | p = 0.1579 | p = 0.0001 | ||
| Change in verbal test score | ||||
| Sibship size of G1 | ||||
| Only child | ref | ref | ref | ref |
| 2 children | 2.6 (0.3, 4.9) | 0.8 (0.3, 1.3) | −0.0 (−2.6, 2.6) | 0.0 (−0.9, 0.9) |
| 3 or more children | 2.4 (0.3, 4.6) | 1.2 (0.7, 1.7) | 1.5 (−0.9, 3.9) | 0.7 (−0.1, 1.6) |
| Missing | excl | excl | excl | 0.8 (−0.1, 1.6) |
| p = 0.0587 | p = 0.0000 | p = 0.1996 | p = 0.0086 | |
| Social group of G0 father in 1960 | ||||
| Non-manual worker | −1.3 (−2.8, 0.4) | 0.2 (−0.2, 0.6) | 1.3 (−0.6, 3.3) | −1.2 (−1.6, −0.7) |
| Manager or self-employed | −3.5 (−6.0, −1.0) | −0.4 (−0.9, 0.1) | 2.7 (0.2, 5.1) | −0.0 (−0.4, 0.4) |
| Manual worker | ref | ref | ref | ref |
| Not working/missing | 0.4 (−3.8, 4.5) | 0.8 (−0.2, 1.8) | −1.6 (−6.0, 2.7) | 0.1 (−0.8, 0.9) |
| Wald test | p = 0. 0356 | p = 0. 0377 | p = 0.0908 | p = 0.0000 |
| Overcrowded home 1960 | ||||
| No | ref | ref | ref | ref |
| Yes | −2.5 (−4.1, −0.9) | 0.2 (−0.2, 0.6) | 1.2 (−0.9, 3.3) | 0.2 (−0.2, 0.6) |
| Wald test | p = 0.0028 | p = 0.3077 | p = 0.2553 | p = 0.3869 |
| Education of G0 parents in 1960 | ||||
| Both less than upper secondary | ref | Variable not included | Variable not available | |
| At least one upper secondary | −2.5 (−5.2, 0.3) | ref | ||
| At least one university | 0.1 (−3.7, 3.9) | −0.7 (−1.1, −0.2) | ||
| Missing | −1.1 (−5.6, 3.4) | −0.8 (−1.4, −0.2) | ||
| Wald test | p = 0.3370 | −0.2 (−0.9, 0.6) | ||
Notes: ref = reference category; excl = excluded category.
Mutually adjusted for all the variables shown in the table as well as conscription office and age at conscription.
G1 men growing up with siblings improved their verbal ability from age 13 to age 18 more than only children. Parents’ (G0) education (high rather than low) when the child (G1) was 7 years old predicted a weaker improvement in verbal ability (Table 2).
The results were similar in data selections A and C, but the confidence intervals around the estimates were wider in selection A due to smaller sample size.
G1 schooling
Table 3 suggests that G1 boys with a longer education improved their logical/numerical abilities from age 13 to 18 less than other G1 boys. Change in verbal abilities was less clearly related to G1 schooling.
Change in test scores (95% confidence intervals) between ages 13 and 18, predicted by number of years at school by age 17 and estimated with linear regression models
| Data selection A n = 258 | Data selection C n = 5,343 | |
|---|---|---|
| Change in logical/numerical test score | ||
| Years at school of G1 by age 17 | ||
| <9 years | 0.2 (−4.2, 4.6) | 1.5 (0.1, 2.8) |
| 9 years | ref | ref |
| >9 years | −2.0 (−3.6, −0.3) | −0.8 (−1.2, −0.4) |
| Change in verbal test score | ||
| Years at school of G1 by age 17 | ||
| <9 years | 0.2 (−3.6, 4.1) | 0.0 (−1.2, 1.2) |
| 9 years | ref | ref |
| >9 years | −0.8 (−2.2, 0.7) | 0.0 (−0.3, 0.4) |
Notes: ref = reference category.
Adjusted for conscription office, age at conscription, sibship size of G1, social group of G0 father 1960, overcrowded home 1960 and education of G0 parents in 1960
Research question Q3 – change from G1 fathers to G2 sons
Comparing raw scores of fathers (G1) and sons (G2) at age 18, we observed that logical/numerical and verbal abilities were correlated both within G1 (r = 0.63 and r = 0.71 in data selections A and D, respectively) and G2 (r = 0.67 and r = 0.70 in A and D, respectively) as well as between generations (logical/numerical r = 0.38, verbal r = 0.32 in selection A; logical/numerical r = 0.35, verbal r = 0.29 in selection D).
Family factors
When comparing G1 to G2, the change in numerical test scores at age 18 was larger if G1 grew up with two or more siblings and in a manual rather than non-manual household. The corresponding change in verbal test score was larger if G1 grew up in a manual rather than a non-manual household (Table 2). In data selection A, the estimates had wider confidence intervals than in selection D. Additional adjustment for parental education (only available in data selection A), did not affect the results (not shown).
Paternal IQ
Strikingly, the observed positive Flynn effect was restricted to father–son pairs, where G1 men had a low rather than high general cognitive ability at age 18. Sons of men with a high general cognitive ability did not differ or differed only marginally from their fathers (Figure 3).
Difference in test scores between fathers (G1) and sons (G2), stratified by general ability of the father, adjusted for conscription office and age at conscription, sibship size of G1, social group of G0 father 1960, overcrowded home 1960 and education of G0 parents in 1960
Citation: Longitudinal and Life Course Studies 2023; 10.1332/175795921X16708793393107
Difference in test scores between fathers (G1) and sons (G2), stratified by general ability of the father, adjusted for conscription office and age at conscription, sibship size of G1, social group of G0 father 1960, overcrowded home 1960 and education of G0 parents in 1960
Citation: Longitudinal and Life Course Studies 2023; 10.1332/175795921X16708793393107
Difference in test scores between fathers (G1) and sons (G2), stratified by general ability of the father, adjusted for conscription office and age at conscription, sibship size of G1, social group of G0 father 1960, overcrowded home 1960 and education of G0 parents in 1960
Citation: Longitudinal and Life Course Studies 2023; 10.1332/175795921X16708793393107
Research question Q4 – relation between adolescent and intergenerational changes
We found a negative relationship between change scores from age 13 to age 18 in G1 men and change scores from G1 to G2 measured at age 18 (Figure 4). Thus, as G1 improvement in relative IQ during adolescence gets larger, so the positive Flynn effect (from G1 to G2) gets successively smaller, even negative. Regression coefficients, adjusted for conscription office and age at conscription, were −0.15, 95% CI (−0.28, −0.02) and −0.32, 95% CI (−1.20, 1.84) for logical/numerical and verbal change scores, respectively.
Relationship between adolescent and intergenerational changes in logical/numerical and verbal test scores: scatter plot of raw change scores and adjusted predictions together with 95% confidence limits, adjusted for conscription office and age at conscription
Citation: Longitudinal and Life Course Studies 2023; 10.1332/175795921X16708793393107
Relationship between adolescent and intergenerational changes in logical/numerical and verbal test scores: scatter plot of raw change scores and adjusted predictions together with 95% confidence limits, adjusted for conscription office and age at conscription
Citation: Longitudinal and Life Course Studies 2023; 10.1332/175795921X16708793393107
Relationship between adolescent and intergenerational changes in logical/numerical and verbal test scores: scatter plot of raw change scores and adjusted predictions together with 95% confidence limits, adjusted for conscription office and age at conscription
Citation: Longitudinal and Life Course Studies 2023; 10.1332/175795921X16708793393107
Discussion
Previous studies of the Flynn effect have usually looked at IQ changes across birth cohorts, without considering direct changes from parents to children. This approach does not distinguish between a compositional effect on IQ change (for example, fertility differences by parental IQ) and a truly generational change based on observed IQ differences between parents and their children. Thomson, as long ago as 1946, stated that ‘Actual measurement of two successive generations is desirable, indeed essential’ (Thomson, 1946: 16). Nevertheless, such comparisons have been very rare.
We found that family determinants of cognitive ability contributed to continuity across generations. A privileged background in childhood, such as a non-manual social class or good parental education, is beneficial for your own cognitive development as well as for that of your children.
We observed that children (G1) with privileged backgrounds had a higher logical/numerical and verbal ability at age 13 than other children. However, this advantage grew smaller during adolescence. We could think about this as a catch-up effect, with disadvantaged G1 children realising an untapped potential to grow.
Similarly, sons (G2) of G1 men from disadvantaged backgrounds improved their abilities (measured at age 18) compared to those of their fathers (also measured at age 18) more than did other G2 men. The potential for intergenerational improvement of cognitive abilities may be larger for sons of disadvantaged men.
Educational expansion of the school system is a prime candidate for the family Flynn effect. Our G2 men, born 1967–76, were the first generation to go to preschool from an early age in large numbers as their mothers entered the labour market (Jonsson, 2004). A compulsory extra year of schooling was introduced between 1962 and 1972, thus covering all of G2 but only part of G1. The cumulative effect of schooling, including preschool, may have been of particular importance for disadvantaged children.
The classic paper by Shavit and Blossfeld (1993) and its critique by Breen et al (2009) differ in their findings about how general the decline in educational inequality has been among compared countries. However, both conclude that in Sweden access to higher education has become less unequal in later cohorts. Manual classes have improved their position more than non-manual classes and the decline in inequality is demonstrated for cohorts born 1908–64 (Breen et al, 2009: Figure 4). Parents of G1 (= G0) would have been more unequal in terms of education than parents of G2 (= G1). When cognitive abilities among men in these two generations (G1 and G2) are compared, there appears to be a convergence between social categories.
Dichotomising father–son pairs (G1–G2) into those where G1 general abilities were high or low suggested that the positive Flynn effect is restricted to the latter. This could in part be a ‘ceiling effect’, restricting the potential for improvement, particularly if initial scores were high. Colom et al (2005) similarly observed that most of the Flynn effect occurs in the lower and middle part of the IQ distribution, with negligible gains at its top. They suggested that this intergenerational catch-up effect may be nutritionally based. It is equally possible that it is related to the better schooling or better home circumstances of disadvantaged children and/or their parents. The two explanations are not mutually exclusive.
Adolescent change in logic/numerical and verbal IQ in fathers (G1) and the corresponding change between fathers and their sons (G1–G2) were negatively correlated. This was an unexpected finding. A plausible explanation is that disadvantaged boys who catch up with other boys in the G1 generation reduce the potential for further improvement on their achievement among their sons, the G2 generation. Such a ceiling hypothesis could be formulated as follows: further improvement in IQ from parents to children becomes progressively more difficult the higher the initial level of parental IQ.
Familial studies of intelligence have suggested that IQ is inherited from both parents in equal measure (Bouchard and McGue, 1981). In our study, mother’s and father’s numerical and verbal abilities at age 13 predicted the corresponding skills of their son at age 18 in a virtually identical way. Since optimal foetal growth and maternal environment in infancy are known to benefit the cognitive development of the growing child (Devlin et al, 1997; Flensborg-Madsen and Mortensen, 2017), we would expect some other paternal factor, social or biological, to counterbalance this maternal influence. The specific nature of this paternal contribution should be an important research problem; it may or may not contribute to the Flynn effect.
Considerations and limitations
The original SBC (G1) included all school children in greater Stockholm, born 1953 and attending school on a particular day in 1966, and is thus largely representative of Stockholm children born 1953. The extension of G1 includes siblings. It also includes the other parent of a G2 child. The study population is therefore predominantly Stockholm based, but due to migration and marriage other parts of Sweden are also represented.
G2 is an unselected group of sons of G1. However, due to cuts in Swedish defence, successively fewer conscripts were called up in the G2 generation. It is unlikely that this resulted in any specific bias in the comparison of fathers and sons (G1 and G2). Further, we have access to a number of individual and family characteristics which we control for in all analyses. Controlling for covariates reduces this and other potential selection problems.
Flynn effects are usually calculated by comparing different birth cohorts who have taken the same intelligence test at different time points (Pietschnig and Voracek, 2015; Bratsberg and Rogeberg, 2018) or by comparing the same group’s results on two tests with different normative bases, an earlier and a later version, and relating the mean scores to the interval between norming and testing (Trahan et al, 2014).
Our approach, linking fathers and sons, made it necessary to consider the ‘regression to the mean’ phenomenon (Galton, 1886). This would occur if a large number of factors, including measurement error, influence G1 and G2 raw test scores, making the process resemble a random process. We would then expect the positive change in the lower half of the ability distribution to be mirrored by an equally large negative change in scores in the upper half. It is clear from Figure 3 that there is a positive cognitive change among those boys whose fathers scored poorly in general ability. However, any opposite change for those whose fathers scored well is either absent (data set A) or much smaller (data set D). Only if these opposite changes were of the same size could we conclude that the change from fathers to sons was a simple ‘regression to the mean’ phenomenon. We conclude therefore that regression to the mean cannot explain our findings of a positive Flynn effect among the sons of men in the lower half of the distribution of cognitive abilities. It can more readily be interpreted as a catch-up phenomenon.
We considered the problem of measuring cognitive abilities. Historically, Swedish conscript testing has been heavily influenced by Thurstone’s ideas about specific dimensions of intelligence (Härnqvist, 1968a), rather than relying on models of general intelligence in a hierarchical system (Gustafsson, 1984). We focus on logical/numerical and verbal abilities, which were available both in school data from age 13 and in conscript data.
The change in the conscript test battery in 1980 necessitates caution when comparing absolute test scores of G1 and G2. It influenced verbal scores more than numerical ones. In the former, the concept discrimination test was replaced by synonyms, in order to improve precision. The numerical test was not changed in any major way. The maximum number of points for both measures in both versions was 40. The exact nature of these changes is classified information. Looking at year by year averages of test scores, we conclude that the introduction of the 1980 test battery caused a small dip on the logical/numerical test and a rise in the verbal test that year. This might cause us to underestimate the Flynn effect in logical/numerical abilities somewhat and overestimate it in verbal abilities. However, Carlstedt and Mårdberg (1993) hold that the 1980 changes of the 1967 verbal test were introduced to increase precision. Such a change would not introduce any bias.
A major weakness in this field, as in our study, is that the evidence for the Flynn effect is almost entirely based on male data. However, Emanuelsson et al (1993) found evidence of a positive Flynn effect among girls. It might therefore be tempting to assume that the effect is broadly similar in men and women. However, some authors argue, based on evolutionary theory, that there is selective pressure against female education and improved IQ during the 20th century (Wang et al, 2016). Clearly there is a need for more data on women to draw correct conclusions about the direction, size and composition of the Flynn effect.
Finally, the idea that cognitive abilities, measured by IQ tests, are merely superficial cultural constructs and therefore that the Flynn effect is a mirage or a methodological artefact (Rodgers, 1998; Beaujean and Osterlind, 2008), contradicts the observation, in a very large number of studies, that IQ scores predict mortality (Calvin et al, 2011). In fact, even IQ change in adolescence, from age 10 to 20, is found to predict later life mortality (Lager et al, 2009).
Conclusion
We had access to a large database of high quality with smaller cohorts embedded. Father–son linkages and mother–son linkages with reference to logical/numerical and verbal abilities were possible. Both parents appeared to have an equal influence on their sons’ cognitive abilities.
We conclude that there is a general improvement in logical/numerical and verbal abilities from one generation to the next, primarily based on improvements among disadvantaged families. The improvements are smaller the higher the initial cognitive abilities, suggesting a ceiling effect and that the family Flynn effect might eventually exhaust itself.
Note
Joint first author.
Conflict of interest
The authors declare that there is no conflict of interest.
References
Almquist, Y.B., Grotta, A., Vågerö, D., Stenberg, S.A. and Modin, B. (2020) Cohort profile update: the Stockholm birth cohort study, International Journal of Epidemiology, 49(2): 367–367e. doi: 10.1093/ije/dyz185
Barker, D.J. and Edwards, J.H. (1967) Obstetric complications and school performance, British Medical Journal, 3(5567): 695–9. doi: 10.1136/bmj.3.5567.695
Beaujean, A.A. and Osterlind, S.J. (2008) Using item response theory to assess the Flynn effect in the National Longitudinal Study of Youth 79 Children and Young Adults data, Intelligence, 36(5): 455–63. doi: 10.1016/j.intell.2007.10.004
Benito, E., Kerimoglu, C., Ramachandran, B., Pena-Centeno, T., Jain, G., Stilling, R.M., Islam, M.R., Capece, V., Zhou, Q., Edbauer, D. et al. (2018) RNA-dependent intergenerational inheritance of enhanced synaptic plasticity after environmental enrichment, Cell Reports, 23(2): 546–54. doi: 10.1016/j.celrep.2018.03.059
Bloom, B.S. (1964) Stability and Change in Human Characteristics, New York: Wiley.
Bocéréan, C., Fischer, J.P. and Flieller, A. (2003) Long-term comparison (1921–2001) of numerical knowledge in three to Five-and-a-half year-old children, European Journal of Psychology of Education, 18(4): 405–24.
Bouchard, T.J. Jr and McGue, M. (1981) Familial studies of intelligence: a review, Science, 212(4498): 1055–9. doi: 10.1126/science.7195071
Bratsberg, B. and Rogeberg, O. (2018) Flynn effect and its reversal are both environmentally caused, PNAS, 115(26): 6674–8. doi: 10.1073/pnas.1718793115
Breen, R., Luijkx, R., Müller, W. and Pollak, R. (2009) Nonpersistent inequality in educational attainment: evidence from eight European countries, American Journal of Sociology, 114(5): 1475–521. doi: 10.1086/595951
Calvin, C.M., Deary, I.J., Fenton, C., Roberts, B.A., Der, G., Leckenby, N. and Batty, G.D. (2011) Intelligence in youth and all-cause-mortality: systematic review with Meta-analysis, International Journal of Epidemiology, 40(3): 626–44. doi: 10.1093/ije/dyq190
Carlstedt, B. (2000) Cognitive Abilities: Aspects of Structure, Process and Measurement, doctoral thesis, Gothenburg: University of Gothenburg, https://gupea.ub.gu.se/handle/2077/9600.
Carlstedt, B. and Mårdberg, B. (1993) Construct validity of the Swedish Enlistment Battery, Scandinavian Journal of Psychology, 34(4): 353–62. doi: 10.1111/j.1467-9450.1993.tb01131.x
Cattell, R.B. (1936) Is national intelligence declining?, Eugenics Review, 28(3): 181–203.
Colom, R., Lluis-Font, J.M. and Andrés-Pueyo, A. (2005) The generational intelligence gains are caused by decreasing variance in the lower half of the distribution: supporting evidence for the nutrition hypothesis, Intelligence, 33(1): 83–91. doi: 10.1016/j.intell.2004.07.010
Dalen, M., Hjern, A., Lindblad, F., Odenstad, A., Ramussen, F. and Vinnerljung, B. (2008) Educational attainment and cognitive competence in adopted men: a study of international and national adoptees, siblings and a general Swedish population, Children and Youth Services Review, 30: 1211–19. doi: 10.1016/j.childyouth.2008.03.006
Devlin, B., Daniels, M. and Roeder, K. (1997) The heritability of IQ, Nature, 388(6641): 468–71. doi: 10.1038/41319
Dutton, E. and Lynn, R. (2013) A negative Flynn effect in Finland, 1997–2009, Intelligence, 41(6): 817–20. doi: 10.1016/j.intell.2013.05.008
Emanuelsson, I., Reuterberg, S.E. and Svensson, A. (1993) Changing differences in intelligence? Comparisons between groups of 13-year-olds tested from 1960 to 1990, Scandinavian Journal of Educational Research, 37(4): 259–77. doi: 10.1080/0031383930370401
Eppig, C., Fincher, C.L. and Thornhill, R. (2010) Parasite prevalence and the worldwide distribution of cognitive ability, Proceedings of the Royal Society B: Biological Sciences, 277(1701): 3801–8. doi: 10.1098/rspb.2010.0973
Flensborg-Madsen, T. and Mortensen, E.L. (2017) Birth weight and intelligence in young adulthood and midlife, Pediatrics, 139(6): art e20163161, doi: 10.1542/peds.2016-3161.
Flynn, J.R. (1987) Massive IQ gains in 14 nations: what IQ tests really measure, Psychological Bulletin, 101(2): 171–91. doi: 10.1037/0033-2909.101.2.171
Flynn, J.R. (2009) What Is Intelligence? Beyond the Flynn Effect, pbk edn, Cambridge: Cambridge University Press.
Galton, F. (1886) Regression towards mediocrity in hereditary stature, Journal of the Anthropological Institute of Great Britain and Ireland, 15: 246–63. doi: 10.2307/2841583
Glymour, M.M., Weuve, J., Berkman, L.F., Kawachi, I. and Robins, J.M. (2005) When is baseline adjustment useful in analyses of change? An example with education and cognitive change, American Journal of Epidemiology, 162(3): 267–78. doi: 10.1093/aje/kwi187
Gustafsson, J.E. (1984) A unifying model for the structure of intellectual abilities, Intelligence, 8(3): 179–203. doi: 10.1016/0160-2896(84)90008-4
Härnqvist, K. (1968a) Relative changes in Intelligence from 13 to 18, I: background and methodology, Scandinavian Journal of Psychology, 9(1): 50–64.
Härnqvist, K. (1968b) Relative changes in intelligence from 13 to 18, II: results, Scandinavian Journal of Psychology, 9(1): 65–82.
Husén, T. and Tuijnman, A. (1991) The contribution of formal schooling to the increase in intellectual capital, Educational Researcher, 20(7): 17–25.
Jonsson, J. (2004) Förskola för förfördelade, in M. Bygren, M. Gähler and M. Nermo (eds) Familj Och Arbete – Vardagsliv i Förändring, Stockholm: SNS, pp 90–121.
Kolk, M. and Barclay, K. (2019) Cognitive ability and fertility among Swedish men born 1951–1967: evidence from military conscription registers, Proceedings of the Royal Society B: Biological Sciences, 286(1902): art 20190359, doi: 10.1098/rspb.2019.0359.
Kuh, D., Ben-Shlomo, Y., Lynch, J., Hallqvist, J. and Power, C. (2003) Life course epidemiology, Journal of Epidemiology & Community Health, 57(10): 778–83.
Lager, A., Bremberg, S. and Vågerö, D. (2009) The association of early IQ and education with mortality: 65 year longitudinal study in Malmö, Sweden, BMJ, 339: art b5282, doi: 10.1136/bmj.b5282.
LeWinn, K.Z., Bush, N.R., Batra, A., Tylavsky, F. and Rehkopf, D. (2020) Identification of modifiable social and behavioral factors associated with childhood cognitive performance, JAMA Pediatrics, 174(11): 1063–72. doi: 10.1001/jamapediatrics.2020.2904
Lucas, A., Morley, R., Cole, T.J., Lister, G. and Leeson-Payne, C. (1992) Breast milk and subsequent intelligence quotient in children born preterm, Lancet, 339(8788): 261–4. doi: 10.1016/0140-6736(92)91329-7
Lynn, R. (1990) The role of nutrition in secular increases in intelligence, Personality and Individual Differences, 11(3): 273–85. doi: 10.1016/0191-8869(90)90241-I
Lynn, R. (1996) Dysgenics: Genetic Deterioration in Modern Populations, Westport, CT: Praeger.
Meisenberg, G., Lawless, E., Lambert, E. and Newton, A. (2005) The Flynn effect in the Caribbean: generational change of cognitive test performance in Dominica, Mankind Quarterly, 46(1), https://mankindquarterly.org/archive/issue/46-1/2.
Mingroni, M.A. (2007) Resolving the IQ paradox: heterosis as a cause of the Flynn effect and other trends, Psychological Review, 114(3): 806–29. doi: 10.1037/0033-295X.114.3.806
Neisser, U., Boodoo, G., Bouchard, T.J. Jr, Boykin, A.W., Brody, N., Ceci, S.J., Halpern, D.F., Loehlin, J.C., Perloff, R., Sternberg, R.J. et al. (1996) Intelligence: knowns and unknowns, American Psychologist, 51(2): 77–101. doi: 10.1037/0003-066X.51.2.77
Nisbett, R.E., Aronson, J., Blair, C., Dickens, W., Flynn, J., Halpern, D.F. and Turkheimer, E. (2012) Intelligence: new findings and theoretical developments, American Psychologist, 67(2): 130–59. doi: 10.1037/a0026699
Pietschnig, J. and Voracek, M. (2015) One century of global IQ gains: a formal Meta-analysis of the Flynn effect (1909–2013), Perspectives on Psychological Science, 10(3): 282–306. doi: 10.1177/1745691615577701
Rodgers, J.L. (1998) A critique of the Flynn effect: massive IQ gains, methodological artifacts, or both?, Intelligence, 26(4): 337–56. doi: 10.1016/S0160-2896(99)00004-5
Schaie, K.W. and Strother, C.R. (1968) A Cross-sequential study of age changes in cognitive behavior, Psychological Bulletin, 70(6): 671–80. doi: 10.1037/h0026811
Shavit, Y. and Blossfeld, H.P. (eds) (1993) Persistent Inequality: Changing Educational Attainment in Thirteen Countries, Boulder, CO: Westview.
Ståhlberg-Carlstedt, B. and Sköld, P. (1981) Inskrivningsprov 80 – Testmanual [Enlistment Battery 80 – Test Manual], report CH55001-H7, Stockholm: FOA [Försvarets forskningsanstalt: Swedish National Defence Research Institute].
Storfer, M.D. (1990) Intelligence and Giftedness: The Contributions of Heredity and Early Environment, San Francisco: Jossey-Bass.
Teasdale, T.W. and Owen, D.R. (2005) A Long-term rise and recent decline in intelligence test performance: the Flynn effect in reverse, Personality and Individual Differences, 39(4): 837–43. doi: 10.1016/j.paid.2005.01.029
Thomson, G. (1946) The trend of national intelligence, Eugenics Review, 38(1): 9–18.
Trahan, L.H., Stuebing, K.K., Fletcher, J.M. and Hiscock, M. (2014) The Flynn effect: a meta-analysis, Psychological Bulletin, 140(5): 1332–60. doi: 10.1037/a0037173
Wang, M., Fuerst, J. and Ren, J. (2016) Evidence of dysgenic fertility in China, Intelligence, 57: 15–24. doi: 10.1016/j.intell.2016.04.001
Zhu, C., Zhang, X., Zhao, Q. and Chen, Q. (2018) Hybrid marriages and phenotypic heterosis in offspring: evidence from China, Economics & Human Biology, 29: 102–14.
