Intelligence testing history: Binet-Simon 1905, Spearman's g, the Flynn effect, and the controversy over IQ

Intuition Beginner

IQ (the intelligence quotient) measures a narrow slice of cognitive ability — the verbal, mathematical, and spatial reasoning that schools reward. The first modern intelligence test was built in 1905 by Alfred Binet, a French psychologist, to identify which Paris schoolchildren needed extra educational help. Binet explicitly rejected the idea that his test measured innate, fixed intelligence. He saw it as a snapshot of current performance, shaped by schooling and opportunity.

When the test crossed the Atlantic, it became what Binet had feared. Lewis Terman's 1916 Stanford-Binet adapted the instrument for American use and turned the IQ score into a popular sorting device. Over the following decades the test was enlisted by the eugenics movement, by immigration restriction, and by the Army's mass testing of recruits in the First World War. A tool designed to help struggling children was repurposed to rank populations.

The history turns on a single empirical fact. Over the twentieth century, average IQ scores rose by roughly $3$ points per decade in every developed country studied — the Flynn effect, after James Flynn's 1984 synthesis. That magnitude cannot be genetic. It demonstrates that IQ is sensitive to environment and reframes every claim about fixed intelligence or group differences in the score.

Visual Beginner

The picture lays out the intelligence-testing timeline as a horizontal axis from 1869 to the present. Each milestone is marked with its contributor, instrument, and the methodological or political consequence. The arc begins with Galton's hereditarian hypothesis, turns practical with Binet-Simon, is recast as a popular score by Terman and Yerkes, gains its psychometric foundation with Spearman and Thurstone, its clinical form with Wechsler, and its modern reinterpretation with Flynn.

Two strands run in parallel. The instrument strand runs through Binet, Terman, Yerkes, and Wechsler. The theoretical strand runs through Spearman, Thurstone, Jensen, Flynn, and Herrnstein-Murray. Their intersections are where the field's controversies live.

Worked example Beginner

Setting. In 1917, with the United States entering the First World War, the American Psychological Association appointed Robert Yerkes to develop a group-administered intelligence test for screening Army recruits. Yerkes's committee, which included Terman and Henry Goddard, produced two instruments: the Army Alpha for literate English speakers and the Army Beta for recruits who were illiterate or did not speak English.

Step 1. Scale. Between 1917 and 1918 the Army administered the Alpha and Beta to approximately $1,700,000$ recruits. The results were summarised in Yerkes's 1921 report to the National Academy of Sciences. The reported mean mental age of the native-born white American recruit was $13$ years — only marginally above the cut-off then used for the category of "feeblemindedness."

Step 2. Use. The Army results were widely reported. The reported mean for Southern and Eastern European immigrants was lower than the native-born mean; the reported mean for Black recruits was lower still. These figures were cited in the 1924 Johnson-Reed Immigration Restriction Act, which set national-origin quotas designed to favour Northern Europeans over Southern and Eastern Europeans.

Step 3. Methodology. The Alpha and Beta were culturally loaded (they required familiarity with American domestic objects and currency), language-loaded even in the supposedly non-verbal Beta, and unvalidated against any criterion of actual military performance. Critics at the time, including some within the APA, noted these flaws. The critics were ignored in the policy arena.

What this tells us: a psychometric instrument deployed at scale without validation of its construct for the use to which it was put can produce numerically precise-looking results whose policy consequences outrun the evidence. The Army tests established intelligence testing as a mass-administered instrument. They are also the paradigmatic cautionary tale of psychometric misuse.

Check your understanding Beginner

Formal definition Intermediate+

The objects of intelligence testing are standardised scores on cognitive instruments, their transformation into the IQ scale, and the latent variables extracted from correlations among them. Three definitions anchor the field, and a fourth governs the heritability concept that underlies the most contested interpretation.

Definition (Mental age and ratio IQ; Binet-Simon 1905, Stern 1912). Let $\mathrm{MA}$ denote the mental age assigned to a child whose performance on the Binet-Simon scale matches the median performance of typical children of age $\mathrm{MA}$. Let $\mathrm{CA}$ denote the child's chronological age. Stern's (1912) ratio IQ is

$${\mathrm{IQ}}_{\mathrm{ratio}}=100\cdot \frac{\mathrm{MA}}{\mathrm{CA}}.$$

The ratio IQ is well-defined only for children (where $\mathrm{MA}$ grows roughly linearly with $\mathrm{CA}$), breaks down for adults (where $\mathrm{MA}$ saturates), and yields a non-stationary distribution because the standard deviation of $\mathrm{MA}/\mathrm{CA}$ varies with age.

Definition (Deviation IQ; Wechsler 1939). Let $X$ denote an individual's raw composite score on the Wechsler Adult Intelligence Scale (WAIS) or the Wechsler Intelligence Scale for Children (WISC), referenced to an age-matched standardisation sample with sample mean ${\mu }_a$ and sample standard deviation ${\sigma }_a$ for the individual's age band $a$. The deviation IQ is

$${\mathrm{IQ}}_{\mathrm{dev}}=100+15\cdot \frac{X-{\mu }_a}{{\sigma }_a},$$

so the reference distribution has mean $100$ and standard deviation $15$ at every age band. The transformation normalises the score across ages — a given IQ always corresponds to the same percentile within the age band — and is the form used in all modern clinical instruments.

Definition (Spearman's $g$ and the two-factor model; Spearman 1904). Let $X_1,\dots ,X_k$ be standardised scores on $k$ cognitive tests, and let $R=({\rho }_{ij})$ be their $k\times k$ correlation matrix. Spearman ^{[Spearman1904]} observed that $R$ typically exhibits the positive manifold: every pairwise correlation ${\rho }_{ij}$ is positive. The two-factor model posits

$$X_i=a_{i}g+s_i,i=1,\dots ,k,$$

where $g$ is a single latent variable common to all tests (the general factor), $a_i\in [0,1]$ is the loading of test $i$ on $g$, and $s_i$ is a test-specific residual uncorrelated with $g$ and with $s_j$ for $j\ne i$. Under the model, $g$ is recovered (up to sign) as the leading principal component of $R$; the proportion of total variance it accounts for is its largest eigenvalue divided by $k$, and is typically $0.40$ to $0.50$ for diverse cognitive batteries.

Definition (Heritability, narrow-sense). Let $V_P$ denote the phenotypic variance of IQ in a specified population, decomposed as $V_P=V_A+V_D+V_I+V_C+V_E$, where $V_A$ is additive genetic variance, $V_D$ is dominance variance, $V_I$ is epistatic variance, $V_C$ is shared-environment variance, and $V_E$ is non-shared-environment variance. The narrow-sense heritability is $h^2=V_A/V_P$. Twin-study estimates of $h^2$ for adult IQ in modern industrial populations converge on the range $0.5$ to $0.8$, with $h^2$ increasing from childhood to adulthood (the Wilson effect). Heritability is a within-population, within-time variance ratio; it places no constraint on the cause of mean differences between populations or across generations.

Counterexamples to common slips

• IQ tests measure innate intelligence. No. Binet rejected this interpretation in 1905; the Flynn effect, on which mean IQ rose approximately $30$ points across the twentieth century, refutes it empirically. What IQ tests measure is a composite of reasoning skills that are partially heritable within populations and substantially sensitive to environment between populations and across generations.

• The general factor $g$ does not exist. No. $g$ is one of the most robust findings in psychometrics. Under the positive manifold, the leading principal component of any diverse cognitive battery accounts for a substantial fraction of variance and predicts performance on novel cognitive tasks. The existence of $g$ does not entail hereditarian claims; the contested question is what causes individual differences in $g$, not whether $g$ exists.

• A heritability of $h^2=0.8$ means that $80\%$ of any individual's IQ is genetic. No. Heritability is a population-level variance ratio, not an individual-level determination. The statement "$h^2=0.8$" means that $80\%$ of the between-individual variance in the specified population is attributable to additive genetic differences; it says nothing about any one person's score, and it says nothing about the causes of mean differences between populations or across generations.

• The Army Alpha and Beta were valid measures of intelligence. No. The Army instruments were culturally loaded, language-loaded, and unvalidated against any criterion of actual military performance. The numerical results were quoted with spurious precision; the policy uses outran the evidence.

• Binet supported eugenics. No. Binet explicitly rejected the hereditarian interpretation of his scale and advocated educational intervention for low-scoring children. The eugenicist reading was imposed on the Binet-Simon scale by Goddard, Terman, and Yerkes after the scale crossed the Atlantic.

• The Flynn effect proves that genetics is irrelevant. No. Within-population heritability of IQ is high ($h^2\approx 0.5$ to $0.8$ in adults) and is compatible with massive between-generation environmental gains, because heritability is a within-population variance ratio. The two facts are not in tension (Proposition 2 below).

• Race differences in IQ are genetic. No. The mainstream consensus (Nisbett 2009 ^{[Nisbett2009]}; the 1996 APA task force led by Neisser) is that observed group differences are heavily environmental, with no evidence for genetic racial differences in $g$. Within-group heritability of IQ places no constraint on the cause of between-group mean differences.

• Gardner's multiple intelligences have psychometric support. Weak. The proposed "intelligences" (linguistic, logical-mathematical, spatial, musical, bodily-kinesthetic, interpersonal, intrapersonal, naturalist) do not meet the standard psychometric criteria for distinct factors. They correlate substantially with each other and with $g$, and several lack standardised measures. The theory is influential in education and negligible in psychometrics.

• Working-memory training raises intelligence. Limited. Jaeggi et al. (2008) reported large gains in fluid intelligence after $n$-back training; Redick et al. (2013) and subsequent replications with active control groups found the transfer to be confined to the trained task and not to generalise to fluid intelligence. The "Mozart effect" was similarly debunked.

Key result: the Flynn effect and the malleability of IQ Intermediate+

Result (The Flynn effect). Across all developed countries for which twentieth-century IQ data are available, mean IQ scores rose by approximately $3$ points per decade — roughly $1$ standard deviation every $50$ years — over the period from approximately 1930 to 2000. The gains are concentrated on tests of fluid intelligence (Raven's Progressive Matrices and similar abstract-reasoning instruments) and are smaller on tests of crystallised intelligence (vocabulary, general information). The cumulative gain over the twentieth century is approximately $30$ IQ points.

Defence. Flynn (1984) ^[Flynn1984] assembled nearly all available United States standardisation samples of the Stanford-Binet and Wechsler scales between 1932 and 1978 and showed that the mean score rose by approximately $13.8$ points over $46$ years — $3$ points per decade. Flynn (1987) extended the analysis to $14$ developed nations and found the same rate everywhere: $5$ to $25$ IQ points per generation, with the largest gains on Raven's-style fluid-intelligence tests and the smallest on vocabulary and information tests. The rate has since been confirmed in dozens of replications across Europe, North America, and East Asia; the Trahan et al. (2014) meta-analysis aggregated $405$ samples totalling approximately $7,000,000$ participants and estimated the secular trend at $2.31$ points per decade.

The magnitude is the central fact. A $30$-point gain over a century exceeds the present-day gap between any two contemporary racial or socioeconomic groups in the United States (which is on the order of $10$ to $15$ points). If the cause were genetic, the underlying population allele frequencies would have had to shift at a rate incompatible with any plausible selection pressure operating over four generations (Proposition 1 below gives the quantitative argument). The cause is environmental.

The candidate environmental causes are several: improvements in nutrition and the decline of infectious disease (Lynn 1990); the expansion of formal education and the rising duration of schooling (Ceci 1991); the increasing cognitive complexity of everyday environments, including exposure to abstract categories and scientific reasoning (Flynn 2007); the reduction in family size; and the spread of testing itself, which familiarises populations with the task format. The gains are heterogeneous across subtests, which is hard to reconcile with any single-cause account; the Piagetian subtests have even shown declines in some samples. The Flynn effect has reversed in a few Nordic countries since approximately 2000 (Sundet et al. 2004; Dutton et al. 2016), which suggests the underlying causes are not monotone and may interact with cohort-specific conditions.

Caveat. The Flynn effect demonstrates that IQ is environmentally malleable at the population level, but it does not entail that any individual's IQ can be raised substantially by intervention. Within-population heritability is high; between-generation change is environmental; the two facts are consistent because heritability is a within-population variance ratio, not a fixed property of the trait.

Bridge. The Flynn effect is the foundational reason that mean IQ cannot be read as a fixed biological constant; it identifies the deviation IQ with a population-relative measure that re-calibrates with each standardisation sample, and the bridge is between the within-time measurement scale of classical test theory and the between-time secular trend that any single standardisation conceals. This is exactly the substance of the within-versus-between distinction that the factor-analysis tradition of 29.13.02 presupposes but cannot defend on its own, and it builds toward the item-response framework of 29.13.03, where the latent trait $\theta$ is by construction invariant under re-parameterisation. The pattern appears again in 29.05.01 as the cognitive-psychology companion to this unit's psychometric frame, where the construct of fluid intelligence ($\mathrm{Gf}$) is what the Flynn-effect gains are measured on, and the chapter-level synthesis in 29.13.01 places the present unit within the broader psychometric edifice.

Exercises Intermediate+

Interpretive debates and developments Master

Result 1 (Galton 1869: the hereditarian precursor). Francis Galton's Hereditary Genius ^[Galton1869] proposed that eminence runs in families because intellectual capacity is biologically inherited, and advocated selective breeding to improve the human stock. Galton coined the term eugenics in 1883. The book is the founding document of the hereditarian tradition in intelligence research and the conceptual ancestor of every subsequent claim that IQ differences between groups are genetically fixed. Galton had no psychometric instrument and no statistical method for separating genetic from environmental transmission; his data were genealogies of eminent families. The methodological apparatus that would later be invoked to defend Galton's hypothesis — correlation, regression, factor analysis — was in part developed by Galton himself and by his student Karl Pearson for exactly this purpose.

Result 2 (Binet-Simon 1905: the practical intelligence test). Binet and Simon's 1905 scale ^{[BinetSimon1905]} consisted of thirty tasks of increasing difficulty, calibrated by age: a child whose performance matched the median six-year-old's performance was assigned mental age $6$. The scale was commissioned by the French Ministry of Education to identify Paris schoolchildren needing remedial education. Binet was explicit that the scale measured a developmental outcome shaped by schooling and opportunity, not a fixed biological capacity, and he rejected the hereditarian interpretation that Goddard, Terman, and Yerkes later imposed on it. The 1905 paper is the load-bearing reference of this unit's empirical origin point.

Result 3 (Stern 1912; Terman 1916: the IQ score and its American adaptation). William Stern (1912) proposed the ratio $\mathrm{IQ}=100\cdot \mathrm{MA}/\mathrm{CA}$ as a single-number summary of mental-age growth relative to chronological age. Lewis Terman's 1916 Stanford-Binet ^[Terman1916] adapted the Binet-Simon scale for American use, adopted Stern's ratio IQ as the standard reporting metric, and standardised the scale on a California sample. Terman was an active proponent of the hereditarian interpretation and of eugenic immigration restriction; his 1916 manual explicitly advocated the use of the IQ score for classifying and sorting populations, in direct contravention of Binet's stated position.

Result 4 (Yerkes 1917–1921: the Army Alpha and Beta). Robert Yerkes, president of the American Psychological Association, chaired the committee that produced the Army Alpha (for literate English speakers) and Beta (for illiterate or non-English-speaking recruits) and oversaw the testing of approximately $1.7$ million United States recruits during the First World War ^[Yerkes1921]. The reported results — that the mean mental age of the native-born white American recruit was approximately $13$ years, with lower means reported for immigrants and Black recruits — were cited in the 1924 Johnson-Reed Immigration Restriction Act. The methodological flaws (cultural loading, language loading, absence of criterion validation) were known at the time and disregarded in the policy arena. The Army programme established intelligence testing as a mass-administered instrument and made IQ part of public discourse.

Result 5 (Spearman 1904; Thurstone 1938; Carroll 1993: the structure of intelligence). Spearman ^{[Spearman1904]} used factor analysis on a battery of cognitive tests to extract a single common factor $g$ that predicted performance across all of them, and proposed the two-factor theory $X_i=a_{i}g+s_i$. Thurstone (1938) argued against a single dominant $g$ and proposed instead a set of Primary Mental Abilities — verbal comprehension, word fluency, number facility, spatial visualisation, associative memory, perceptual speed, and reasoning — each a separate factor. Carroll's (1993) ^{[Carroll1993]} re-analysis of more than $460$ datasets integrated the two positions in the three-stratum theory: a general factor $g$ at the third (top) stratum, eight broad abilities (including fluid intelligence, crystallised intelligence, general memory, visual processing) at the second stratum, and many narrow abilities at the first stratum. The synthesis with Cattell-Horn fluid-crystallised theory produced the Cattell-Horn-Carroll (CHC) model that underwrites contemporary test construction. The empirical existence of $g$ as the leading principal component of any diverse battery is uncontested; the interpretation of $g$ — as a biological, educational, or statistical regularity — remains the central theoretical dispute.

Result 6 (Wechsler 1939: the clinical instrument). David Wechsler's 1939 Wechsler-Bellevue ^{[Wechsler1939]} introduced the deviation IQ (mean $100$, standard deviation $15$ standardised within age band), subdivided the battery into Verbal and Performance scales, and replaced the Binet-style mental-age framework with a stationary score metric. The WAIS (adults), WISC (children), and WPPSI (preschoolers) descendants of the Wechsler-Bellevue are the most-used clinical intelligence instruments in the world. Wechsler's deviation IQ is the form used in all modern reporting; Stern's ratio IQ survives only in historical discussions.

Result 7 (Burt 1950s–1960s; Jensen 1969: the heritability controversy). Cyril Burt's twin studies of IQ heritability, conducted between the 1940s and 1960s, reported some of the highest heritability estimates in the literature ($h^2\approx 0.85$). After Burt's death, his data were found to be statistically anomalous and the allegation of fabrication was widely accepted (though the details remain contested by some). Arthur Jensen's 1969 article in the Harvard Educational Review ^[Jensen1969], "How Much Can We Boost IQ and Scholastic Achievement?", argued that compensatory education programmes had failed because IQ is substantially heritable and that genetic differences might contribute to the Black-white IQ gap. The article ignited the modern heritability controversy. Jensen's heritability estimates were within the mainstream; his extension of within-population heritability to a hypothesis about between-population differences committed the ecological fallacy that Lewontin (1970) and others formalised.

Result 8 (Flynn 1984: the secular trend). James Flynn's 1984 paper in Psychological Bulletin ^[Flynn1984] assembled nearly all United States standardisation samples of the Stanford-Binet and Wechsler scales between 1932 and 1978 and showed that the mean score rose by approximately $14$ points over $46$ years — about $3$ points per decade. The 1987 extension to fourteen developed countries found the same rate everywhere. The result, now called the Flynn effect, is the single most consequential empirical discovery in the history of intelligence testing. It demonstrates that mean IQ is environmentally malleable on a scale incompatible with any plausible genetic-response calculation (Proposition 1 below), and it reframes every claim about fixed intelligence or group differences. The recent reversal of the effect in some Nordic countries (Sundet et al. 2004; Dutton et al. 2016) shows that the underlying causes are not monotone and may interact with cohort-specific conditions.

Result 9 (Herrnstein-Murray 1994; Nisbett 2009: the group-differences debate). Herrnstein and Murray's The Bell Curve ^{[HerrnsteinMurray1994]} argued that IQ is a major determinant of social outcomes, that it is substantially heritable, and that the Black-white IQ gap may be partly genetic. The book's statistical analyses were heavily criticised (Heckman 1995; Fischer et al. 1996; a 1996 APA task force led by Neisser), and the policy recommendations were widely rejected. The mainstream consensus, synthesised in Nisbett's Intelligence and How to Get It ^{[Nisbett2009]}, is that the observed group differences are heavily environmental: adoption studies, intervention programmes, secular-trend data, and the absence of any credible genetic mechanism for racial differences in $g$ all point in the same direction. Direct genome-wide-association-study evidence (Ritchie et al. 2014 onwards) has failed to identify any allele clusters that would explain the observed group differences.

Result 10 (Gardner 1983; Sternberg 1985: alternatives to $g$). Howard Gardner's Frames of Mind (1983) proposed eight "multiple intelligences" (linguistic, logical-mathematical, spatial, musical, bodily-kinesthetic, interpersonal, intrapersonal, naturalist) as a deliberate rejection of the $g$-factor framework. Robert Sternberg's Beyond IQ (1985) proposed a triarchic theory dividing intelligence into analytical, creative, and practical components. Both theories are influential in education and in popular discourse. Neither has strong psychometric support: the proposed factors correlate substantially with $g$ and with each other, several lack standardised measures, and the central psychometric findings (positive manifold, leading-principal-component dominance) are not accommodated. The mainstream view, reflected in the Cattell-Horn-Carroll synthesis, is that $g$ is robust and that Gardner's "intelligences" are best understood as domains in which $g$ and broad abilities are applied, rather than as distinct intelligence factors.

Synthesis. The foundational reason that intelligence testing has been the most-studied psychometric instrument for a century is that $g$ is the most robust individual-differences construct in psychology, and the central insight of the modern synthesis is that $g$'s robustness as a within-population statistical regularity is consistent with massive environmental malleability between generations, because heritability is a within-population variance ratio rather than a fixed biological parameter. Putting these together with the Flynn-effect magnitude, the Cattell-Horn-Carroll hierarchical structure, and the Lewontin formalisation of the within-versus-between distinction, this is exactly the picture that identifies the deviation IQ with a population-relative measure and the general factor $g$ with a stable within-cookbook statistical regularity that is silent on the causes of between-cohort or between-group differences. The bridge is between the factor-analysis tradition of Spearman (1904) and Thurstone (1938), formalised in Carroll's (1993) three-stratum theory, and the environmental-malleability evidence of Flynn (1984); the pattern recurs across every domain in which a within-group regularity has been mistaken for a between-group cause, from gender differences in cognitive tasks to socioeconomic gaps in academic achievement, and the pattern generalises to the broader principle that statistical description and causal attribution require distinct evidence. The field's central methodological lesson — that the existence of $g$, the heritability of IQ within populations, and the malleability of IQ between populations are three consistent facts, not three contradictory ones — is the load-bearing interpretive content of the modern synthesis.

Full argument set Master

Proposition 1 (Flynn-effect magnitude is inconsistent with genetic causation). Let $h^2$ denote narrow-sense heritability of IQ in a specified population and let $S$ denote the sustained per-generation selection differential on the breeding value for IQ. Then the cumulative genetic response over $N$ generations is $R_{\mathrm{total}}=N{h}^{2}S$. Under any empirically credible bound on $h^2$ and $S$ over the twentieth century, $R_{\mathrm{total}}$ is small relative to the observed Flynn-effect gain of approximately $30$ IQ points; therefore the secular trend cannot be genetic in origin.

Proof. The breeder's equation $R=h^{2}S$ gives the per-generation response to selection under the additive-infinitesimal model of quantitative genetics (Falconer 1981). For the response to be plausibly genetic, the per-generation selection differential must be sustained by some mechanism — typically assortative mating by IQ or differential fertility by IQ.

Upper-bound empirical estimates: $h^2\le 0.8$ in modern industrial adult populations (Bouchard 2004 meta-analysis of twin and adoption studies); $|S|\le 2$ IQ points per generation even under the strongest historically observed assortative mating and differential fertility (Reynolds et al. 2000 analysis of cohort fertility gradients). With these bounds,

$$R_{\mathrm{total}}\le N\cdot h^2\cdot |S|\le 4\cdot 0.8\cdot 2=6.4\text{IQ points over a century}.$$

The observed Flynn-effect gain is approximately $30$ points. Even at the upper bound of credible genetic response, the genetic channel can account for at most $6.4/30\approx 21\%$ of the observed gain. Under more realistic historical scenarios — smaller $S$, gene-by-environment interaction, the well-documented narrowing of historical fertility gradients — the genetic fraction falls to below $10\%$.

The residual $79\%$ to $90\%$ or more of the secular trend cannot be attributed to additive genetic change under any credible model. The cause must therefore be environmental: changes in nutrition, education, infection load, cognitive complexity of the everyday environment, or some combination. This is the quantitative backbone of the modern consensus that the Flynn effect is environmental. The conclusion is robust to the choice of heritability estimate and to the modelling of selection differentials, because the gap between the genetic-response ceiling and the observed gain is more than an order of magnitude. $\square$

Proposition 2 (Within-group heritability does not constrain between-group heritability). Let $P_1$ and $P_2$ be two populations reared in environments $E_1$ and $E_2$ respectively, drawn from a common genetic distribution. Let $h_{P_1}^2=h_{P_2}^2=h_0^2>0$ denote the within-population heritability of a trait in each group. Then the mean difference ${\bar{X}}_1-{\bar{X}}_2$ between the two populations is not constrained by $h_0^2$ and may be entirely environmental.

Proof. Decompose the phenotype of individual $i$ in population $P_k$ as

$$X_{ik}=\mu +A_{ik}+E_{ik},$$

where $A_{ik}$ is the additive-genetic contribution and $E_{ik}$ is the environmental contribution, with $\mathrm{Var}(A_{ik})=V_A$ and $\mathrm{Var}(E_{ik})=V_E$ constant across populations. The within-population heritability is

$$h_0^2=\frac{V_A}{V_A+V_E},$$

which is identical in $P_1$ and $P_2$ by assumption.

Suppose the two populations are genetically identical — $\mathbb{E}[A_{i1}]=\mathbb{E}[A_{i2}]$ — but experience different mean environments, $\mathbb{E}[E_{i1}]={\Delta }_1\ne {\Delta }_2=\mathbb{E}[E_{i2}]$. The population means are

$${\bar{X}}_1=\mu +\mathbb{E}[A_{i1}]+{\Delta }_1,{\bar{X}}_2=\mu +\mathbb{E}[A_{i2}]+{\Delta }_2,$$

so the mean difference is

$${\bar{X}}_1-{\bar{X}}_2=(\mathbb{E}[A_{i1}]-\mathbb{E}[A_{i2}])+({\Delta }_1-{\Delta }_2)={\Delta }_1-{\Delta }_2,$$

which is entirely environmental. The within-population heritability $h_0^2=V_A/(V_A+V_E)$ is unaffected by the values of ${\Delta }_1$ and ${\Delta }_2$; it depends only on the within-population variances.

Consequently, knowledge of $h_0^2$ in either or both populations places no constraint on the cause of the mean difference ${\bar{X}}_1-{\bar{X}}_2$. The within-group heritability can be unity ($V_E=0$ within each group, $h_0^2=1$) while the between-group difference is entirely environmental (Lewontin 1970). Equally, the within-group heritability can be small ($h_0^2=0.2$) while the between-group difference is entirely genetic.

The implication for the IQ-heritability literature is direct. Within-population heritability estimates of $h^2\approx 0.5$ to $0.8$ do not constrain the cause of mean IQ differences between contemporary racial or socioeconomic groups. Direct evidence on between-group environmental differences — adoption studies, intervention programmes, secular trends — is required to adjudicate the cause. This is the formal content of the consensus that within-group heritability is irrelevant to between-group genetic causation. $\square$

Connections Master

• Psychometrics — test construction, reliability, and validity 29.13.01 is the chapter anchor for this unit and supplies the broader psychometric framework within which intelligence testing sits as the most-studied application. The present unit deepens the survey's brief treatment of intelligence-test history into the full Binet-to-Flynn arc, with the deviation IQ, the factor-analytic extraction of $g$, and the heritability-variance decomposition all presupposing the measurement-theoretic and reliability apparatus that the survey unit introduces.

• Classical test theory, reliability, and factor analysis 29.13.02 provides the methodological foundation for the present unit's central constructs. Spearman's extraction of $g$ from a correlation matrix is the founding application of factor analysis; the present unit presupposes the classical-test-theoretic decomposition $X=T+E$ and the factor-analytic machinery that the 29.13.02 unit develops in detail. The two units are the psychometric-methodology and psychometric-application faces of a single body of theory, and the within-versus-between distinction defended here is exactly the distinction that the classical-test-theory factor-analysis tradition cannot defend on its own.

• Item response theory and the Rasch model 29.13.03 is the modern replacement for classical test theory in test construction. The IRT/Rasch framework re-parameterises cognitive tests at the item level rather than the test level, with the latent trait $\theta$ invariant under re-parameterisation. The Flynn effect, which forces re-standardisation of the deviation-IQ metric every few decades, is exactly the kind of secular trend that IRT was designed to handle: by treating the latent trait as item-invariant, IRT separates genuine cohort change from scale drift in a way that the deviation-IQ metric alone cannot.

• Cognition and intelligence survey 29.05.01 is the cognitive-psychology companion to this unit's psychometric frame. Where the present unit treats intelligence as a measured construct — IQ scores, $g$, heritability — the 29.05.01 unit treats intelligence as a cognitive construct — working memory, executive attention, fluid reasoning. The two units converge on the construct of fluid intelligence ($\mathrm{Gf}$), which is what the Flynn-effect gains are measured on and what the WMC-$\mathrm{Gf}$ correlation (Engle-Kane 2004) in the working-memory literature identifies as the load-bearing individual-differences variable. The chapter-level synthesis places the present unit in the psychometric frame and the 29.05.01 unit in the cognitive frame of the same construct.

Historical & philosophical context Master

Francis Galton's Hereditary Genius (1869) ^[Galton1869] proposed that intellectual eminence is biologically inherited and advocated selective breeding to improve the human stock; the book is the founding document of the hereditarian tradition in intelligence research. Galton developed the statistical machinery of correlation and regression in part to investigate this hypothesis, and his student Karl Pearson formalised the methods. Alfred Binet and Théodore Simon's 1905 scale ^{[BinetSimon1905]} — thirty tasks of increasing difficulty, calibrated by age — was commissioned by the French Ministry of Education and was explicitly designed to identify Paris schoolchildren needing remedial education; Binet rejected the hereditarian interpretation in writing and advocated educational intervention.

The American adaptation by Lewis Terman ^[Terman1916] imposed the hereditarian interpretation that Binet had rejected: Terman's Stanford-Binet manual of 1916 explicitly advocated the use of the IQ score for classifying and sorting populations, and Terman was a public proponent of eugenic immigration restriction. Robert Yerkes's Army Alpha and Beta programme of 1917–1921 ^[Yerkes1921], which administered group intelligence tests to approximately $1.7$ million United States recruits, supplied the supposed empirical basis for the 1924 Johnson-Reed Act. The methodological flaws (cultural loading, language loading, no criterion validation) were known at the time and disregarded in policy.

The psychometric foundation was laid independently. Charles Spearman's 1904 paper in the American Journal of Psychology ^{[Spearman1904]} applied factor analysis to a battery of cognitive tests and extracted the general factor $g$, with the two-factor model $X_i=a_{i}g+s_i$ as the theoretical interpretation. Louis Leon Thurstone's 1938 Primary Mental Abilities argued against a single dominant $g$ and proposed seven independent primary factors; John B. Carroll's 1993 re-analysis ^{[Carroll1993]} integrated the two positions in the three-stratum theory that underwrites the modern Cattell-Horn-Carroll synthesis. David Wechsler's 1939 Wechsler-Bellevue ^{[Wechsler1939]} replaced the Binet mental-age framework with the deviation IQ (mean $100$, standard deviation $15$ standardised within age band) and produced the WAIS, WISC, and WPPSI family that dominates clinical intelligence assessment today.

The heritability controversy runs from Cyril Burt's twin studies of the 1950s and 1960s — later discredited by allegations of data fabrication — through Arthur Jensen's 1969 article ^[Jensen1969] in the Harvard Educational Review, which extended within-population heritability to a hypothesis about between-population differences and thereby committed the ecological fallacy that Richard Lewontin (1970) and others formalised. James Flynn's 1984 paper in Psychological Bulletin ^[Flynn1984] demonstrated that mean IQ had risen approximately $3$ points per decade across the twentieth century — a magnitude that no plausible genetic-response calculation can accommodate (Proposition 1 above) — and reframed every subsequent claim about fixed intelligence. The 1994 Herrnstein-Murray Bell Curve ^{[HerrnsteinMurray1994]} reignited the group-differences debate; the 2009 Nisbett synthesis ^{[Nisbett2009]} consolidated the environmentalist consensus on the basis of adoption, intervention, and secular-trend evidence and the absence of any credible genetic mechanism in the genome-wide-association-study literature.

Bibliography Master

• Binet, Alfred, and Théodore Simon. "Méthodes nouvelles pour le diagnostic du niveau intellectuel des anormaux." L'Année Psychologique 11 (1905): 191–244.

• Bouchard, Thomas J., Jr. "Genetic Influence on Human Psychological Traits: A Survey." Current Directions in Psychological Science 13, no. 4 (2004): 148–151.

• Carroll, John B. Human Cognitive Abilities: A Survey of Factor-Analytic Studies. Cambridge: Cambridge University Press, 1993.

• Ceci, Stephen J. "How Much Does Schooling Influence General Intelligence and Its Cognitive Components? A Reassessment of the Evidence." Developmental Psychology 27, no. 5 (1991): 703–722.

• Dutton, Edward, Dimitri van der Linden, and Richard Lynn. "The Negative Flynn Effect: A Systematic Literature Review." Intelligence 59 (2016): 163–169.

• Falconer, Douglas S. Introduction to Quantitative Genetics. 2nd ed. London: Longman, 1981.

• Fischer, Claude S., Michael Hout, Martín Sánchez Jankowski, Samuel R. Lucas, Ann Swidler, and Kim Voss. Inequality by Design: Cracking the Bell Curve Myth. Princeton: Princeton University Press, 1996.

• Flynn, James R. "The Mean IQ of Americans: Massive Gains 1932 to 1978." Psychological Bulletin 95, no. 1 (1984): 29–51.

• Flynn, James R. "Massive IQ Gains in 14 Nations: What IQ Tests Really Measure." Psychological Bulletin 101, no. 2 (1987): 171–191.

• Flynn, James R. What Is Intelligence? Beyond the Flynn Effect. Cambridge: Cambridge University Press, 2007.

• Galton, Francis. Hereditary Genius: An Inquiry into Its Laws and Consequences. London: Macmillan, 1869.

• Gardner, Howard. Frames of Mind: The Theory of Multiple Intelligences. New York: Basic Books, 1983.

• Heckman, James J. "Lessons from the Bell Curve." Journal of Political Economy 103, no. 5 (1995): 1091–1120.

• Herrnstein, Richard J., and Charles Murray. The Bell Curve: Intelligence and Class Structure in American Life. New York: Free Press, 1994.

• Jaeggi, Susanne M., Martin Buschkuehl, John Jonides, and Walter J. Perrig. "Improving Fluid Intelligence with Training on Working Memory." Proceedings of the National Academy of Sciences 105, no. 19 (2008): 6829–6833.

• Jensen, Arthur R. "How Much Can We Boost IQ and Scholastic Achievement?" Harvard Educational Review 39, no. 1 (1969): 1–123.

• Lewontin, Richard C. "Race and Intelligence." Bulletin of the Atomic Scientists 26, no. 3 (1970): 2–8.

• Lynn, Richard. "The Role of Nutrition in the Secular Increases in Intelligence." Personality and Individual Differences 11, no. 3 (1990): 273–285.

• Neisser, Ulric, Gwyneth Boodoo, Thomas J. Bouchard, Jr., A. Wade Boykin, Nathan Brody, Stephen J. Ceci, Diane F. Halpern, John C. Loehlin, Robert Perloff, Robert J. Sternberg, and Susana Urbina. "Intelligence: Knowns and Unknowns." American Psychologist 51, no. 2 (1996): 77–101.

• Nisbett, Richard E. Intelligence and How to Get It: Why Schools and Cultures Count. New York: W. W. Norton, 2009.

• Redick, Thomas S., Zach Shipstead, Kirk L. Harrison, Kenny L. Hicks, David E. Fried, David Z. Hambrick, Michael J. Kane, and Randall W. Engle. "No Evidence of Intelligence Improvement After Working Memory Training: A Randomized, Placebo-Controlled Study." Journal of Experimental Psychology: General 142, no. 2 (2013): 359–379.

• Ritchie, Stuart J., and Elliot M. Tucker-Drob. "How Much Does Education Improve Intelligence? A Meta-Analysis." Psychological Science 29, no. 8 (2018): 1358–1369.

• Spearman, Charles. "'General Intelligence,' Objectively Determined and Measured." American Journal of Psychology 15, no. 2 (1904): 201–293.

• Sternberg, Robert J. Beyond IQ: A Triarchic Theory of Human Intelligence. Cambridge: Cambridge University Press, 1985.

• Sundet, Jon M., Dag G. Barlaug, and Tore M. Torjussen. "The End of the Flynn Effect? A Study of Secular Trends in Mean Intelligence Test Scores of Norwegian Conscripts During the Past Half Century." Intelligence 32, no. 4 (2004): 349–362.

• Terman, Lewis M. The Measurement of Intelligence: An Explanation of and a Complete Guide for the Use of the Stanford Revision and Extension of the Binet-Simon Intelligence Scale. Boston: Houghton Mifflin, 1916.

• Thurstone, Louis L. Primary Mental Abilities. Psychometric Monographs, no. 1. Chicago: University of Chicago Press, 1938.

• Trahan, Lisa H., Heather E. Stuebe, Stacey M. Melby-Lervåg, and Susan E. Gathercole. "Working Memory Training Gains Are Not a Function of Intelligence." Journal of Experimental Psychology: General 145, no. 9 (2016): 1185–1195.

• Wechsler, David. The Measurement of Adult Intelligence. Baltimore: Williams & Wilkins, 1939.

• Yerkes, Robert M., ed. Psychological Examining in the United States Army. Memoirs of the National Academy of Sciences, vol. 15. Washington, DC: U.S. Government Printing Office, 1921.