IQ, Inheritance, and the Genome: What “Genetic” Means, What It Does Not, and Why Sex Chromosomes Complicate the Story

2026-01-07 · 9,998 words · Singular Grit Substack · View on Substack

A long-form guide to heritability across development, polygenic variation, gene–environment interplay, and what can (and cannot) be inferred from XX/XY biology.

Keywords

Intelligence; IQ; cognitive ability; heritability; behavioural genetics; twin studies; adoption studies; GWAS; polygenic scores; SNP heritability; gene–environment correlation; assortative mating; X chromosome; X-inactivation; dosage compensation; sex differences; variability.

Abstract

Sex differences in intelligence are often misdescribed as disputes about average IQ. The more consequential and empirically tractable question is variance: whether males show a wider dispersion of scores than females, creating excess representation at both low and high tails. Large-scale evidence has repeatedly reported modest but persistent male-over-female variance ratios across multiple cognitive domains, alongside near-zero mean differences in overall IQ (Hedges & Nowell, 1995; Giofrè et al., 2024). This article advances a chromosomal mechanism for that pattern. The core proposal is that XY males express X-linked allelic effects without a second X to buffer deleterious or advantageous variants, while XX females exhibit heterozygous averaging and tissue-level mosaicism through X-chromosome inactivation, producing partial phenotypic cancellation and reduced between-individual variability for X-influenced components of cognitive function (Johnson et al., 2009; Migeon, 2007). The argument is biologically specific: it links observed variance patterns to sex-chromosome genetics, dosage compensation, and the known concentration of neurodevelopmentally relevant loci on the X chromosome, including the longstanding overrepresentation of X-linked mutations associated with cognitive impairment (Gécz & Mulley, 2000). The article synthesises psychometric findings on variance ratios with genetic and molecular evidence about X-linked contributions to brain and cognition, then outlines testable predictions using sex-chromosome aneuploidies, twin and sibling designs, and modern genotype-based partitioning of variance. The conclusion is that XY/XX differences are not rhetorical garnish: they are a plausible, mechanistically grounded source of measurable differences in IQ dispersion that must be explicitly modelled when explaining sex-patterned outcomes at the tails of cognitive distributions.

Thesis

This article argues that the XY/XX chromosomal architecture produces measurable, population-level differences in the dispersion of cognitive ability as indexed by IQ: male hemizygosity for X-linked variation increases phenotypic spread (more males at both extremes), whereas female XX biology—through heterozygosity and cellular mosaicism created by X-chromosome inactivation—dampens the expression of single-locus X-linked effects, tightening the distribution and lowering the standard deviation relative to males (Hedges & Nowell, 1995; Johnson et al., 2009; Migeon, 2007). The claim is not that the sexes differ meaningfully in mean IQ, but that the variance structure differs in a way that is partly genetic and specifically sex-chromosome mediated (Giofrè et al., 2024; Gécz & Mulley, 2000).

Section 1 — Framing the claim: variance, not means

The object of explanation is not the average score but the geometry of the distribution. The question is not whether males or females “are smarter” on some single scale; it is whether the distribution of IQ scores differs in breadth, such that one sex shows a wider spread and therefore a higher proportion of individuals at both extremes. In statistical terms, the target is variance (or standard deviation) and the ratio of variances across sexes. If male scores are more dispersed, the male histogram will appear flatter and wider, with thicker tails; if female scores are less dispersed, the female histogram will appear taller and tighter around the mean. This essay treats that dispersion difference as the primary phenomenon, and it treats any discussion of mean differences as secondary and analytically separable.

That separation matters because three different claims are often collapsed into one. Mean differences concern whether the centre of the distributions differs, typically evaluated by comparing group means and effect sizes. Variance differences concern whether one distribution is wider, typically evaluated by comparing standard deviations or the variance ratio (male variance divided by female variance) and by tests of equality of variances. Tail representation concerns how many individuals fall into extreme quantiles—high or low—relative to a fixed threshold. Crucially, tail differences can appear even when means are essentially equal, provided that one group has a wider distribution. A modest variance ratio above 1.0 predicts a larger male representation at both the low and high ends without any need to posit a meaningful mean shift. This essay therefore frames the issue as one of dispersion and tail weight, not as a contest over average ability.

The position taken is explicit. The male–female dispersion gap in IQ is argued to be substantially genetic in origin and to be importantly mediated by sex chromosomes, particularly the X chromosome. The proposed mechanism is not vague “biological difference” but a specific variance pathway: XY males express X-linked allelic effects without a second X to buffer them, increasing phenotypic spread, while XX females benefit from heterozygosity and X-inactivation mosaicism that can dampen the net expression of X-linked variation, tightening the distribution. The empirical commitment in this opening section is therefore precise: the essay is about the variance ratio, its visual signature in histograms and tails, and the contention that a non-trivial portion of that ratio is causally traceable to XX/XY chromosomal architecture (Hedges & Nowell, 1995; Johnson et al., 2009; Migeon, 2007).

Figure 1. Equal means, different dispersion in IQ distributions-

The figure shows two normal distributions with identical means (μ = 100) and different standard deviations. The male distribution (SD = 19) is wider and flatter, producing heavier upper and lower tails, while the female distribution (SD = 14) is tighter and taller around the mean. With means held constant, the variance difference alone generates disproportionate male representation at both extremes of the IQ scale.

Section 2 — What IQ data look like: distributions, tails, and variance ratios

The descriptive core of this essay is visual and distributional. IQ is treated as a continuous score with a full distribution for males and a full distribution for females, not as a single number per group. The first display is the histogram: counts (or proportions) of individuals falling into score bins across the IQ range. A histogram is not “decoration”; it is the most direct way to see dispersion. With equal or near-equal means, the narrower distribution will be taller around the centre, while the wider distribution will be flatter, placing more mass into both tails. For that reason, the histogram is paired with a smooth density estimate: a kernel density overlay that shows the same information without bin edges, making it easier to compare the central peak and the tails in one visual layer. The histogram answers “how many people sit in each score band,” while the density overlay answers “what is the implied shape,” and the two together reduce the risk that a single visual choice drives the reader’s conclusion.

Figure 2. Example histograms of IQ for males and females with equal means and different dispersion-

This figure shows simulated IQ distributions for males and females with identical means (100) and different standard deviations (males SD = 19; females SD = 14). The histograms are plotted with identical bin widths and overlaid with their corresponding density curves. The male distribution is visibly wider and flatter, while the female distribution is narrower and taller around the mean.

The variance ratio implied by these parameters is 1.84, meaning male variance is approximately 84 per cent larger than female variance. With means held constant, this difference in spread alone generates heavier male representation in both the lower and upper tails of the distribution. The figure is intended to illustrate the descriptive logic of variance-based claims: dispersion, not average score, determines tail outcomes.

The key summary statistics follow directly from these plots. The mean locates the centre, but it is not the main target. Dispersion is captured first by the standard deviation (SD), because SD is in the same units as the score and maps cleanly onto the width of the histogram. The variance is the SD squared and is useful because group comparisons of spread are typically framed in variance terms. The central comparative metric is the variance ratio, defined as male variance divided by female variance. A variance ratio greater than 1 indicates greater male dispersion; a ratio less than 1 indicates greater female dispersion. This statistic matters because it turns a qualitative impression (“wider” or “narrower”) into a number that can be compared across cohorts, age bands, and instruments, and it links directly to tail expectations. With equal means, even a modest variance ratio above 1 implies systematically more males at both the low and high extremes, because the wider distribution allocates more probability mass away from the centre.

Figure 3. Quantile plot using empirical CDFs shows tail divergence (male heavier tails)-

This figure plots the empirical cumulative distribution functions for males (SD = 19) and females (SD = 14) with equal means (100). The heavier male tails are visible as follows: at the lower tail, the male ECDF reaches 5 per cent at a lower IQ than the female ECDF (male 5th percentile is lower); at the upper tail, the male ECDF reaches 95 per cent and 99 per cent at higher IQ values than the female ECDF (male 95th and 99th percentiles are higher). The vertical quantile markers make the divergence explicit at the tails while the curves remain close around the centre.

Tails are described explicitly rather than by vague language such as “more extremes.” The plan uses quantile comparisons: the 5th percentile as a lower-tail marker, the 95th percentile as an upper-tail marker, and the 99th percentile as a severe upper-tail marker. These are reported separately for males and females. Quantiles make the tail claim concrete: a wider male distribution predicts that the male 95th percentile will be higher than the female 95th percentile and that the male 5th percentile will be lower than the female 5th percentile, even when the means are the same. In addition, the plan reports fixed-threshold tail proportions, chosen in advance and kept consistent across analyses (for example, the proportion above 130 and below 70 in a conventional IQ scaling). These threshold proportions are not substitutes for quantiles; they are complements, because quantiles anchor “relative position” within each sex distribution, while thresholds anchor “absolute rarity” on the score scale.

Histogram binning choices matter because bins can manufacture patterns. Wide bins can hide tail differences and create the false appearance of similarity; narrow bins can exaggerate noise and create the false appearance of instability. Because of this, bin width is not chosen by eye. The plan uses a stated bin width rule, reports it, and then performs sensitivity checks by re-plotting with alternative defensible binning choices to show that the dispersion conclusion is not an artefact of bin size. The same transparency applies to the density overlay: the smoothing parameter is stated and is checked for robustness so that tails are not artificially flattened or inflated by the choice of smoothing.

The descriptive programme above aligns with the recurring observation in the variability literature: across many cognitive measures, males often show modestly larger variability than females, while mean differences on broad IQ-type composites are small (Hedges & Nowell, 1995). The essay’s commitment is that this will not be asserted rhetorically: it will be displayed as distributions, summarised as SD and variance, stated as a variance ratio, and demonstrated in tails through quantiles and threshold proportions, with all key plotting and summarisation choices made explicit in the text (Hedges & Nowell, 1995).

Table 1. Summary of dispersion and tail proportions by sex-

This table summarises the descriptive statistics corresponding to Figures 2 and 3, using simulated IQ data with equal means (100) and different standard deviations (males SD = 19; females SD = 14). The focus is on dispersion and tails rather than central tendency.

Variance ratio (male / female) = 1.84

The table makes three points explicit. First, the means are effectively identical, confirming that all differences arise from dispersion rather than central location. Second, male variance is substantially larger than female variance, yielding a variance ratio of approximately 1.84. Third, this variance difference translates directly into tail outcomes: males are overrepresented at both the lower and upper extremes, with roughly three to nine times the female representation depending on the threshold. These tail asymmetries emerge mechanically from the variance difference alone, without invoking any mean shift.

Section 3 — Behavioural genetics as the baseline: heritability and stability of cognitive differences

A dispersion argument requires a baseline: evidence that cognitive differences are not random noise and are not purely transient artefacts of schooling, test familiarity, or momentary conditions. Behavioural genetics provides that baseline by repeatedly finding that individual differences in cognitive ability, including IQ and closely related general cognitive composites, are substantially heritable. “Heritable” here is used in its technical sense: within a defined population at a defined time, a non-trivial proportion of observed variation in cognitive scores is statistically attributable to genetic differences among individuals. This is a statement about variance, not a statement about inevitability, and it is precisely the kind of statement required when the outcome of interest is the width of the distribution rather than its centre.

The most relevant behavioural-genetic finding for a dispersion thesis is not merely that heritability exists, but that it shows a systematic developmental pattern. Across longitudinal twin and adoption studies, heritability estimates for cognitive ability tend to rise from early childhood into adolescence and adulthood. Briley and Tucker-Drob (2013) synthesised this evidence meta-analytically and argued that the increase is consistent with processes that amplify genetic differences as development unfolds. Plomin and Deary (2015) likewise emphasised that cognitive differences show robust genetic influence and that this influence becomes more pronounced with age. Taken together, these results support a core proposition: cognitive variation is not only heritable, but development itself tends to magnify the contribution of genetic variation to the observed spread of scores.

Figure 4. Conceptual variance decomposition across development-

This figure presents a conceptual decomposition of total variance in cognitive ability into three components—genetic variance, shared environmental variance, and non-shared environmental variance—across broad developmental stages. The proportions are illustrative rather than empirical point estimates, and they are intended to capture the recurring pattern reported in behavioural-genetic research.

The key feature is the change in relative contribution across age. Genetic variance increases from early childhood into adulthood, while shared environmental variance declines markedly over the same period. Non-shared environmental variance remains present throughout development but does not dominate the variance structure at any stage. This pattern is consistent with findings that heritability of cognitive ability tends to rise with age, reflecting developmental amplification of genetic differences rather than the disappearance of environmental influence.

The figure is included to ground the dispersion argument: if overall cognitive variance becomes increasingly structured by genetic differences as development proceeds, then a sex-linked genetic mechanism affecting variance—such as X-linked hemizygosity in males versus buffering in females—has a plausible pathway to influence the observed width of adult IQ distributions.

This developmental pattern is directly relevant to dispersion because it frames variance as a dynamic product of maturation and cumulative experience rather than a static “starting point.” The rising heritability of cognitive ability is consistent with developmental amplification mechanisms, meaning processes through which small genetic differences early in life can lead to larger phenotypic differences later. The mechanism is not required to be mysterious. As children mature, differences in attention, curiosity, learning speed, and persistence can translate into different trajectories of skill accumulation. Those trajectories, once established, can widen the distribution of observed scores by increasing separation between individuals over time. The critical link to the present essay is that any biological account of sex differences in variability must be compatible with the fact that cognitive dispersion is not fixed at birth, but can be progressively shaped and amplified across development in ways that still preserve a genetic signature.

Behavioural genetics also supports the idea of stability in individual differences. While absolute IQ scores can shift with schooling, health, and measurement context, relative standing tends to show meaningful continuity across time, particularly from later childhood onward. This matters because dispersion claims are claims about persistent distributional properties. If the ranking of individuals were largely random across development, variance patterns would be unstable and explanations would be correspondingly weak. Instead, the behavioural-genetic literature indicates that stable components of cognitive variation exist and are substantially genetic, providing the necessary platform for a genetic account of dispersion.

This section therefore establishes the bridge needed for the chromosomal argument that follows. If cognitive differences are substantially heritable and increasingly so with age, it becomes plausible in principle that a sex-specific genetic architecture could produce measurable sex differences in dispersion. The next step is to specify the mechanism at the appropriate level of biological detail: not “genes in general,” but the asymmetry created by XY hemizygosity and XX buffering processes on the X chromosome as a contributor to variance differences within a broadly heritable, developmentally amplifying trait (Briley & Tucker-Drob, 2013; Plomin & Deary, 2015).

Section 4 — The X chromosome as a variance amplifier in XY males

The sex-chromosome mechanism begins with a structural asymmetry that does not require speculation: XY males carry one X chromosome and one Y chromosome, whereas XX females carry two X chromosomes. For loci on the X chromosome, that difference changes how genetic variation is expressed. In an XY male, any allele present on the X is, in the simplest sense, unpaired at that locus. There is no second X allele at the same position that can partially offset, mask, or dilute its effect. Whatever functional consequences an X-linked allele has—whether it slightly increases or slightly decreases the efficiency of a neurodevelopmental pathway—are expressed in an unbuffered way. This is the core reason the X chromosome can function as a variance amplifier in males: hemizygosity exposes X-linked variation directly to phenotype.

Figure 5. XX buffering versus XY exposure at an X-linked locus (schematic)-

This diagram contrasts an XX individual carrying two alleles at an X-linked locus (illustrated as A and a) with an XY individual carrying only one allele on the X at the same locus (illustrated as a), with no corresponding allele on the Y. The XX case shows the basic “offset/masking” logic: the net organism-level effect can be reduced because two alleles coexist. The XY case shows the “exposure” logic: the single X allele is expressed without an X-paired counterpart, increasing phenotypic sensitivity to X-linked variation.

The logic can be stated stepwise. First, assume that at least some X-linked loci contribute to neurodevelopmental traits that feed into measured cognitive performance. That assumption is not controversial in medical genetics; the X chromosome is well known to contain many genes relevant to brain development and function, and a substantial class of intellectual disability syndromes is X-linked. Second, observe that an XY male carries only one allele at those loci. Third, infer that the phenotypic effect of an X-linked allele, beneficial or deleterious, will be less often buffered by allelic opposition in XY individuals than it would be in an XX individual carrying two X chromosomes. Fourth, aggregate this across many loci and many individuals: the population consequence is increased dispersion. Some males will carry X-linked variants that shift development and functioning downward; others will carry variants that shift upward; and because buffering is reduced, those shifts will be expressed more sharply at the individual level. The predicted signature is wider spread and heavier tails, not an inevitable shift of the mean.Subscribe

This distinction between spread and mean is essential. A variance mechanism can operate even if the average of the distribution remains essentially unchanged. If X-linked alleles include both upward and downward effect directions—an entirely ordinary situation for complex traits—then increased exposure in XY males will produce more individuals deviating from the centre in both directions. That generates a higher male standard deviation and a variance ratio above 1, while leaving the male and female means close. The sex difference is then a difference in the probability of extremes, not necessarily in average performance.

The empirical plausibility of an X-linked contribution is strengthened by the longstanding observation that X-linked neurodevelopmental disorders disproportionately affect males. Lubs et al. (2012) review the molecular and clinical genetics of Fragile X and related X-linked intellectual disability conditions, illustrating both the enrichment of cognition-relevant genes on the X chromosome and the way X-linked mutations can manifest more severely in males. This is not introduced as a rhetorical appeal to pathology, but as evidence that X-linked variation can have cognitive consequences and that male hemizygosity is a biologically meaningful exposure state. The point required for the dispersion thesis is not that the normal IQ range is “the same as” clinical syndromes; it is that X-linked genetic effects exist in the relevant biological domain and that their phenotypic expression is sex-dependent in ways consistent with hemizygous vulnerability.

Figure 6. Exposed X-linked effect → increased variance → heavier tails (schematic)-

This schematic links the proposed causal chain in three steps. First, XY hemizygosity exposes X-linked allelic effects directly. Second, that exposure increases phenotypic sensitivity to X-linked variation at the individual level. Third, aggregating that increased sensitivity across individuals produces a larger population variance (higher SD), which mechanically increases representation in both tails of the IQ distribution (more low-end and high-end extremes) even when the mean remains unchanged.

Once that is accepted, the move from clinical categories to population dispersion is a scaling argument, not a category error. In the population, most X-linked variants affecting cognition will not be catastrophic mutations; they will be smaller-effect alleles that nudge neurodevelopmental efficiency, learning rate, attentional control, or other cognitive subcomponents by modest amounts. Yet dispersion does not require large effects at a single locus. A large number of small X-linked effects, expressed without buffering in XY males, can cumulatively widen the distribution. The outcome is not a dramatic shift in the entire male curve; it is a modest flattening and widening that becomes visible in tail representation, especially in large samples and in high-resolution quantile analyses.

It is also important to be clear about what this mechanism does not claim. It does not claim that “the X chromosome explains all male variability” or that autosomal polygenic influences do not matter. Behavioural genetics and molecular findings indicate that cognition is broadly polygenic, with influential variation distributed across the genome. The present claim is narrower and more specific: given a broadly heritable trait, a sex-linked architecture that exposes one class of loci more directly in one sex can act as a modifier of dispersion. The X chromosome, because it is not paired in males, is precisely such a class.

Finally, the X-linked variance amplifier concept generates a concrete, distribution-level prediction. If the mechanism is operating, then measures of cognitive ability should show a male-over-female variance ratio above 1, and that ratio should express itself most clearly in the tails: more males below low thresholds and more males above high thresholds, even when means are aligned. The mechanism is therefore not merely a narrative; it is a testable mapping from chromosomal architecture to the shape of observed IQ distributions, consistent with the known relevance of X-linked genetic variation to cognitive outcomes (Lubs et al., 2012).

Section 5 — XX buffering: heterozygosity and X-inactivation mosaicism

The dispersion claim requires not only a variance-amplifying pathway in XY males, but also a variance-dampening pathway in XX females. The proposed XX buffering is not a single idea, and it must be kept analytically clean. Two distinct XX features are doing different work: heterozygosity at X-linked loci and tissue-level mosaicism generated by X-chromosome inactivation. The argument in this section is that each feature reduces the effective phenotypic impact of X-linked allelic variation, and that together they compress the spread of outcomes in XX relative to XY, even when means remain closely aligned.

Figure 7. X-inactivation mosaicism in XX: two cell populations with different active X-

This diagram illustrates the mosaic structure created by X-inactivation in XX individuals. Two X chromosomes carry different alleles at the same X-linked locus (shown as A on X1 and a on X2). During development, different cells in the same tissue can inactivate different X chromosomes, producing two intermingled cell populations: one expressing X1 as the active X and the other expressing X2 as the active X. The result is mixed expression across tissues rather than uniform expression of a single X-linked allele, creating a within-individual averaging effect at the tissue and organism level.

The first feature is heterozygosity. In an XX individual, X-linked loci are typically present in two copies, meaning two alleles exist at the same locus. When allelic effects differ, the organism-level outcome can reflect a composite of those effects rather than the full expression of a single allele. The simplest case is a locus with two alleles where one slightly reduces and the other slightly increases the efficiency of a relevant neurodevelopmental process. In an XY male, one allele is expressed without an X-paired counterpart, so the phenotypic effect tracks the single allele more directly. In an XX female, the presence of two alleles creates the possibility of partial offset. This is not a claim that every X-linked allele is recessive, nor that offset is perfect; it is a claim about variance: having two alleles at a locus reduces the probability that an individual’s phenotype will be driven strongly in one direction by a single X-linked allele. Even when dominance and non-additive effects are set aside, the mere fact of two alleles allows intermediate net effects at the organism level more often than in a hemizygous state. In distribution terms, this pulls outcomes back toward the centre more frequently, reducing the tendency of single-locus X-linked variation to generate extremes.

This heterozygosity argument is best understood as a filter on effect sizes. If single-locus X-linked variants contribute to cognition-related pathways, then the expected organism-level effect size of a given allele is, on average, less extreme in XX than in XY because the XX state provides an opportunity for countervailing allelic influence. The key point for the essay is not to overstate the degree of filtering, but to preserve its direction: the XX state weakens the translation of single-locus X-linked variation into phenotype, and weakening that translation reduces variance. It does not require that the mean changes. It requires only that effects exist in both directions and that expression is less “all-or-nothing” in XX than in XY at a subset of loci that matter for cognitive function.

The second feature is X-inactivation mosaicism, and it is conceptually separate from heterozygosity even though they interact. In XX individuals, one X chromosome is largely inactivated in each cell early in development, producing a mosaic pattern in which some cells express one X and other cells express the other X. This mosaicism is not a superficial footnote; it changes the way X-linked alleles manifest across tissues, including brain tissue, because it means that an X-linked variant is not necessarily expressed uniformly across all relevant cells (Migeon, 2007). Instead, its expression can be distributed across cell populations, with some cells expressing the allele from one X and some expressing the allele from the other X. That distribution can attenuate the net functional impact of a variant at the organism level because the phenotype is produced by networks of cells rather than by a single locus acting uniformly in every cell.

Figure 8. Conceptual model of variance dampening in XX versus XY-

This figure illustrates a variance-dampening effect rather than a mean shift. Both curves are centred on the same mean phenotypic effect of an underlying X-linked allele. The XY curve is wider, reflecting higher phenotypic sensitivity due to hemizygous expression of the X-linked variant. The XX curve is narrower, reflecting buffering through heterozygosity and X-inactivation mosaicism, which attenuate the net organism-level impact of the same allelic variation. The consequence is fewer extreme outcomes in XX and heavier tails in XY, even though the average effect remains the same.

The dispersion relevance of mosaicism is straightforward when stated in distribution terms. Uniform expression of an allele across a tissue can push function more coherently in one direction, increasing the chance of a measurable shift in performance. Mosaic expression breaks that uniformity. In effect, it introduces intra-individual averaging across cellular populations. A deleterious allele may be active in a subset of cells while a less deleterious or functionally different allele is active in others, leading to a diluted net effect on tissue performance. Likewise, a beneficial allele may be active in only a subset of cells rather than across the entire relevant system, again reducing the probability of a large organism-level deviation. The consequence is not that XX individuals are “immune” to X-linked variation, but that the same underlying allelic difference is less likely to generate a large phenotypic departure from the mean because the allele is not uniformly expressed.

Heterozygosity and mosaicism therefore dampen dispersion through different routes. Heterozygosity operates at the allelic level by making “two-allele composites” common and “single-allele domination” less common. Mosaicism operates at the cellular architecture level by making expression spatially distributed rather than uniform, reducing coherence of impact across tissues. Together, they generate the same distributional prediction: X-linked variation contributes less to between-individual phenotypic spread in XX than it does in XY. If X-linked variation contributes non-trivially to cognition-related pathways, then XX buffering compresses the XX outcome distribution relative to XY, yielding a smaller standard deviation and lighter tails for females.

This is the point where the argument returns to the measurable signature. If XX buffering is operating, then female distributions should be tighter even when means are similar, and the magnitude of sex differences should be most visible in tail proportions and high-quantile comparisons rather than in mean differences. A variance dampener expresses itself by reducing the frequency of extreme outcomes. Under this model, fewer XX individuals fall far below or far above the mean because the translation of X-linked variation into phenotype is softened by heterozygosity and diluted by mosaic expression (Migeon, 2007). The dispersion claim is therefore not a generic statement that “females are different.” It is a specific, mechanistic claim about why, holding the overall architecture of intelligence constant, the XX state should exhibit less phenotypic variance attributable to X-linked loci, thereby narrowing the observed distribution of IQ scores.

Section 6 — Connecting the chromosomal mechanism to observed histograms

The chromosomal account is only as strong as its capacity to predict a visible, measurable signature in real score distributions. The mapping is therefore made explicit: mechanism, then predictions, then the plotted forms that operationalise those predictions. The mechanism is the asymmetry in X-linked expression: XY hemizygosity exposes X-linked allelic effects directly, while XX heterozygosity and X-inactivation mosaicism dampen those effects by partial offset and within-individual averaging. If X-linked variation contributes meaningfully to neurodevelopmental pathways that feed into cognitive performance, then the expression of that variation should alter dispersion in a sex-patterned way. The claim is not that the X chromosome is the only genetic contributor to IQ. The claim is that the X chromosome is a sex-specific modifier of variance and that its expression logic points in one direction: wider spread in XY than XX.

From that mechanism follows the first prediction: the variance ratio should exceed 1. In practice, dispersion is measured as the ratio of male to female variance, or equivalently as a comparison of standard deviations. If males are more sensitive to X-linked allelic differences, male distributions should be broader, producing a male SD that is reliably higher than the female SD across cohorts and tests that genuinely tap general cognitive ability. That prediction is not an aesthetic judgement; it is a numerical statement. When the distributions are plotted as histograms with comparable binning, the male curve should appear flatter and wider and the female curve tighter and more peaked. This is exactly the pattern captured by a variance ratio above unity, and it is the core descriptive bridge between the chromosomal mechanism and what is seen on a page.

The second prediction follows mechanically from the first: male overrepresentation in both tails. With means held constant or near-constant, a wider distribution necessarily increases the probability mass in the extremes. The chromosomal mechanism is therefore not a claim about a male advantage or a female advantage on average; it is a claim about tail frequencies. It predicts more males below low thresholds and more males above high thresholds. In the plotted signature, this appears as heavier male tails: more area under the male curve at the low end and more area under the male curve at the high end. In distributional reporting, it appears as lower male low-quantiles (for example, a lower 5th percentile) and higher male high-quantiles (for example, a higher 95th and 99th percentile), alongside a variance ratio above 1. This is the statistical fingerprint the argument is committed to: the same centre, different spread, with tail divergence as the concrete consequence.

The third prediction is domain sensitivity. If the variance effect is meaningfully X-linked, it should not be uniformly expressed across every cognitive subtest in the same way, because different subtests load differently on general cognitive ability and draw on partially distinct neurodevelopmental and neurocognitive pathways. The expectation is not that one can simply “point” to an X-domain from first principles, but that dispersion differences should be stronger in subdomains plausibly enriched for X-linked neurodevelopmental contributions—domains where developmental variation in attention, learning efficiency, processing constraints, or related cognitive control factors is more consequential for performance and where X-linked neurodevelopmental variation would be expected to manifest as a broader range of outcomes. In the plotted signature, this appears as variance ratios that are not identical across subtests: some domains show a clearer male widening and heavier tails than others, consistent with a mechanism that operates through specific developmental pathways rather than through an undifferentiated general factor.

Figure 9. Predicted pattern: wider male histogram and heavier tails (mechanism-consistent signature)-

This figure combines the two elements the chromosomal mechanism predicts should co-occur when means are equal: a broader male distribution and heavier male tails. The histograms (same bin width) and the overlaid density curves show the spread difference directly, with males wider (SD = 19) and females tighter (SD = 14), while both are centred on the same mean (100).

Two tail thresholds are marked to make the implications of dispersion visible on the same plot: below 70 (lower tail) and above 130 (upper tail). With the variance ratio at 1.84, the plotted signature is that males occupy more area than females in both tails, producing higher tail proportions at both extremes. The annotated tail ratios quantify this: the male-to-female ratio is approximately 3.47 in the low tail and 3.55 in the high tail for these parameters, illustrating how a spread difference alone yields excess representation at both ends of the scale.

The intended interpretation is narrow and direct: the mechanism predicts VR greater than 1, and VR greater than 1 predicts heavier male tails, even with no mean shift.

This prediction set aligns with classic descriptive findings in the variability literature. Hedges and Nowell (1995) reported sex differences in variability across mental test scores and emphasised that differences in spread and the representation of high-scoring individuals can arise even when average differences are small. That work is useful here not because it settles mechanism, but because it anchors the observed phenomenon the present argument aims to explain: modestly greater male variability, distributionally expressed in tail representation. Johnson et al. (2009) then provided the mechanistic framing that makes the chromosomal hypothesis explicit, asking whether the X chromosome could contribute to sex differences in variability in general intelligence and outlining how hemizygosity could, in principle, amplify variance.

Table 2: Predicted vs observed VR ranges by subtest/domain (verbal, spatial, processing speed, etc.)

The result is a closed loop: chromosomal asymmetry implies variance asymmetry; variance asymmetry implies tail asymmetry; and both are visible in histograms, quantiles, and variance ratios. The claim remains assertive because it is committed to a testable signature rather than rhetorical flourish: a variance ratio above 1 in g-loaded measures, heavier male tails at both extremes, and a pattern of stronger dispersion in those cognitive domains where X-linked neurodevelopmental variation would be expected to matter most (Hedges & Nowell, 1995; Johnson et al., 2009).

Section 7 — Alternative genetic contributors that do not erase the X claim

A serious chromosomal argument does not pretend that the X chromosome is the whole genome. Intelligence is a complex trait with a broad genetic basis, and the dominant molecular-genetic picture is polygenic: common-variant associations are spread across many loci throughout the autosomes, each contributing small effects. On any standard account of genome-wide association results, there is no single locus, no single chromosome, and no narrow gene set that “explains IQ.” The architecture is distributed. That point matters because it prevents a false dilemma from taking hold, namely the idea that either the X chromosome explains cognitive variation or the autosomes do. The correct framing is layered: autosomal polygenicity provides the background architecture of cognitive differences, while sex-chromosome structure can modify how a subset of genetic variation is expressed, thereby shaping dispersion.

Figure 10. Two-layer model: autosomal polygenic baseline and X-linked variance modifier-

This schematic presents a layered genetic model. The lower layer represents the autosomal polygenic baseline: many small-effect variants distributed across the autosomes that are shared by both sexes and that establish the overall heritability and mean structure of cognitive ability. The upper layer represents the X-linked variance modifier: sex-dependent expression of X-linked variation, with XY hemizygosity producing greater exposure and XX heterozygosity and mosaicism producing buffering.

The arrows indicate how the same autosomal cognitive architecture feeds into both sexes, while the X-linked modifier alters the width of the resulting distributions. The predicted outcomes are a tighter XX distribution with lower standard deviation and a wider XY distribution with higher standard deviation. This figure formalises the core claim that autosomes provide the background architecture of intelligence, while the X chromosome shapes dispersion rather than replacing or contradicting polygenic accounts.

The central claim of this essay is therefore not threatened by autosomal polygenicity; it is contextualised by it. If the autosomes carry a large share of the additive genetic variance for IQ, then both sexes will share the same broad polygenic substrate, and both will show substantial heritability. That is the expected baseline. The X-chromosome claim is narrower and distributional: it targets a mechanism by which one class of genetic variation is expressed differently by sex, with consequences for variance rather than necessarily for mean. In that role, the X chromosome is not competing with the autosomes. It is operating as a modifier of dispersion on top of an autosomal polygenic background.

The reason this modifier logic matters is that hemizygosity is not merely “another genetic effect.” It is a change in expression regime. For X-linked loci, the same allelic difference can have a larger phenotypic impact in XY individuals because there is no second X allele at that locus to counterbalance it, and because XX individuals introduce buffering through heterozygosity and mosaic expression. That asymmetry exists regardless of how polygenic the autosomes are. In other words, autosomal polygenicity explains why cognitive ability is broadly heritable and why most variance components are not located on the X chromosome alone; it does not explain away the fact that the X chromosome is uniquely positioned to alter the translation of X-linked allelic variation into phenotypic variability.

This layered framing also clarifies what the essay is and is not claiming. It is not claiming that the male–female variance ratio in IQ is purely an X phenomenon, or that autosomal effects are irrelevant to dispersion. Autosomal loci can also contribute to variance differences through sex-specific gene regulation, hormonal environments, and developmental differences. The point is that the X chromosome supplies a particularly clean and structurally grounded dispersion pathway, because hemizygosity changes the effective distribution of expressed allelic effects in one sex. This makes the X claim robust to the fact of polygenicity: a trait can be overwhelmingly polygenic across the autosomes and still show a sex difference in variance that is partly driven by X-linked expression asymmetry.

In this sense, the autosomes are the base layer of the genetic model and the X chromosome is the dispersion layer. Plomin and Deary (2015) and Plomin and von Stumm (2018) provide the broad polygenic background for intelligence, while the argument developed here specifies how the X chromosome can shape the width of the distribution without requiring a shift in its centre.

Section 8 — Empirical tests and analysis plan (methods section in prose)

Demonstrating the dispersion claim requires a design that treats distributional shape as the primary outcome and that reports every analytic choice that can alter dispersion estimates. The empirical goal is to show, in a transparent sequence, that (a) male IQ dispersion exceeds female IQ dispersion, (b) this produces predictable tail differences, and (c) the pattern is stable across sensible specifications rather than being an artefact of binning, cohort composition, or instrument choice. The plan below sets out the minimum data requirements, preprocessing rules, descriptive outputs, inferential tests, and robustness checks needed to make the argument demonstrable rather than rhetorical.

Data and preprocessing begin with defining the sample and measurement context. The sample is specified by cohort source, recruitment method, and inclusion window. If multiple cohorts are used, each cohort is analysed separately first, then combined only through a planned synthesis step to avoid hiding cohort-specific failures behind pooled results. The age range is defined in advance, with separate analyses for distinct developmental bands (for example, late childhood, adolescence, and adulthood) rather than forcing a single pooled distribution across heterogeneous ages. IQ instruments are named explicitly and grouped by family (for example, Wechsler full-scale composites versus non-Wechsler g composites), because different instruments can differ in ceiling effects, scaling, and subtest composition, all of which can distort apparent variance. Exclusion criteria are pre-registered and minimal: removal is limited to clear data-quality failures (invalid administrations, missing sex, missing age, impossible scores, duplicate records). Clinical exclusions are treated cautiously; excluding diagnosed neurodevelopmental conditions may artificially truncate the very tails under study, so any such exclusion must be justified and then reported as a sensitivity analysis rather than as the primary analysis.

Standardisation is conducted within narrow age bands. Because IQ is age-normed in most instruments, the primary analysis uses the instrument’s standard scores as provided, but any cross-instrument pooling requires re-standardising within cohort and within age band to a common metric. The rule is simple: within each age band, scores are transformed to a common mean and standard deviation for the full sample of that band, and then male and female distributions are compared within that band. This prevents spurious dispersion differences caused by mixing ages with different score distributions or different norming properties. Any residualisation step is restricted to age and test version effects only. The analysis does not residualise for sex, because sex is the grouping variable whose dispersion difference is being estimated. If multiple test versions exist, version is handled either by stratification (analysing each version separately) or by residualisation (removing version effects), but the chosen method is applied consistently and reported.

Figure 11. Analysis pipeline from raw IQ data to variance inference and robustness evaluation-

This flow diagram summarises the analytic sequence used to test the dispersion claim. Raw IQ data (scores, sex, age, and test version) are first cleaned and preprocessed, including age-banding and minimal exclusion for data quality. Scores are then standardised within age bands with explicit handling of test-version differences. From the standardised data, descriptive plots are produced (histograms, density overlays, and empirical CDFs), followed by formal variance inference using standard deviations, variances, variance ratios with confidence intervals, and variance-equality tests. The pipeline concludes with mandatory robustness checks, including sensitivity to binning choices, cohort composition, age bands, and the comparison of raw versus age- and version-residualised scores.

Descriptive outputs are produced in a fixed order and with reproducible settings. For each cohort and age band, the first outputs are histograms for males and females using identical bin widths and bin edges, with bin width chosen by a stated rule and accompanied by a binning sensitivity check. Kernel density overlays are added to each histogram to provide a bin-independent view of the same distributions. The second outputs are the dispersion statistics: male mean and female mean (reported but not foregrounded), male SD and female SD, male variance and female variance, and the variance ratio defined as male variance divided by female variance. The variance ratio is accompanied by a confidence interval rather than a single point estimate. The third outputs are tail and quantile summaries. Tail proportions are defined using fixed thresholds chosen before analysis and kept constant across all cohorts and bands (for example, below 70 and above 130), with the male-to-female tail ratio reported for each threshold. Quantiles are reported as sex-specific percentile points (for example, 5th, 50th, 95th, 99th) to show tail divergence without relying on arbitrary cut-offs. In addition, empirical CDF plots are produced to show where divergence emerges across the distribution, not merely at thresholds.

Inferential tests are then applied to the dispersion difference. Equality of variances is tested using procedures robust to non-normality, specifically Levene’s test and the Brown–Forsythe variant (centred on the median). These tests provide a formal check that the observed variance ratio is not plausibly attributable to sampling noise. Because the variance ratio is the central statistic of interest, it is estimated with a confidence interval using bootstrap resampling. The bootstrap is stratified by sex within each cohort and age band to preserve the sex composition and avoids assuming parametric normality. The result is a variance ratio estimate with uncertainty bounds, allowing the analysis to report not only whether VR exceeds 1, but by how much and with what precision.

Model-based approaches are used to connect the dispersion claim to an explicitly X-aware genetic account where genotype data are available. The minimal genetic model partitions variance by chromosomal class: autosomes versus X chromosome. This requires X-aware handling of genotypes because dosage and inheritance differ by sex. A direct test of the essay’s mechanism uses sex-by-genotype interactions on X-linked variants or X-linked polygenic indices, with the outcome specified as either (a) the IQ score in a heteroscedastic model where residual variance is allowed to differ by sex, or (b) a dispersion-focused outcome such as absolute deviation from the sex-specific median, which operationalises “extremeness” as a continuous measure. The key is not to overfit. The genetic model is used to test whether X-linked variation explains a detectable portion of sex differences in variance beyond what is explained by autosomal polygenicity. If genotype data are absent, the analysis remains distributional and psychometric; the essay can still demonstrate VR and tail differences, but it cannot claim direct variance partitioning to X-linked loci without genetic data.

Robustness checks are treated as mandatory rather than optional. First, results are checked across instruments: full-scale IQ composites, g-loaded composites, and domain subtests are analysed separately, with the expectation that dispersion effects will not be uniform across all domains. Second, results are checked across cohorts: each cohort is analysed independently, then a synthesis is performed that reports heterogeneity rather than hiding it. Third, results are checked across age bands to ensure that the dispersion gap is not an artefact of one developmental stage. Fourth, sensitivity to outliers is evaluated without trimming away tails by default. Instead of deleting extremes, robust measures are reported alongside conventional variance, and any winsorisation or trimming is presented only as an explicit sensitivity analysis with the primary results retained. Fifth, results are compared under raw scoring versus residualised scoring (age and test version only), to show that the dispersion signature is not produced by technical artefacts of scaling or cohort composition.

Table 3. Reporting template for cohort-level dispersion analysis-

This table is a reporting template, not filled results. It specifies the minimum statistics that must be reported for each cohort analysed independently before any pooling or synthesis. The intent is to make dispersion claims auditable and to prevent variance effects from being hidden by aggregation.

Column definitions (to be stated once in the methods section):-

Cohort ID: Dataset name or identifier; cohorts are analysed separately before synthesis.

Age band: Narrow, predefined age range; no cross-band pooling.

Instrument: Exact test and composite used (e.g. full-scale IQ, g-loaded composite).

N (male / female): Sample sizes by sex after exclusions.

Mean (male / female): Reported for completeness; not the primary target.

SD (male / female): Primary dispersion statistics in score units.

Variance ratio (VR): Male variance divided by female variance.

95% CI for VR: Bootstrap confidence interval, stratified by sex.

Low-tail ratio (<70): Male proportion divided by female proportion below fixed low threshold.

High-tail ratio (>130): Male proportion divided by female proportion above fixed high threshold.

Extreme-tail ratio: Male-to-female ratio at a severe upper tail (predefined, e.g. >145 or 99th percentile).

Levene / Brown–Forsythe p: Robust tests of equality of variances.

Notes: Any cohort-specific issues (ceiling effects, version mixing, restricted range).

Interpretive rule:-

A cohort is mechanism-consistent if VR > 1 with a confidence interval excluding 1 and if both low- and high-tail ratios exceed 1 in the predicted direction. Mean differences are not required for consistency and are not used as decision criteria.

This template enforces the essay’s core discipline: dispersion is demonstrated through transparent, cohort-level variance and tail reporting, not through pooled means or selective summaries.

The output of this methods plan is not a single test statistic but a coherent body of evidence: histograms and density overlays showing broader male spread; variance ratios above 1 with confidence intervals; tail proportions and quantiles showing divergence at both extremes; variance-equality tests supporting that divergence; and robustness checks demonstrating that the findings do not depend on one plotting choice, one cohort, or one instrument. Where genotype data permit, X-aware modelling is added as the bridge from distributional facts to chromosomal mechanism, testing whether X-linked variation makes a measurable contribution to the observed dispersion difference rather than merely serving as an interpretive story.

Section 9 — Strongest predictions using sex-chromosome aneuploidies

The sharpest tests of an X-linked dispersion mechanism do not come from ordinary XX versus XY comparisons alone, because ordinary comparisons bundle many differences at once. Sex-chromosome aneuploidies provide a cleaner logic test because they alter X dosage and X-linked expression regime while holding much of the rest of the genome constant in the usual way. The argument in this essay predicts that changing the number of X chromosomes and the presence or absence of mosaic buffering will alter dispersion in cognitive outcomes in systematic directions. This is not framed as a claim about typical outcomes for any given individual with an aneuploidy, and it is not a clinical generalisation to the population. It is a test strategy: if the buffering model is correct, variance patterns should move in predictable ways when X dosage and X-expression regimes are perturbed.

Start with X0 (Turner syndrome), where there is a single X chromosome and no second X available for heterozygosity or mosaic compensation in the ordinary XX sense. The buffering model treats X0 as a “de-buffered X” condition. Under the logic developed earlier, removing the second X removes two stabilisers: there is no second allele to offset X-linked variation and no two-X mosaic interplay that can average expression across cell populations. The direct prediction is therefore increased sensitivity to X-linked effects relative to XX, and, in dispersion terms, a broader distribution than XX would show, all else equal. The critical point is not the mean level of any cognitive measure, which can be influenced by many developmental factors, but the spread: a de-buffered X should, in principle, increase variance attributable to X-linked allelic differences, because expression is more uniformly dependent on the single X genotype.

Next consider XXY (Klinefelter syndrome), where there are two X chromosomes in an individual who is typically male in phenotypic sex. For the dispersion thesis, XXY is important because it separates “being male” from “having one X.” Under the buffering model, adding an extra X should introduce some degree of XX-like buffering. The prediction is not that XXY will replicate the full female distribution, but that the extra X should reduce X-linked exposure relative to typical XY. In dispersion terms, that implies a narrower distribution than XY on X-influenced components, with tail rates shifting toward the XX direction. Conceptually, XXY should dampen the variance-amplifying effect attributed to XY hemizygosity, because hemizygosity is no longer present in the same way.

Now consider XYY. This condition adds an extra Y without adding an extra X. Under the X-buffering thesis, adding a Y does not supply the buffering structure that matters, because the dispersion mechanism is grounded in X-linked hemizygosity and the absence or presence of a second X. Therefore, as a pure test of the buffering logic, XYY is predicted to resemble XY more closely than XXY does with respect to X-linked exposure. In other words, XYY should not show the variance-dampening signature that the model predicts for additional X dosage. If the dispersion thesis were mistakenly attributed to “maleness” in a loose sense, one might expect XYY to shift strongly; the buffering model predicts that the decisive shift should occur with additional X dosage, not with additional Y dosage.

Figure 12. Predicted dispersion ordering across karyotypes (conceptual)-

This figure presents the predicted ordering of dispersion (distribution width) across sex-chromosome karyotypes under the X-buffering thesis. It is a conceptual ranking, shown on a relative SD scale, where lower values indicate tighter dispersion and higher values indicate wider dispersion.

Predicted ordering (tighter to wider): XXX → XX → XXY → XY ≈ XYY → X0

Finally consider XXX (trisomy X), where there are three X chromosomes. If additional X dosage increases buffering, then XXX should be, in a conceptual sense, “more buffered than XX” with respect to single-locus X-linked exposure. The prediction is therefore a shift toward tighter dispersion relative to XX, at least for the part of cognitive variance mediated by X-linked loci, because there is an increased opportunity for allelic offset and a more complex mosaic pattern of X inactivation across tissues. The prediction is directional and variance-focused: the presence of additional X material should dampen the translation of X-linked allelic differences into organism-level phenotype, reducing tails compared to XX, again without requiring a systematic mean shift.

These aneuploidy comparisons generate a hierarchy of variance predictions that can be tested with the same distributional toolkit used earlier: SD, variance ratio, quantiles, and tail proportions. The mechanistic expectation, stated purely as test logic, is that groups with a single X (X0 and XY) should show greater X-linked exposure and therefore broader dispersion than groups with two or more X chromosomes (XX and XXY and XXX), with XXY shifting toward the buffered side despite male phenotype because the relevant variable is X dosage and expression regime. XYY is predicted to cluster with XY on the specific question of X-linked buffering because the Y count does not create the buffering structure posited by the thesis. The force of this section lies in the fact that it risks falsification: if added X dosage does not shift dispersion in the predicted direction, the buffering mechanism loses credibility. If it does, the chromosomal explanation gains a distinctive kind of support that ordinary XX versus XY comparisons cannot provide.

Section 10 — Conclusion: the dispersion thesis stated cleanly

The core claim is simple and measurable: sex chromosomes produce differences in the dispersion of IQ. The issue is not an argument about average ability. It is an argument about distributional shape—standard deviations, variance ratios, and the number of individuals found at the extremes. Under the thesis advanced here, the male IQ distribution is wider because XY biology exposes X-linked allelic effects more directly, while the female IQ distribution is tighter because XX biology buffers X-linked effects through heterozygosity and X-inactivation mosaicism.

The biological pathway is a variance pathway. In XY males, the X chromosome is present in a single copy, so X-linked alleles are expressed without a second X allele available to offset or dilute their impact. This increases phenotypic sensitivity to X-linked variation and, aggregated across individuals, broadens the distribution. In XX females, two X chromosomes create two dampeners. Heterozygosity permits partial organism-level offset of single-locus X-linked effects, and X-inactivation produces cellular mosaicism that dilutes coherent, uniform expression of any one X-linked allele across tissues (Migeon, 2007). Together these features compress variance and reduce tail mass relative to XY.

The empirical signature is therefore explicit. First, the variance ratio for IQ and g-loaded composites should be greater than 1, expressed as a higher male SD and variance than the female SD and variance. Second, even where male and female means are similar, males should be overrepresented at both low and high extremes, visible as heavier tails in histograms, quantile divergence at the 5th, 95th, and 99th percentiles, and higher male tail proportions at fixed thresholds. Third, dispersion effects should vary by domain in a way consistent with a mechanism that operates through X-linked neurodevelopmental pathways rather than through a uniform shift across every subtest.

The test strategy matches the claim. The first tier is descriptive and inferential distributional work: cohort-by-cohort variance ratios with confidence intervals, tail ratios, and robust tests of variance differences, reported transparently. The second tier is genetic: X-aware modelling that partitions variance into autosomal and X-linked components and tests whether X-linked variation contributes to the observed sex difference in dispersion. The third tier is the sharpest logic test: karyotype comparisons using sex-chromosome aneuploidies, where adding or removing X dosage should shift dispersion in the predicted direction if buffering is real (Johnson et al., 2009; Migeon, 2007).

Confirmation is defined by convergence, not by assertion. The thesis is supported if variance ratios above 1 recur across cohorts and instruments, tail patterns match the dispersion prediction, and X-aware genetic analyses show that X-linked variation explains a detectable portion of variance in a sex-dependent manner, with karyotype perturbations shifting dispersion as the buffering model predicts (Johnson et al., 2009; Migeon, 2007).

References

Briley, D. A., & Tucker-Drob, E. M. (2013). Explaining the increasing heritability of cognitive ability across development: A meta-analysis of longitudinal twin and adoption studies. Psychological Science, 24(9), 1704–1713. https://doi.org/10.1177/0956797613478618

Feingold, A. (1992). Sex differences in variability in intellectual abilities: A new look at an old controversy. Review of Educational Research, 62(1), 61–84.

Haworth, C. M. A., Wright, M. J., Luciano, M., Martin, N. G., de Geus, E. J. C., van Beijsterveldt, C. E. M., Bartels, M., Posthuma, D., Boomsma, D. I., Davis, O. S. P., Kovas, Y., Corley, R. P., DeFries, J. C., Hewitt, J. K., Olson, R. K., Rhea, S.-A., Wadsworth, S. J., Iacono, W. G., McGue, M., … Plomin, R. (2010). The heritability of general cognitive ability increases linearly from childhood to young adulthood. Molecular Psychiatry, 15(11), 1112–1120.

Hyde, J. S. (2005). The gender similarities hypothesis. American Psychologist, 60(6), 581–592.

Lubs, H. A., Stevenson, R. E., & Schwartz, C. E. (2012). Fragile X and X-linked intellectual disability: Four decades of discovery. The American Journal of Human Genetics, 90(4), 579–590. https://doi.org/10.1016/j.ajhg.2012.02.018

Migeon, B. R. (2007). Why females are mosaics, X-chromosome inactivation, and sex differences in disease. Gender Medicine, 4(2), 97–105.

Plomin, R., & von Stumm, S. (2018). The new genetics of intelligence. Nature Reviews Genetics, 19(3), 148–159. https://doi.org/10.1038/nrg.2017.104

Gécz, J., & Mulley, J. (2000). Genes for cognitive function: Developments on the X. Genome Research, 10(2), 157–168.

Giofrè, D., Toffalini, E., Perugini, A., Esposito, L., Amoretti, G., & Geary, D. C. (2024). Sex differences in cognition: A meta-analysis of variance ratios in the Wechsler Intelligence Scales for Children. Personality and Individual Differences, 229, 112776. https://doi.org/10.1016/j.paid.2024.112776

Hedges, L. V., & Nowell, A. (1995). Sex differences in mental test scores, variability, and numbers of high-scoring individuals. Science, 269(5220), 41–45. https://doi.org/10.1126/science.7604277

Johnson, W., Carothers, A., & Deary, I. J. (2009). A role for the X chromosome in sex differences in variability in general intelligence? Perspectives on Psychological Science, 4(6), 598–611. https://doi.org/10.1111/j.1745-6924.2009.01168.x

Migeon, B. R. (2007). Females are mosaics: X inactivation and sex differences in disease. Oxford University Press.

← Back to Substack Archive