Ten Myths About Human Genetics That Refuse to Die
What ancient DNA, polygenic traits, and recent evolution actually show
Introduction
Modern discussions of genetics and anthropology often repeat claims that sound scientific, but do not hold up well once they are compared with population-genetic data and ancient DNA. School-level Mendelism is still used to explain human pigmentation. Ancestry plots are treated as full descriptions of appearance and behavior. Ancient populations are described with modern labels, as if biological change stopped somewhere in the Paleolithic. These ideas persist not because the data are unclear, but because different concepts, ancestry, traits, selection, and time, are repeatedly conflated.
This post addresses ten such myths. They range from pigmentation and ancestry to intelligence, evolution, and human origins. Each survives because it offers a simple story. Many stop working once they are examined using population genetics, ancient DNA, and polygenic models.
Several of the points below build on analyses previously published on this Substack or in my academic papers. Where relevant, I refer back to earlier posts and use this essay as a synthesis, connecting results that are often discussed separately, such as time series of polygenic scores in ancient DNA, trait divergence beyond neutral expectations, and the frequent confusion between ancestry metrics and trait distributions.
Myth 1: Mendelian oversimplifications
A persistent mistake in popular genetics is the attempt to explain complex human traits using simple Mendelian categories. Blond hair and blue eyes are routinely described as “recessive,” as if that label were sufficient to explain their inheritance, evolution, and distribution. This framing obscures more than it clarifies, because it collapses very different genetic architectures into a single cartoon.
To see why this is misleading, it helps to distinguish between three broad categories of trait architecture: monogenic traits, oligogenic traits with a major locus, and polygenic traits.
Monogenic traits are controlled primarily by variation at a single gene, often with a clear dominance relationship. Classic Mendelian disorders fall into this category. These traits can often be predicted accurately from genotype alone, and simple inheritance patterns are genuinely informative.
Most visible human traits do not belong to this category.
Eye color provides a useful bridge case. Eye color is not monogenic, but in Europeans it is strongly influenced by a major locus in the regulatory region of the HERC2 gene that controls expression of OCA2. Variation in this region explains a large fraction of the blue versus brown eye color contrast in European populations. Because of this large effect, eye color can appear Mendelian in simplified teaching examples, and blue eyes often behave approximately like a recessive phenotype relative to brown in restricted datasets.
At a biological level, eye color reflects the amount and distribution of melanin in the iris. Blue eyes do not contain blue pigment. They result from low melanin content in the anterior iris layers, which increases light scattering in the iris stroma and produces a blue appearance through Rayleigh scattering. Brown eyes reflect higher melanin concentration, which absorbs light and reduces scattering. This means that eye color is fundamentally a quantitative trait related to melanin levels, even if one locus has a large regulatory effect.
The simplified Mendelian approximation breaks down quickly once this quantitative nature is taken seriously. The HERC2/OCA2 region does not contain a single universal “blue-eye allele,” but multiple variants and haplotypes with regulatory effects. In addition, several other loci contribute to eye color variation, particularly for intermediate phenotypes such as green or hazel eyes and for variation outside Europe. These intermediate colors usually reflect intermediate melanin levels and, especially in the case of hazel, uneven spatial distribution of melanin across the iris rather than a distinct pigment.
This architecture explains several observations that are incompatible with a strict single-gene recessive model. First, eye color does not fall cleanly into two categories. Intermediate colors are common and genetically meaningful. Second, while uncommon, it is possible for two blue-eyed parents to have a child with darker eyes. Under a true one-gene recessive model this would be impossible. In reality, additional loci can increase iris melanin enough to shift the phenotype, even when both parents have blue eyes.
For these reasons, eye color is best described as oligogenic with a major locus. One region has a disproportionate effect, but it does not act alone, and dominance relationships are approximate and context-dependent rather than absolute. This is precisely why eye color prediction works reasonably well but not perfectly.
This distinction becomes even clearer with hair color, which is unmistakably polygenic. Hair color is influenced by many loci involved in melanin synthesis, transport, and regulation, including genes affecting both eumelanin and pheomelanin pathways. Genome-wide association studies have identified dozens of loci contributing to hair color variation, with no single locus exerting the kind of dominance seen for eye color. Blond hair, in particular, is not the product of one mutation but the cumulative effect of multiple variants, each with small to moderate effects, shaped by selection and drift over time (Morgan et al., 2018).
The practical consequences of these different architectures are visible in forensic genetics. The HIrisPlex and HIrisPlex-S systems predict eye, hair, and skin color using panels of trait-associated SNPs. Eye color prediction performs best because of the strong contribution of the HERC2/OCA2 region. Hair color prediction is less precise because it depends on many loci with distributed effects. Skin color prediction requires still more information, reflecting its highly polygenic nature. These systems work not because traits are Mendelian, but because enough trait-relevant variants are included to approximate a polygenic score.
A concrete example appears in Mendel’s Dwarf, a science-themed novel in which a key plot twist relies on the claim that two blue-eyed parents cannot produce a brown-eyed child, because brown is treated as a dominant trait. That inference only follows from the one-gene cartoon, and it fails once eye colour is treated as the quantitative, multi-locus trait described above, especially once intermediate colours and modifiers are allowed.
Myth 2: Skin color, hair color, and eye color are tightly linked
Pigmentation traits are correlated in many populations. The mistake is treating correlation as equivalence. Some correlations are genuinely high. In one large UK Biobank analysis, the highest genetic correlation reported was between basal skin colour and ease of tanning, around 0.88, which indicates extensive shared genetic basis for those two phenotypes (Pairo-Castineira et al., 2022).
Other correlations are more trait-specific. Hair colour is partially correlated with both skin and eye colour, but GWAS show it has a broad multi-locus architecture and does not map perfectly onto skin pigmentation gradients (Morgan et al., 2018). Eye and hair colours can show very strong genetic correlations for particular combinations in European datasets, such as blue eyes with blond hair, yet the correlations are not perfect and weaken when intermediate categories like green or hazel are considered (Lin et al., 2016).
Blue eyes provide a concrete example of partial decoupling in ancient DNA. Western European hunter-gatherers (WHG) are a well-known case where alleles associated with blue eyes appear at high frequency, even though these populations did not yet carry the full set of alleles that later pushed European skin pigmentation to modern levels, and there is no reason to assume they matched modern northern Europeans in hair colour either. In other words, light eyes can arise and spread without automatically bringing along light skin and blond hair, because these traits overlap biologically but are not a single genetic package. For the ancient DNA evidence and timeline, see:
Myth 3: Ancestry equals phenotype
One of the most persistent conceptual errors is treating ancestry as a proxy for traits. PCA plots, Fst, and related metrics summarize shared genetic history across the genome. They do not describe how selection has acted on specific loci.
A clear illustration comes from Jewish populations. On West Eurasian PCA plots, Ashkenazi Jews often cluster closely with southern Italians, including Sicilians, and with Levantine populations. Despite this proximity, specific traits and loci can differ substantially. Some Jewish populations show higher frequencies of blond hair than neighboring Mediterranean groups. At the monogenic level, founder conditions such as Tay–Sachs disease occur at much higher frequencies despite modest genome-wide divergence.
These patterns are not paradoxical. Selection, drift, and bottlenecks act on particular loci and traits. Genome-wide ancestry can remain similar while trait-relevant allele frequencies diverge.
The same logic applies to polygenic traits. Height provides a clean example because it is measurable, highly polygenic, and geographically structured. Polygenic scores for height predict population-level differences above and beyond genome-wide genetic distance (Piffer & Kirkegaard, 2024).
Genetic proximity is not a sufficient statistic for trait similarity.
For example, Dutch people and British are extremely close genetically, yet the Dutch are substantially taller, and so is their height PGS, as I have shown here:
Conversely, populations that are genetically very distant, such as Sardinians and East Asians, can have very similar average height and height polygenic scores.
This point is developed quantitatively here, using Qst statistics and regression analyses that incorporate both genetic and environmental effects:
The sections that follow generalize this logic to selection, evolution speed, cognition, and human origins.
