Genes, Meat, or Soy: What Best Predicts Human Height?

Average height at age 18 is partly inherited and is also an imperfect cumulative marker of growth conditions from prenatal life through adolescence. Maternal and early-childhood undernutrition predict later growth (Victora et al., 2008), while repeated diarrhoea contributes to stunting (Checkley et al., 2008). NCD-RisC’s 200-country trajectories show how much childhood and adolescent height varies across place and time.

Linear growth requires adequate energy and a digestible supply of indispensable amino acids. The FAO’s Digestible Indispensable Amino Acid Score, or DIAAS, evaluates protein quality amino acid by amino acid. Mathai and colleagues’ growing-pig assay found substantial digestibility differences among dairy and plant proteins; Herreman and colleagues’ review found that animal-sourced proteins often score higher on average, although variation within both categories is substantial. DIAAS rates individual foods, not whole-diet health or effects on human height.

Height is also highly polygenic. Yengo and colleagues mapped thousands of common-variant associations, allowing part of the measured genetic signal to be summarized in a polygenic score. A PGS is not a complete measure of genetic potential, and standardizing it as a Z score does not solve cross-ancestry portability or population structure. Omitting genetics can distort an ecological comparison; including a PGS does not eliminate genetic confounding.

Here, “meat” and “soy” are title shorthand. The actual exposures are national animal-protein supply—including meat, dairy, eggs, fish, and seafood—and plant-protein supply from the full food system, including grains and legumes.

This replication asks four questions:

• Across the 18 years preceding height measurement at age 18, are animal- and plant-protein supply associated differently with national height?

• Do either protein source’s returns diminish as supply rises—and, if so, where does the curve begin to flatten?

• How much predictive information does a standardized height polygenic score add on its own and alongside diet?

• Do the answers survive stricter population-size thresholds and adjustment for broader national development?

The answers do not fit neatly into a genes-versus-nutrition contest.

Measuring genes and growth conditions together

Height is the cumulative result of inherited differences and the conditions in which children grow. Nutrition matters, but so do infection, sanitation, prenatal health, medical care, inequality, and many other features of development.

The original study (Piffer & Kirkegaard, 2024) compared height with food supply measured at a single point in time. I instead paired national height at age 18 in 2019 with average food supply during 2001–2018—the full 18 calendar years preceding the measurement. A country entered the primary analysis only if all 18 annual observations were available. Animal protein, plant protein, and calories were measured separately.

The model also included a standardized height polygenic score. The primary genetic data came from jointly scored whole-genome-sequenced and imputed samples. Populations represented by fewer than 25 people were excluded from the primary country averages; a stricter 40-person threshold served as a sensitivity test.

The resulting primary dataset contained 36 countries with complete height, nutrition, and genetic estimates. These country PGS values are equal-weighted averages of sampled genetic populations, not nationally representative genomic surveys. That limitation matters, but the score still produced a consistent conditional association across the principal analyses.

The strongest coefficient was dietary. The strangest result was genetic

To compare unlike predictors, height, the PGS, and each dietary variable were converted to standard-deviation units. The coefficients therefore describe a typical cross-country difference in each predictor, holding the others constant.

In the primary diet-focused model, the standardized associations with national height were:

• Animal protein: 0.72

• Height polygenic score: 0.37

• Plant protein: 0.24

• Calories: 0.24

Animal protein had the largest coefficient. Its association was also clearly separated from zero, as was the PGS association. The positive plant-protein and calorie estimates were much less certain.

The model-fit comparison made the genetic result more surprising. A model containing only the PGS had an R-squared of 0.0013: by itself, the score captured just 0.13 percent of the variation among these countries. Diet alone produced an R-squared of 0.686. Yet the full diet-plus-PGS model reached 0.803. Adding the genetic score after diet therefore increased in-sample R-squared by 0.117.

Figure 1. The genetic and dietary estimates depended on model specification.

These values are predictive statistics, not causal percentages. It would be wrong to say that genes “explain 11.7 percent” of height differences, just as it would be wrong to call the diet-only R-squared a nutritional share of height. The genetic and dietary predictors overlap in a suppression pattern, so their incremental contributions do not add up like slices of a pie.

Why the genetic score became informative after environmental adjustment

Suppression sounds exotic, but the logic is simple. A predictor may look uninformative by itself because it is correlated with another predictor in a direction that masks its relationship with the outcome. Holding that second predictor constant can reveal information that was hidden in the simple comparison.

In the primary model, the PGS coefficient was 0.368 with a robust p-value of 0.00015. The result persisted when the minimum population sample was raised from 25 to 40: the coefficient was 0.296, with p=0.00049. In the separate aggregate-frequency analysis, it was 0.413, with p=0.00057.

The score remained associated when the 2019 Human Development Index (HDI) joined the dietary variables. Its standardized coefficient was 0.479 in the primary analysis, 0.384 under the stricter population threshold, and 0.485 in the aggregate-WGS analysis. All three estimates were statistically clear.

The primary HDI-adjusted result created an important asymmetry. The PGS coefficient rose from 0.37 to 0.48. Animal protein fell from 0.72 to 0.32 and became imprecise. Plant protein was 0.23. The measured genetic signal was more stable across these specifications than the animal-protein signal.

Animal protein was not interchangeable with plant protein

Across the observed national food-supply distributions, the baseline standardized coefficient was 0.72 for animal protein and 0.24 for plant protein. A direct HC3 test distinguished those standardized coefficients (p=0.00077). The animal-protein point estimate remained numerically larger under the stricter threshold—0.63 versus 0.20—and in the 27-country aggregate-WGS regression—0.83 versus 0.31. Those checks did not establish a larger per-gram or continent-adjusted association.

That contrast is partly a consequence of standardization. Animal-protein supply had a standard deviation of 21.5 grams per person per day in the primary sample, compared with only 8.5 grams for plant protein. A one-standard-deviation animal-protein contrast therefore represented roughly two and a half times as many grams.

On the original scale, the linear estimates for another 10 grams were much closer: 1.52 centimetres for animal protein and 1.31 centimetres for plant protein. Their difference was not statistically distinguishable (p=0.64). The corresponding difference tests were also null under the stricter threshold and in the aggregate-WGS analysis.

I then added continent fixed effects. North and South America were combined because North America contributed only one complete country in two tracks; retaining a singleton category made HC3 covariance undefined. With Africa as the reference, the overall-SD coefficients became 0.66 for animal protein and 0.35 for plant protein in the primary sample. Animal protein remained numerically larger in all three tracks, but the animal-minus-plant contrasts were no longer statistically clear: p=0.153 in the primary sample, p=0.275 under the stricter threshold, and p=0.162 in the aggregate-WGS sample.

Because continent indicators remove between-continent variation, I also re-standardized the remaining within-continent variation. In the primary sample those coefficients were 0.65 for animal protein and 0.51 for plant protein, with p=0.52 for their difference. About 62 percent of the original animal-protein dispersion remained within continents, compared with 91 percent for plant protein.

The continent indicators were not jointly significant and they worsened leave-one-country-out prediction. But they substantially increased collinearity—animal protein’s variance-inflation factor rose to 11.9—and widened the uncertainty around both protein estimates. The defensible conclusion is therefore narrower: animal protein had the larger standardized association in the baseline models, but the data do not establish a larger per-gram effect or a continent-adjusted advantage over plant protein.

Why protein quality is a plausible part of the explanation

One plausible biological contribution is protein quality. Crude protein is usually estimated from total nitrogen using a conversion factor; it does not describe amino-acid composition or digestibility. Growth depends on the indispensable amino acids that can actually be digested and absorbed.

The Food and Agriculture Organization recommends the Digestible Indispensable Amino Acid Score, or DIAAS, for evaluating this. DIAAS considers each indispensable amino acid separately, corrects it using digestibility at the end of the small intestine, and assigns the food the score of its most limiting digestible amino acid.

Animal proteins generally score higher because they tend to combine high ileal digestibility with an amino-acid pattern closer to human requirements. The FAO report’s young-child example gives whole-milk powder a DIAAS of 122, compared with 64 for peas and 40 for wheat. In a controlled study using growing pigs, many dairy proteins showed higher ileal amino-acid digestibility than pea, wheat, and several soy products; extrapolation to humans and whole diets remains a limitation.

This is a tendency, not a law separating two kingdoms of food. Soy and potato protein can score well; processing changes digestibility; and complementary plant proteins can compensate for one another’s limiting amino acids.

The present analysis did not calculate a DIAAS-weighted national food supply. Protein quality therefore offers a plausible mechanism, not a demonstrated explanation of the country-level coefficient.

Then the animal-protein curve bent

A straight-line model assumes that moving from 10 to 20 grams of animal protein is associated with the same height difference as moving from 60 to 70 grams. Several models allowed that relationship to flatten instead. Every version included the standardized height PGS, calories, and the other protein source, but not HDI.

The confirmatory evidence for diminishing returns was specific to animal protein. In the primary analysis, the quadratic curve bent downward with an HC3 robust p-value of 0.009. It did so again under the stricter population threshold, with p=0.005. Plant protein showed no comparable curvature.

Figure 2. PGS-adjusted, diet-focused nutrition curves in the primary analysis; HDI is not included.

The selected plateau models placed the bend at about 53 grams of animal protein per person per day in the primary analysis and about 43 grams under the stricter threshold. Across 500 bootstrap resamples, the median estimates were about 50 and 42 grams, but the uncertainty was broad.

The separate aggregate-genome analysis retained the stronger animal-versus-plant contrast but did not confirm statistically clear curvature. The responsible conclusion is therefore evidence of diminishing returns in the two nested pooled analyses, not a universal plateau at 50 grams.

Development confounds both stories

Animal-protein availability is not randomly distributed. Countries with abundant meat, dairy, eggs, and fish also tend to differ in income, sanitation, vaccination, prenatal care, food security, childhood disease, and access to medicine. The HDI attenuation suggests that part of the animal-protein signal travels with this broader developmental environment.

The genetic signal is entangled too. Ancestry, geography, development, and diet are not independently assigned to countries. The PGS suppression result may reflect height-related genetic information, population structure, gene–environment correlation, unequal score portability, or some combination of these.

HDI does not magically solve the problem. It is a broad composite, and highly correlated predictors become difficult to estimate separately. Adding it may remove confounding, but it may also absorb part of the pathway through which development improves nutrition and growth.

Country averages cannot determine which interpretation is correct. An ideal follow-up would combine individual genotypes with repeated childhood dietary measurements, parental height, infections, household conditions, and socioeconomic circumstances. Within-family genetic comparisons and within-country nutritional changes would help separate mechanisms that this ecological design cannot.

What survives the genes-versus-environment debate

Four conclusions are defensible.

First, the measured genetic score contained conditional information about national height. It added substantially to the diet model and remained associated after HDI was included, although its precision weakened after continent adjustment. Its incremental R-squared is neither a heritability estimate nor a causal genetic share.

Second, animal-protein supply carried a larger standardized signal than plant-protein supply in the baseline models. That was not a threefold difference per gram, and the standardized contrast was no longer distinguishable after continent adjustment.

Third, the animal-protein association did not continue linearly without limit. It flattened in the pooled analyses, although the data did not establish a universal nutritional threshold.

Fourth, neither the genetic nor nutritional result can be interpreted at the individual level. National food supply is not personal intake, and population PGS averages are not representative national genomes.

The point is not that animal protein beat genes, or that genes became important only after diet was measured. The model cannot sustain either claim.

The surprising result was not whether genes or diet won. It was why neither could be interpreted separately.

Methods and sources

Height is the arithmetic mean of the NCD Risk Factor Collaboration’s modeled national estimates for boys and girls at age 18 in 2019. Protein and calorie supply come from FAO food-balance data distributed by Our World in Data. These series describe food available at the end of the supply chain, before consumer waste—not measured intake. Nutrition exposures are complete-country averages for 2001–2018. The FAO series crosses a documented methodology revision beginning in 2010, so part of the 18-year average may reflect a measurement discontinuity.

All 9,627 WGS and imputed samples were processed in one joint run using a common 12,422-variant scoring catalogue; the MIX-height score used 8,858 nonzero-weight variants. Missing dosages were mean-imputed from the pooled sample. The primary inclusive population threshold was N>=25; N>=40 was a sensitivity analysis. The separate frequency analysis used trait-specific SNP intersections present and allele-compatible in all 51 original WGS panels.

The PGS was calculated from the study’s harmonized GWAS weight file with PLINK 2; Yengo et al. (2022) provides the recent large-scale height-GWAS context.

Subscribe if you are interested in genetics, intelligence, and the future and past of humanity.

References

Yengo, L., Vedantam, S., Marouli, E. et al. (2022). “A saturated map of common genetic variants associated with human height.” Nature, 610, 704–712. https://doi.org/10.1038/s41586-022-05275-y

NCD Risk Factor Collaboration (NCD-RisC). (2020). “Height and body-mass index trajectories of school-aged children and adolescents from 1985 to 2019 in 200 countries and territories.” The Lancet, 396, 1511–1524. https://doi.org/10.1016/S0140-6736(20)31859-6

Victora, C.G., Adair, L., Fall, C., Hallal, P.C., Martorell, R., Richter, L. & Sachdev, H.S.; Maternal and Child Undernutrition Study Group. (2008). “Maternal and child undernutrition: consequences for adult health and human capital.” The Lancet, 371(9609), 340–357. https://doi.org/10.1016/S0140-6736(07)61692-4

Checkley, W., Buckley, G., Gilman, R.H., Assis, A.M., Guerrant, R.L., Morris, S.S., Mølbak, K., Valentiner-Branth, P., Lanata, C.F. & Black, R.E.; Childhood Malnutrition and Infection Network. (2008). “Multi-country analysis of the effects of diarrhoea on childhood stunting.” International Journal of Epidemiology, 37(4), 816–830. https://doi.org/10.1093/ije/dyn099

Food and Agriculture Organization of the United Nations. Food balance-sheet protein and calorie supply series, distributed by Our World in Data. https://ourworldindata.org/food-supply

Food and Agriculture Organization of the United Nations. (2013). Dietary protein quality evaluation in human nutrition: Report of an FAO Expert Consultation. FAO Food and Nutrition Paper 92. https://www.fao.org/4/i3124e/i3124e.pdf

Mathai, J.K., Liu, Y. & Stein, H.H. (2017). “Values for digestible indispensable amino acid scores (DIAAS) for some dairy and plant proteins may better describe protein quality than values calculated using the concept for protein digestibility-corrected amino acid scores (PDCAAS).” British Journal of Nutrition, 117, 490–499. https://doi.org/10.1017/S0007114517000125

Herreman, L., Nommensen, P., Pennings, B. & Laus, M.C. (2020). “Comprehensive overview of the quality of plant- and animal-sourced proteins based on the digestible indispensable amino acid score.” Food Science & Nutrition, 8, 5379–5391. https://doi.org/10.1002/fsn3.1809

The 1000 Genomes Project Consortium. (2015). “A global reference for human genetic variation.” Nature, 526, 68–74. https://doi.org/10.1038/nature15393

Chang, C.C., Chow, C.C., Tellier, L.C.A.M., Vattikuti, S., Purcell, S.M. & Lee, J.J. (2015). “Second-generation PLINK: rising to the challenge of larger and richer datasets.” GigaScience, 4, 7. https://doi.org/10.1186/s13742-015-0047-8

Piffer D, Kirkegaard EOW. Polygenic Selection and Environmental Influence on Adult Body Height: Genetic and Living Standard Contributions Across Diverse Populations. Twin Res Hum Genet. 2024 Dec 6:1-18. doi: 10.1017/thg.2024.43. Epub ahead of print. PMID: 39639460.