Family planning
The parents’ heights predict a child’s height better than their DNA
In the largest genetic study of height so far, the average of the two parents’ heights accounted for about 44 percent of the differences in their adult children’s height, measured in 981 UK families of European ancestry. The best DNA score matched that only when it was read from a person’s own genome. A score worked out from the parents falls short of both, because each child inherits a random half of each parent’s variants and no test on the parents can say which half.
The mid-parental formula
The usual way to estimate a child’s target adult height from the parents is a formula Tanner, Goldstein and Whitehouse published in 1970, alongside height standards for children that allow for the parents’ heights. Doctors who assess children’s growth still use it widely, despite criticism and several proposed alternatives. It adds the mother’s and the father’s adult heights, adds 13 cm for a son or takes 13 cm away for a daughter, and halves the total. The 13 cm is the average difference in height between adult men and women.
For a father of 180 cm and a mother of 166 cm, the target is 179.5 cm for a son and 166.5 cm for a daughter. Each target carries a range of 8.5 cm either side, about two standard deviations, so roughly 95 in 100 children end up inside it. For that son the range runs from 171 to 188 cm.
A range 17 cm wide is the honest size of the uncertainty. The formula sees two numbers. It cannot see which of the parents’ variants a child inherits, or the nutrition, illnesses and hormones that shape growth in childhood. A review of the formula published in BMC Pediatrics in December 2025 lists what else limits it: the difference in height between the sexes, the rise in average height from one generation to the next, a large gap between the two parents’ heights, the tendency of tall people to pair with tall people, and the strength of the correlation between parents’ and children’s heights. It also notes that the formula tends to underestimate the adult height of children whose parents are very short.
What a DNA height score is
A polygenic score adds up many common genetic variants, each of which on its own shifts height by a tiny amount. Every variant is weighted by its effect in a large study, the weighted variants are summed into one number, and that number is placed against a reference group as a percentile. Polygenic scores in plain words explains the idea at more length.
Height has given genetics more associated variants than any other human trait. In 2022 the GIANT consortium and 23andMe pooled data from 5.4 million people and found 12,111 independent variants linked to height. Tested on people the study had not used to build it, a score from those variants accounted for about 40 percent of the differences in height among people of European ancestry, and about 10 to 20 percent among people of other ancestries.
The score in an Aimosti report is an earlier and smaller one: 3,286 variants from GIANT’s 2018 study of about 700,000 people of European ancestry. In that study’s independent check, in the US Health and Retirement Study, a score from these variants correlated with measured height at about 0.44, which works out to about a fifth of the differences in height between people. Its percentile is placed against a European-ancestry reference group, so for anyone of other ancestry it is less accurate.
What a score from the parents misses
A child’s own score can only be read from the child’s own genome. From the two parents, the most anyone can work out is the midpoint of their two scores, which is what a child would carry on average. Each child inherits a random half of each parent’s variants, so brothers and sisters land at different points around that midpoint, and the variation among them is about half the variation in the whole population.
Measured heights cannot tell which half a child inherits either. They start from more, though. A parent’s height already reflects every variant that parent carries, including the many no study has found yet, while even the best score captures under half of the differences in height between people.
The 2022 study put numbers on both. In 981 families in the UK Biobank, all of European ancestry and each a mother, a father and their adult child, the parents’ average height accounted for 43.8 percent of the differences in the children’s height. The best DNA score did about as well, 44.7 percent in unrelated people of European ancestry, a gap the authors found not significant, but that score was read from each person’s own genome and not from their parents’. Combining the parents’ heights with the child’s own score reached about 55 percent, because a child’s own score picks up differences between brothers and sisters that the parents’ average cannot.
For two people planning a family, then, the two measured heights are the better predictor. A score from the parents’ genomes reads part of what their heights already show, and it shares the heights’ blind spot: which half each child inherits.
Both parents need the same version of the score
A percentile means something only against the reference group of the score that produced it. Two scores built from different variant lists, or with different weights, rank people on different scales, and averaging a percentile from one with a percentile from the other gives a number that describes neither. The partner view therefore works out a child’s expected score only when both reports carry the same version of the height score, the same 3,286 variants with the same weights, and leaves it out when they differ.
It also leaves the score out unless each file read at least 90 percent of those variants. A gVCF can show this, because it records every position that was sequenced, including those that match the reference genome, and the report counts a variant as read only where the sequencing reached a depth of 10 reads.
Which files can carry the height score
| Your file | In your own report | In a child outlook |
|---|---|---|
| A whole-genome gVCF | A precise percentile when at least 90 percent of the 3,286 variants were read at a depth of 10 or more, and an estimate when 50 to 90 percent were | Yes, when both partners’ files reach 90 percent and carry the same version |
| A whole-genome plain VCF | An estimate. The file lists only where you differ from the reference genome, so it cannot prove the other positions were read | No |
| A chip export23andMe, AncestryDNA, MyHeritage | Not reportable. The arrays we checked read 14 to 43 percent of the variants, below the 50 percent the report needs | No. A chip report cannot join the partner view |
The chip figures come from four real exports run through the report’s own code: 23andMe versions 3 and 5, an AncestryDNA file from 2024 and a MyHeritage file. The free file check tells you in your browser which kind of file you have, and supported files explains each type.
How the partner view works it out
The height calculator and the height score both sit in Aimosti’s partner view, where two people compare their whole-genome reports. Each partner needs their own whole-genome report, from a VCF or a gVCF. One of you can invite the other by email so that two accounts are linked, or one account can hold both files once the partner has confirmed by email. How the partner view works, with a sample couple.
Once both reports are in, the view shows the calculator: two heights in, a target for a son and a daughter out, with the 8.5 cm range. The heights you type are not stored. If both of you also switch on the couple outlook, which is marked as just for fun, and both reports carry the height score from a gVCF that passes the bar above, the view adds each of your score percentiles, the score a child would be expected to carry, and the range that 95 in 100 of your children would fall in. It stays a score: the view never turns it into centimetres.
Two parents both at the 90th percentile, for example, give an expected child score at the 90th percentile, and 95 in 100 of their children would score between the 46th and the 99th. That spread is the part no test on the parents can predict.
A whole-genome report is $39 per person, or $59 for two with the couple pack.
Common questions
Can a DNA test predict how tall my child will be?
Only roughly, and less well than the parents’ measured heights. A test on the parents can give the score a child would carry on average, but each child inherits a random half of each parent’s variants, so children of the same two parents spread widely around that average. In the largest genetic study of height so far, the parents’ average height accounted for about 44 percent of the differences in their adult children’s height.
How accurate is the mid-parental height formula?
About 95 in 100 children reach an adult height within 8.5 cm of the target, so the range is 17 cm wide. The formula uses only the two parents’ heights, so it cannot allow for a child’s nutrition or health while growing, and it tends to underestimate the adult height of children whose parents are very short.
Can 23andMe or AncestryDNA raw data predict height?
Not with the height score in an Aimosti report. On the four real chip exports we checked, from 23andMe, AncestryDNA and MyHeritage, the array read between 14 and 43 percent of the score’s 3,286 variants, and the report needs at least 50 percent before it places anyone on the score. A chip report also cannot join the partner view.
Do height scores work for every ancestry?
Not equally. In the 2022 study a score from 12,111 variants accounted for about 40 percent of the differences in height among people of European ancestry and about 10 to 20 percent among people of other ancestries. The score in an Aimosti report is placed against a European-ancestry reference group, so for anyone of other ancestry its percentile is less accurate.
Why do both parents’ scores have to be the same version?
A percentile is a position on one score’s own scale. Two versions with different variants or weights rank people on different scales, so averaging them gives a number that describes neither. The partner view works out a child’s expected score only when both reports carry the same version of the height score, and leaves it out otherwise.
Does Aimosti store the heights typed into the calculator?
No. The calculator works out the target from the two heights in the form and saves neither of them.
Sources
- Yengo L, Vedantam S, Marouli E, et al. A saturated map of common genetic variants associated with human height. Nature. 2022;610:704–712. The 5.4 million people, the 12,111 variants, the 40 and 10 to 20 percent, and the 981-family comparison.
- Yengo L, Sidorenko J, Kemper KE, et al. Meta-analysis of genome-wide association studies for height and body mass index in ∼700000 individuals of European ancestry. Human Molecular Genetics. 2018;27(20):3641–3649. The study behind the score in the report, and its 0.44 correlation with height in the Health and Retirement Study.
- Tanner JM, Goldstein H, Whitehouse RH. Standards for children’s height at ages 2–9 years allowing for heights of parents. Archives of Disease in Childhood. 1970;45(244):755–762.
- Ciancia S, Cajas PR, Cools M. How accurate is Tanner’s formula in estimating target height? BMC Pediatrics. 2026;26:30, published online 6 December 2025.
- MSD Manual Professional Edition. Height Potential: A Child’s Target Height Based on Midparental Height. The 13 cm adjustment and the 8.5 cm range as two standard deviations.