Hacker News new | ask | show | jobs
by throwaway9917 653 days ago
Many of these studies don't account for test/retest reliability of IQ tests. Height is pretty easy and reliable to measure, and it doesn't change based on how much sleep you had the night before, or whether you ate breakfast, what the conditions of the test were, the time of day, etc.

The test-retest correlation on most IQ tests is around 0.7-0.8, whereas for height, it's almost 1.0. That means if you're measuring intelligence by a single IQ test, the correlation with genetics has a maximum of 0.7-0.8 due to noise in the test.

Studies that either explicitly correct for this, or look at averages of multiple tests taken over time show higher correlations between genetics and intelligence.
2 comments
Jensson 653 days ago
> whereas for height, it's almost 1.0

You grow and shrink about 1% per day, as you get compressed when you stand up and decompress when you sleep. That is about 1.5 cm, and with a human height standard deviation is about 6cm you get around 0.25 standard deviation variation when measuring the height of a single human at different times of day, or about 4 point difference if we have 15 points per standard deviation.

link
throwaway9917 653 days ago
Standard deviation of human height (within a single sex) is about 7cm. If human height changes over the course of a day by 1.5cm, assuming it linearly declines over the day, that's a stddev of about 0.6cm. You also can reasonably assume that people are normally going to be measured at the doctor/researcher's office during business hours, so it's probably more like 0.4cm. The measurement error of just standing up straight or not and reading the numbers is probably more.
link
bumby 653 days ago
Not to make a nitpick in your point because it's a good one, but it made me wonder:

How different would the height measure be if it were standardized like IQ? E.g., how likely would someone who is in the 90% percentile in 6th grade be in the 90% percentile as an adult? From that perspective, I imagine the standardized height "re-test" correlation is a lot less than 1.0 (although it would certainly be close to that with two measures of height as an adult).

link
dmurray 653 days ago
0.8 correlation with adult height for boys aged 4, according to [0]. The real study [1] is paywalled but looks legit. I'm guessing it's even higher at age 12, though it might be confounded by puberty soon after.

[0] https://www.theatlantic.com/ideas/archive/2020/02/can-you-pr...

[1] https://pubmed.ncbi.nlm.nih.gov/21767107/

link