Question on the breeder equation - I recently tried to calculate what's possible in 8 generations of human selection (driven by assortative mating).
I was getting absolutely crazy results - like even if the top 10% assortatively mates, so not terribly selective, in 8 generations the 8th generation have mean IQ's of ~220, and then get positive and negative buffs from there, so the geniuses in this crowd would all be 250+.
That...seems wrong to me. We should see it in the world, for one thing.
Okay, so where are the breakdown points? h^2 theoretically decays per generation, as you select out the "random luck" baseline components. That can nerf it a little - but also, we might expect it to rise. If Greg Clark has taught us anything, it's that "lineage mean" is meaningfully different than "population mean," and your pool regresses to their lineage mean, which is increasing each generation.
And lineage means really ARE much stronger, empirically - see "social competence" persisting at ~75% everywhere, significantly higher than the actual intelligence and educational attainment h^2's of ~0.5 and ~0.4.
So what else? Is the top 10% assorting too tight? It doesn't seem to be? After all, if we look at the top ~0.5% - 1% that get into T20 schools in America, the child of a T20 alum is 70-75x more likely to get in than a given random person (Chetty). So if the assortation is THAT tight, 10% seems extremely generous.
So I'm kind of at an impass there - why aren't we seeing IQ 200+ families out there? Sure, there's high attainment lineages like the Galton / Darwin / Wedgwoods, the Bohrs and Curies, etc. Scott at ACX had a post about several (https://www.astralcodexten.com/p/secrets-of-the-great-families). But it's not "200 IQ mean" levels of distinction.
I ran this puzzle by Claude, and it came up with some cockamamie ceiling defined by dynasty recurrence z_ss = b * z_spouse / (2-b). So if we plug in Clark's .75 persistence, and a 2sd spouse, you get a 1.2sd cap.
But the whole point of the assortation over generations is that your and your spousal lineage means are increasing! You're not capped at 2sd!
Further, we know that there can definitely be multiple distinct lineages assorting to this level - with Kulin Brahmins famously attaining persistence nearly at ~1, Samurais diverging only ~5% from platonically perfect assortation over hundreds of years, and so on.
So is the breeder equation simply not suitable for multiple generations for some reason? IS there a ceiling on what assortation can achieve, even down at only top decile assortation? Why isn't it whacking T20 alums, if so?
So anyways, I'm kind of puzzled, and hoped you might chime in.
Whenever I have questions like that I go to Genetics and Analysis of Quantitative Traits by Lynch and Walsh. They have a section on assortative mating where they have models for this. Particularly, pages 155-160.
My intuition is that what’s going wrong in your model is either:
A. You are modelling perfect matching; realistically speaking, T20 students are 120 IQ or so, and do not match heavily on intelligence.
B. You are not taking into account regression to the mean (average of two parents → child correlation is 0.6).
Another note: the distribution of intelligence is left-tailed, so people who are 4SD above the mean in percentile terms are maybe 3SD above the mean in real terms. Intelligence does not have diminishing returns, but I think that the genes that cause it do.
Particularly, equations 7.18 and 7.19 -- which allow you to model how much population variance changes with assortative mating, and then an upper limit of how much the variance can increase with each assortatively mating generation.
My guess would be physiological limitations regarding the brain, in terms of neuron density and waste heat management.Perhaps the morphological changes needed to operate mean IQs of such magnitude, would require a combination of very strict lineage sorting and selection incentives that hasnt come about so far.
The solution is easy. Key attributes like attractiveness, social functionality and mental operationality just drop off after around 140 IQ. Its like a built in kill switch in our DNA. A closely related reason would be having those 3 attributes allow you to be socially dominant and use other people's brainpower for your own goals, which is more beneficial personally despite not being productive socially or civilisationally.
This is still true in russia (or so ive been told) probably since that is seen as the moral status quo over there. It was also certainly true in de gaulle’s france
@Leon Voß
Question on the breeder equation - I recently tried to calculate what's possible in 8 generations of human selection (driven by assortative mating).
I was getting absolutely crazy results - like even if the top 10% assortatively mates, so not terribly selective, in 8 generations the 8th generation have mean IQ's of ~220, and then get positive and negative buffs from there, so the geniuses in this crowd would all be 250+.
That...seems wrong to me. We should see it in the world, for one thing.
Okay, so where are the breakdown points? h^2 theoretically decays per generation, as you select out the "random luck" baseline components. That can nerf it a little - but also, we might expect it to rise. If Greg Clark has taught us anything, it's that "lineage mean" is meaningfully different than "population mean," and your pool regresses to their lineage mean, which is increasing each generation.
And lineage means really ARE much stronger, empirically - see "social competence" persisting at ~75% everywhere, significantly higher than the actual intelligence and educational attainment h^2's of ~0.5 and ~0.4.
So what else? Is the top 10% assorting too tight? It doesn't seem to be? After all, if we look at the top ~0.5% - 1% that get into T20 schools in America, the child of a T20 alum is 70-75x more likely to get in than a given random person (Chetty). So if the assortation is THAT tight, 10% seems extremely generous.
So I'm kind of at an impass there - why aren't we seeing IQ 200+ families out there? Sure, there's high attainment lineages like the Galton / Darwin / Wedgwoods, the Bohrs and Curies, etc. Scott at ACX had a post about several (https://www.astralcodexten.com/p/secrets-of-the-great-families). But it's not "200 IQ mean" levels of distinction.
I ran this puzzle by Claude, and it came up with some cockamamie ceiling defined by dynasty recurrence z_ss = b * z_spouse / (2-b). So if we plug in Clark's .75 persistence, and a 2sd spouse, you get a 1.2sd cap.
But the whole point of the assortation over generations is that your and your spousal lineage means are increasing! You're not capped at 2sd!
Further, we know that there can definitely be multiple distinct lineages assorting to this level - with Kulin Brahmins famously attaining persistence nearly at ~1, Samurais diverging only ~5% from platonically perfect assortation over hundreds of years, and so on.
So is the breeder equation simply not suitable for multiple generations for some reason? IS there a ceiling on what assortation can achieve, even down at only top decile assortation? Why isn't it whacking T20 alums, if so?
So anyways, I'm kind of puzzled, and hoped you might chime in.
Whenever I have questions like that I go to Genetics and Analysis of Quantitative Traits by Lynch and Walsh. They have a section on assortative mating where they have models for this. Particularly, pages 155-160.
My intuition is that what’s going wrong in your model is either:
A. You are modelling perfect matching; realistically speaking, T20 students are 120 IQ or so, and do not match heavily on intelligence.
B. You are not taking into account regression to the mean (average of two parents → child correlation is 0.6).
Another note: the distribution of intelligence is left-tailed, so people who are 4SD above the mean in percentile terms are maybe 3SD above the mean in real terms. Intelligence does not have diminishing returns, but I think that the genes that cause it do.
https://www.technotheoria.org/p/is-intelligence-really-normally-distributed
Particularly, equations 7.18 and 7.19 -- which allow you to model how much population variance changes with assortative mating, and then an upper limit of how much the variance can increase with each assortatively mating generation.
Link to the book:
https://annas-archive.pk/md5/41d9cca5e44028effca96d9415adf6ba
My guess would be physiological limitations regarding the brain, in terms of neuron density and waste heat management.Perhaps the morphological changes needed to operate mean IQs of such magnitude, would require a combination of very strict lineage sorting and selection incentives that hasnt come about so far.
The solution is easy. Key attributes like attractiveness, social functionality and mental operationality just drop off after around 140 IQ. Its like a built in kill switch in our DNA. A closely related reason would be having those 3 attributes allow you to be socially dominant and use other people's brainpower for your own goals, which is more beneficial personally despite not being productive socially or civilisationally.
Makes sense. Only an environmental change would make people more conservative in aggregate.
You forgot to add the most important part:
Woman are much more likely to be liberal than conservative!
This is important when you are measuring TFR (total fertility per WOMAN).
They are much more likely to be collectivist*
It wasn't that long ago when roles were reversed and women were more conservative than men.
This is still true in russia (or so ive been told) probably since that is seen as the moral status quo over there. It was also certainly true in de gaulle’s france