site banner

Small-Scale Question Sunday for May 7, 2023

Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?

This is your opportunity to ask questions. No question too simple or too silly.

Culture war topics are accepted, and proposals for a better intro post are appreciated.

Jump in the discussion.

No email address required.

I realise this may come across as stirring the pot, but I hope I've been here long enough to have earned the benefit of the doubt.

In the context of the HBD debate, could someone please ELI5:

  • The concept of heritability and how it relates or doesn't relate to genetic causes of individual or group differences. I am aware of the "books at home" example. Is that all there is to it?

  • What precisely g is?

  • Steelman(!) Turkheimer's position. No, I don't want to hear about his politics.

  • Roughly summarise the position of Kirkegaard et al.

This whole debate always gets technical so quickly that I very often just get lost. I don't want to rehash the arguments here, I would like to understand the basics. But the waters are often so damn muddied (purposely so, I suspect) that it's very hard to get a grasp of what people are even fighting about.

Low-rigor response because I think you do have to study the object level a bit to evaluate the sides.

What precisely g is?

Precisely what its definition says. As Jensen himself put it:

It… reflects individual differences in performance on tests or tasks that involve any one or more of the kinds of processes just referred to as intelligence. The g factor emerges from the fact that measurements of all such processes in a representative sample of the general population are positively correlated with each other, although to varying degrees. A factor is a hypothetical source of individual differences measured as a component of variance. The g factor is the one source of variance common to performance on all cognitive tests, however diverse.

This is the definitive blog post refuting a popular methodological criticism.

Do you mean what it corresponds to in reality? I suppose it's just a holistic brain performance index, that's also predictive of general health. There isn't one physical thing that creates g, but the sum of diverse brain factors (half of our genome gets expressed in the brain) ensures that it emerges in the factor structure of our mental abilities, conditional on similar amount of training useful for each, and that predicts both general functioning and peak achievement. It isn't surprising that our abilities are highly correlated. We have a very homogenous brain made up of extremely complex computational elements (neurons) implementing simple task-agnostic learning algorithms, so most of the complexity and variation of our basic architecture (though not specific structure acquired over the lifetime) is shared between neurons or in their generic connectivity. Random genetic errors or (non-localized) environmental insults create different perturbations of the neuronal function, but on the whole-brain level they all push the system away from its optimal regime, no matter what it's trying to learn or to perform. You're unlikely to score in the 99th percentile on arithmetic and 20th on vocabulary if your axonal conduction is shot or your synapses are too sparse or do not get pruned well or your total cell count is too low or if your cell migration was too noisy or intracellular metabolism is somehow defective. And even if you've somehow developed specific tricks to cope very well with your shortcomings in a given skill, on the population level the power of the general factor becomes overwhelming.

Much of the issue is reducible to signal/noise ratio. High-performing brains ride the edge of metastability and wrangle representations easily; low-performing ones waste power fending off chaos and lose track of the context.

The concept of heritability and how it relates or doesn't relate to genetic causes of individual or group differences. I am aware of the "books at home" example. Is that all there is to it?

Steelman(!) Turkheimer's position. No, I don't want to hear about his politics.

But it is politics ultimately: we should be super, duper, ultra skeptical of HBD-related stuff and probably censor it for good measure, because Holocaust. He is leery of political implications of HBD acknowledgement regardless of specific merits of a given study.

Closer to the object level, I think he was recently steelmanned by Tailcalled:

People might read that intelligence is genetically correlated with myopia, or that homosexuality is genetically correlated with depression, and conclude that these are due to a direct biological link, rather than due to smarter people straining their eyes reading or staying indoors more, or homosexual people being discriminated against. Yet as we saw with education, this assumption is unwarranted; phenotypic causality leads to heritability.

Or put another way: the whole causal chain between genes and outcomes matters, for it may have links that depend on contingent properties of the environment. If proven, this would invalidate our assumption about this trait's heritability in a more general case. In our ancestral environment, smarter people probably hadn't been more myopic, because there had not existed the contingent segment «smart – reads a lot – sits indoors more than others – stupid monkeys haven't yet built good indoors lighting – not enough dopamine signaling in the retina – extended axial elongation period – myopia». It can be highly heritable, we can even find explanatory genetic polymorphisms – but the relationship disappears whenever we stop having our smartest children spend more time indoors with poor illumination. Likewise for all contingent chains. Therefore, even though everything is heritable, the first law of behavioral genetics does not mean that genetics is destiny or that people's phenotypes are molded by their genes in the intuitive strong sense the texture of their hair is.

This line of critique is pretty old, used by Lewontin etc. already, and is fair enough on its face, but assumes that some common mitigations psychometrists know about are insufficient, and it's only when it comes to proving this that we see clearly how Turkheimer is politically driven.

Roughly summarise the position of Kirkegaard et al.


We are living in a saturated and humane age. Human brain development is actually surprisingly robust and insensitive to inputs that are scarce in developed economies (even if some of them were scarce not so long ago in the past), and human cognitive development mostly proceeds close to the optimal way so long as you don't severely and obviously fuck up brain development or deviate far from low-effort common sense in nurture. Thus, developed societies have exhausted reasonable interventions that target contingent hazards disproportionately affecting g in different demographics; probably all hazards that bring down g in the population at large. While some differences in positive factors remain, they're long in the diminishing returns regime for more advantaged groups, as far as intelligence is concerned; and negative factors are similarly minor and maintained not by any iniquity, scarcity or coercion but out of their free choice by people who are worse off; and the rest is completely unsystematic. There's no low-hanging fruit left. No more toxic lead paint we can scrape off walls of homes where redlined minorities live, no malnutrition we can solve with food stamps, no miracle iodine-enriched brain-enhancing diet that only upper class kids are getting, no worms in ponds peasants have to drink from, no education reform that can remove some unnecessary cultural pressure. We've optimized and homogenized our environment to the point that practically all subpopulations realize the same, very high, percentage of their genotypic potential for g that is possible at the current technological/infrastructural level, and so differences in outcomes predicated on differences in g are explained by true differences in heredity, and are not amenable to elimination via any social policy change we'd recognize as fair.

Mostly the same logic applies to non-cognitive factors that influence outcomes, if to a lesser extent.

This model is corroborated by an extreme wealth of evidence and by the complete failure to create a competitively powerful model that rejects its premises, despite nearly a century of trying to do so, generous investment and genuine desire of talented researchers. This knowledge is being obfuscated and suppressed by political means.

Thank you! Very helpful.

Two more questions if you'll indulge me:

  • What are the most common anti-HBD arguments you encounter that aren't sophistry?

  • Gun to your head, what are the most persuading arguments against the position that the observable inter-group IQ gap is genetic in origin to a significant degree?

For me it was Chandy's research with IRS data that showed that income reversion to mean was the same for Black girls as for the White population. I suspect he's just doing something stupid but I can't dismiss the results especially since they are based on the largest sample ever collected in this space.

Thanks! Could you elaborate a bit? What makes this a good argument? What does it imply?

First off, it's Chetty. Sorry! I got him confused with a paper on distributed deadlock detection.

Chetty managed to convince the IRS to give him detailed income data on... everybody. Yes, literally everybody. It's a sociologist's wet dream. Anyway, he did a bunch of analysis comparing parental vs. child incomes. The idea is that populations tend to mean-revert but they do so differently across races, etc... Whites mean-revert to a higher income level than blacks. It's a neat way of getting rid of the influence of starting conditions. The study itself is here: The precise graph I'm talking about is here:

A lot of HBD arguments rely on the assumption* that Black underperformance is due to genes. However, there isn't much of a reason that Black women should consistently have better IQ-related genes than Black men (or at least none that I've heard yet). If outcomes between Black women and Black men diverge substantially, that implies that Black underperformance might not be related to their genetics after all.

  • An assumption with a lot of at least circumstantial evidence behind it (imo)...

That's not "the same for Black girls as for the White population", it's the same for black women and white women, the white male curve is noticeably higher. And on a meta level, it's always suspicious when a paper drops categories midway through. Whatever conclusions you want to infer from this graph, they would be much firmer if it also had asians.

Good catch on white women vs. white men. I wonder what causes the difference. I do think your comment on Asians is a bit of a non-sequitur.

Why a non-sequitur? Earlier parts of section IV show a higher steady state for asians than for whites, then they get dropped from the comparison. If they were included in this graph, it would either show that their advantage is also mostly male and asian women match white women, in which case maybe this graph says a lot more about gender than about race. Or the asian over white advantage is maintained for both genders, which would make for a much stronger anti-HBD argument. Because one of the more appealing HBD talking points imo is that by Occam's razor the black/white gap and the white/asian gap have the same basis.

More comments