🏅 Top traders

Curious what % other people would resolve this question to. Please post as a reply to this comment.

@levifinkelstein can I ask, what was the strongest argument you saw in favor of resolving towards yes?

@levifinkelstein Hmm I think my primary alternative hypothesis would be that there's some sort of social network thing going on, but I haven't seen any serious attempts at quantifying it so I don't know how much probability to give it.

An obvious argument would be that due to the fact that it's super easy to study, the fact that no anti-racists have done any good regressions must imply that it's not real, because anti-racists try to prove racism wrong and this would be a great approach for doing so. So maybe I should give a probability of like 95% or so.

However, this argument assumes that our civilization is even halfway adequate at investigating psychometric questions, and after developing a bunch of psychometric techniques myself that I haven't seen discussed much in these sorts of debates, I'm starting to doubt our adequacy at this.

But if we're inadequate at other psychometric questions, then the people pushing for the idea that there are race differences in IQ are presumably also inadequate. Heck, the idea of race differences in IQ is pushed by the same sorts of people who push ridiculous ideas like that IQ test scores aren't affected by effort. So based on that maybe I should give it a probability of 5% or so.

I think 70% is a fine compromise between "this certainly looks true if you consider the differential psychology on the topic" and "but social science is a dumpster fire so who really knows". (Differential psychology is Highly Replicable but that doesn't necessarily mean they use the right methods.) I might have chosen a higher number, like 80% or 90% or so, but I can't complain about this outcome.

@tailcalled "Heck, the idea of race differences in IQ is pushed by the same sorts of people who push ridiculous ideas like that IQ test scores aren't affected by effort. So based on that maybe I should give it a probability of 5% or so."

haha what, who are these people that think putting in an effort doesn't affect performance?

@levifinkelstein Timothy Bates (a leading intelligence researcher) recently did a study arguing that effort only matters a tony amount for performance:

https://twitter.com/tailcalled/status/1520314311285030912

This study has regularly been promoted around the human biodiversity sphere, e.g. by Emil Kirkegaard. For a more familiar name, Bryan Caplan likes it.

As you can see in my thread, I think Bates's study is kind of bullshit. But I think I've found out what historically happened to produce this bullshit:

So obviously black people score lower on IQ than white people, and there's been a lot of drama about why that is. It is common for leftists to do random asspull theories to explain it away, and one such theory is that black people weren't motivated enough to score high.

Given an asspull theory, there's of course a need for a bullshit study to prove the theory. Usually I think this involves disregarding questions of effect size, direction of causality, and statistical power, but supposedly (haven't checked) specifically in the case of the importance of effort, they instead shortcutted such issues by producing a fraudulent study showing that effort/motivation is such a big factor as to plausibly explain it.

I suspect online leftists then started posting the fraudulent study everywhere (or possibly posting meta-analyses based on it?), and that lead to human biodiversity people feeling a need to respond to it. But for some reason rather than responding with a reasonable answer of "the study you are spamming everywhere is fraudulent and if you consider the effect sizes involved then they can only explain a small fraction of the gap", they instead for some reason decided to do their own bullshit study disproving that effort matters.

(I suspect this sort of thing is an additional factor, beyond political bias against HBD, in why HBD is marginalized. A lot of social science is bad, and the HBD community is mostly reactionary, trying to push back against progressive ideas that they consider overreach. So a lot of HBD stuff involves taking bad progressive stuff and assembling a response/counter to it. But this means that progressive stuff can bait HBD into using bad methods, by framing the debate in terms of those bad methods.)

When HBDers are challenged on it, my impression is that they have a threefold argument:

• Many tests are administered in situations where it matters a lot for your future (e.g. SAT), so people should be applying max effort to those tests, and therefore effort ought not to matter for them

• Knowledge tests ought to not be dependent on effort, since either you know stuff or you don't, yet they probably correlate just as much with effort as performance tests do, showing that the effort correlation is actually reverse causality

• People bring varying levels of intelligence to testing situations because intellingence is genetic, byt there's no corresponding factors influencing effort which could plausibly lead to bringing different levels of effort to testing situations

I don't think any of these arguments are good. But I do agree with them that effort seems unlikely to explain the racial gaps.

@tailcalled There are corresponding factors but some of them are the very same factors - if you're smart, you're likely to apply more effort, whereas if you're really dumb, you might not realise that effort would be a good idea. I think I saw this in school many times, though the cases I thought I saw might have been more due to severe untreated ADHD or other mental issues, rather than low intelligence per se.

hate to see it

pls resolv

Very curious to see how this resolves

I would be curious what your current areas of uncertainty are, so we can share evidence about those.

A paper that supports the position of no genetically driven racial differences in intelligence between Black and White kids
https://www.nber.org/system/files/working_papers/w12066/w12066.pdf

Using a newly available nationally representative data set that includes a test of mental function for children aged eight to twelve months, we find only minor racial differences in test outcomes (0.06 standard deviation units in the raw data) between Blacks and Whites that disappear with the inclusion of a limited set of controls

It's reasonable to expect differences at 1 year to not be particularly predictive of differences later in life. This doesn't provide much infomation either way.

... I don't think this study, even if its results were in the other direction, should move the needle much.

A correlation of .3 between the BSID and future measures of IQ means that the
BSID score explains only nine percent of the variation in future test scores for a
particular individual.

One could easily see the BSID mostly measuring environmental factors of intelligence like nutrition/general health. Tailcalled's study, which I argued still doesn't move the needle much, is relatively much more convincing. X correlates with Y + Y as a proxy for Z -> X as a proxy for Z can go wrong!

@AmmonLam Note that since IQ is more environmental and less genetic in childhood then in adulthood, this paper arguably supports the notion that racial differences in adult IQ are genetic rather than envuronmental. Though I am not familiar with the details of the methods and suspect the result to mostly just be noise.

I created another market with better resolution criteria imo:

@JonathanRay It's not asking about the IQ gap though

@levifinkelstein Not directly, but it's something that environmentalists and hereditarians should sharply disagree on.

If I try to make a good-faith forecast, then there are a few things to consider. First, if they've been scoring around 0.8 std to 1.0 std lower than whites, then 4 IQ points is roughly ~30% of that. It should be pointed out that this is below estimates for the genetic contributions in the broader population (when not comparing races). Which for the trait of cognitive ability is probably >50%, and some results claim 80% (but it depends heavily on the dataset and how exactly "heritable" is being measured", I prefer closer to 50% out of uncertainty).

As a prior, it doesn't make much sense for it be lower than the background rate. That is, the percent attributable to genes shouldn't be lower when picking out more genetically-distant groupings, and comparing those. If anything, you would expect genetic variance to exert more unambiguously-causal effects onto the traits, as the genetic distance increases. And we would expect that relationship to continue past the black-white comparison, all the way out to comparing different species. That would probably end up looking obviously true, if we tried objectively measuring multi-species measures of intelligence. For this not to be true would be very weird.

Secondly, there's the fact that the black-white IQ gap is actually larger on the test subitems which are more g-loaded. This is much more compatible with genetic factors than the test being culturally biased, since the more culturally-biased subitems should be less g-loaded.

Thirdly, the closing of the gap seems to have stalled in recent years. It used to be around ~1.2 std, and IIRC made it to slightly below 1.0. Which is excellent, but my understanding is this has stalled for many years. The environmental-centric proponents were implicitly hoping the trend would continue. Hopefully down to 0.5 std and below, toward 0 with effort and improving conditions. But as far as I can tell, they seem stuck in the 0.8 to 1.0 std area. This seems more compatible with there being some genetic component, which isn't improved by the many social interventions that have been tried. People have tried very hard at this for many years. None seem to make lasting differences (that were g-loaded). Not beyond those initial improvements from things like not-being-poisoned by lead contamination, and proper nutrition.

Fourth, there's good reason to suspect that the Jewish - white IQ gap is >=0.5 std, and probably has some genetic components. This is related to their extremely high rates of some types of nervous system disorders, for which genetic carriers score higher on IQ tests. I doubt that Jews are just eating better food or "having stimulating discussions and that gives them higher IQ" or something. And I strongly doubt the Nazi interpretation that they are actively harming whites, and somehow causing such an IQ gap. I'm convinced environment plays some role, but it seems highly likely that genes account for some of the Jewish - white gap. This reinforces the point here, because there's no reason to think the black - white pairing would be particularly special, compared to the Jewish - white gap.

What probably will actually happen is that blacks will get substantially richer in the coming decades, just as they have in the past. And this is compatible with a world where genes explain >30% of the variance in intelligence. The above points make it pretty unlikely that less than 4 points are attributable to genes. Yet the whole time, things will keep getting better for most people. And people insisting it's 100% environment will secretly be the most-pessimistic, because that would leave fewer, less-replicable ways for improving the wellbeing and ability of blacks. But I digress, this is a forecast.

This is related to their extremely high rates of some types of nervous system disorders, for which genetic carriers score higher on IQ tests

iirc this bit didn't really end up being true (do the specific genes for those nervous sytem disorders have big iq pgs contributions)?, but cba to look it up atm

Every time relevant experts are surveyed anonymously, the median response is that both genes and environment contribute to the gap. The 100% environment view is held by fewer than 20% of respondents.

I would be cautious about trusting intelligence researchers, IME it is questionable how reliable they are. It is better to directly read the papers and maybe even analyze the data to see if their conclusions are supported.

E.g. one time a fairly major/leading intelligence researcher argued that IQ test scores don't depend on effort. I think that seems intuitively absurd to most people, who have taken a test and noticed lots of questions of intermediate difficulty that they don't immediately solve withhout effort, but can solve if they apply a bit of effortful thinking. And the intuitive feeling of absurdity is valid, because when I looked closer at his claim, it turned out that he did the statistics wrong and correct statistics supported the notion of effort mattering:

https://twitter.com/tailcalled/status/1520314311285030912

However despite the study being flawed, it still went viral among a bunch of IQ researchers and race science types, and the researchers in question don't seem to have properly acknowledged the flaws.

emil weighs in here: https://emilkirkegaard.dk/en/2020/08/motivation-and-iq-scores/ His plot of standard error vs effect size makes effort effects look like mostly publication bias.

@JonathanRay Hedge's g is a measure of an effect size for binary variables, which means that it is probably comparing incentivized vs incentive-less performance. But as my tweet chain explains, this does not necessarily compare the effect of effort, because effort may not be very affected by incentives.

I guess we should post evidence? This is actually interesting because there's some emerging evidence that seems like it has the potential to be strong to me but which might have some sort of subtle flaw that I don't understand. Lemme explain:

Behavior geneticists regularly construct something called "polygenic scores". At its most basic, a polygenic score is constructed by taking an absurdly large number of people, measuring some trait in them (such as IQ), as well as sequencing their genes, and then using regression to predict the trait from the genes.

In principle, as the sample sizes get large enough, this should recover the genetic factors that affect IQ. So naively, we should just be able to compare the polygenic scores for IQ (or IQ-related traits), to make a clean comparison of genetic predisposition to IQ. Unfortunately, the naive method isn't actually that clean, because polygenic scores constructed in this way are not just going to isolate the genetic causes, but are also going to be confounded by various effects. From what I understand, most of the possible confounding factors aren't going to break the racial comparisons, but there are two kinds that might:

• Population stratification: If there were genetically different populations (say, different races) which differ in important environmental factors (say, nutrition), then the genes that indicate group membership in this populations would end up functioning as proxies for the environmental factors, and because the environmental factors affect the trait in question, the proxies would end up being included as predictors in the polygenic score, even though the genetic variants are not actually causal for IQ.

• Gene-environment interactions: If there were genetic factors (say, dark skin color) which due to some environmental factor (say, racial prejudice) have a negative causal effect on IQ, then these genetic factors would get included in the polygenic score for IQ, even though they would disappear under equalized environments.

I think people who construct polygenic scores are usually aware of these problems and more, so they tend to do various things to adjust for them. For instance:

• Constructing the polygenic scores in an ethnically homogeneous sample, and controlling for a number of principal components of genetic variance. I think that should eliminate much of the population stratification effect, at least on the racial level (though it might continue to exist on a smaller level, due to stuff like assortative mating - but I think this shouldn't confound racial comparisons, unless I'm missing some mechanism?).

• It is possible to look at gene expression to get some clues for the mechanism, e.g. this can distinguish factors that operate through physiological mechanisms like skin color or bone structure (as would be expected from certain kinds of racism hypotheses) from factors that operate through neurological mechanisms (as would be expected from the innate IQ hypothesis).

Anyway, as I understand things, researchers have used large samples to construct polygenic scores that predict IQ as well as IQ-related variables (such as educational attainment, which is probably the one with the largest sample sizes). These polygenic scores AFAIK pass the two tests above.

The scores are not equally valid within different races; they correlate more strongly with IQ among europeans than among asians or africans. This is to be expected from the methods used to construct them, because:

• They don't actually measure all of a person's DNA bases when collecting the data; instead they collect a smaller number of reference positions which correlate with large segments of DNA. These reference positions are used as a proxy for the actual DNA, and so "tags" the causal variants, but because of the differences in genetic population structure by race, the correlations will be different across races, and therefore the tags will not work as well in other races than the ones the polygenic score was originally fit on.

• Current methods don't have sample sizes large enough to detect all of the genetic variants that are relevant for IQ, so instead they detect the variants that are sufficiently common or have sufficiently high effect size to be obvious, given the sample sizes they work with. (Also they just entirely ignore variants that are too rare; IIRC they only include variants with a frequency of more than 1-in-1000.) However, since different races differ in their distributions of genetic variants, this means that the variants that are statistically obvious in one race might not be statistically obvious in another race, and so we would expect the correlation to be weaker as we switch from the race we've fit the score on to the race we haven't, even if the scores were entirely based on causal variants.

• There is going to be some bog-standard overfitting going on, where e.g. a genetic variant just happens to be more common in people of higher intelligence for no particular reason other than chance, and this is going to affect its coefficient in the model.

I've seen people suggest that these sorts of factors would bias racial comparisons in polygenic scores, but I don't understand how or why. I would think that whatever biases would be introduced by these factors would be random, and so they would cancel out in expectation. Like for instance, maybe bias 2 means that a genetic variant that affects intelligence among black people but is nonexistent or has reached fixation among white people would not be part of the score, but that doesn't seem like a big deal because I would think the variant would be equally likely to increase or decrease IQ and so the expected gap it would induce would be 0.

These three reasons might not be the only reasons that polygenic scores vary in accuracy by race; gene-environment interactions might also be relevant in some way. But the sort of gene-environment interaction I mentioned before wouldn't lead to differences in polygenic score accuracy by race I think, it would have to be some subtle one that I don't think I've seen any proper explanation for why would be relevant in this context.

Ok, so enough theory, what about practical stuff? Among race scientists, there's sprung up a tradition of ranking the races on the latest version of whichever IQ/education polygenic scores come out. So for example, one notable study is Global Ancestry and Cognitive Ability, compares racial groups across polygenic scores for education constructed in different ways and finds that white people score higher than black people. But I'm kind of confused about their exact methodology - they have two polygenic scores, one of which seems to have been "cleanly" constructed to avoid the sorts of concerns I wrote about in the beginning of the comment, and one of which seems to just have been optimized for maximizing predictive power. As far as I can tell, they are mainly ranking the races on the latter score, and I can't find the race ranking of the former score (I might have overlooked it, idk).

(They also have a bunch of other interesting investigations, for instance looking into whether skin color predicts IQ after controlling for ancestry.)

My impression just from seeing various discussions is that the racial rankings in education/IQ polygenic scores are fairly robust. However, the rankings have mainly been produced by people who try to prove racial difference in IQ, so it is conceivable that they are biased, and mainstream behavioral geneticists don't want to touch this debate other than to say that they oppose racism, so they don't make their own alternative rankings.

For a contrasting perspective, I guess it might be relevant to link the paper No support for the hereditarian hypothesis of the Black–White achievement gap using polygenic scores and tests for divergent selection, where a biologist argued that the IQ gaps are not genetic. I don't understand his argument as well as I understand the argument on the opposite side, but you might want to read his paper and see if it gives anything informative.

Actually I've just been told that Divergent selection on height and cognitive ability: evidence from both genetic distance (Fst) and polygenic scores is a better study for this than Global Ancestry and Cognitive Ability. It uses polygenic scores that were created in an ethnically homogenous sample, and finds a correlation across ethnicities, e.g. here:

plausibly meaningless speculation:
weird that US blacks / LA mexicans are significantly above trend (indicating the trendline overestimates genetic contribution to iq, although not that its not genetic), but that US blacks are so close to other blacks on the PGS - and yet all the african ethnicities have such similar PGSes (does that make sense? why would the othe rethnic groups be so varied but the african ones e so clustered?).

they seem to use lee et al's eduactional pgs, which is as you said:

All association analyses were performed at the cohort level in
samples restricted to European-descent individuals

Maybe something weird about the way the pgs generalizes makes that happen?

I think this is very weak evidence, but it's less very-weak as any of the anti-genetic-cause evidence I've seen. Seems very plausible something subtle breaks it, even in the case the iq-is-genetic conclusion is true.

weird that US blacks / LA mexicans are significantly above trend (indicating the trendline overestimates genetic contribution to iq, although not that its not genetic)

If I understand the data correctly, many of the datasets are from different countries. Nationality appears to affect IQ scores just as much as genetics does, for non-genetic reasons.

So if you compare black people in less-developed african countries to white people in the us, you are going to also get the nationality effect and not just the genetic effect.