« Fired for mentioning he believes in Christian stuff | Main | Trains higher class than buses »

November 11, 2009

Comments

It's funny, because that's what I'm afraid of. I have a 130+ IQ (gifted kid, law school), yet my parents are frustrating simpletons, so I think that I may be a cognitive genetic fluke, in which case I genuinely fear that I will produce moronic children. lol.

I think the narrow sense heritability of IQ is considerably lower than .8, .3 or .4 are probably more reasonable guesstimates. I don't know if that's for childhood or adult IQ though.

All the more reason to seek a partner who is intelligent and has intelligent parents and family. Even with regression towards the mean, you are still significantly more likely to have children with high IQs. You generally see this in people who come from good families. It does seem like people who get "lucky" and are able to gain financially because of their fluke intelligence might be more likely to seek a partner based only on looks, like a hot waitress or nanny.

If I remeber correctly, the IQ is supposed to regress towards the mean of the parents' "group". I never really understood what this group consisted of - humans in general, ones race, ones nation, ones class or caste or some kind of historical family average. The other issue is that there must be some correlation of the environmental component between parents and children as well - perhaps even beyond factors that don't seem to impact IQ such as class or home environment.

I can't help but think that you need some expert advice on this topic.

As for the IQ of your children, it's all a matter of chance, it's just that the odds differ for each of us.

I am a liberal, and by that I mean I am a person that genuinely wishes that all races could have the same IQ. I think the world would be a much better place if all races had the same IQ. I am eager to pay higher taxes if there was some way my taxes could be used for a social program that would produce a society where all races have the same IQ

Unfortunately, my reading of this blog and my independent study of the issue has led me to conclude with certainty that the races will never have the same IQ. Social spending is mostly wasted.

I think that we need to address regression to the racial mean. The reason is that in our society you routinely see East Asian parents with low IQs producing grandchildren with very high IQ and you routinely see black parents with high IQ producing grandchildren with low IQ. The effect is so startling, and so demoralizing to those high IQ black grandparents, that it really destabilizes things.

I would like to discuss this - how fast does the regression to the mean work - one generation, two generations, etc.

But if you had a group of 130+ IQ Jews, Northern Europeans, SE Asian Indians, and NE Asians and cross bred them you'd be specifically breeding for high IQ and get less regression to respective group means. What you'd get is what occurs any top ranked US grad school in the hard sciences, physics or math.

Tomatohead, if optimizing the IQ of your offspring is a top priority, marry someone who is at least as smart as you but due to as different genetic basis as yours as possible.

You're offspring will also be more resistent to the many deleterious human genetic diseases that are recessive, including the most deadly ones which only persist in recessive form.

I, literally, have an I.Q. of 142, and I find this blog to be trite and, quite honestly, immature. Your post on Frank McCourt writing anti-Bible papers in high school vs. writing a children's book that is pro-bible is absurd. Is it not the purpose of life to....FIND the meaning of life??? Doesn't one mature with age, and with age acquire wisdom that takes us closer to the Truth?

Good luck with your blog. I hope you, too, grow with age from mere cerebral intellectual to a person of wisdom. Wisdom trumps the intellectual every time, as wisdom is beyond the mind.

[HS: I assume he wrote the children's Christmas book to make money. No one's going to pay him to write an atheist Christmas book.]

Ask Razib (at gene expression) what the regression to the mean theory means.

My non-expert guess is that regression to the mean refers to the average IQ of one's four grandparents.

IQs much higher or lower than that figure are likely a fluke or some kind of a recessive manifestation.

"gentiles" huh?

"Tomatohead, if optimizing the IQ of your offspring is a top priority, marry someone who is at least as smart as you but due to as different genetic basis as yours as possible."

Again, wouldn't this go back to hybrid vigor? Perhaps Jews should be setup with NE asians?

As I understand it regression towards the mean works as follows: Assuming both parents are of the same ethnicity, and one parent has an IQ of, say mu+3*sigma, the likelihood of the other parent having such high IQ is small. Consequently their children will have an IQ in between the parents. But regression towards the mean can be minimized if both parents have a high IQ. This is backed by the little we know of SNPs that increase intelligence. The Alleles that bring about high IQ usually come from being homozygous (two letters being the same). So, in order to maintain the high-IQ trait both parents need to be homozygous.

"I never really understood what this group consisted of - humans in general, ones race, ones nation, ones class or caste or some kind of historical family average. "

be a bayesian.

"For example, if you gathered together several thousand white gentiles with an average IQ of 130, and they all had children, then according to some theories, the average IQ of their children might be only 124. But assuming there’s no sexual selection for high or low IQ, and they all mate only within the group, all future generations would also have an average IQ of 124."

yes. this is called *evolution* :-) the less intelligent members of the HBD-sphere talk about regression to the mean as if it was some mystical force. kind of ironic that dumb people don't understand the nature of IQ as a heritable trait very well (to be fair, francis galton has this attitude too, though karl pearson kept trying to explain him to the essence of the issue he didn't get it).

in any case, re: narrow-sense heritability,

1) heritability is higher for higher SES in the developed world

2) assortative mating for trait-value increases heritability


finally, there are estimates for middle class & up for narrow-sense IQ heritability on the order of 0.8. but others object that this is an overestimate that is collapsing in gene-env correlations and such (perhaps malloy will comment since he knows this stuff).

i highly recommend d. s. falconer's *introduction to quantitative genetics* for people who are curious.

Regression toward the mean is like an IQ test for HBD enthusiasts. I estimate that fewer than 5% of commenters and bloggers understand it.

I guess Galton himself did not, but he didn't know about genes. It's trivial if you have any math chops -- just the addition of uncorrelated random variables, one of which represents genetic potential.

Here is a discussion in the context of athletic ability:
http://www.gnxp.com/blog/2009/10/version-20-of-montana-gretzky.php

I am sure that a lot of IQ is inherited. But a simple question. If I have a higher IQ than my two siblings, or I can see difference in my children's IQ's (and they have the same parents), how can it all be a matter of a simple mathematical formula? The same parents give some of their children higher IQ genes than they do their other children. this is a complex genetic formula, not a simple one.

I, literally, studied the WAIS-III. You can't "literally" have an IQ of 142 simply because the test isn't so accurate as to give you a particular number. It's not like a ruler where you can determine height down to the millimeter. The IQ score is always a range, say 110-115 or 138-142.

At any rate, giving your IQ as a calling card is, quite honestly, trite & immature.

you can assume most readers of this blog have an IQ of at least 1.5 to 2 SD's above the mean simply due to the large amount of abstract material presented.


"I, literally, studied the WAIS-III. You can't "literally" have an IQ of 142 simply because the test isn't so accurate as to give you a particular number. It's not like a ruler where you can determine height down to the millimeter. The IQ score is always a range, say 110-115 or 138-142."

Indeed. Especially considering IQ scores regress to the mean to. People who score 142 on one IQ test such as the WAIS-III will average an IQ equivalent of "only" 129 when they take another IQ test such as the Raven, SAT, LSAT etc. And people who qualify for Harvard law on the LSAT, are often shocked to discover they don't even qualify for Mensa when they sit down and take the WAIS-III. Their scores have regressed to the mean. This happens because the higher you score on an any IQ test, the more likely it is that you lucked into a test that just happens to measure an aspect of intelligence you are especially gifted in, but when confronted with a second IQ test (especially one measuring other aspects of intelligence), odds are your luck will run out & you're not as much of an outlier as the first test suggested. It's known for example that the average IQ at Harvard as measured by the WAIS is 130, even though the average IQ as measured by their SATs is about 143.

I was under the impression that regression towards the mean still continued, but at smaller intervals. In the example above, the first generation has an IQ of 130 and half of their genes are not expressed. The second generation has an IQ of 124 and a quarter of their genes are not expressed. I thought the third generation would have an IQ of 121 and have an eighth of their genes not expressed. The terminal generation would have an IQ of 118.

*The effect is so startling, and so demoralizing to those high IQ black grandparents, that it really destabilizes things.*

It's the main reason behind my choice to no longer have children. Why take the risk in bringing a child into the world that isn't smart enough to become successful in it? Why bring a child that's isolated from his racial peers, and from the people of his intellectual level? Even marrying outside of my race doesn't solve the problem since it doesn't purify the potential low IQ genes.

I'm almost tempted to think that the anti-white rantings of some in the black community can be interpreted as hate of those with better IQ, and the realization that they'll never be able to compete with whites and Asians.


"Tomatohead, if optimizing the IQ of your offspring is a top priority, marry someone who is at least as smart as you but due to as different genetic basis as yours as possible.

You're offspring will also be more resistent to the many deleterious human genetic diseases that are recessive, including the most deadly ones which only persist in recessive form."

Excellent point, Elmer.

It would be interesting to see IQ data of the kids whose a parents are high IQ people from significantly different genetic backgrounds.

Even if it's the same test, one can expect regression to the mean.

Suppose you take 1000 college students and give them all the LSAT. You then take the 10 kids who got the highest scores and have them re-take the LSAT a month later. One can be reasonably confident that the average scores of those 10 kids will go down.

It seems to me that there are really a fair amount of mental abilities that we call IQ -- they might often relate to each other but perhaps not always.

The SATs/GREs, for instance, are clearly a test of aptitude, but the questions themselves seem to require just the teeniest bit of creativity. They're mostly easy, but of course some people will be able to solve them faster than others. I would say that you can take an individual with an IQ of an 100, train him or her on the SATs/GREs, and, giving them as much time as they need (say, 5 hours instead of 2.5) you can get them to earn a score equivalent to what kids who go to top twenty schools earn. But that's ONLY if you give them enough time -- under normal time constraints, this person will *not* be able to speed up his brain.

On the other hand, if you look MENSA type IQ questions, they're some seriously tricky/difficult problems. A person can stare at them as long as possible and not be able to answer. The same goes for a lot of those problems were they ask to predict the next item in a serious. You can simply be incapable of solving them.

But the logic games on the LSATs, and any analytical/quantitative question on the SATs/GREs, always seem to be easily solvable -- the question is, can you solve them in the amount of time they give you?

Thus I'm not surprised that some people who score well on collegiate aptitude exams might not end up scoring spectacularly that traditional hardcore IQ exams (like Mensa's)...even though they still end up with decent scores. We're talking about different sort of intellectual abilities here.

If you attempt to measure someone's IQ, you'll get some measured value, call it X.
X is the combination of the true underlying value (call that T) and a random variable. The tighter the correlation (i.e., the more valid and precise the test), the smaller the magnitude of the random component. Regression to the mean is basically the observation that someone who scored really high, or really low on a measurement probably has a very high or very low value for said random variable. On something like the SAT at the high end, making or not making a few mistakes due to fatigue or inattention can make quite a bit of difference. Accordingly, that person is likely to have a random variable closer to the mean on their next administration. The concept is far easier to explain to people who have experience with role playing games with random, but weighted attribute generation systems.
When you're talking about a person's kids, regression to the mean doesn't refer so much to the dilution of your passed-on intellect due to your mate being slower than yourself. Instead, it refers to the fact that if one or both of you are freakishly smart (or dumb) relative to your respective families (who are the best estimate of the heritable component of the IQ that you personally possess), your children are likely to have a more average throw of the dice on the random component of their intelligence---i.e., they'll have an IQ closer to the mean of your respective families than you and your spouse do.
So if you're looking for smart kids, go for a spouse that is smart and who has a similarly smart family---i.e., she's not an outlier in it.

a friend who is a statistician gave me three examples which helped make it clearer for me:

1) it's "noise"

2) it's measurement error

3) thinking of a bivariate distribution (one variable being additive genetic variance, the other everything else)

that error can be environmental, developmental, or due to breaking down of extremely favored genotypes through sexual reproduction (you can think of lots of "errors"). if you think of it as an "error term" i think it is clearest why i suggested that one should be a bayesian; update your information when possible, but know you aren't going to be perfect. keeping in mind that with something like height or IQ there's going to be a residual that you're not going to be able to account for (though with height in developed world situations the heritability is at least 80% now, and some estimates place it closer to 90%). there's a ~15 point standard deviation in IQ among siblings.

note that heritability increases for height & IQ when you have developed world levels of nutrition. you're squeezing some of the noise out of the prediction, and concomitantly regression isn't as much of an issue since parent-offspring correlations loom larger (instead of variables which you aren't aware of, and which are distributed across the population randomly).

but like i said, read d. s. falconer. not that long, the algebra isn't too hard, and it's *interesting*.

"If I remeber correctly, the IQ is supposed to regress towards the mean of the parents' 'group'. I never really understood what this group consisted of - humans in general, ones race, ones nation, ones class or caste or some kind of historical family average."

Look, it's quite simple:

(1) Your grandparents' recessive traits can impact you even if they "skipped" your parents.

(2) There are children who are less intelligent than either of their parents.

Give it a few moments thought and the concept of regression to mean should become intuitive.

I don't mean to imply in my last post that the problem is strictly one of recessive genes. It is a matter of the distribution of genes in a population.

Imagine two kinds of genes contribute to intelligence and that the dominant version of each gene provides the greatest intelligence. 'AB' confers the greatest intelligence, but is found at a low frequency. 'Ab' and 'aB' confer less intelligence. Your mother has 'aB' and your father has 'Ab.' You might get 'AB' and end up smarter than your parents (25% probability). You might also get 'ab' and end up dumber than your parents (25% again). There is a 50% chance you'll be as smart as one of your parents. So far, so good.

The problem arises in that you are in a population where 'a' and 'b' are far more common than 'A' or 'B' and you marry from within that population. If you won the genetic lottery and got 'AB' while your spouse has 'ab' (most common), 'aB,' or 'Ab' then your children will be less intelligent than you.

Now spread this effect over the hundreds or thousands of genes that govern intelligence and you can see the problem: you can only increase intelligence among your descendants by finding an intelligent partner, but to the extent that potential partners are exceptional compared to their own ancestors, you run the risk of losing your gains among your children. Even if your partner is your equal in intelligence, he or she will be harboring plenty of very common, less than optimal genes (as will you). They might be dominant or recessive, but to the extent they differ from your own set, and to the degree you both come from mediocre backgrounds, you and your descendants will have problems holding your gains as generations pass. The only way to overcome the problem is as HS just mentioned: find intelligent partners, accept a leftward shift in the bell curve for a generation, and don't marry outside this new population thereafter.

Besides the obvious (marrying someone with worse genes than yours), regression to the mean has several sources:

1-Uncertainty in measurement. Your score may be affected by random error. If you score very high, it is more likely (AOTBE) that you were "lucky" than "unlucky" (i.e. that the error was positive). At the next measurement, you (or your kids) are therefore more likely to score lower than higher.

2-On any given scale, there are physical limits to how much you can score. When you have very high-score genes, there will be very few events (recombination, epistatic interactions, mutations) that can make your score much better at the next generation. OTOH there are many possible events that can make it much worse. There are diminishing returns on the numbers of "good" genes that you can have, and when you already have many good genes, you have more to lose than to gain - even if you marry someone smart. That's a case of weak epistasis: genes effects are almost additive, except at the extremes.

3-Of course, there's an even simpler explanation, namely strong epistasis: in the real world, genes do not act additively, they interact with each other. So very smart people are the result of a "just right" combination of interacting genes, and any modification is likely to disturb these interactions, leading to poorer performance regardless of how smart your spouse is (because her own smartness may result from a different set of genes interacting in a different way, and mixing the two sets will not result in the benefits of either).

Or in short: mixing half the ingredients of a bacon cheeseburger and half the ingredients of a chocolate cake is not going to result in anything nearly as good as either.

Yeah, I know, "IQ is purely additive", blah. Whatever.

For what it's worth, the High IQ Society online test appears to be extremely reliable.

The first time I took it, I received a 133. For the hell of it, I took it twice more and received 131 and 134. Only on the fourth try did I get the discussed-above regression dip into the low 120s.

I suspect the widely popular SAT-IQ conversion charts are utter crap. E.g. I received a 600V/800M pre-recentering SAT score, which converts to an IQ in the mid-140s. I don't think the Math SAT g loading is nearly as high as that for Raven's type tests.

But, it does appear if you take a test several times, the average will be something close to your "real IQ."

On second thought -- perhaps the SAT-IQ conversion tables ARE fairly accurate -- if they are read as SAT-VIQ conversions.

Exactly what kind of SAT problems measure performance/non-verbal IQ?

Maybe this is what explains the discrepancy.

FWIW, I see a lot of east asian-black hybrid kids on campus.. marrying an east asian should address david alexander's concerns.

"Of course, there's an even simpler explanation, namely strong epistasis: in the real world, genes do not act additively, they interact with each other. So very smart people are the result of a "just right" combination of interacting genes, and any modification is likely to disturb these interactions, leading to poorer performance regardless of how smart your spouse is (because her own smartness may result from a different set of genes interacting in a different way, and mixing the two sets will not result in the benefits of either). "

That's a good point. The proteome is much large than the genome and there are plenty of scenarios where protein interactions could impact intelligence in non-additive ways. These might involve differences in protein geometry or in the availability of "building block" proteins.

The effect will still appear largely additive if intelligence is expressed as the sum of a great many genes, each of which has a very small impact on an individual's sum intelligence.

"The first time I took it, I received a 133. For the hell of it, I took it twice more and received 131 and 134. Only on the fourth try did I get the discussed-above regression dip into the low 120s."

I haven't seen this IQ test, but I expect that it is less extensive than a standard IQ test. Lengthier tests should provide greater stability. Like tests of genetic ancestry, IQ tests are more revealing in the aggregate than in individuals.

There is an interesting line of inquiry here: intelligence in adulthood is very stable for most people, but we may find that there does exist a minority that exhibits unusual instability in intelligence. We may also find that some people do witness gains in g through mental exercise, at least for short periods of time. There are a lot of questions that could be asked about these kind of higher order intelligence functions.

It's the fact that an IQ test remains the single best way of evaluating a person's intelligence.

“Happiness in intelligent people is the rarest thing I know.”

"I haven't seen this IQ test, but I expect that it is less extensive than a standard IQ test."

It's a timed, broad, simplified version with 6 subtests:

Verbal Analogies, Mathematics, General Knowledge

Spatial Visualization, Visual Sequencing, Visual Memory.

This post is wrong.

A group of white gentiles with an average actual IQ of 124 would not have children with average IQs of 124 even if they exclusively bred with each other.

The problem is that in order to have your offspring regress to the mean of a group you're a part of, the group must be sufficiently inbred (like a race or certain types of clan) and comprehensive.

A group of white gentiles selected for the quality of having high IQs doesn't cut it.

Thus the children of the interbreeding whites with 124 IQs would actually have children with an average IQ of 112. In other words, they will regress halfway to the white mean.

Now if they were a group of interbreeding African-Americans with IQs of 124, they'd have children with an average IQ of 104.5. In other word, they’ll regress halfway to the black mean of 85.

And a group of interbreeding Ashkenazi Jews with IQs of 124 would have children with an average IQ of 119.5 (assuming an average IQ of 115 for Ashkenazi’s as a comprehensive group).

Michael Ventura,

Your race hangup is preventing you from appreciating what is actually being said here. I think you need to reread the post, and think about what a "group" actually means. Once the regression happens, thus removing the possible "extra" environmental boost that helped get the initial high IQ's, the regressed population becomes a NEW "group", provided that they only breed with each other. With your interpretation, there is no room for new "groups" to ever evolve! We'd all still be Africans, in that case. There is nothing magical about any human race that prevents them from having high IQ genes, once those genes become present in the group through outside interbreeding or mutation. They just have to be selected for in order to significantly change the group characteristics. The hypothetical scenario outlined here illustrates such selection.

My only quibble with this post is that it may take more than one generation to completely regress to the "new group" mean, but the bulk of the regression would occur 1st generation so the overall concept expressed here is right.

Sigh,

You don't seem to understand that a group of white gentiles with 124 IQs would have, on average, parents and grandparents that had lower IQs than them.

This makes it very likely that they managed to get such high IQs partly through fortuitous epistatic gene interactions; and due to meiosis these gene interactions will have their positive effects diluted in their offspring, thus causing regression to the mean.

"With your interpretation, there is no room for new 'groups' to ever evolve! We'd all still be Africans, in that case."

Well, regression to the group mean doesn't always happen, it's just that it happens in a strong majority of cases.

Thinking about it though, I think in my post I overstating the need for a group to be inbred for it to be a meaningful group from the standpoint of regression to the mean.

A small outbreeding family group can have regression to the familial mean as well.

Also it could be that the category "white" is too broad for determining what mean the offspring of whites will regress to. There certainly could be meaningful IQ related genetic substructure in the white population caused by a long period of assertive mating, or selection effects being stronger on one group than on others.

For instance, maybe there's a town where the white men worked as engineers, and the best engineers got the most opportunities to mate.

At the same time there's a town where the white men were coal miners, and the man best at splitting rocks (a task not requiring an overly high IQ) got the most opportunities to mate.

Over a long period of time such selection pressure could produce two groups, both white, with very different IQs their offspring would be regressing towards.

It would take a whole lot more than one generation to do it, though, which is what Half Sigma doesn't seem to get.

The comments to this entry are closed.