IQ and Magnus Carlsen, Leo Messi and the Decathlon

First thesis: IQ is more akin to a composite measure of performance such as the decathlon than it is to a single characteristic such as height or speed.

Second thesis: When looking at extraordinary performance in any specific field, IQ will usually be highly correlated with success, but it will not fully explain or predict top-end performance, because extraordinary performance in a specific field is a result of extraordinary talent in a sub-category of intelligence (or even a sub-category of a sub-category), rather than truly top-end achievement in the composite metric.

I don't understand what you are trying to argue here; many of your claims seem to be dubious or wrong, and in particular, you seem to be covertly redefining and using in non-statistical ways words like 'predict' which have accepted statistical meanings, especially around here. IQ is not going to correlate r=1 with any measure of success because no measure of success is identical with intelligence. The existence of successful scientists or chess players whose rank-percentile in IQ is less than their rank-percentile in scientific accomplishment or chess playing does not shed any light on what the neurological basis of IQ, as this is a general phenomenon of regression to the mean and the extreme order statistics of two variables which are correlated r<1: http://lesswrong.com/lw/km6/why_the_tails_come_apart/ I'm not sure how this is relevant to any discussions of what neurological features IQ is caused by or what the causal nature of IQ is - but looking at the extremes doesn't tell us anything I can think of. Consider two traits X and Y which are correlated r<1, 0.99 say; in world A, you measure 100 people, and in world B (with the same X and Y and all causal aspects the same), you measure 1 billion people, and you note that the person ranked #1 for X is not ranked #1 for y, and in world B, the #1 X guy is in fact hundreds of thousands or millions of spots down the list of Yers; what have you learned about world B which is different from world A? Nothing, as far as I can see.

Or take the example of height. You know it's heritable, polygenic, due to greater size in all parts of the body, affected by primarily generalist genes (as well as variants with body-part-limited effects), and the various environmental & genetic effects generally have overall effects due to affecting metabolism & nutritional resources etc, and so it is a very real factor; but the best pro basketball player is not the tallest pro basketball player - does this show that height is somehow not real or not a general body size factor? If points scored in the NBA correlated r=0.90 with height, would this show that height is actually caused by a bunch of independent genes, and r=0.99 would show it's really a latent factor? How is there any connection here between points scored in the NBA championship and what height 'really' is, surely the correlation is going to be entirely driven by, you know, how useful height is playing basketball compared to other things like musculature or aerobic endurance?

Whether you measure height in centimeters or feet, however, the ranking of the people you measure will be identical no matter how you measure it.

No, it won't; it will depend on how you measure it and when and where and by whom, just like a LSAT will be highly correlated with but not identical with an IQ test. There is measurement error in measurements of height, reflecting issues of the person taking the measurement, how the measured person stands, what time of day it is, what shoes they are wearing, what phase of life they are in, whether they just came back from the International Space Station, whether you are using self-reports or doctor measurements, and so on. Many objective measurements, from blood panels to brain volume, have measurement error in them; every morning I weigh myself 3 times on my bathroom scale and record the data, and you know what, maybe once a month all 3 weighings will agree exactly on my current weight/BMI/body-fat-percentage/resting-metabolism/visceral-fat, and the rest of the time they'll disagree, sometimes by half a kilogram. My weight and body fat percentage hadn't changed in the 30 seconds it takes, but nevertheless. Everything has measurement error in it. Some things are easier to measure, is all.

We’ve been using IQ as a measure of intelligence for over 100 years and it has never predicted an Einstein, a Musk, or a Carlsen.[4] Who is the best counter-example to this argument? Terence Tao? Without obvious exception, those who have been recognized for early-age IQ are still better known for their achievements as prodigies than their achievements as adults.

Of course IQ scores predicted them, or would have. Even if you take the bogus examples people like to cite like Feynman, the IQ scores are still well >100. Every IQ study shows predictive power. The Terman study famously just missed some future Nobelists because Terman could only follow a small number of kids. Note that early-age IQ is not the same as adult IQ (any more than childhood height is identical to adult height!), so it's not even the same thing in the first place (which is probably one reason that, for example, Hunter Elementary School has disappointed its administration - after all, after regression to the mean of their stable adult IQ, the students are not that special). Or, forget Roe or TIP or SMPY, just think about university admissions standards! No one gets into MIT, Stanford, Harvard, or Yale (much less their grad programs) with a 1000 SAT and a 3.0 GPA. Surely you admit at least one important person who wasn't Terence Tao graduated from a college like Oxford or CMU and had high SAT scores, which you note correlates highly with IQ?

While IQ is almost certainly highly correlated with high-end performance, IQ fails a metric to explain or, more importantly, to predict top-end individual performance

Not controversial at all; seems like a straightforward application of the law of diminishing returns. If performance in a certain area depends partially on IQ and partially on some other X, then IQ will correlate with performance, but mere high IQ will not be enough to achieve best performance, because when you already have the high IQ, the amount of X becomes the critical factor.

IQ has to be measured by multiple sub-tests.

Wrong. For example, Raven's Progressive Matrices only have one category.

Phlogiston and Albert Einstein’s IQ

I agree that trying to indirectly calculate IQ of famous dead people is complete bullshit. Talking about values of IQ above 150 doesn't make sense statistically. Anything using Eistein as an example is automatically suspicious. And "he must have had high IQ, because he was awesome, and we believe high IQ is what makes people awesome" is circular logic.

But we should distinguish between the following statements...
1) People love to make stupid claims about IQ.
2) The concept of IQ is stupid.
...and a good start would be to look at what scientists are saying about IQ, as opposed to what clickbait journals do.

You might be interested in this paper on profoundly gifted children which says in part "different patterns of profound intellectual talent uncovered in their youth were predictive of qualitatively different educational, occupational, and creative outcomes"

The tallest player to ever play in the NBA was Gheorghe Mureșan, who was 7'7". He was not very good. Manute Bol was almost as tall and he was good but not great. By contrast, the best basketball player of all time was 6'6" [citation needed]. In fact, perhaps an athletic quotient would be better for predicting top-end performance than height, since Jordon, Lebron and Kareem are all way more athletic than Muresan and Bol.

I will attempt to explain the strongest counterargument that I'm aware of regarding your first thesis. When you take a bunch of tests of mental ability and you create a correlation matrix, you obtain a positive manifold, where all the correlations are positive. When you perform a factor analysis of these subtests, you obtain a first factor that is very large, and secondary through n-iary factors that are small and vary depending on the number of factors you use. This is suggestive that there is some sort of single causal force that is responsible for the majority of test performance variation. If you performed a factor analysis of a bunch of plausible measures of athleticism, I think you would find that, for example, bench press and height do not participate in a positive manifold and you would likely find multiple relevant, stable factors rather than 1 athletic quotient that accounts for >50% of the variation. Cardio ability and muscular strength are at odds, so that would be at least two plausible stable factors. This argument is on Wikipedia here#Factor_structure_of_cognitive_abilities). Personally, in light of the dramatic differences there are between the different parts of an IQ test battery, I find this fact surprising and underappreciated. Most people do not realize this, and the folk wisdom is that there are very clear different types of intelligence.

The second point I would make regarding your first thesis is that there are plenty of researchers who don't like g, and they have spent decades trying to come up with alternative breakdowns of intelligence into different categorizations that don't include a single factor. Those efforts were mostly fruitless, because every time they were tested, it turned out that all the tests individually correlated with g still. Many plausible combinations of "intelligences" received this treatment. Currently popular models do have subtypes of intelligence, but they are all viewed sharing g as an important top-level factor (e.g. CHC theory) rather than g simply being a happenstance correlation of multiple factors. In this case absence of evidence is evidence of absence (in light of the effort that has gone into trying to uncover such evidence).

To be honest, I very much doubt that actual IQ researchers would disagree with your second thesis. My argument would be that for most fields there is enough randomness that you would not expect the most intelligent person to also be the most lauded. Even Einstein had to have the luck to have the insights he did, and there were undoubtedly many people who were just as smart but had different circumstances that led to them not having those insights. Additionally, there is a thing called Spearman's law of diminishing returns, which is the theory that the higher your g is, the less correlated your subtype intelligences are with your g factor. That is, for people who have very high IQs, there is a ton more variation between your different aspects of intelligence than there is for people with very low IQs. This has been measured and is apparently true, and would seem to support your thesis. It is true that these two observations (the factor decomposition and Spearman's law) seem to be in tension, but hopefully one day someone will come through with an explanation for intelligence that neatly explains both of these things and lots more besides.

Unrelated to your two theses, I think the fact that IQ correlates with SO MANY things makes it interesting alone. IQ correlates with school performance, job performance, criminality, health, longevity, pure reaction speed, brain size, income, and almost everything else (it seems like) that people bother to try correlating it with. If IQ hadn't originally come from psychometric tests, people would probably simply call it your "favor with the gods factor" or something.

There are enough correlations that any time I read a social sciences paper with statistics on outcomes between people with different characteristics, I always wish they would have controlled for IQ (but they never do). This may seem silly, but I think there is definitely an argument that can be made that IQ is "prior to" most of the things people study. We already know that IQ can't be meaningfully changed. It's pretty much set by the time you are an adult, and we know of nothing besides iodine deficiency that has a meaningful impact on it in the context of a baseline person in modern society.

scooped up Leo Messi from the hinterlands of Argentina at age 12 and he went and became Leo Messi. Lebron James was on the cover of Sports Illustrated when he was in high school.

Identifying Messi at age 12 is a lot more impressive than identifying James a year before he goes pro, when the only thing holding him back from going pro immediately is the NBA's diploma requirement.

Some people say that Newton was "scooped up" at 12, although I'm not sure how accurate that is.

Even if it were true about Newton, it wouldn't be useful to compare him to Messi without knowing the base rate of how many kids are scooped up, in both fields. Here is something where we can look up the base rate: the IMO. Something like half of Fields Medalists won a gold medal at the IMO in high school. About 50 gold medals are handed out each year, fewer in the past. You should be really impressed by that rate of prediction. Of course, it's comparing math to math, not g.

I appreciate you writing this post.

The question of whether IQ is more like height, where the concept cuts reality by its joints, or more like athleticism, where it's just an aggregated characteristic, is indeed very interesting. It seems that people have strong opinions about the matter, but I'm not exactly sure how to disentangle these two hypotheses. What kind of methodology can we use in principle to answer this question?

When we think of great athletes, we think of the top performers in one individual discipline, rather than the composite.

When people think of the best athlete in the world, they think of Leo Messi or Lebron James, not Ashton Eaton.

I'm not sure it's very relevant to your point, but I think it's pretty weird to call soccer or basketball "one individual discipline." I thought you were going to point to something obviously narrower than the decathlon, like one of its components. Whereas, popular spectator sports seem to me to involve a lot more skills than the decathlon. At the very least they involve teamwork and a lot more perception, but I think that they involve a greater variety of physical skills. On the other hand, OJ Simpson was a world-class sprinter, not a decathlete.

If we dug deep into the sub-components of Carlsen’s IQ (or perhaps the sub-components of the sub-components), we’d probably find some sub-component where he measured off the charts.

It's still probably a combination of sub-components + memory (often underrated in IQ discussions) + physical fitness (relevant in long tournaments), not any single one. Quotes on his childhood:

At two years, he could solve 500-piece jigsaw puzzles; at four, he enjoyed assembling Lego sets with instructions intended for children aged 10–14.

Simen Agdestein emphasises Carlsen's exceptional memory, stating that he was able to recall the locations, populations, flags and capitals of all the countries in the world by age five. Later, he memorised the locations, populations, coats-of-arms and administrative centres of "virtually all" 356 Norwegian municipalities. 

and on his distinctive style:

Kasparov expressed similar sentiments: "[Carlsen] has the ability to correctly evaluate any position, which only Karpov could boast of before him.

In a 2012 interview, Vladimir Kramnik stated that Carlsen's "excellent physical shape" was a contributing factor to his success against other top players as it prevents "psychological lapses", which enables him to maintain a high standard of play over long games and at the end of tournaments, when the energy levels of others have dropped.

Levon Aronian said in 2015: "Magnus' main secret is his composure and the absence of any soul-searching after mistakes during a game." (i.e. chess tilt)

Tyler Cowen gave a point of view on Carlsen's playing style: "Carlsen is demonstrating one of his most feared qualities, namely his 'nettlesomeness,'... Using computer analysis, you can measure which players do the most to cause their opponents to make mistakes. Carlsen has the highest nettlesomeness score by this metric, because his creative moves pressure the other player and open up a lot of room for mistakes. In contrast, a player such as Kramnik plays a high percentage of very accurate moves, and of course he is very strong, but those moves are in some way calmer and they are less likely to induce mistakes in response." 

The favored approach is,unfortunately, rather tautoligical: IQ measures 'intelligence' so imperfections with real world correlations can be explained by the notion that real world tasks are diluted with factors that are irrelevant to what we want to call 'intelligence'. This is opposed to the highly viable, but painful, explanation that IQ tests just dont pick up on other factors that are critical to generative thinking. One of the most hoary ideas is that because of some positive manifold, predicting a common underlying feature, it must reflect 'intelligence'. Absurd and akin to saying that because height is positively correlated to performance in practically all sports, that height = athleticism (interestingly, height is about 80 percent genetic, like IQ). There is absolutely no reason to view intelligence rooted in a single factor, especially, in light of late developments in cognitive psychology with intuition/heuristic processes which mediate analytical thinking.

A suggestion: IQ might be better understood less like Height and more like "how high can you jump". Where for example, people can train to jump higher, people can practice IQ tests. People can wear fancy shoes and jump a lot, causing different jumping results. People can fail to jump high unless they are wearing certain shoes, and you can claim that certain aids are cheating. For example - he's only a great musician and not really good at anything else, therefore not really a high IQ.

While this isn't a great analogy, it's more fitting than "how tall".