Tuesday, May 23, 2006

g and Creativity   posted by agnostic @ 5/23/2006 02:29:00 PM

I've been meaning to write this up for awhile now, so here goes. It's more of a pointer to what books & articles to read, so it may not cover every question you have about g and creativity. [1]

In Carroll's (1993) encyclopedia of cognitive abilities, Ch. 10 surveys psychometric data on the ability of "idea production" -- basically, creative thinking. It's a 2nd-stratum factor (called Gr for "general retrieval" ability), dominated by the 3rd-stratum general factor (g), thus a sister of the more well known visuospatial and verbal factors. Two of the 1st-stratum daughter factors that compose Gr are called FA (Associational Fluency) and FO (Originality/Creativity), which respectively measure how well you can draw associations between items in long-term memory (e.g., how interconnected your idea-web is), and how well you can devise novel solutions to problems (e.g., take ordinary objects & fashion new uses for them). Real-life tasks will also tap into Gc, or "crystallized" intelligence (the knowledge acquired as a result of applying g in academic / intellectual domains, e.g., vocabulary items, scientific ideas, etc.). So, given two individuals whose Gr is the same, the one with a larger store of ideas will have a richer web of associations.

Simonton (1999 Ch. 2 & 3; 2004 Ch. 3 & 5) argues convincingly that at root the creative process is like Darwinian selection: ideas are combined in blind variation, largely unconsciously; most of these combinations are useless, but a few show promise. These are then selected and elaborated on consciously, putting flesh on the inspiration. Simonton says that aside from the usual suspects of high intelligence (g) and a healthy store of ideas & facts to draw on (unspecified, but must mean Gc), another part of creativity is having a "flat" associative hierarchy -- connections between ideas criss-cross a lot, rather than respect a "steep" hierarchy like Linnaean classification schemes. This trait is essentially Gr, though he does not recognize Carroll's terminology nor that it is dominated by g.

This idea is originally due to Mednick (1962), whose Remote Associates Test measures how fluent one is in making remote associations. Ex: given the words "rat," "cottage", and "blue," they are all remotely related to "cheese" (the answer). French, Price, & Ekstrom's (1963) test has a subsection for idea production, one of whose tasks requires the subjects to fill in blanks to complete similes: e.g., "She was as pale as..." This test is a measure of diverse cognitive abilities, not just creativity, reflecting the consensus that there is no cognitive ability that is not influenced by g, and that all 2nd-stratum abilities correlate with each other. See Gottfredson (2003), where she tears Robert Sternberg a new orifice for lazily trying to argue that "practical intelligence" is different from and not correlated with g, one prong of his Triarchic theory of intelligence.

Now, a skeptic like Michael from 2blowhards could say: OK, sure, if you restrict "creative products" to those endeavors that demand braininess, by definition creative people will be brainy, whereas if you took a more liberal & inclusive view in which "creative products" included more popular forms where the cut-offs for g are more forgiving, this effect would shrink or vanish. So, is it possible to avoid this tautology? Yes: define "creative products" roughly as "forms which inspire a sense of awe or marvel in the beholder, as if such products were not thought possible to come from mere mortals." If that sounds too reverential, you can use more folksy language, but you get the idea. Call this the "Wow!" definition. It avoids any mention of cognitive ability, favoring only the ability of the work to inspire, rather than to provoke the beholder to complain, "Meh, my 3rd grade son coulda done that." It also cuts across disciplines: it captures why we find the Michelangelos & Beethovens more creative than illustrators & composers of advertising jingles, and the Newtons & Gausses more creative than "lab men" & those who report another example of a phenomenon with thousands of attested examples.

It may just so happen that humans are more wowed by feats of cognitive difficulty, but we didn't build this in -- it's just a quirk of human psychology. Martians might well be wowed by similes such as "The printer paper was as white as white printer paper." And it really is the cognitive part that we're wowed by -- no one but a boor would claim that Stephen King is a creative genius or super-intelligent based on his popularity / best-selling status. Even those whose personal tastes lead them to prefer King to Goethe are not confused about who is more creative or intelligent. They simply feel that King speaks to their tastes more than that arty-farty stuff.

So, rather than a creative / non-creative dichotomy, we have a spectrum of more or less creativity. The farther one moves toward the creative end, the higher the demands on g in general and Gr in particular; the more one moves toward the less original end, the more relaxed the demands, as one is no longer re-inventing the wheel. I should also amend the definition to include only those forms which more or less originate with the creator, rather than forms which result from the individual "following a script." Thus we exclude actors, orchestra members, and individuals who solve math homework problems by means of common algorithms (like long division), as they're fundamentally different from playwrights, composers, and mathematical discoverers / pioneers. We expect the latter to be smarter, while we don't expect the former to be smarter (which is not the same as expecting them to be dull!).

Lastly, intelligence -- whether g, Gr, or anything else -- is certainly not sufficient for high status in creative fields. Beginning with Galton, researchers of genius have noted that the distribution of "eminence" (a proxy for genius, as there are no unrecognized geniuses) is not the normal bell curve that we know and love, but rather log-normal. The less mathematically inclined can see pictures of what this looks like here; the more quantitative can read a nice pdf here on its use in the sciences, which has an excellent log-normal version of the Galton board. The key difference is that, while a normal curve is symmetric, the log-normal curve is skewed; for eminence, it is highly skewed, like the green or pink curves on the NIST link above. At first, you might think: "Well, maybe that's just the far, far-right tail of a bell curve -- the super-duper nerds." But the shape is not the same. If you look at the "slope" of the right half of a bell curve, it changes from "steep" descent to "shallow" descent pretty quickly -- by the time you got to the far-right tail, you would only see a shallow descent. With the log-normal curve, however, you see a change from steep to shallow, so it can't be the same as the far-right tail of a bell curve. ("Steep" and "shallow" refer to the absolute value of the slope (rise / run) at a point: if it's greater than 1, i.e. "rising" more than it's "running", call it steep; if it's between 0 and 1, i.e. rising less than it's running, call it shallow.)

So what's the big deal? Well, log-normal curves usually imply that there's a synergy among various components that produce the effect -- where the whole is greater than the sum of its parts. (Or: where the effects of the random variables involved are multiplicative rather than additive -- if the effects are additive, the result is also a normal bell curve.) To take a more concrete example, consider female facial beauty -- let's say there are 5 variables involved: eyes, lips, hair, prominence of bone structure, and rough facial geometry (e.g., long or short face). Assume there is a universal scale of less-to-more beautiful values for each variable (e.g., fuller vs thinner lips, larger vs smaller eyes, etc.). Also assume each variable would yield a normal bell curve. Now, in an additive model, you would take the points earned from each variable and sum them up -- so, the lips variable doesn't "see" or "care about" the eyes variable; you get X points for full lips no matter what. By the Central Limit Theorem, the sum of these 5 variables would itself be a normal bell curve -- on a scale of 1-9, most people would be a 5, and the percentages would drop off in both directions, so that there would be very few 2s or 8s. That's the opposite of reality, though: the percentage of 2s is far greater than that of 8s! This is because the variables "see" or play off of each other -- large eyes and full lips produce a more beautiful effect than if you just added their values together. Likewise, large eyes, full lips, lustruous hair, and prominent cheekbones -- 4 great qualities -- can all be brought crashing down by a long horse-face, rather than only suffer a slight reduction in beauty.

In the case of creativity, there are several ideas for what these other factors are that interact synergistically with intelligence -- e.g., the Big Five personality trait Openness, Eysenck's personality trait Psychoticism, schizotypal personality, and so on. This is another interesting area of creativity research, but it doesn't mean that intelligence is any less important as a result. Consequently, claims to the effect that "IQ doesn't matter in creative fields past a threshold of 120" are nonsense. It's hard to get large data sets for such situations since there are damn few Newtons and Beethovens alive to test. Moreover, let's say it turns out that when you do the multiplication of the variables for arts vs sciences, the intelligence variable is weighted more in the sciences than in the arts -- nevertheless, in any cognitively demanding area higher *g* always helps.

[1] For more background on g, sub-g factors, and brain correlates of g, see these two posts from the GNXP archives.


Carroll, J.B. (1993). _Human cognitive abilities: A survey of factor-analytic techniques_. Cambridge: CUP.

French, J.W., Ekstrom, R.B., & Price, L.A. (1963). _Manual and kit of reference tests for cognitive factors_. Princeton, NJ: ETS.

Gottfredson, L. S. (2003). Dissecting practical intelligence theory: Its claims and evidence. _Intelligence_, 31(4), 343-397.

Mednick, S.A. (1962). The associative basis of the creative process. _Psychological Review_, 69, 220-32.

Simonton, D. (1999). _Origins of Genius: Darwinian Perspectives on Creativity_. New York: OUP

------------- (2004). _Creativity in Science: Chance, Logic, Genius, and Zeitgeist_. Cambridge: CUP

Addendum: In the comments people are bringing up what I hope they wouldn't bring up, since it's a whole 'nother post, but I saw it coming! So a word or two about racial differences in creativity. First, unlike the copious data on the 1-SD difference in the means between Af-Ams and whites, there is no similar data on tests only of Gr, so all we have is speculation. That said, my guess is that, while not doing so well on g, Af-Ams do pretty well on Gr -- lopsidedness in the 2nd-stratum factors isn't unheard of. On average, Af-Ams and Ashkenazi Jews have greater verbal than spatial sub-scores; NE Asians are the reverse. Commenters have mentioned hip-hop music, but that's the wrong place to look -- Jazz for sure. Also, the popular phenomenon of "yo momma" jokes is basically a modified remote associates / similes test. True, you can cheat by stealing someone else's joke, but word gets around fast, and unoriginal jokes are quickly booed. You can see all this play out on the new MTV show "Yo Momma." The more remote the assocation, while still making sense, the higher the score; ditto for verbal cleverness. For example: "Yo momma sweat butter and syrup and got a job at Denny's wiping pancakes across her forehead." "Yo breath smell so bad the only dis I'm gonna give you is dis-infectant." And so on.

However, all of this is at the popular level, while what I was writing about was high culture -- so Jazz would still survive, but not the other stuff. If you want to include popular culture as well, these products don't inspire as much marvel in the beholder, so the cognitive difficulty (which is what really inspires awe) isn't as demanding. Thus, the g variable is weighted less, but remember: creativity results from multiplying together a host of factors. All I was arguing was that g was one of them, and likely heavily weighted. Some of these other factors may favor Af-Ams -- for example, people who reach high eminence in creative fields are usually more disagreeable than agreeable; at such heights, diplomacy is for suckers. Again, I'll take it for granted that the Af-Am mean is more in the disagreeable / confrontational direction than is the NE Asian mean. So that's in the Af-Ams' favor. But I also mentioned Openness to experience, and Af-Ams seem to be more conventional and less tolerant of novelty, fantasy, thrill-seeking, and so on (this difference is often the seed for black comics' jokes about how whacky white people are). On the plus side, that renders them more immune to New Age flimflam, yet it also is a penalty when we consider the factors involved in creativity.

As for NE Asians, they have higher g on average, their strength coming from superior visuospatial skills. So, look at how good they are at innovating in visual areas of pop culture (their poor turn-out as comedians, etc. would be due to lower verbal skills). I've mentioned before that they're pretty innovative when it comes to visual tasks, provided they're financially secure: graphic design, hell, any design, film, video games, and so on. But as for the Openness, Agreeableness, and schizotypal / eccentric-nutty behavior, I'd guess they tend to score in the direction penalized in the creativity multiplication. I'm sure some of this is cultural, but still, exposure to & incorporation into mainstream American culture still leaves the (correct) stereotype that NE Asians are more conformist than whites or Af-Ams. Less Open and more Agreeable individuals will appear more conformist. To the extent that there is a genetic component to these personality traits, then there will be a ceiling that they'll hit even when the cultural obstacle is removed.

These observations pertain as well to the male-female gap in creativity. This is particularly apropos given the Larry Summers fiasco. In order to show discrimination, you'd not only have to take into account the different variances in the male & female curves for g, but the potential differences in means and/or variances for all the other curves involved in the multiplication. Not only are males more likely to have an IQ of 155, but they're certainly more likely to be Disagreeable and schizotypal / eccentric (my guess: difference in means), as well as to be the more Open / daydreamer sort or exhibit nonconformity (my guess: difference in variances). If you look at the real loonies among scientists and artists -- mathematicians and composers -- they're so overwhelmingly male, the culprit can't be g only. In these fields, females are penalized not only for lower likelihood of reaching say IQ 155, but also for on average being more Agreeable & diplomatic, not Open enough (too practical), and not nutty or eccentric enough -- and remember, these penalties are multiplied or compounded, not merely added together.

It's hard to avoid the obvious: at the high culture level, males of Eurasian origin (as in, west of the Himalayas / Siberia) dominate more than would be expected just based on g, though that certainly plays a role too. My hunch is that, for whatever obscure reasons, their average values for each of the variables involved in the multiplication result in greater creativity than other demographic groups. Cultural factors can dampen or amplify this pattern to an undetermined extent, but those are the differences we're starting with. When we move down to the popular, less awe-inspiring level of culture, where the demands on g are less intimidating, then other groups will make greater headway -- again, following the pattern of their average cognitive profile (more verbal for Af-Ams and Ashkenazis, more visual for NE Asians).