How do you guys deal...

[Jon Miller]

A large part of this thread is me disagreeing with those who think that IQ tests can be reasonably used to compare intelligence for the top ~1% (Alby's post was not worth replying too).

An article in support of my position (which I was surprised was not obvious to people on this site):

http://www.assessmentpsychology.com/genius2.htm

"Where IQ tests are less useful is in making meaningful distinctions between different IQ levels at the extremes of ability, both above and below the mean, but especially the former. This is due in part to the decreasing discrimination between subtest scale scores at the extremes of performance. For example, on the Vocabulary subtest of the Wechsler Intelligence Scale for Children-III, (this subtest correlates more strongly with overall intellectual ability than the other 12 subtests), the raw score difference between a scale score of 10 and 13, which represents a one standard deviation difference, is 7-10 points for a 16-year old; while the difference between a scale score of 16 and 19, which is also a one standard deviation difference, but two to three standard deviations above the mean, is only three points. Correct responses on the Vocabulary subtest are scored as either one or two points, depending on the quality of the response. Using Wechsler's system of classification, the difference between Average and High Average intelligence on this subtest may reflect a difference of as much as 5-10 correct responses; while the difference between Superior and Very Superior intelligence may reflect a difference of only two correct responses. The latter is hardly a substantial difference, and at the higher levels of IQ, not a particularly meaningful one."

He doesn't even go into the serious statistical and classification problems involved.

JM

[MikeH]

I never disagreed with that (I said I'd seen no evidence either way, so thanks for the link). I was arguing that you hadn't proposed a better alternative.

[Elok]

Like you said, this entire thread was fundamentally started as a troll and at least in part so you could eventually point out this mistake to Alby.

What mistake? Alby said something about the 99th percentile of IQ for the whole world. He made no claims about the statistical accuracy, reliability, worth or whatever of such a hypothetical and obviously unfeasible test. If we're going to "unbundle" words to have extra meanings that aren't even hinted at, it'd be just as easy and far more reasonable to read "intelligence" for Alby's "IQ" and make the whole question go away.

[MikeH]

Exactly. Which is why KH's blanket denial of that point was wrong. It's irrelevant whether or not he was anticipating later errors. The responsibility in 'efficient communication' is on the communicator.

I think your XKCD sums it up nicely.

[Elok]

Ah. Sorry, I thought you were saying Al had actually made a real mistake. Don't mind me.

[MikeH]

He did post some data about the US IQ percentile stats later on.

[Elok]

I wasn't even looking at those, TBH; I was focusing on his and KH's original claims, which are clear-cut. I can't be arsed, as usual, to get to the bottom of the whole thread. Let's hear it for laziness!

[Jon Miller]

I never disagreed with that (I said I'd seen no evidence either way, so thanks for the link). I was arguing that you hadn't proposed a better alternative.

I have.

In experimental science you start with your clear signal.

Especially with the methods to measure intelligence are:
1. tests which are not accurate beyond superior (which should be obvious) and only measures about 1/2 of intelligence
2. personal measurement (always somewhat subjective) which is only accurate measurement of those who are less intelligent than you (and with similar types of intelligence)
3. identification based on achievements

(3) is the only method to get a sample of extreme intelligence without being completely swamped with those of only superior and very superior intelligence (which is a large, whether it is more than 70 million or only more than 7 million).

Identification of a small, pure sample is also the method that experimental science always uses.

If it were valuable to identify those with extreme intelligence (And even give a rough ranking), the method would be to take a group that had been identified by achievement and who also were trained in psychology/etc and have them rank/identify the rest.

By the way, based on the (historic) rankings of genius as determined by those who study intelligence, the Flynn effect can not cause a shift of the distribution by a sigma (at the greater than superior end).

JM

[MikeH]

You don't seem to understand my concern.

What you haven't demonstrated yet is that achievement is a valid measure of extreme intelligence.

Is it possible that any of the 70 million, or 7 million could achieve a 'genius' work given the right conditions of time, place, education and exposure to random external events?

Is it possible that a large percentage of people with extreme intelligence never achieve a 'genius' work, despite being capable of it?

Your problem with current testing is that it is extremely unreliable at the extremes, I don't dispute that, I'm saying your method is also extremely unreliable at the extremes, and of no use outside the extremes.

If it were valuable to identify those with extreme intelligence (And even give a rough ranking), the method would be to take a group that had been identified by achievement and who also were trained in psychology/etc and have them rank/identify the rest.

This is fundamentally the same methodology that was used to generate the first IQ tests, the next step was to produce tests that produced results that matched the observed rankings. You are just increasing the level of observers to only those of high intelligence and stopping short of developing it into a repeatable test.

(using IQ numbers in the following paragraph just to make the point)

It is conceivable that you could devise a test that someone of IQ ~ 160 would barely be able to answer any of the questions, someone of IQ 170 could achieve a pass and someone of IQ 180 would be able to ace.

My point is that because current tests give you poor resolution at high levels doesn't mean you can't devise a test that would give you very high resolution at high levels, even if it gave you no resolution at any level below that.

[Elok]

I haven't been following the JM argument, but has anybody mentioned that not everyone with ability will necessarily have ambition to match? KH is apparently quite brilliant, and he's apparently got the drive to make a lot of money doing very challenging work. But, while I allow that I'm not at his level, that life wouldn't suit me. There may well be some utter geniuses out there who live out their lives doing rather ordinary jobs while they get their mental stimulation from hobbies, friends, or trying to solve ancient mathematical proofs when they're not teaching at the local college. How do you define "achievement" such that it covers everybody's attempts at self-fulfillment?

[MikeH]

Yes, most of us.

[Elok]

All right, carry on then.

[MikeH]

AFAIK KH is making a bit of money and doing a bit of finance, he's not making billions. He is certainly underachieving for one of the top 20k brains ever to live.

[Elok]

AFAIK KH is making a bit of money and doing a bit of finance, he's not making billions. He is certainly underachieving for one of the top 20k brains ever to live.

I suppose, but the whole concept of achievement is nebulous and not correlated with intelligence very closely. Tesla comes to mind. Did he achieve a lot, or almost nothing?

[MOBIUS]

AFAIK KH is making a bit of money and doing a bit of finance, he's not making billions. He is certainly underachieving for one of the top 20k brains ever to live.

I think you may have hit the nail on the head there.

Despite being relatively successful, he realises that since he has chosen materialistic worth as he chosen benchmark, he is still massively underachieving by his standards...

Sucks to be him.