I find this study extremely interesting. Before anyone starts to spout anti-IQ rhetoric - let me say that I realize that the predictive effects of IQ (which are extensively documented in research journals) are only as good as they are simply because most people aren't that psychologically different from each other (owing both to genetic factors and socialization), which will natural reduce the variance in performance due to other factors (and increase the associated variance in performance attributed to IQ). In fact, one of the major findings in the IQ literature is that in the general population, the primary sources of intelligence are highly correlated with each other. Verbal intelligence is correlated with mathematical intelligence, and both of those are are even highly correlated with reaction times. But among people who are autistic, this finding may be less accurate than it is among neurotypicals. Some of us Aspies are exceptionally talented at certain things while simultaneously being incredibly incompetent at other things.

Of course, findings only apply to the population at large and may not apply to specific subsets of people. Most research shows that people who sleep more get higher grades - but this is definitely not true for certain subgroups of people - there are plenty of intelligent people at MIT/Caltech who sleep far less than the average student in a state school. The same logic applies to skipping classes and lower grades (Caltech's classes have notoriously high absence rates since the students there are independent studiers). Similarly, IQ tests (and other assessments normalized to the general population) may not necessarily predict performance among certain subgroups of people, especially those who are non-neurotypical. This logic could apply to GPAs and SAT scores too (I know a mathematical genius with asperger's at UChicago, who only got 600s on his SATs for example). And I also suspect that it may apply to subgroups of people with Attention Deficit Disorder. There's s a good chance that other factors may interfere with IQ, which may increase variance in performance due to other environmental factors (and decrease variance in performance due to autism).And simply increasing the variation in environment will naturally reduce the variation in IQ due to comparatively immalleable factors (early childhood influence is just as immalleable as genetics once you start measuring IQs of older children and teenagers, for example).

I'm still pretty sure that IQ tests will measure *something* in Aspies (as the article says, it's that Aspies tend to have uneven performance that tends to make them stick out). Scores on the individual scales will still say something about the Aspie's range of strengths and weaknesses.

Anyways, I think this topic would be very interesting to this group, since it has a high population of non-neurotypical thinkers. In particular, I'd like to encourage the discussion of other metrics that may predict performance among neurotypicals, but not necessarily among non-neurotypicals. In particular, it has a lot of significant relation to signalling theory. If someone's a high school dropout and didn't continue onto college, for example, chances are that he probably isn't the type of person who would get along with people on lesswrong. But if you learn that he has Asperger's and other traits of an unusual background, then you might figure that the "high school dropout" signal has significantly less validity in that case, and he might have a much higher chance of getting along with people on lesswrong.

New Comment
9 comments, sorted by Click to highlight new comments since:

This is almost a tautological result. Your IQ score is a single number. You can meaningfully reduce IQ to a single number because cognitive skills tend to be correlated within an individual meaning that people who tend to be "intelligent" at some academic tasks tend to be "intelligent" at all tasks. A high functioning autistic, however, must necessarily be much worse at some cognitive tasks than others and so necessarily has a high variance in cognitive abilities.

Okay that's true. But the cognitive tasks Aspies do worse on are usually social tasks, which IQ tests usually don't capture. So that's why I still think of this result as interesting/novel.

When I took my IQ test, at least two oddities stood out to me which may have affected its (already nonexistent given that I hit the ceiling) accuracy.

1) There was a segment of the test in which I was instructed to try to memorize a paragraph of text read aloud to me so I could repeat it back "exactly as [the tester] said it". For the first third or so of this test, I thought "exactly" meant that I would be marked not only on exact wording, but also on things like inflection [edit: apparently this is a technical term; I didn't mean the technical definition, I meant the reference class which includes things like tone, pitch, emphasis, etc.] and speed, and tried very hard to imitate those aspects of her reading. Then I asked for clarification and she told me that it was only the words that mattered. A neurotypical might not have misunderstood the instructions as I did; someone more autistic than me who made the same error might not have thought to check for such a misunderstanding (and would probably also have taken much more processing power to try to remember inflection etc. while under the misapprehension).

2) There was a section where I was shown pictures of sheets of paper with pieces cut out of it (a la paper snowflakes except rectangular), both folded and unfolded (they were to be matched to each other; this was multiple-choice but I don't remember if the choices were the unfolded or folded versions). I asked for a pencil and paper. The proctor remarked that there wasn't an official policy about whether pencil and paper were allowed for the section, and made a judgment call and gave them to me. I'm not sure how, if at all, my autism interacted with this irregularity.

I remember doing the folded-sheets part you mentioned as part of a test required to enter the military*. It was a test on paper, and I completed the language/reading/writing section so quickly that I could afford to take my time, but I don't remember if I drew diagrams or pictures to help myself. I do remember that apparently I did very well and have the spatial skills to be a fighter pilot if I want to (I laughed).

*I was rejected later on for health reasons. Which I'm still sore about, but that's unrelated.

things like inflection and speed


right, intonation.

And I also suspect that it may apply to subgroups of people with Attention Deficit Disorder

It seems the case, the study that comes to mind is Executive Function Impairments in High IQ Adults With ADHD by Brown, Reichel & Quinlan. People with ADHD were much more likely to have Working Memory Index and Processing Speed Index (WMI & PSI) scores two standard deviations below their Verbal Comprehension Index or Perceptual Organisational Index (VCI & POI). As a side note, VCI is considered the best indicator of premorbid IQ.

I've actually been meaning to ask this for a while, can you (assuming this study is accurate) use Bayes theorem to argue that where VCI/POI>WMI/PSI is two standard deviations out, ADHD is a possibility worth considering? (and the numerical value of that possibility if I'm being ambitious?)

DOI: 10.1177/1087054708326113 for those interested.

My Bayes' is not very strong, so forgive me. This is about as naive as it gets.

Prior for adult ADHD is about 4%.

For working memory:

2SD difference given ADHD is 35%, and for the rest of the population it's 2.4%. I'll use LWM to represent a significantly lower working memory.

This suggests an overall population of 3.7% having a 2SD difference, since 4% x 35% + 96% x 2.4% = 3.7%

So P(ADHD | LWM) = 35% x 4% / 3.7% = 37.8%

For processing speed:

2SD difference given ADHD is 44.9%, and for the rest of the population it's 8.7%. I'll use LPS to represent a significantly lower processing speed

This suggests and overall population of 10.1% having a 2SD difference, since 44.9% x 4% + 8.7% x 96% = 10.1%

so P(ADHD | LPS) = 44.9% x 4% / 10.1% = 17.8%

Each of those increase the odds of ADHD significantly above the population baseline, and if you're experiencing ADHD symptoms in conjunction with results like that, it's probably worth seeing a psychiatrist.

Thanks for taking the time to math that out :) I have seen a few psychiatrists in the past 5 years and unfortunately medication wasn't an option. I do think I'm performing better with age however!