Correlation between IQ and Autism test results

Post Reply New Topic

So as explored in http://www.wrongplanet.net/postt158261.html, the correlation between verbal IQ, performance IQ and Asperger test results from the sample derived from Wrong planet was -.18! That implies, a more intelligent person tends to score less 'autistically', on average, and implies that the more intelligent person tends be less autistic. I don't know if I believe that as I think there could be other affects, but it is an interesting relationship to explore.

So, what would be the ideal study and assessment be like in order to get a good sense of the correlation?

I'm thinking of a short test so that it doesn't discriminate against those with shorter attention spans, and I'm thinking 10 IQ questions across the cognitive difficulty range and 10 AQ test questions would make for a good test to administer to the general public to investigate the relationship. As for who to approach to administer the test, I'm thinking of advertising it on a popular social website -- Twitter? Facebook? Something like that? And then provide an incentive for taking the test such as the "Take this test for a chance to win an iPod and help out the Autism community!", and then everybody gets a 1 in a million chance of winning an iPod, but you restrict the amount of people allowed to take the test to 100,000 so there's only a 1 in 10 chance of giving away an iPod.

It's not statistically significant...



Check out those R-squared values. It's not a very good fit at all. You already have n=55, so the effect size is going to be very small no matter whether you can find another thousand people to take the test or not.

Last edited by Callista on 22 Apr 2011, 1:08 pm, edited 1 time in total.

A suggestion - in the AQ question, choose two from each sub-scale (social, communication, attention switching, attention to detail and imagination). Could be interesting to see how the different components of AQ relate to IQ.

Your critique is well taken, although I noticed the invalid data was retained in making the graph. Here's the newest version of the spreadsheet where each test and its scores has its own tab, and the missing and invalid entries for each row has been removed.

http://www.mediafire.com/file/5b70sbdib ... icFull.ods

I'm pretty sure it's still not statistically significant, though. I wonder if a significantly greater sample size would increase the statistical significance? I know you suggested otherwise, but wikipedia says "For example, a sample Pearson correlation coefficient of 0.1 is strongly statistically significant if the sample size is 1000" which seems to be the 'ball park' this statistical relationship could be in.

Last edited by swbluto on 22 Apr 2011, 1:28 pm, edited 2 times in total.

Interesting -- do you know where information about each question and their corresponding subscale might be on the net? I'm guessing I'll have to check the original AQ test's research paper.

http://www.autismresearchcentre.com/doc ... tal_AQ.pdf

You've got a lot of different measurements there... did you do a post-test to correct for that?

Sorry--I just realized my fascination with statistics is taking over. I should stop being so picky and let you do your research. Don't mind me... I get this way sometimes when there are large amounts of numbers involved.

Don't worry, you're absolutely no problem. If anything, your critiques have been informative and at least educational as I could use a refresher. I'm not really a "professional researcher" or anything, I'm just a computer-science-major oriented amateur fascinated by mathematics and psychometric tests, and I want to help out others in screening their own probability / severity of possessing autistic traits and I want to explore the connection between various seemingly prominent mental characteristics, especially where it seems intriguing.

Having said that, I actually don't know what a "post test" is. I assume that's referring to the rho variable that's related to the level of statistical significance? I see it in the online pearson correlation calculator I used, but I didn't record the values and I'd have to rerun it to get the values. At this point, I wasn't worried about it, as I was simply finding the highest correlated tests to determine which tests should have more data gathered from them to determine their suitability as autistic-screening tools, and once I have enough data from those tests, to conduct the multiple-regression analysis to create the "final" diagnosis_score formula which incorporates many different tests.