Asperger test data and analysis

Post Reply New Topic

In this thread, I'm going to take the data that I've gathered from http://www.wrongplanet.net/posts157848-highlight.html and http://www.wrongplanet.net/postt113459.html and search for patterns and trends and I will analyze it.

Here's the numerical data gathered so far: http://www.mediafire.com/?i6ulfjc6885amzt

You'll have to rename it so that the ".jpg" extension is deleted and you're left with an .ods file that can be opened within Open Office's Calc.

So, as a brief explanation of the data, the IQ scores for everyone who has a "culled from the "set of scientific experiments"" comment in a rightward cell were guessed. There's no particular reason why they were guessed other than to have something that would "fill the void", and they certainly aren't relevant within any IQ regression analyses. Also, the PIQ for everyone else was obtained from iqtest.dk and it appears that the average performance IQ of the iqtest.dk's normed sample was 115 (Inferring from the average difference between people's claimed PIQ and their tested PIQ on iqtest.dk.), so the "actual nonverbal IQ" will be about 15 points higher than average. Also, those with an age of 16-17 will have 5 nonverbal points added to match the "adult population" distribution curve that iqtest.dk was normed on. A similar age-IQ adjustment will happen with the verbal IQ test if it isn't age-adjusted, which requires further exploration to determine.

So... the analysis will be completed, correlations will be explored and the equations will be constructed soon.

After examining the data, it became rather clear that the number of samples from the "neurotypical population" was the main limiting factor in conducting a reliable multiple regression analysis. While it would've been nice to conduct the analysis on all the tests, the only combination of tests that had a sufficiently large number of people taking them were RDOS's Aspie/Neurotypical test and Wired's AQ test. What I need to do at this point is run a correlation analysis on all the test data gathered so far for each test and correlate them with the highest "asperger"-loaded test, the AQ test. Once I do that, I need to take the most highly correlated tests and try to encourage neurotypical individuals to take those tests such that we have a sufficiently large sample size to conduct the multiple regression analysis on more than 2 tests.

Anyway, here's the semi-final formula derived from the analysis.

DIAGNOSIS_SCORE = -0.255856672631116 + 0.0223559704161317*AQ_TEST_SCORE + 0.00201304824238126*ASPIE_TEST_SCORE -0.000615104830914709*NEUROTYPICAL_TEST_SCORE

Since the average AQ test score from the officially diagnosed aspergers group was 40.36, and the sum of ASPIE_TEST_SCORE and NEUROTYPICAL_TEST_SCORE averages to 200 (Rounded down), the AQ test has a weighting roughly .223*40.36 / (.00201-.00062)*200 = 2.12 times higher than the Aspie/Neurotypical test for the typical person with aspergers. So, while the Aspie/Neurotypical test is predictive, it's nearly twice as less predictive of aspergers as the AQ test.

Here's the DIAGNOSIS_SCORE distribution for the "officially diagnosed" aspergers group:

0.8826208261
0.7334650647
1.0497371349
0.6143367729
1.047086247
1.1023893651
1.1786391428
1.0862849792
0.5624565342
0.6834296202
0.8689207861
1.0020835446
1.0020835446
0.9981246457
1.0400745587
0.9229040054
1.0770697741
0.9901954805
1.1326631333
0.9978450084
0.9385271612
0.7319552177

Here's the "DIAGNOSIS_SCORE" distribution among known neurotypicals (Sampled from WrongPlanet, so there's going to be a positive bias or otherwise known as a "negatively skewed" distribution -- in other words, individuals in this sampling will over-represent members from the actual neurotypical population with high scores; To get a score distribution more representative of the general population requires random sampling from a large number of individuals.):

0.0878695711
-0.1806483726
0.410091374
0.4334092841
0.3282831612
0.2259309539
-0.0725477821
-0.1759066352
-0.0811592498
0.0081302367
0.2007005317
0.080790442


From here, one could find their "DIAGNOSIS_SCORE" ranking/percentile among officially diagnosed Aspies. How to turn this into a more meaningful statement regarding ones likelihood of possessing aspergers requires substantially more test data and, ideally, from a random population. Until that amount of test data becomes available, this is the most meaningful comparison one can make.

So, now I just need to make the web-calculator and host it on my web server.

--------------------------
DAY 3 (April 17, 2010)
--------------------------

I just ran the following variables through the Pearson Correlation calculator at http://www.wessa.net/rwasp_pairs.wasp#output along with the DIAGNOSIS_SCORES. There were a few interesting findings.

===============================
Correlation with DIAGNOSIS_SCORE
===============================
ASPIE_SCORE: .943
SQ/EQ - EQ: -.928
Emotional Intelligence: -.718
AS (Conforms to DSM quite well): .702
Highly Sensitive Person: .667
SQ/EQ - SQ: .65
Facial memory: -.487
Facial recognition: -.258
VIQ: -.182
PIQ: -.18
VIQ-PIQ: -.102

As you can see, the SQ/EQ - EQ test has a correlation with the DIAGNOSIS_SCORE about as high as the ASPIE_SCORE itself, which I think is a bit interesting. Its counterpart, the SQ index, has a correlation of only .65. Also interestingly, the next highest correlating factor was the "Emotional Intelligence test" which is surprising because the Emotional Intelligence test is simply a self-assessment of ones behavior and abilities, so it would be open to intrapersonal biases. Past that, the DSM version of the AS test has a correlation of .702, while the Highly-Sensitive-Person test(HSP) has a correlation of .65. The Facial Memory test isn't highly correlated with aspergers and neither is facial recognition, VIQ, PIQ or the difference between ones VIQ and PIQ scores.

So, we need more scores from neurotypical individuals from the SQ/EQ - EQ test, the Emotional Intelligence test, the AS (DSM style) test, the HSP test and SQ/EQ - SQ test. The facial memory, facial recognition and the VIQ, PIQ tests need to be removed from the "Core tests" list.

In retrospect, I'm noticing that the tests that are most highly correlated with aspergers are self-assessment like tests and not tests of ability (Like the Facial memory or Facial recognition test or VIQ or PIQ test.). That's pretty interesting, because self-assessment is open to biases and potentially mistaken thinking. Since it is potentially vulnerable to said biases, I wonder if there are certain perceptions that would predispose one to score more highly on "Highly associated Asperger tests" than otherwise, and that might account for the unexpectedly high correlations, at least partially? Of course, actual behavior and thinking will still mostly determine ones score on "Highly associated asperger tests", but still, I wonder how much bias is playing into this data... (Like, for example, if someone *thinks* they have aspergers and answers in a way that tends to influence their scores so that they score in a "more aspergian" way.)

-------------------

New suggestions: Since the AQ test was found to be the most highly correlated test with aspergers, I wonder if detailed score distribution information is available on the AQ test? If so, then we should be able to get a sense of the "DIAGNOSIS_SCORE" distribution among the normal population by expressing the ASPIE_TEST score using AQ test score / ASPIE_TEST averages, and then find out how much overlap there is between the officially diagnosed aspergers population and the normal population and using that information to approximate a likelihood of possessing aspergers for a given combination of scores.

How are you dealing with autistic people who are not specifically AS?

I just ran the correlations using the calculated DIAGNOSIS_SCORE for everyone (Which is a combination of the AQ test score and the ASPIE/NEUROTYPICAL test scores.) and correlated that with the other tests and had a few interesting findings (Such as facial memory only being fairly moderately correlated with aspergers, and facial recognition barely correlating.). The results are detailed in the 2nd post of this thread.

So, erm, can we make whatever information has been gathered make sense for non mathy people like me?

Are there any translators here who can translate mathese into English?

I can try, but I'll readily admit I'm a rather low quality translator.

So, two tests with a correlation coefficient (That number you see) close to -1 or 1 are said to have a "high correlation" - that is, if you score high on one, you're likely going to score high on the other. If you score low on one, you're likely going to score low on the other. The closer the number is to 0, the less "predictive" it is - that is, if it has a correlation below .5, then scoring highly on one test doesn't automatically mean you scored highly or lowly on the other as you're less able to accurately predict it. Think of "Correlation coefficient" as a number indicating predictability.

So, a score on the EQ portion of the EQ/SQ test predicts quite well how high your AQ test score and ASPIE/NEUROTYPICAL score will be, while the Facial Recognition test doesn't predict your "Autistic test" scores. Put another way, a given person with aspergers isn't necessarily much worse at recognizing faces than NTs, at least according to the Facial Recognition test results, even though people with a greater severity of aspergers are slightly worse at recognizing faces *on average*. The same thing is true about the Facial Memory test.

So, I was thinking back to the correlations with aspergers and these tests, and the most highly correlating tests were self-assessment tests instead of ability tests. In fact, the highest correlating ability test was the Facial memory test, but yet that had *only* a correlation of -.487. Since aspergers is at least partially defined by lacking "certain social abilities such as social insight", it's surprising that there isn't a higher correlating ability test. And, ability tests are important because self-assessment tests are vulnerable to biases and potentially unreliable subjective interpretations and assessments.

So, the quest is ON for looking for an ability test that highly correlates with aspergers!

So, any ideas where to look?

One way to get this aspergian ability test would be to create it. Essentially, create (or borrow...) questions and then release the test to the public, and then correlate the questions with other known scores on known-to-be-reliable aspergers tests such as the AQ test and against official diagnoses, and then take the highest correlating questions and reduce the test down to a reasonable size.

Another way is to steal an existing emotional ability test. Does anyone know of any? I liked Queendom's Emotional Intelligence test, but I honestly have no clue how they scored their questions, so scoring it would be a bit difficult. I suppose the "right answer" would likely be the one that's most frequently answered by those do "well on the test", and so one could infer a way to grade it, but it wouldn't necessarily be the same as Queendom's scoring algorithm.

This site at http://www.emotionaliq.org/Test.htm details a fairly reliable way to create a "scoring system" using an "expert rater" group system and analyzing scoring frequencies.