Repository navigation
WISC Exploration
This page details the results from analyzing the WISC cognitive measures on their own, disregarding anything related to the brain.
My sanity checks mostly consist of checking the data distribution for imbalances in age, IQ, sex, and diagnosis status across the usable dataset.
/Data%20Distribution/Entire%20Dataset%20(595)/Diagnosis%20Distribution.png)
We notice that the age distribution has a slight skew towards younger subjects, IQ looks normally distributed, there are twice as many males as females, and that almost every subject has some diagnosis.
I did a quick check on the relationship between age and IQ.
/Data%20Distribution/Age%20vs%20IQ.png)
It's interesting that the relationship isn't flat but that IQ actually decreases with age. While the decrease isn't large, the fact that it does decrease may reflect the Flynn effect (since there's a 10-year difference between the youngest and oldest subject) or that WISC doesn't characterize older subjects well (maybe that's why there's the adult version of WISC called WIAS). Another possible explanation is that fluid intelligence decreases as the subjects gain experience, thus increasing their crystalized intelligence.
I also checked the IQ distribution for each diagnosis status.
/Data%20Distribution/Diagnosis%20vs%20IQ%20Distribution.png)
From the 2123 subjects that have WISC scores, here are the scores (raw and scaled) correlated with age using r^2 as the measure.
| Subtest | Raw | Scaled |
|---|---|---|
| Block Design (BD) | 0.2691 | 0.0120 |
| Similarities | 0.3725 | 0.0165 |
| Matrix Reasoning (MR) | 0.3045 | 0.0000 |
| Digit Span (DS) | 0.3018 | 0.0009 |
| Coding | 0.4260 | 0.0175 |
| Vocab | 0.4500 | 0.0087 |
| Figure Weights (FW) | 0.2605 | 0.0194 |
| Visual Puzzles (VP) | 0.2137 | 0.0286 |
| Picture Span (PS) | 0.2563 | 0.0019 |
| Symbol Search (SS) | 0.1615 | 0.0078 |
As we can see, the r^2 for some scores are high, while all of the scaled versions are near zero (as expected).