A composite and its own subtests are the same testing, twice.

What it does to the number

Suppose working memory is genuinely weak and everything else is average. The index is low, its two subtests are low, and all three land in the working memory domain mean. The domain now leans on that one weak ability three times as hard as a domain represented by a single measure — so the profile shape is partly a record of which instruments report composites.

Across a cohort it is worse, because different people had different batteries. A group where half the cases have index-plus-subtests and half have subtests alone has two different weightings inside one mean.

The fix: fold the composite out when its parts are present

Casebook knows which measures compose which. When a composite and its constituent parts are both present for one testing, the composite is folded out of the domain mean and the parts are used — one testing, counted once. When the parts are absent, the composite stands in for them.

This is why the catalog records composition at all: dozens of instruments and hundreds of measures, each with the domain it belongs to and the measures it contains.

Where this is honestly incomplete

Rating-scale composition is not shipped, and the reason is worth stating plainly: the structures are published in the manuals, and filling them in from memory would silently change every profile that used them.

The ABAS-3 General Adaptive Composite contains three domain composites which contain eleven skill areas. With nothing recorded, all fifteen count into one domain mean and the composite is weighted as though it were a twelfth skill area. That is wrong, and it is wrong in a direction nobody would notice.

Two things stand in for it, neither of which is a fix:

• It is reported. Any domain that one instrument is deciding says so — on the case page and in the report itself, with the numbers, and with the summary scores sitting beside their probable parts.
• It is recordable. The composition can be entered once from the Domains page by somebody with the manual open. After that the composite folds out exactly as a Wechsler index already does.

One real overlap has no clean answer: the MASC-2 GAD Index shares items with other scales rather than containing them, and the model expresses containment. So it is counted alongside them, which is a small inaccuracy stated rather than hidden.

The related trap

The other thing that quietly distorts a domain mean is a rating scale pointing the opposite way to an ability measure.