Would varying the inter-subtest correlations change the result?
Asked by: the committee, in Session 5 — Simulation Studies on Subtest Indicators:
Correlation between subtests are all the same because real data wouldn’t show this. You only have three points in the plot – would it make a difference if you didn’t have the same correlation between the three?
And, separately: “Why is life science different than the others; are other subtests more related to the total score?”
Answer given: none recorded.
Why it matters. Batch 1 imposes a single correlation value across all three subtest pairs — 0.65, 0.70, or 0.75 — which real data would not produce. If the true correlation structure is asymmetric, the simulation may be systematically favourable to whichever method handles symmetric dependence better, and the uniform CT-PO advantage in CT-PO vs SBAC Simulation Results Deck would need re-testing.
This is the same objection as Model the joint density of the three dimensions arriving from a different direction, and it is why Review item parameters and total-score correlations before Batch 3 was set as a precondition for a third batch.
A route to an answer. copula-approach-to-subtest-indicators notes that under a copula formulation this stops being a limitation and becomes a specification choice — a t-copula with a full correlation matrix, or a vine if the pairwise structure is genuinely asymmetric — and that the per-pair Kendall’s tau falling out of the fit answers the life-science question directly.
Related. M-STEP Subscore Reporting