Use automated item generation to enlarge the pool

The advice. Automated item generation “would help generate a large number of questions as it creates cousin items” — the committee’s point being that if forms necessarily share most of their items, attention should go to the minority that differ. (The remark names a specific overlap figure; it is held verbatim in the restricted layer.)

Why. The overlap constraint is downstream of pool size. Automated item generation produces families of cousin items from a common model, which is well suited to licensure content where the construct is stable and the item shells are repeatable. It attacks the cause rather than the symptom.

The caveat the committee embedded. Cousin items are cousins. Generating variants of the same item model enlarges the pool without necessarily making forms meaningfully different — which is why the recommendation is phrased around focusing on the non-overlapping 25%, and why it does not by itself satisfy Define ‘meaningfully different’ per incidence group.

Status. Open.