Computerized Adaptive Testing

How computerized adaptive testing works

A CAT begins with an initial estimate or prior distribution. After each response, the scoring model updates what it knows about the respondent. A selection rule then chooses a useful next item from an item bank. The process stops when the estimate is precise enough, a maximum length is reached, or another stopping rule is met.

The exact algorithm varies. Many educational CATs use item response theory and select items near the current ability estimate. A personality CAT can select items that best separate plausible trait profiles or reduce uncertainty across several dimensions.

Why CAT can be shorter

A fixed form asks everyone the same questions, including items that may add little information for a particular person. An adaptive system redirects that question budget toward unresolved parts of the estimate.

Efficiency is not guaranteed. It depends on a calibrated item bank, a sound model, enough content coverage, and selection constraints. A small or biased bank can make adaptation worse, not better.

CAT safeguards

Practical systems often balance statistical information with:

• coverage of required content areas;
• limits on repeated item exposure;
• fairness and accessibility constraints;
• protection against asking near-duplicate questions;
• transparent stopping criteria.

Optimizing only for mathematical information can narrow what the assessment measures.

Adaptive does not mean AI-generated

Computerized adaptive testing refers to dynamic item selection. The questions can be carefully prewritten and validated. Generating new questions with an AI model is a separate design choice and introduces different quality-control problems.