Computerized Adaptive Testing

Computerized adaptive testing selects questions in response to the evidence collected during a test. Its purpose is usually to estimate a learner's proficiency or support a classification efficiently. It differs from an adaptive tutor, which selects activities to help the learner improve. A question that is informative for measurement is not necessarily the best next teaching activity, and an adaptive test need not use a generative language model.

Many adaptive tests use item response theory to connect the probability of a response with an estimated proficiency and calibrated item characteristics. After an answer, the system updates its estimate and selects another item that should provide useful information. Selection also respects constraints such as curriculum coverage, item exposure, accessibility, and test length. 'Correct means harder, incorrect means easier' captures only a rough intuition and misses these design requirements.

The process needs an item bank with empirical evidence about how questions function. If the bank contains few suitable items at an important proficiency level, adaptation cannot create information that is absent. Calibration must fit the intended population and construct sufficiently well. A newly generated question with an estimated difficulty label is not automatically interchangeable with a calibrated item, even if a teacher judges it sensible.

Imagine a numeracy assessment beginning with a question about interpreting a percentage discount. After a correct response, the test may choose a more informative percentage item while ensuring that later items also sample other required content. Another learner receives a different sequence. Comparing their raw numbers correct would be misleading because they answered different questions. The scoring model places response patterns on a common scale, subject to its assumptions and supporting evidence.

A stopping rule determines when enough information has been collected or a limit has been reached. A test might stop after reaching a specified precision, after supporting a classification relative to a threshold, or after a maximum number of items. Operational programmes such as NCLEX explain their own decision rules; those rules should not be assumed to apply to every adaptive test. A shorter test is not automatically a sign of stronger or weaker performance.

Teachers can demonstrate adaptation using a small fictional item bank and a transparent paper simulation. Learners track how each response changes an estimate and why content requirements sometimes override a simple difficulty choice. The exercise should be presented as a model of the process, not a validated assessment instrument. It helps explain why test takers may find many questions challenging and why question count alone does not reveal a result.

Quality assurance includes item security, subgroup functioning, accommodations, stable delivery, and clear handling of interruptions. Reports should communicate uncertainty and avoid treating a proficiency estimate as a permanent description of a person. Adaptive delivery can improve efficiency when the bank, model, and operational rules are sound. It cannot compensate for a poorly defined construct or an inappropriate decision. Educators therefore ask what is being measured, how the bank was calibrated, how stopping works, and what evidence supports using the resulting score.