Computer-Adaptive Testing

Computer-adaptive testing (CAT) is a computer-based assessment in which the questions presented to a candidate change according to the candidate’s responses. It repeatedly estimates the candidate’s ability and selects suitable questions from a calibrated item bank, rather than giving every candidate an identical fixed paper.

Adaptive testing cycle

The system follows an iterative process of estimating ability, selecting an item and updating the estimate.

• The test begins with an initial ability estimate, often using a question of moderate difficulty or available prior information.
• After each response, an item response theory model updates the candidate’s estimated ability using the item’s known difficulty and other calibrated characteristics.
• The algorithm generally selects the next question that provides high statistical information near the current ability estimate, while observing content and test-design constraints.
• A correct answer usually raises the estimate and may lead to a harder item; an incorrect answer usually lowers it and may lead to an easier item. The progression is not mechanical because each response is interpreted probabilistically.

Completion and scoring

Adaptation continues until a specified stopping rule is reached. The test may stop when the ability estimate attains the required measurement precision, when a classification decision becomes sufficiently reliable, or when the maximum number of questions or testing time is reached.

• The final score is based on the estimated ability and the characteristics of the administered items, not merely on the number of correct answers.
• Candidates may receive different questions, but their scores can remain comparable when items are properly calibrated on a common scale.

Conditions for credible use

Effective CAT requires a large, secure and well-calibrated item bank, reliable software and adequate digital infrastructure.

• Content-balancing rules ensure that adaptation does not omit required subjects or skills.
• Item-exposure controls reduce repeated use and possible leakage of particular questions.
• Accessibility provisions and careful validation are necessary to prevent technology, language or disability-related barriers from distorting performance.
• Poor calibration or an unrepresentative item bank can produce biased or unstable ability estimates.

How UPSC asks this