Abstract

Latent traits – mental abilities, qualities and attitudes unobservable by direct measurement – are challenging to quantify. Psychologists, medical and educational researchers try to learn about latent traits by administering tests and relating the score of the test to the latent trait. Nevertheless, if a four year old is presented with an intelligence test for an adult, little is to be learned about the child’s intelligence, even if the child is exceptionally bright. On the other hand when the same four year old child prodigy is tested with a standard instrument appropriate for his age group, one does not learn much about the child’s IQ either. Even if some reasonable conclusions could be drawn, the precision of these would be questionable.

Adaptive tests provide a solution to this problem: the difficulty of the test adapts while the test is performed. Based on estimates of the latent trait obtained during testing, the next test item is chosen from a pre-calibrated pool of items, so that in a rigorous sense some notion of information is maximized. Many tests that adapt to the respondent have been developed in the last 30 years, personal computers making adaptive testing feasible. The underlying theory of these test procedures is an active field of research. I will give an overview of basic and advanced procedures and of problems in adaptive testing and strategies to cope with them.