A generative approach to the modeling of isomorphic hidden-figure items

Abstract

A generative approach to psychometric modeling consists of encoding information about the cognitive processes and structures that underlie test performance into an item-generation algorithm in such a way that the generated items have known psychometric parameters. An important by-product of the approach is that the knowledge about the response process is tested every time a test is administered. Validation thus becomes an ongoing process rather than an occasional event. This approach is illustrated through an analysis of hidden-figure items, and is shown to be feasible. Index terms: construct validity, generative modeling, isomorphic problems, item difficulty, spatial ability, validation.