Abstract:
|
In the past, most health outcomes instruments were developed and evaluated using classical test theory. Using these methods, items were analyzed based on sample-dependent statistics, such as proportion endorsed. In recent years, item response theory (IRT) has gained popularity because it provides the methodology to evaluate an item or a set of items without sample dependence and supplies item (or scale) information across the continuum of the trait being measured (i.e., functional disability, psychosocial quality of life, degree of motivation, general mental health). For example, IRT provides information about the expected proportion endorsed for those subjects at low-, middle-, high-levels, and levels between, rather than just an overall proportion. The best items can be selected after evaluating their performance at all levels. Because of this advantage, scales can be tailor-made for the task at hand. Items needed for a diagnostic tool can be selected so that they provide the most information around the diagnostic cut-point; whereas, items needed for more general instruments can be selected so that they provide information across the continuum.
|