Assessing measurement invariance in questionnaires within latent trait models using item response theory

Abstract

Using questionnaires or inventories. researchers often perform mean comparisons between different populations (e.g .• males vs. females) in order to draw inferences about actual differences in the constructs being measured. However, such comparisons are not meaningful unless the assessments obtained in each of the populations are conmesurable or invariant across populations. Most researchers simply assume that measuremnt invariance holds. However, the extent to which this assumption is a reasonable one for specific measures and specific populations should be tested empirically. Using item response theory, the present study shows how gender measurement invariance can be determined when. as is most common, a psychological construct is assessed by means of a questionnaire or inventory composed of categorical items. To illustrate our method, the Positive Problem Orientation scale of the Social Problem-Solving Inventory-Revised (D' Zurilla. Nezu & Maydeu-Olivares, 1996) was assessed and found to be reasonably gender invariant. whereas the Negative Problem Orientation scale was not.