How Do You Prefer Your Item Analyses: Traditional or Rasch?

Abstract

Item analyses of a 22-item attitude questionnaire were undertaken using both a traditional approach and the Rasch technique. These analyses, obtained from the responses of a sample of 158 Year 5 children indicated, at first glance, discrepant results.

Initial inspection of the traditional item analysis suggested that, due to extremely low discrimination indices, all of the items were unsatisfactory. For the Rasch model, however, all items appeared to refer to a unidimensional trait and formed a very comformable set.

Our reconciliation of these seemingly disparate results is based upon three considerations: first, whether the items do or do not measure a single trait; second, whether the interaction of the subjects with the items has resulted in a set of random responses, and third, whether there is an adequate dispersion of subjects along the scale continuum.