ESTIMATING THE RELIABILITY OF A TEST CONTAINING MULTIPLE ITEM FORMATS

The reliability estimation problem for tests composed of multiple item formats, a common classroom instrument, has largely been ignored. Based on their underlying test models, the two most common internal consistency estimation techniques, the split-halves approach and Cronbach's alpha coefficient, would generally be inappropriate for multiformat instruments. In this article, three models representing varying degrees of part-test parallelism and their appropriateness for ''mixed'' tests are discussed. The models of focus are: classically parallel, tau-equivalently parallel, and congenerically parallel. It was recognized that a test composed of multiple item formats would most likely result in parts with differing functional lengths; such an instrument can be represented only through the adoption of a congeneric model. An appropriate reliability estimate for this type of instrument can be derived using a technique that combines Raju's (1977) and Feldt's (cited in Feldt & Brennan, 1989) procedures. An illustrative example is included to clarify the need for an appropriately modeled reliability technique.