The Effects of Categorical Data on Correlated Error Terms and Subsequent Effects on Coefficient Alpha

• In Event: 60.031 - Reliability: New Approaches

Abstract

Under Classical Test Theory, every item score has a “true” component that represents the actual level of the construct being measured, and an “error” component. One of coefficient alpha’s assumptions is the independence of error terms. It has been suggested that categorical data violates this assumption, although the mechanisms through which this occurs are unclear. The present paper uses a simulation study to examine the mechanisms by which categorical data affects estimates of coefficient alpha, McDonald’s omega, and a non-linear SEM estimate of reliability. Preliminary results found that a smaller number of categories and a greater discrepancy between the skew of the true scores and categorical observed scores lead to greater discrepancies between true reliability and alpha.

Authors

• Kate Pinder, James Madison University

• Deborah L. Bandalos, James Madison University