Correcting Internal Consistency Estimates Inflated by Correlated Item Errors

Coefficient alpha is perhaps the most widely employed estimate of reliability for self-report measures in communication research. However, coefficient alpha misestimates internal consistency when measures are not congeneric. Congeneric measures meet the assumption of classical test theory in that all item errors are uncorrelated or random; when item errors are correlated, the measure fails to be congeneric, signaling the presence of latent factor(s) other than those the researcher intended to measure. Scholars have alerted to the fact that many self-report measures used are noncongeneric and that coefficient alpha might be yielding inaccurate, often inflated, reliability estimates. In this essay, we propose a formula for estimating the internal consistency of a measure that reflects the degree to which reliability is affected by correlated item errors.