Congeneric and (Essentially) Tau-Equivalent Estimates of Score Reliability

James M. Graham

Educational and Psychological Measurement · 2006 · 688 citations · 19 references

Concepts

TL;DR

Coefficient alpha is a lower‑bound estimate of reliability that assumes essentially tau‑equivalent measurement; violations of this assumption often cause alpha to underestimate true reliability. The article proposes a hierarchy of measurement models for reliability estimation and demonstrates how structural equation modeling can test their fit to data. The authors discuss how test and data characteristics affect the degree of tau‑equivalence violation. Heuristic and applied examples illustrate the concepts discussed.

Abstract

Coefficient alpha, the most commonly used estimate of internal consistency, is often considered a lower bound estimate of reliability, though the extent of its underestimation is not typically known. Many researchers are unaware that coefficient alpha is based on the essentially tau-equivalent measurement model. It is the violation of the assumptions required by this measurement model that are often responsible for coefficient alpha's underestimation of reliability. This article presents a hierarchy of measurement models that can be used to estimate reliability and illustrates a procedure by which structural equation modeling can be used to test the fit of these models to a set of data. Test and data characteristics that can influence the extent to which the assumption of tau-equivalence is violated are discussed. Both heuristic and applied examples are used to augment the discussion.

References

19