A Simulation Study for Comparing Three Lower Bounds to Reliability

Reliability is the quantification of the consistency of results from a measurement procedure across replications. By far, Guttman’s Lambda 3 or coefficient alpha, as a lower bound to the reliability, is the most frequently used and reported index of reliability because of its simplicity in computation and accessibility in statistical programs. However, coefficient alpha tends to be negatively biased when the internal structure of the measurement is heterogeneous. This study is designed to use simulated data to evaluate and compare two alternatives for coefficient alpha: the greatest lower bound and Lambda 2 with coefficient alpha. In light of our simulation results, lambda 2 is recommended because it shows the least amount of bias under most of the simulation conditions.