Evaluating the performance of different procedures for constructing confidence intervals for coefficient alpha: A simulation study

Reliability is one of the most important aspects of testing in educational and psychological measurement. The construction of confidence intervals for reliability coefficients has important implications for evaluating the accuracy of the sample estimate of reliability and for comparing different tests, scoring rubrics, or training procedures for raters or observers. The present simulation study evaluated and compared various parametric and non-parametric methods for constructing confidence intervals of coefficient alpha. Six factors were manipulated: number of items, number of subjects, population coefficient alpha, deviation from essentially parallel condition, item response distribution and type. The coverage and width of different confidence intervals were compared across simulation conditions.