Variability of Estimated Variance Components and Related Statistics in a Performance Assessment

Generalizability theory provides a conceptual and statistical framework for estimating variance components and measurement precision. The theory has been widely used in evaluating technical qualities of performance assessments. However, estimates of variance components, measurement error variances, and generalizability coefficients are likely to vary from one sample to another. This study empirically investigates sampling variability of estimated variance components using data collected in several years for a listening and writing performance assessment. This study also evaluates stability of estimated measurement precision from year to year. The results indicated that the estimated variance components varied from one study to another, especially when sample sizes were small. The estimated measurement error variances and generalizability coefficients also changed from one year to another. Measurement precision projected by a generalizability study may not be fully realized in an actual decision study. The study points out the importance of examining variability of estimated variance components and related statistics in performance assessments.