Effects of autocorrelation on single-subject single-facet crossed-design generalizability assessment

The effects of first-order serial dependence on estimates of variance components and common reliability indices in single-subject single-facet crosseddesign generalizability analyses were investigated by adding autoregressive and moving-average components to four sets of direct observation data known not to contain serial dependence. Results indicated that, whereas universe score, relative error, and expected observed score variance estimates were systematically biased, the amount of bias with signal/noise ratios, generalizability coefficients, and the D-study analog of Cronbach alpha were negligible.