A COMPARISON OF TECHNIQUES WHICH TEST FOR JOB DIFFERENCES
用蒙特卡洛方法比较了单变量和多变量方差分析在检验工作差异时的统计功效和第一类错误控制，发现方法优劣取决于数据是否满足循环性和齐性假设，并给出了选择指南。
There has been a recent trend in research seeking the most appropriate statistical technique for determining job similarities/differences. Monte Carlo methods were used to analyze more closely the repeated measures analysis of variance and the multivariate analysis of variance in order to add further insight into the viability of these techniques for this purpose. The conventional univariate analysis of variance, the ε‐adjusted univariate F test, and the ε‐adjusted univariate F test were compared to three multivariate tests (Roy's largest‐root criterion, Wilk's likelihood ratio, and the Pillai‐Barlett trace) in terms of power and control for Type I error when (1) circularity and homogeneity were met, (2) homogeneity was met but circularity was violated, (3) homogeneity was violated but circularity was met, and (4) both homogeneity and circularity were violated. The efficacy of the techniques was shown to be contingent upon whether the assumptions were met or not. The univariate test proved to be the better technique when circularity was met. The multivariate technique proved to be the better test when homogeneity was met while circularity was violated. The results were mixed when both circularity and homogeneity were violated. Guidelines for selecting a statistical technique which tests for job differences are offered.