Type I Error Rates and Power Estimates for Selected Two-Sample Tests of Scale
Estimated Type I error rates and power are reported for a modified Fligner-Killeen test, O’Brien’s test, O’Brien’s test using Welch’s modified ANOVA, the Brown-Forsythe test, and two tests developed by Tiku. Normal and nonnormal distributions and a two-group design were investigated. O’Brien’s test and the Brown-Forsythe test had estimated Type I error rates near the nominal level for all conditions investigated. To maximize power when the sample sizes are equal, O’Brien’s test should be used with platykurtic distributions and the Brown-Forsythe test with leptokurtic distributions. Either test can be used when the kurtosis is zero. When the sample sizes are unequal, O’Brien’s test should be used with platykurtic distributions and the Brown-Forsythe with symmetric-leptokurtic distributions. With other distributions, the tests have similar power.