PSYC 394 Lecture Notes - Lecture 6: Family-Wise Error Rate, Statistics Education, Microsoft Powerpoint
Document Summary
As with other parametric tests, there are three main assumptions: homogeneity of variance, normal distribution, independence of observations. Anova tests are generally robust: the f ratio controls type 1 errors well for violations of non-normality (for two-tailed tests, but not one-tailed tests) Anova tests lose accuracy when the group sizes are largely different. F values depend on variance and can thus become either very liberal or very conservative. The f ratio indicates whether there is an overall difference between group means, but does not indicate which specific means differ. Family-wise error rate: compounding of error rates (5%) when running multiple analyses eg. ) comparisons between three groups yields a total error rate of 20% F ratio error = 5% t-test1 error = 5% (group 1 group 2) t-test2 error = 5% (group 1 group 3) t-test3 error = 5% (group 2 group 3) This high error rate explains why a series of t-tests are not used to determine specific differences.
