ABSTRACT
Comparisons of groups can be performed without the assumption of equal variance across groups. This chapter considers the consequences of testing for mean differences while ignoring differences in variances. Weighted analysis of variance is possible when variances depend on some independent measurement, or covariate, in an obvious way. Comparison of means while allowing for unequal variance is quite straightforward if the variances are known up to a constant, by considering a weighted analysis. Weighted analysis of variance ideally involves choosing weights proportional to the inverse of the variance, as this essentially reduces to the equal variance case. The process of conducting variance tests changes the significance and interpretation of subsequent formal tests in ways that are poorly understood. There have been many ideas for nonparametric assessment of unequal variance. The most commonly used and accepted approach is H. Levene's test.
