Ping Sa
Department of Mathmathics and Statistics
University of North Florida
Testing the Variance of Non-Normal Distributions
Currently, the most commonly used test of a single variance is the
chi-square test when the population is normal. However it is quite sensitive
to departures from normality. A new test procedure is proposed for the
upper-tailed test for the variance of non-normal populations. Edgeworth
expansion is used to derive the new test statistic. A Monte Carlo study
investigates the properties of the new procedure for a variety of
distributions. It is shown that the new test yields controlled type I error
rates as well as good power performances when the sample size is moderate or large.
Back to Colloquium Series