(This result is based on simulations with 100,000 replications.) In contrast, with 20% trimming, the actual probability of a Type I error is 0.022. Wilcox (1994a) reports situations where H 0: μ 1 > μ 2 is tested with α = 0.025, but the actual probability of a type I error is 0.054. As an illustration, suppose the first group has a normal distribution, and the second group is skewed with κ 1 = 2 and n 1 = n 2 = 12. Theoretical results, supported by simulations, indicate that as the amount of trimming increases from 0 to 20%, Yuen's method yields confidence intervals for μ t 1 − μ t 2 with probability coverage closer to the nominal level ( Wilcox, 1994a). ( Luh & Guo, 2010, report results on strategies for determining the sample sizes.)Īs previously indicated, when two distributions differ, it can be difficult getting a confidence interval for the difference between the means that has probability coverage reasonably close to the nominal level.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |