Chapter 17: Robust Data Analysis

Rand R. Wilcox

Robust data analysis Traditional methods for comparing means perform well in terms of Type I errors when the corresponding distributions do not differ in any manner. But three major insights indicate that when distributions differ, under general conditions, routinely used methods can perform poorly in terms of power, measuring effect size, and achieving accurate confidence intervals. Many new and improved methods have been derived with the goal of addressing known problems, some of which are outlined in this chapter. These major insights also have important implications for Pearson's correlation and least squares regression and are briefly outlined. Routinely used methods for making inferences about means were once thought to perform relatively well when violating the assumptions of normality or equal variances. But three major insights have revealed that under general conditions, these classic techniques perform poorly in terms of power and achieving accurate confidence intervals, and they suffer from other ...

Users without subscription are not able to see the full content on this title. Please, subscribe or login to access all content on this website.

• Buy The Book