From Null Hypothesis Signicance Testing to Effect Sizes
There are two main parts to the statistical reform argument: the negatives and the positives. The negatives are criticisms of NHST, and the positives refer to advantages of estimation and other recommended techniques. Most of this book concerns the positives, but in this chapter I’ll first consider the negatives: NHST and how it’s taught and used. This chapter focuses on
• NHST as it’s presented in textbooks and used in practice • Problems with NHST • The best ways to think about NHST • An alternative approach to science and the estimation language it uses • The focus of that language, especially effect sizes (ESs) and esti-
mation of ESs • Shifting from dichotomous language to estimation language • How NHST disciplines can become more quantitative
Suppose we want to know whether the new treatment for insomnia is better than the old. To use NHST we test the null hypothesis that there’s no difference between the two treatments in the population. Many textbooks describe NHST as a series of steps, something like this:
1. Choose a null hypothesis, H0: μ = μ0, where μ (Greek mu) is the mean of the population, which for us is the population of difference scores between the new and old treatments for insomnia. It’s most common to choose H0: μ = 0, and that’s what we’ll do here.