ABSTRACT
Critical attacks on null hypothesis testing over the years have not greatly diminished its use in the social sciences. This chapter tells why the continued use of hypothesis tests is not merely due to ignorance on the part of data analysts. In fact, a null hypothesis that an effect is exactly zero should be rejected in most circumstances; what investigators really want to test is whether an effect is nearly zero, or whether it is large enough to care about. Although relatively small sample sizes typically used in psychology result in modest power, they also result in approximate tests that an effect is small (not just exactly zero), so researchers are doing approximately the right thing (most of the time) when testing null hypotheses. Bayesian methods are even better, offering direct opportunities to make statements such as "the probability that the effect is large and negative is .01; the probability that the effect is near zero is .10; and the probability that there is a large positive effect is .89. "
