I've sort-of got a psychology degree (part of a double-major), and null hypothesis significance testing is all I know. I've known there were problems with it for a while, but I have no idea what the alternatives are (sure, Bayesian inference, but is that really a solution? I don't have the background to know,m but that journal doesn't think so). What are the alternatives?
Null hypotheses aren't inherently bad if n is large and the researchers apply the correct test. Part of the trouble comes with when they: - Don't define their tests in advance (We looked at the data and then decided to apply a one-tailed t-test between these groups) - Test multiple hypotheses testing (Is this subgroup A significantly different from subgroup B? What about A and C? B and C?) - Use the wrong type of test (i.e. t-test on a population that isn't normally distributed) Previous to mk's post, I'd have said (1) fix the issues above, (2) report the confidence interval instead, or (3) look at the data for "obvious" changes. Apparently (2) is problematic, too, so others may have a better answer >_>