To calculate p-values properly requires paying a lot of attention to how you choose the null hypothesis and whether it is really appropriate for your problem and the state of the art. I do not have a lot of experience in ecology, but in bioinformatics people often choose null hypotheses because they make the p-values easy to compute, or because everyone does it that way, or (more cynically) because they make their results appear significant. One can get a good p-value by choosing a null hypothesis that is almost certain to be wrong, regardless of the fact that the consensus was already that this null hypothesis was almost certain to be wrong before any of the reported experiments were undertaken. That doesn't mean the reported experiments advanced scientific understanding.
--Ruchira On Tue, Mar 1, 2011 at 6:24 AM, Jeff Houlahan <jeffh...@unb.ca> wrote: > Hi Chris and all, I actually think that it's a mistake to diminish the role > of p-values. My opinion on this (stongly influenced by the writings of Rob > Peters) is that there is only one way to demonstrate understanding and that > is through prediction. And predictions only demonstrate understanding if > you make better predictions than you would make strictly by chance. The > only way to tell if you've done better than chance is through p-values. So, > while there is a great deal more to science than p-values, the ultimate > tests of whether science has led to increased understanding are p-values. > Best. > > Jeff Houlahan > Dept of Biology > 100 Tucker Park Road > UNB Saint John >
