P. Values and all that

I feel I should write something about the  ASA pronouncement on p-values, but reached the conclusion that to say anything that's in any way  useful and that won't be misconstrued by fools or knaves will be so time consuming that it will seriously interfere with doing stuff I really need to be getting on with.

It's an important issue, requires careful discussion and, perhaps most importantly, requires that some attention be paid to disciplinary context. It also requires an understanding that science doesn't take place in a knowledge vacuum. Usually no single experiment or piece of data analysis is decisive (at least in the fields I work in).

Already the usual cranks and buffoons are bellowing nonsense about it down their megaphones, but since a defining feature of crankiness is incorrigibility in the face of  reason it is a waste of time to include them in the conversation.

For what it is worth I  find  the ASA statement in itself  to be entirely reasonable. It doesn't condemn the use of p-values; it does point out that p-values are often used or interpreted inappropriately. Who could disagree with that? There are good  comments by Senn, Mayo and Benjamini among others. You can catch the latter on Mayo's blog. 

I even liked Gelman's 'garden of forking paths' routine though I fear where that would take us with regard to most of the quantitative empirical work in sociology.  I tend to think of most of the work I do in terms of estimation and data smoothing rather than hypothesis testing but whatever you are doing you do need to think about all of the data dependent decisions you have made and how those might have gone the other way if the data had been different.