Binging on Gelman again while I was traveling. Here is a gem (link).
If you ever dreamed of having "all the data", be sure to give this a read and give it a few moments' thought. Skim through the main post quickly. The good stuff is in the comments.
Can we please kill hypothesis testing already?
http://wmbriggs.com/blog/?p=11261
"Hypothesis tests confuse making a decision about the truth of some proposition with the probability the proposition is true. HTs cannot give the probability any hypothesis is true. HTs conflate probabilities and decisions, which is always suboptimal."
Posted by: Nate | 09/09/2014 at 10:49 AM
Nate: I take a slightly less incendiary position on this topic. Firstly, the concept of statistical significance is central to statistics, regardless of methodology--the idea that the observed sample is but one realization of many alternatives is very important.
Secondly, it is wrong to say "hypothesis tests confuse making a decision about the truth of some proposition with ..." -- the theory of hypothesis testing is fine; its common misinterpretation is the point of contention. I do agree that it is a hopeless enterprise to explain to nontechnical people how to use it.
Posted by: junkcharts | 09/19/2014 at 03:53 AM