forum 8: week of 5 March: tests and evidence

Fragment of a discussion from Course talk:Phil440A
Jump to: navigation, search

I'm also interested in the mis-specification testing, especially in the context of respecifying a statistical model (on p. 9). From the way the authors put it, it almost sounds like one should look for a statistical test that fits the data in a way that confirms the hypothesis, even if several other models have already failed to provide good results. I may be misunderstanding this, but, from personal experience, when the data fails to be informative, it seems counterintuitive to "assess a candidate model considered for use in drawing a primary inference"; that just sounds like trying to make the data provide the results I wanted, instead of whatever I actually got.

07:14, 6 March 2012