I was reading a post from Andrew Gelman. Every now and again he gives statistical advice to reader who write in with questions. Today he had a question about a rare drug exposure in a large population.
This reminds me of one idea that I wish I'd known when I was first doing statistical in the financial world (before deciding to get a PhD). It was the idea of "off support inference" or of trying to extend inference beyond the bounds of the data given. This can happen when the few exposed participants are in a portion of the data where there are few controls and there is effect measure modification. Some fairly severe errors can result.
Which makes me think about to all of my risk modeling and makes me wonder if I worried enough about these issues at the time?
No comments:
Post a Comment