Post date: Sep 26, 2015 7:45:24 PM
The pictures at the bottom of https://sites.google.com/site/advancedbiascorrection/ are perhaps "woodpile", but the ones at the top are close to what we want I think. There is also a grab bag of files at https://sites.google.com/site/advancedbiascorrection/powerpoint-presentations. Unfortunately, they are not organized or even annotated, so I don't recommend going there unless we're desperate.
In terms of the message of the paper, I would suggest that that the slide presentations convey the basic elements of the argument we want to articulate in the paper. (Annoyingly, there are multiple presentations, fashioned a bit differently for different audiences over the years. Consider the union of these slides.) In this union of slides there are several images that Jack does not use in the draft that we might consider recruiting into the draft. I guess this selection will depend in part on how big we can allow the paper to be. What are your opinions on this?
In addition to the theses about needing to generalize the notion of bias when applying it in risk analysis, the presentations introduce four ABC corrections:
using validation studies to increase the uncertainty for model predictions in the face of poor model performance,
jacketing naked (scalar) estimates (which is just a special case),
deconvolution to remove measurement errors that blur an estimate, and
the nonparametric c-box as an analog of deconvolution for a small-sample-size estimate of a constant.
This is a bit messier than the two ideas we originally were talking about, which were just jacketing and deconvolution in general terms. We don't have to do everything in the first paper. Maybe we should save some parts for a follow-on manuscript.