This problem might not occur to those of you who don't devise new measurements, but it is an interesting topic anyway, and the article introduced me to a new validity concept: "incremental validity"
It's what you get when you can make an established screening test into a better predictor by adding new information. Or, as they cast the problem, when you want to evaluate prediction from one already accepted test and you think your test will perform better.
I'm not entirely convinced that p-values and significance testing are the right tools to solve this problem in the first place. I think more in terms of AUC parameter estimation.
Nevertheless, read it and see if you can convince me that the significance testing approach is the correct one.
No comments:
Post a Comment
Comments to this blog are moderated. Urgent or other time-sensitive messages should not be sent via the blog.