Syndicate content

Add new comment

Submitted by Elizabeth on

I'm very interested in the index vs. PCA question. Though I'm not sure why you are worried about the unit size for the PCA. Presumably your index will be in standard deviations, which will require some harder thinking as well for people who don't have an intuitive understanding of an SD. And you can always convert the predicted values of the PCA factors into SD, so you would have the same units as your current index. For illustration, you could show a kernel density plot for the two groups. But maybe I'm wrong.

Anyway, Kling et al talk a bit about this issue for a paragraph or so, I think in their appendix to the MTO paper. My interpretation (perhaps incorrect) was that a PCA is appropriate when you are trying to get at an unknown factor (as you seem to be) but an index is important when you care about all the individual outcomes (eg, diabetes, hypertension, etc, you may all care about equally in a health index).