2 thoughts on “Economic freedom and statistical measurement

  1. Andrew, there's a technique called "improper linear models" or a special case called "unit weighted regression" in which a bunch of variables are thought up that seem to have a correlation with what you want, they are standardized in some way (perhaps nondimensionalized so that they all have some common measurement scale), and then rather than choosing coefficients in an "optimal" manner (ie. least squares) some other method of choosing coefficients is used.

    (Here I am telling YOU about statistical techniques, I assume that you are probably totally familiar with this, but perhaps some of your readers are not)

    It seems that the discussion on the linked blog is relevant to the question of whether an optimal weighted regression or an improper regression is better. In general, it's known that the optimal weighted regression is better (it is "optimal" after all) but the context for unit weighted regression is where the "outcome" variable is not measurable and hence it is impossible to find the optimal coefficients.

    In this context I guess there is some literature on the use of "improper" regression, especially for improving clinical judgment, and the big name in this field was Robyn Dawes back in the late 70's and early 80's

    http://psycnet.apa.org/journals/amp/34/7/571/

    It's been a while since I looked at that stuff, and there may be better methods that have been discovered, but I am generally against the idea that a method needs to be optimal or not at all… so The Heritage Foundation is doing something good even if it isn't optimal.

  2. Dan:

    Yes, we discuss this sort of combination of variables in chapter 4 of ARM and I alluded to it recently here. (But you're probably right that many blog readers have not fully internalized our chapter 4.)

    P.S. Robin Dawes's paper in the Kahneman, Slovic, and Tversky book is indeed a classic.

Comments are closed.