12-31-2014 12:48 PM
A couple of points: R squared and adjusted R squared are oddly defined with models without intercepts. Still, you could plug the R-squared value obtained from GLM into the formula for adjusted R squared (no intercept):
ADJRSQ(no int) = 1 - n * (1 - R^2)/(n - p), where n is the number of observations and p is the number of parameters fit.
04-28-2015 01:17 PM
Steve et al.,
It was not clear to me if the model the R^2 value came from had to have any type of option in regards to the intercept or not. Do I run a normal proc glm when getting the R^2 value for the calculation. Also, do you know the name of this particular R^2 adjustment?
04-28-2015 02:49 PM
The formula I gave for the adjusted R^2 is just the standard adjustment, except that the n-1 in the numerator and denominator is replaced by n, since the intercept is not estimated. I don't remember where I found that, but Draper and Smith would be a good place to start. It needs to be the value obtained with the noint option, for it to be consistent. It is a measure of the association between the response and the X values, both considered as deviations from zero, rather than deviations from the mean.