Linear Reg (Press Statistic)

hlsmith

Omega Contributor
#1
I was not familiar with the Press Statistic. I am currently looking it up. Seems to be a leave-one-out procedure.


Can any one provide some insight on its interpretation or actual usefulness?


Thanks!
 

Dason

Ambassador to the humans
#2
I'd say it's most useful when comparing models. Interpreting it on it's own doesn't get you too far other than "lower is better". When comparing across models though you can use it to see which models are more sensitive to omitting single observations. If you're overfitting (or just close to overfitting) the data with a complex model then omitting an observation might drastically change the predictions in that area which will most likely result in a large contribution to the press statistic.

I guess one could compare the press statistic to the typical residual sum of squares to get an idea if the press statistic is "large" but I've typically just used it for model comparison.
 

hlsmith

Omega Contributor
#3
Yes, your latter point is what keeps coming up:


"A fairly close agreement between the PRESS statistic (see Table 55.5 on
page 2969) and the Sum of Squared Residuals indicates that the MSE is a reasonable
measure of the predictive accuracy of the fittedmodel (Neter,Wasserman, and Kutner,
1990)." - From SAS Documentation