Xref: utzoo comp.ai.neural-nets:959 alt.cyb-sys:31 Path: utzoo!attcan!uunet!mcsun!ukc!warwick!ecctp From: ecctp@warwick.ac.uk (Dr J A K Cave) Newsgroups: comp.ai.neural-nets,alt.cyb-sys Subject: Re: Generalization Criteria Message-ID: <263@orchid.warwick.ac.uk> Date: 27 Sep 89 12:56:50 GMT References: <506@uvaee.ee.virginia.EDU> Reply-To: ecctp@warwick.ac.uk (Dr J A K Cave) Organization: Computing Services, Warwick University, UK Lines: 17 The psychometric literature has spawned an approach known as G-theory to quantify the generalizeability of structural information to bigger "universes" - in practical terms it looks like a gussied-up analysis of variance/covariance, but that's OK for some purposes. I will try to dig up some references, but for starters you might search the published works of Rich Shavelson: I know he did a summary for the NAS that appeared in a paper on job performance measurement within the last 3 years. Econometrics has come up with a variety of measures grossly related to MSE. Examples include Mallow's Cp, Amemiya's PC, mean-square prediction error on an independent sample; or Rbar-squared (the correlation coefficient adjusted for degrees of freedom, using the correlation between predicted and actual values on a disjoint set of data. Don't know if this is even close to what you are seeking (different disciplines, different meanings for the same words...)