<- file stat 97jack.html ->

Leave-one-out vs jackknife

=======================Rich Ulrich, 20 Mar 1997==========ssm Subject: Re: jackknife vs LOO Message-ID: <5gro7b$8si@usenet.srv.cis.pitt.edu> Rodney Sparapani (spara002@mc.duke.edu) wrote: : Roberto Cappuccio wrote: : > : > I would like to know what's the difference between : > JACKKNIFE and LEAVE-ONE-OUT cross validation : > Thanks : > roberto : Leave-one-out is just the first step of Jack-knifing. -- I agree with that. But... : You take : your dataset and append on to the end of it, i.e. suppose you have : N observations, then append N(N-1) leave-one-out observations. Now, : perform your calculations on this data, retrieving N+1 estimates. : This is where you create what are called the pseudo-values. -- I don't understand this description of SOMEONE's computational method. A different computation uses one pass through the sample, after obtaining the Whole-sample statistics: For each Case (i), an revised estimate of the parameter is obtained by comparing the Whole-sample estimate and the leave-one-out estimate. The average of those revised estimates is relatively unbiased, and its variance is relatively robust, compared to the whole-sample estimate or a Leave-one-out average, for cases when those would give you biased estimates. (In the simplest models, they all might come out the same.) ================= : I can't : do it justice in ASCII and my TEX is very rusty. See the following : paper: : RG Miller, The jackknife - a review, biometrika (1974), 61, 1, pp. 1-15 : -- * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Document by Rich Ulrich. E-mail to wpilib+@pitt.edu

FAQ top.

Ulrich home page.

Ulrich FAQ. http://www.pitt.edu/~wpilib/stats99.html