Date: Thu, 14 Oct 1999 16:03:48 +0200 Torben Haslund "SAS(r) Discussion" Torben Haslund Jack-knifing and External Studentization for ML-estimation text/plain; charset="us-ascii"

Dear SAS-Lers,

I am wondering if it is possible to calculate externaly studentized (=full leave-one-out cross-validated or jack-knifed) residuals in Proc Mixed (ML-estimation) in one run, in stead of during resampling (collecting estimates from each left out sample during as many runs as I have samples)?

As far as I can see (R. Dennis Cook and Sanford Weisberg: Residuals and Influence in Regression) it is possible for least squares estimation (GLM) to calculate external studentized residuals in one run in stead of during resampling. Cook and Weisberg write at p18-20 for Least Squares estimation:

ri = ei / (sigmahat * sqrt(1-vii))

ti = ei / (sigmahat(i) * sqrt(1-vii))

and

ti = ri * sqrt( (n-p'-1)/(n-p'-ri**2) )

in these formulas

ei = residual ri = internally studentized residual ti = externally studentized residual sigmahat is the standard deviation and sigmahat(i) is the standard deviation calculated without the i-th case

vii is the diagonal element of the hatmatrix n is number of observations (cases) and p' is number of terms in the model (I guess it would be number of fixed+random terms in a mixed model - comments to this are wellcome).

If anybody has ideas about calculating externally studentized residuals for mixed models (ML-estimation) using these equations I will appreciate hearing from you.

torben

::::::::::::::::::::::::::::::::: Torben Haslund Department of Plant Biology Swedish University of Agricultural Sciences P.O. Box 7080 S - 750 07 UPPSALA, Sweden

E-mail address: Torben.Haslund@vbiol.slu.se Home and postal address: Ostre Paradisvej 7B, DK-2840 HOLTE, Denmark Phone home: +45 4541 1198

Back to: Top of message | Previous page | Main SAS-L page