s-news
[Top] [All Lists]

Re: mixed models

To: "Paul, David A" <paulda@BATTELLE.ORG>
Subject: Re: mixed models
From: "Osman" <osman.al.radi@utoronto.ca>
Date: Fri, 30 May 2003 14:56:34 -0400
Cc: <s-news@lists.biostat.wustl.edu>
References: <940250A9EB37A24CBE28D858EF077749136DA0@ws-bco-mse3.milky-way.battelle.org>
Thanks for you reply..
 
I have the Pinheiro/Bates book and it is seriously an excellent read even for a non-statistician like my self.. However, I cannot find a function or option to get confidence limits for the predicted values (not the random or fixed coefficients) !!
 
Thanks
 
Osman
----- Original Message -----
To: 'Osman'
Sent: Friday, May 30, 2003 2:47 PM
Subject: RE: [S] mixed models

There are many useful functions you may apply to a fitted lme object:
 
summary( )
fitted( )
resid( )
predict( )
intervals( )
update( )
coef( )
fixef( )
ranef( )
 
For the intervals( ) function, you can use which = "var-cov" to extract
the confidence intervals for the random effects and within-group fixed
effects variance components.
 
Also, for multilevel random effects models, you can use the level =
option in (for example) ranef( ).  This allows you to extract the random
effects at various levels.  The same comment applies to the fixed effects
which may be extracted using fixef( ).
 
You might want to invest in the excellent book "Mixed Effects Models
in S and Splus" by Jose Pinheiro and Douglas Bates.  It is well worth the
money and will save you a lot of grief.  Someone on the R listserv once
said (I don't remember who) that mixed models in R/Splus are a "space
shuttle, not a bottle rocket."  That's certainly true.  There are lots of useful
applications for these models, but they are extremely sophisticated and
require a correspondent amount of training.
 
Best,
  david paul
 
-----Original Message-----
From: Osman [mailto:osman.al.radi@utoronto.ca]
Sent: Friday, May 30, 2003 12:57 PM
To: s-news@lists.biostat.wustl.edu
Subject: [S] mixed models

Hello,
 
After using m1<-lme(....) I can get linear predictors by predict(m1,newdata=new).. However the results is a vector of predicted valued only.. there no options to get confidence limit.. Is there a way of getting confidence limits for linear predicted values from an lme object?
 
Thanks
 
Osman 
<Prev in Thread] Current Thread [Next in Thread>