Skip to content
Back to formatted view

Raw Message

Message-ID: <412E6177.9050308@stat.wisc.edu>
Date: 2005-04-20T01:48:44Z
From: Douglas Bates
Subject: gls: Newton-Raphson or EM?
In-Reply-To: <00e101c48b90$ef60b360$ad133a86@www.domain>

Dimitris Rizopoulos wrote:
> Hi Brendan,
> 
> according to
> 
> @Book{pinheiro.bates:00,
>   author    = {J. Pinheiro and D. Bates},
>   title     = {Mixed-Effects Models in S and S-PLUS},
>   year      = {2000},
>   address   = {New York},
>   publisher = {Springer-Verlag}
> }
> 
> Section 2.2.8, the optimization procedure for lme (so I suspect also
> for gls) is a hybrid algorithm which starts as EM for 25 iterations
> and then switches to Newton-Raphson.
> 
> Best,
> Dimitris
> 
> ----
> Dimitris Rizopoulos
> Doctoral Student
> Biostatistical Centre
> School of Public Health
> Catholic University of Leuven
> 
> Address: Kapucijnenvoer 35, Leuven, Belgium
> Tel: +32/16/396887
> Fax: +32/16/337015
> Web: http://www.med.kuleuven.ac.be/biostat/
>      http://www.student.kuleuven.ac.be/~m0390867/dimitris.htm

True, but it looks as if Brendan is asking about the gls function, which 
uses one of R's unconstrained optimizers (either nlm or optim) without 
any EM iterations.

> 
> 
> ----- Original Message ----- 
> From: "Brendan A. Klick" <bklick at jhsph.edu>
> To: <r-help at stat.math.ethz.ch>
> Sent: Thursday, August 26, 2004 7:08 PM
> Subject: [R] gls: Newton-Raphson or EM?
> 
> 
> 
>>Hello,
>>
>>Does anyone know whether the gls function in the nlme library uses
> 
> the Newton-Raphson or EM algorithm to find the restricted
> log-likelihood or maximum log-likelihood estimates?
> 
>>Brendan Klick
>>bklick at jhsph.edu