R help - Mar 2000 - the function lme in package nlme

Dear people,

A somewhat clueless question follows:

I just discovered that the lme function in contrib package nlme for R,
while similar to the lme function in Splus, does not use the cluster
function option. This difference does not appear to be documented in the
V&R `R Complements' file.

I have data which is divided into 6 groups
The lme model is of the form (simplified from the actual model a bit)

y  = a + b x_{ij} + \alpha_i + \beta_i x_{ij} + \epsilon_{ij} 

i = 1,...,6.

where a and b are fixed effects and \alpha and \beta are random effects.

I have a vector of the form

clus = (1,1,1,2,5,6,4,4,4,3,...)'

where the 1, 2 3 are all factors corresponding to different groups (or
clusters in Splus terminology).

In Splus I would use the argument 

(...,cluster = clus,...) 

I thought that the equivalent thing to do in R is to use a argument of the
form

(...,correlation = groups,...)

where groups is a CorStruct object of the class corSymm, where groups is
obtained by

groups <- corSymm(form = ~ 1 | clus)
groups <- initialize(groups, data = labor.df)

Unfortunately this appears to be wrong, since I get the error

**********************************************************************
Error in lme.formula(fixed = earning ~ edu + job.pres + age, random = ~1 +
: 
	Incompatible formulas for groups in "random" and
"correlation"
***********************************************************************

I also don't quite understand what the time covariate in the corSymm
documentation means.

*****************************************************************
form
       a one sided formula of the form ~ t, or ~ t | g, specifying a time
covariate t 
******************************************************************

I am also wondering why the cluster argument was not retained for
compatibility with Splus. Was there a problem with this?

                           Sincerely, Faheem Mitha.

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

On Tue, 28 Mar 2000, Faheem Mitha wrote:
> Dear people,
> 
> A somewhat clueless question follows:
> 
> I just discovered that the lme function in contrib package nlme for R,
> while similar to the lme function in Splus, does not use the cluster
> function option. This difference does not appear to be documented in the
> V&R `R Complements' file.
This is only a temporary difference.  The nlme library underwent a major
revision between versions 2 and 3.  R has version 3 but many current
versions of S-PLUS still use version 2. I believe you can obtain version 3
for S-PLUS from nlme.stat.wisc.edu.

Mathsoft has slower release cycles with more detailed planning, so it
will take them longer to incorporate the new nlme.
 
> I have data which is divided into 6 groups
> The lme model is of the form (simplified from the actual model a bit)
> 
> y  = a + b x_{ij} + \alpha_i + \beta_i x_{ij} + \epsilon_{ij} 
<snip> 
> I thought that the equivalent thing to do in R is to use a argument of the
> form
> 
> (...,correlation = groups,...)
> 
> where groups is a CorStruct object of the class corSymm, where groups is
> obtained by
> 
> groups <- corSymm(form = ~ 1 | clus)
> groups <- initialize(groups, data = labor.df)
> 
You trying to fit a random intercept _and_ compound symmetric
correlation of the residuals. This is not a good idea.

I think you want

  lme(y~x,random=~x|clus,data=labor.df)

that is, a random intercept and random slope within each cluster, and
independent residuals epsilon_ij

The example in help(lme) is of this type.

	-thomas

Thomas Lumley
Assistant Professor, Biostatistics
University of Washington, Seattle

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at
stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._