[R] summary.lme: argument "adjustSigma"

[Christoph Buser]

Dear R-list

I have a question concerning the argument "adjustSigma" in the
function "lme" of the package "nlme". 

The help page says:

"the residual standard error is multiplied by sqrt(nobs/(nobs - 
npar)), converting it to a REML-like estimate."

Having a look into the code I found:

stdFixed <- sqrt(diag(as.matrix(object$varFix)))

if (object$method == "ML" && adjustSigma == TRUE) {
    stdFixed <- sqrt(object$dims$N/(object$dims$N - length(stdFixed))) * 
        stdFixed
}

tTable <- data.frame(fixed, stdFixed, object$fixDF[["X"]], 
    fixed/stdFixed, fixed)


To my understanding, only the standard error for the fixed
coefficients is adapted and not the residual standard error. 

Therefore only the tTable of the output is affected by the
argument "adjustSigma", but not the estimate for residual
standard error (see the artificial example below). 

May someone explain to me if there is an error in my
understanding of the help page and the R code? 
Thank you very much.  

Best regards,

Christoph Buser

--------------------------------------------------------------
Christoph Buser <buser at stat.math.ethz.ch>
Seminar fuer Statistik, LEO C13
ETH Zurich	8092 Zurich	 SWITZERLAND
phone: x-41-44-632-4673		fax: 632-1228
http://stat.ethz.ch/~buser/
--------------------------------------------------------------


Example
-------

set.seed(1)
dat <- data.frame(y = rnorm(16), fac1 = rep(1:4, each = 4),
                  fac2 = rep(1:2,each = 8))

telme <- lme(y ~ fac1, data = dat, random = ~ 1 | fac2, method = "ML")
summary(telme)
summary(telme, adjustSigma = FALSE)