R help - Apr 2006 - SE estimates for treatment groups from nlme

I am wondering how to obtain SE estimates for fixed effects from a nonlinear
mixed effects model?

I have fixed effects corresponding to three factors A, B and C with 2, 3 and 3
levels respectively. I have fit a model of the following general form:

nlme1<-nlme(y~ SasympOrig(x, Asym, lrc), data=df, fixed=list(Asym~A*B*C,
lrc~A*B*C),
	start=c(fixef(ETR.nlme)[1], rep(0,17), fixef(ETR.nlme)[2], rep(0,17))) 

I am using the default ("contr.treatment" and "contr.poly").
The summary table (tTable) gives me the baseline coefficients and a list of
differences and I have no trouble calculating the coefficients for any of my
treatment groups by forming the correct linear combination of coefficients.
However, I don't understand how to obtain the standard errors for these
linear combinations from the summary report. What I want to obtain would be 18
parameter estimates with appropriate SEs for my 18 different groups
(A*B*C=2*3*3=18).

In response to a similar posting, it was recommended to use the function
"estimable" in the package gregmisc. I have looked at this and cannot
figure out how to apply it in my situation. Any help would be greatly
appreciated.

Thanks,
Katie

____
Katie Grieve
Quantitative Ecology and Resource Management
University of Washington

Katherine A Grieve <grieve <at> u.washington.edu> writes:
> 
> I am wondering how to obtain SE estimates for fixed effects from a
nonlinear
mixed effects model? 
> 
It's a bit of fiddling, but there is an example coming close to what you
want
on page 373 of Pinheiro/Bates, and in file \library\nlme\scripts\ch08.R, 
section 8.2. And don't forget that you can use "intervals" on
nlme-results.

Dieter

I have looked at that example on p. 373, however it still does not help me to
  get the stand errors i need. In my case, with the 18 different groups, I am 
combining more than just the (intercept) term with one other row in the Summary 
tTable. It is no problem to form linear combinations of the coefficients, but 
the SE in the tTable cannot just be combined. Similarly, intervals just gives me
  CI for each entry (row) in the tTable but these are not the group estimates 
that I am looking for.

An example to illustrate what i mean the rows of the tTable would be labeled 
something like this (A=2 levels, B=3 levels, C=3 levels):
Fixed Effects: (Asym~A*B*C)
                   Value      SE
Asym.(intercept)    x1
Asym.A1             x2
Asym.B1             x3
Asym.B2             x4
Asym.C1             x5
Asym.C2             x6
Asym.A1B1           x7
Asym.A1B2           x8
Asym.A1C1           x9
Asym.A1C2           x10
Asym.B1C1           x11
Asym.B2C1           x12
Asym.B1C2           x13
Asym.B2C2           x14
etc...

So, to get an estimate for the groups I form linear combinations e.g.:
Group A2,B1,C1 = x1 + x3 + x5 + x11

I am combining 4 coefficients to get the group mean -- is there a way to 
get the corresponding SE. It is definitely not a linear combination of the SE
listed in the tTable.
And, intervals just gives me a CI for each entry (row) in the table, not the 
linear combinations which form my groups.

Thanks again.
Katie
-----------------------------
It's a bit of fiddling, but there is an example coming close to what you
want
on page 373 of Pinheiro/Bates, and in file \library\nlme\scripts\ch08.R, 
section 8.2. And don't forget that you can use "intervals" on
nlme-results.

Hi Katie,

maybe the easiest solution is to create a new factor that corresponds to
the combinations of the three factors A, B and C. A quick and dirty way
to create such a factor is:


ABC <- factor(paste(A, "x", B, "x", C, sep =
""))
ABC


and then fit the model using the variable ABC instead of A*B*C.

Christian