R help - Apr 2011 - multiple comparisons with generalised least squares

Dear R users,

I have used the following model:

M1 <- gls(Nblad ~ Concentration+Season + Concentration:Season, data=DDD,
weights=varIdent(form=~ 1 | Season*Concentration))

to assess the effect of Concentration and Season on nitrogen uptake by
leaves (Nblad). I accounted for the difference in variance across the factor
levels by using the varIdent function.

Then I wanted to perform multiple comparisons with the glht function of the
multcomp package.

glht(M1,linfct = mcp(Season = "Tukey"))

However, here I got an error message "Error in terms.default(object) : no
terms component. Error in factor_contrasts(model) : no ?model.matrix? method
for ?model? found!".

Does the glht function work with a gls model? And if not, is there an other
way to perform multiple comparisons for a gls model?

I've searched this forum for an answer to this question, but I could only
found someone with the same question which remained unanswered. I hope
someone can provide an answer now!

Many thanks in advance!

Sandy



--
View this message in context:
http://r.789695.n4.nabble.com/multiple-comparisons-with-generalised-least-squares-tp3441513p3441513.html
Sent from the R help mailing list archive at Nabble.com.

Hi Sandy,

I was wondering if you ever recieved an answer regarding the use of multiple
comparisons for gls models? 
I have also been searching various forums and books to see if there are any
methods I could use and have only found people, such as yourself, asking the
same question. 

Many thanks if you can give any assistance,

Rachel

--
View this message in context:
http://r.789695.n4.nabble.com/multiple-comparisons-with-generalised-least-squares-tp3441513p4632412.html
Sent from the R help mailing list archive at Nabble.com.

racmar wrote
> 
> I have also been searching various forums and books to see if there are
> any methods I could use and have only found people, such as yourself,
> asking the same question. 
> 
I was looking into this recently, as well, and found that the problem has to
do with building the model.matrix/terms/model.frame for gls objects when
using the glht function.  I ended up creating three gls-specific functions
and was able to get estimates/confidence intervals for the toy example
below.   You may find these functions useful (under the caveat that I only
checked that I got estimates and not that the estimates were correct ;) ).

Ariel

Example:
library(nlme)

Orthodont$fage <- factor(Orthodont$age)

#toy example with Orthodont data using age as a factor
fitgls <- gls( distance ~ fage, data=Orthodont, 
	weights = varIdent(form =~1|fage))



library(multcomp)

#notice the error about the model.matrix for gls objects when using glht
confint( glht (fitgls, mcp(fage="Tukey") ))

#create model.frame, terms, and model.matrix functions specifically for gls
objects
model.matrix.gls <- function(object, ...) 
	model.matrix(terms(object), data = getData(object), ...)


model.frame.gls <- function(object, ...) 
	model.frame(formula(object), data = getData(object), ...)


terms.gls <- function(object, ...) 
	terms(model.frame(object),...)


#now run glht again
confint( glht(  fitgls, mcp(fage="Tukey") ))




--
View this message in context:
http://r.789695.n4.nabble.com/multiple-comparisons-with-generalised-least-squares-tp3441513p4636009.html
Sent from the R help mailing list archive at Nabble.com.

Ariel THANK YOU for posting this. worked great for me just pasting this part
into R


Ariel wrote
> model.matrix.gls <- function(object, ...) 
> 	model.matrix(terms(object), data = getData(object), ...)
> 
> 
> model.frame.gls <- function(object, ...) 
> 	model.frame(formula(object), data = getData(object), ...)
> 
> 
> terms.gls <- function(object, ...) 
> 	terms(model.frame(object),...)
and running ghlt again. it seemed to me like the output was correct.



--
View this message in context:
http://r.789695.n4.nabble.com/multiple-comparisons-with-generalised-least-squares-tp3441513p4648972.html
Sent from the R help mailing list archive at Nabble.com.