[R] post hoc test for lme using glht ?

David Winsemius dwinsemius at comcast.net
Thu Jul 22 00:53:24 CEST 2010


On Jul 21, 2010, at 6:36 PM, SNAFU1 wrote:

>
> Hi,
>
> I have a fairly simple repeated measures-type data set I've been  
> attempting
> to analyze using the lme function in the nlme package.  Repeated  
> searches
> here and other places lead me to believe I have specified my model
> correctly.
>
> However, I am having trouble with post-hoc tests.  From what I  
> gather, other
> people are successfully using the glht function from the multcomp  
> package to
> perform post-hoc tests.  I've tried multiple iterations, but can't  
> seem to
> get it to work.  Here is (a subset of) what I have been trying.  Any  
> help
> would be greatly appreciated:
>
>> model.3<-lme(fixed=Totnum~Wk*Pop, random=~1|nUID, data=nona)
> #Wk has 6 levels, 0-5; Pop has 2 levels; nUID is a unique ID given  
> to each
> individual (multiple individuals from each population were re- 
> measured each
> week)
>> anova(model.3)
>            numDF denDF  F-value p-value
> (Intercept)     1   592 649.7753  <.0001
> Week            5   592 302.9706  <.0001
> Pop             1   222  70.7268  <.0001
> Week:Pop        5   592  36.8576  <.0001

Come on... are you cutting and pasting fro disjoint sections of your  
console? The formula says Wk and the parameter name in the anova  
result is Week. What gives?

>> summary(glht(model.3, linfct=mcp(Wk = "Tukey")))
> Error in mcp2matrix(model, linfct = linfct) :
>  Variable(s) ‘Wk’ of class ‘integer’ is/are not contained as a  
> factor in
> ‘model’.
>> is.factor(Wk)

Well, this object should not even exist at the top level environment  
unless you attached it and didn't tell us about it.

Try:
Wk <- NULL
nona$Wk <- factor(nona$Wk)

Then re-run your code.


> [1] FALSE
>> Week<-factor(Wk)
>> is.factor(Week)
> [1] TRUE
>> model.3<-lme(fixed=Totnum~Week*Pop, random=~1|nUID)
>> summary(glht(model.3, linfct=mcp(Week = "Tukey")))
> Error in contrMat(table(mf[[nm]]), type = types[pm]) :
>  less than two groups
>
> Thanks,
>
> Rob
-- 
David Winsemius, MD
West Hartford, CT



More information about the R-help mailing list