Skip to content

How to specify arguments in lme() ?

2 messages · w jj, Douglas Bates

Original
w jj
# 
Hi,

I have a question about the function lme() in R.

I have a 2*2*3 layout with some missing data (labelled as *).  These 3 
factors are labelled as A,B,C, the response is Score. The layout is as 
follows:-

A              B                 C                    Score
1               1                  1                         5
1               1                  2                         *
1               1                  3                         1
1               2                  1                         4
1               2                  2                         4
1               2                  3                         *
2               1                  1                         3
2               1                  2                         *
2               1                  3                         4
2               2                  1                         2
2               2                  2                         *
2               2                  3                         5

Suppose these data are stored in a data frame called "test".

If all these 3 factors are fixed, then I can fit a model without the 3-way 
interaction as:-
fit1<-lm(Score~A*B+A*C+B*C,data=test)        

If one of these factors, say A, is a random effect variable, then I need to 
fit a mixed effect model using lme(). I have read the R documention on 
lme(), but I am still not clear how to specify the random argument. I tried 
to do:-

fit2<-lme(Score~A*B+A*C+B*C,data=test,random=~A, na.action=na.pass) 

but the system give a message as follows:-
Error in getGroups.data.frame(dataMix, groups) : 
        Invalid formula for groups

So how should I specify the arguments? 

Thank you very much for your help!

Jiajie
Douglas Bates
#
On 1/18/07, w jj <jiajiehere at hotmail.com> wrote:

            

      
      
      
    

        
You could do it but you don't really want to try to fit a model with
several random effects generated by a factor with only two levels.
Estimating variances, which is what is done for a random effect, is
more difficult than estimating means or other linear combinations of
the responses, which is what fixed effects parameters end up being
expressed as.  Trying to estimate a variance when observing a factor
at only two levels is overly optimistic.

Just for the record, the call to lmer in the lme4 package would be

fit2 <- lmer(Score ~ B*C+(1|A/B)+(1|C:A), data = test)

            

      
      
      
    

        
I don't think you want to use na.pass here.  The underlying C code for
fitting lme or lmer models doesn't take kindly to finding NA's in the
data.