Skip to content

repeated measurement using mixed model

3 messages · Jinsong Zhao, Christoph Scherber

#
Hi there,

We have got a valuable data set from a long-term experiment. The 
experimental design is following:

Trt     Time     Y
   1     1983
   1     1986
   1     1993
   1     1998
   1     2003
   1     2010
   2     1983
   2     1986
   2     1993
   2     1998
   2     2003
   2     2010
   3     1983
   3     1986
   3     1993
   3     1998
   3     2003
   3     2010

Because the data was collected in a way of repeated measurement. we 
think it should be analyzed using linear mixed model. However, there is 
no group variable, and as you have noted, there is no replication for 
each treatment (Trt). So, we don't know how to deal with this data set.

Any suggestions or comments will be really appreciated. Thanks in advance.

Regards,
Jinsong
#
Dear Jinsong Zhao,

In this case, you may wish to use a generalized least squares model (gls), also to be found in the nlme library. You can start with an initial model
containing only fixed effects, and then update it with an appropriate correlation structure.

Try this first:

library(nlme)
?gls
?corStruct

All the best
Christoph
on 18.11.2011 07:29, Jinsong Zhao wrote:

            

  
    
#
On 2011-11-18 21:32, Christoph Scherber wrote:
Dear Christoph,

Thank you very much for the kindly reply.

Dose gls() fit the linear model without random effects? Is corStruct 
used with correlation argument to explore the structure of residual 
variance?

Another question, someone suggests that replications of response 
variable, e.g., determined concentration on 3 samples obtained from same 
treatment, should be include in the design. Then, the random effects 
could be estimated (I don't know how), and it will improve the power of 
statistical analysis. I don't think so, however, I can't find a way to 
explain that.

Regards,
Jinsong