Defining random statement for PROC GLIMMIX repeated measures

[iatoloye]

Hello All,

I am uncertain which of the statements I have for the random statement in my code is correct. I have a data set collected annually from the same set of plots and one treatment factor (with two levels) and three blocks. I am trying to determine the differences between the treatment levels using repeated measures.

Sample data 

Treatment Block Year Yield @@;

0 1 2008 50

50 1 2008 75

0 2 2008 65

50 2 2008 85

0 1 2010 45

50 1 2010 65

...

Below are the codes:

First code:

proc glimmix data = Hist plots = studentpanel;
class block treatment Year ;
model lScore_Mean = treatment|year ;
random block block*treatment;
run;

 

Second code:

proc glimmix data = Hist.SHAPE plots=residualpanel;
class block treatment Year ;
model lScore_Mean = treatment|year / ddfm=kr2;
random intercept / subject=block ;
random year / residual subject=block*treatment;
covtest 'Is GLM OK?' glm;
run;

 

The results of the first code showed the treatment effect was not significant (p>0.1) while the results of the second code showed the treatment effect was significant (p<0.1).

I was hoping to get similar results but it seems I am not clear about the denominator degrees of freedom. I got different answers and I am not sure which one is right. 

Thank you very much!😊

[sbxkoenk]

Hi,

I have moved your post to the "Statistical Procedures" board.

More chance to get a good answer here.

Good luck,

Koen

[Rick_SAS]

Can you clarify your experiment? The data seem to say that each block received Treatment=0 and Treatment=50 in the same year?  Is that correct? I don't understand how the block can get both treatments at the same time.

[iatoloye]

Thank you for the question. I forgot to mention that the treatments 0 and 50 were applied only once in the same year, say the year 2000, to different plots. So we have 3 blocks and one treatment with 2 levels (0 and 50) (6 plots in all) and data collected from the same 6 plots annually (2000 till 2020).

[iatoloye]

I meant data collected biennially (2000 till 2020).

[iatoloye]

I have reposted the sample data

Treatment Block Year Yield @@;

0 1 2008 50

50 1 2008 75

0 2 2008 65

50 2 2008 85

0 3 2008 65

50 3 2008 85

0 1 2010 45

50 1 2010 65

0 2 2010 65

50 2 2010 85

0 3 2010 65

50 3 2010 85

[Rick_SAS · Accepted solution]

This is a linear mixed model. Is there a reason you want to use GLIMMIX instead of PROC MIXED? 

The documentation for PROC MIXED contains examples of block designs and repeated measures.

[iatoloye]

Thank you for the link to the documentation. There is no obvious reason for using GLIMMIX.

[SteveDenham]

You will find that GLIMMIX does some rather unusual things with denominator degrees of freedom, at least in my opinion (see the documentation - Details>Degrees of Freedom Methods for a full discussion).  Check the output - in the Model Information table, it will indicate which method is used.  That is most likely the source of the difference you are seeing.

SteveDenham