Help With Analysis for Replicated Latin Square Design containing nested data and repeated measures

[ChKayAb]

I'm having trouble debugging my analysis.

Am I using the right procedure? 

I designed a Latin square observational experiment that was replicated for 3 years.

3 new colonies were assigned to each farm each year, and I had 3 farms in each of my 3 treatments.  

Colonies are my sub replicate (nested in farm). 

I then measured colony weight weekly. 

However, my data is not entirely complete. 

In year 1, I only have measurements for weeks 1, 5, 9, and 13.

In year 3, I only assigned 2 colonies to each farm.

I've managed to come up with the below model. However, my results don't show a test statistic for Treatment effect. 

Can you tell me if I'm using the right procedure and is my model is incorrect? 

proc mixed;
class Treatment Colony Farm Year Week Weight;
model Weight = Treatment Week Treatment*Week Treatment*Year Treatment*Week*Year Farm Year Week / ddfm=kr;
random int / subject=Colony(Farm);
random int / subject=Year;
repeated week / subject=Treatment(Colony Farm Year) type=ar(1);
run;

[SteveDenham]

This is a "not sure if it will work" answer, but the degrees of freedom look a bit wonky, and it is probably because you are using the Kenward-Roger denominator degrees of freedom with a very sparse dataset due to the missing values.

Try, in order, ddfm=satterthwaite, ddfm=bw, or using ddf= with fixed values based on a skeleton ANOVA table. You could then add in an EMPIRICAL option to get the shrinkage estimates.

SteveDenham