Split- plot mixed with split block as repeated measures (6 factors)

[palolix]

Dear SAS communities,

I have an experiment with 2 factors that are fixed (depths and date), date is also the factor analyzed as repeated measures. 4 factors are random (treatment, irrigation, dose and method of application). The response variable is nematode count (RKN). How would our model looks in Glimmix?

Here is what I have so far:

proc glimmix data=one;
class Block  Depths    Date trt    dosis    Irrigation method;
model RKN=Depths|Date|trt|dosis|Irrigation|method/dist=lognormal ddfm=kr;
random intercept Depths trt dosis Irrigation method  Depths*dosis Depths*Irrigation Irrigation*dosis Depths*Irrigation*dosis/Subject=Block;
random Date/residual subject=Block(Depths trt dosis Irrigation method Depths*dosis Depths*Irrigation Irrigation* dosis Depths*Irrigation*dosis) type=ar(1);
run;

I would greatly appreciate if you could help me with this model.

Thank you very much!

Caroline

[sld]

Wow, what an ambitious experiment. In my opinion, a full 6-way factorial design is treacherous and typically excessively optimistic. Whatever will you do if your 6-way interaction is significant? (Which it could be, just due to random noise particularly if the number of replications is small.)

I would think that treatment, irrigation, dose and method of application would be fixed effect factors, not random effects factors. And in fact, your MODEL statement includes these factors and so your model is incorporating them as fixed effects.

You appear to have a blocked design, but I suspect the design might be much simpler than your RANDOM statements would imply. To provide more help, the Community will need more information about your experimental design, particularly the physical layout.

This looks like a fairly complex model for you, and I highly recommend studying SAS System for Mixed Models, 2nd ed or the (hopefully) soon-to-be-released update SAS for Mixed Models: An Introduction . 

[palolix]

Thank you so much for your feedback on this! Yes you are right, the model was far too complicated, that is why we decided to leave only 4 factors, so now it worked.