08-04-2017 04:19 PM
We conducted a trial with a one-way treatment structure with 2 trt levels. We applied the two treatments to pens of pigs in a randomized block design. However, we ended up with 45 pens, thus one of the blocks out of 23 blocks had only 1 treatment. How do I code this unbalanced incomplete block desing in Proc MIXED? Or is the code exactly the same as a RCBD and there is something we need to compute on the datalines?
08-04-2017 04:25 PM
I am somewhat unclear on what exactly do you mean by a "block" in this case, is it a pen of pigs which gets two treatments, or is it two pens of pigs that are somehow paired (if so, how are they paired) or is it something else?
Depending on your answer, I could see removing one block that didn't get two treatments and doing the analysis, or alternatively analyzing all 45 blocks and ignoring the imbalance.
08-04-2017 04:31 PM
08-04-2017 05:25 PM
That makes sense.
I would then analyze all 45 pens, with the average weight in the pen as the coviarate (assuming average weight has some theoretical effect on the results).
08-07-2017 10:01 AM
Yes the pen average weight indeed has an effect on the results, that is why we decided to block. However, I am a bit concerned about analyzing all 45 pens with BW as a covariate. I don't think it is appropriate to remove the block factor and analyze the it as a covariate since blocking was already part of the design structure.