I am studying experimental design this morning.

I did some SAS coding and must write this up now. I need to study the SAS printout some more, but I did notice that I have unequal variances with two blocking variables in my latin square design. I am wondering if I can look at the trends in these blocks and perhaps chose a transformation that might correct the unequal variances. I also know there is a technique for finding an appropriate transform but I may just try guessing.

Just posting an update now. I tried the standard transformations of the response variable, squaring, taking the square root, taking the log, taking the log to base 10 and none of these worked to reduce the unequal variances of the residuals. I also tried coding the data by subtracting 10 and this did not work. I will try subtracting 16 now and then write this whole transforming attempt up.

I then need to check the reduced and full model F test. I also need to show how to solve this for a missing value using the Q prime analysis. But then I am done with this and I should be able to do it today.