Best approach to analyze data skewed to the left with GLM

[palolix]

Thank you so much for keep helping me with this StatDave, I really appreciate it! You are right, including weeks in the model will get me normal residuals. However, the issue that I am facing with this is that since my data is very unbalanced since not all avocado varieties were evaluated at all weeks for every harvest, I am getting a lot of inestimable lsmeans. So I am in a dilemma of whether I include weeks but get inestimable lsmeans, or I remove weeks and don't get normal residuals. 

[StatDave · Accepted solution]

Without having the data and clarity on all of the variables involved in the study (season is new), it's hard to say. But you might be able to obtain estimable results by fitting the model in PROC GLM and using the E option in the MODEL statement to see the general form of the estimable functions. With that info, you should be able to write ESTIMATE or CONTRAST statements that are estimable. It might help to treat Week as a continuous variable by not including it in the CLASS statement. 

[palolix]

I will do that, thank you so much for your help StatDave!

[Ksharp]

Did you try to LOG this Y variable ?

[palolix]

If you mean to transform it, then no.