574345_en-US

liliconcom

Latin Square Design Analysis using GLM - Need Help

Hi Team,

I am trying to replicate Proc GLM solution and unable to succeed. Appreciate your help on this.

I have attached the dataset and the SAS output.

 

Thanks

 

 

Re: Latin Square Design Analysis using GLM - Need Help

Hi Murali,

Here you are. I took a few guesses and it looks like two things were off in your data table. First, there was a variable missing, which was the actual "Rows" variable in the analysis (it wasn't the rows variable in the original table). I assumed that a variable like that would be ordered in the table, which would mean 4 levels, and 4 rows per level, to get 3 degrees of freedom. That guess turned out correct based on the output. Then, I saw that the variable called "Rows" in your initial data table was actually the "Column" variable in the original analysis based on the sums of squares, so I relabeled it. For a Latin square the analysis is a 3 factor main-effects only mode, screenshot included below.

I hope this helps!

Julian

 

 

 

 

Re: Latin Square Design Analysis using GLM - Need Help

In the dataset, change Row to Column and add the Row blocking factor (based on this link). It looks like they're treating both Row and Column as fixed effects. I have always understood these design components to be random blocking factors in Latin Squares and have analyzed them as such. I am attaching the data table with Row and Column as they appear in the link above with table scripts giving the PROC GLM analysis as well as the analysis with Row and Column as random. 

I did a Discovery Talk in 2012 that talked about analyzing Latin  and Graeco-Latin Square designs. The zipped file should have a few examples. If it doesn't, let me know and I'd be happy to share them.