topic Re: Determine a good matrix of experiment in Statistical Procedures

Determine a good matrix of experiment

Mathis1 — Mon, 25 May 2020 13:47:57 GMT

Hello,

I ran a proc GLM on a unbalanced dataset. I would like to determine what could be a good matrix of experiment for the next data collection, according to the most significant variables outputed by the Proc GLM. I have in total 10 variables. 3 among the 10 are very significant with 10 levels. I thought maybe there exists a way in SAS to specify that I want for instance a complete design for those 3 variables (which would amount to 1000 experiments), and an optimal (although not complete) design for the 7 remaining variables....

I precise that I'm very newcomer to the subject...

Thank you by advance 🙂

Re: Determine a good matrix of experiment

StatDave — Mon, 25 May 2020 14:19:25 GMT

You can use PROC OPTEX in SAS/QC to create an optimal design that can estimate the model effects that you specify. See the discussion and examples in the OPTEX documentation.

Re: Determine a good matrix of experiment

Mathis1 — Mon, 25 May 2020 14:52:25 GMT

Thank you for your message. I read about this procedure, but the point is that I don't have SAS/QC ( I only work on SAS 9.3 at my work). I wondered if there were a way to do the same work on my current version of SAS 😕

Thank you for your help 🙂

Re: Determine a good matrix of experiment

PaigeMiller — Mon, 25 May 2020 15:49:21 GMT

A DO loop in a data step will give you the 10*10*10=1000 conditions in your experiment. I don't think there's any way in SAS to get an optimal design on 7 factors without PROC OPTEX.