DOE with covariate

[#1]

I have done some basic searching already, but can't find an example matching this scenario.

Suppose I wanted to run a typical full or fractional 2-level factorial DOE but there was a covariate I wanted to account for. I was expecting that I could just go to custom design and add the factors and specify one of those as a covariate. It would make the table and fill in the +1 -1 as appropriate for the manipulated factors and leave the covariate column blank. Then I would run my treatments and measure the response as well as where the covariate was at and then fill in the two columns and analyze. JMP would know I'm using a covariate and present the results appropriately pulling out the effect of the covariate first. As well as maybe some results regarding a possible covariate x factor interaction and I'd react accordingly.

This doesn't seem to be how it works though. When I try to add a covariate factor it's asking for a file. Or if I have a worksheet open with some columns filled in I could specify which column is the covariate. 

What if I don't have the data yet? I just want JMP to give me the design matrix and then I'll go get the data and fill it in.

Now I know I can just generate my own matrix run it and then specify the fit after the fact and turn on sequential tests, Type 1, Type 3, etc. But that all is very manual, and if JMP has covariate stuff built in already how do I use it?

[#2 · Accepted solution]

You are on the right track. A covariate is our name for the case where you can pre-measure the level of the covariate for each run. Your case is different. Use the Uncontrolled factor type for the case where the covariate must be determined during the run.

[#3 · Accepted solution]

ok thank you. Yes I worked through the JMP help example where it provided a file to open and set as a covariate. I'll check out the Uncontrolled factor and see how that goes.

[#4 · Accepted solution]

Here is what I do (old school).  Just as you suggest, create the factorial as you wish.  Add a column for the covariate and column(s) for your response variable(s).  For analysis, set the design factors as nominal (regardless of their data type).  Leave the covariate as continuous.  

When doing the analysis (fit model), put the covariate term first in the model, then enter the appropriate model for your factorial.  Click on the red triangle next to the response>Estimates>Sequential Tests.  Use the sequential test (type 1) for the covariate and partial tests (type 3) for the factorial terms of the model.

I haven't found a quicker way to do this, though one may exist.

[#5]

ok thanks. yea, I think this is just the method that still applies to my type of use case. The built in covariate feature of JMP seems to target something slightly different as Mark described.