Design of Experiments based on covariates factors from two tables

[#1]

Hi dear JMP enthusiasts and DoE lovers,

I face a difficult situation and would like to know if there are any easier options. I would like to setup a DoE with 5 covariates :

• 2 covariates, from a table A, are predicted physico-chemical properties of formulations (thanks to the model of a first DoE),
• 3 covariates, independent from the first ones, from a table B, are characteristics of different packagings.

I would like to create a model (screening main effects and interactions) with these 5 covariates to evaluate relationships and interactions between formulation covariates and packaging characteristics on the response, but when I use "Custom Design" and want to add covariates, I only have the option to add covariates from the table on which I create a Custom DoE (table A OR table B). I only see two options to solve my problem (but feel free to propose any others I may have missed):

1. I have to create a new table with all covariates, and so create all possible combinations A/B of the 2 factors from table A with 3 factors from table B (but this may require some time as table A has 10000 rows, and this is not very practical...),
2. There may be an easier solution, where I can somewhere change the current datatable during DoE creation and switch to another one to introduce new factors (or script this) ?

Any inputs or help will be greatly appreciated !

Thanks in advance for your help,

[#2]

Hi @Victor_G ,

  I believe you should be able to do this manually  and probably scripting as well. I made a mock set of three continuous X factors, X1, X2, X3, with 10k rows and then another mock table of factors X4 and X5, with 600 rows. In Custom DOE, when you go to select the covariates, just be sure to select the top level data table before adding the covariate term. This will allow you to select from the data table of interest. I did this for a DOE where I opted for 100 rows and it pulled covariates from both data tables and distributed their levels appropriately in the new DOE, which had factors X1 to X5 in it.

  The only drawback here is that the covariate row index is only provided from the second data table that you have selected. Presumably, if you script this and set the current data table to dt1 or dt2, you should be able to get the row number of the selected covariates and save that as a column. For example, you could have a column named Covariate Row Index dt1 and likewise for dt2. JMP will not do this for you when you do it manually, unfortunately. It would be nice if it did.

Hope this helps!,

DS

[#3]

Hi @SDF1,

Thanks a lot for your response !

I tried this option by switching the table which is "on top" (I have seen it in another topic in the forum), it does enable to enter other covariates from another table, but I don't have all possible combinations that are done between the "big" table A and the "small" table B, so it ends up with a poorly efficient design, and covariates row number is not higher than the number of rows from the small table, which is an indication that not all combinations have been done.

Moreover, doing it this way does create a correlation/link between covariates from table A and covariates from table B (JMP seems to join the covariates in the design creation by row number), which is not the situation I have experimentally, as covariates A are independent from covariates B.

I have found a workaround to prepare the covariates datatable (with all combinations, so nA x nB rows) quickly with another low-code tool, but it would be great if there was a solution where JMP would be able to store the data in memory and do the combinations in the background when building the design based on two covariates tables. 

Thanks a lot for the interest and this option !

[#4]

Hi @Victor_G ,

  This could be a really good submission to the JMP wish list. It seems to me that it's not a fundamentally difficult problem to solve and could be a very nice feature to have that would bring about more flexibility within the DOE platform. At least you have some kind of intermediate work-around, even if it's not perfect/ideal. I definitely think this should be added to the wish list. I can see my organization benefiting from this added capability as well.

Thanks!,

DS

[#5]

Thanks a lot for your input and time @SDF1.
I will create a new item in the JMP Wish List linked to this specific need for covariates in DoE (and link this topic for more details).

Wish list updated with this topic : Custom design based on covariates from two tables - JMP User Community