Subscribe Bookmark RSS Feed

Create a RSM using already collected historical data

soumyar

New Contributor

Joined:

Jun 4, 2017

Hi,

 

I have a dump of historical values for the explanatory and response variables. How do i build an RSM given this data? Can i build a custom design based on data already collected for a series of runs?

1 ACCEPTED SOLUTION

Accepted Solutions
txnelson

Super User

Joined:

Jun 22, 2012

Solution

It should not be an issue

Jim
6 REPLIES
txnelson

Super User

Joined:

Jun 22, 2012

Yes.  You can build your custom design that matches your historical data, and then once the model is built, you just cut and paste the historical data into the design

Jim
soumyar

New Contributor

Joined:

Jun 4, 2017

Thanks for the reply! Any material that I can refer to for building the custom design?

txnelson

Super User

Joined:

Jun 22, 2012

Help==>Books==>Design of Experiments Guide

Help==>Tutorials==>DOE Tutorial

Jim
soumyar

New Contributor

Joined:

Jun 4, 2017

Creating custom factorial design can handle historical data as random as the below sample data?

-> X1 - X5 - Independent Variable

-> Y - Response Variable

 

 Run OrderX1X2X3X4X5Y
1690430-31123
2730421-21221
3861426-11227
4830428-31329
5581391-31124
6860385-41328
7741446-41429
8850405-41326
9930368-21521
10751382-11228
11931414-31127
12981422-21421
13631397-21529
14731355-21228
15610420-41123
16922390-21522
17752427-21425
18930379-11028
19782431-21030
20770390-11522
txnelson

Super User

Joined:

Jun 22, 2012

Solution

It should not be an issue

Jim
Peter_Bartell

Joined:

Jun 5, 2014

Maybe I'm missing something but since you already have the x and y matrix from your historical data, why do you need to create a custom design when your primary goal is for RSM model evaluation? You can just use the JMP data table for your historical data and then go straight to the Fit Model platform, and then pick the appropriate Fit Model personality, effect specification, etc. You can still use the Evaluate Design platform on your design matrix to evaluate for Power, correlation among effects, etc.

 

And how exactly are you going to create the custom design? I would find it highly unlikely that the custom design platform for an I optimal design is going to have a set of treatment combinations that you can find within your historical data collection of combinations?

 

I'd just be mindful of correlation among the predictor variables for personalities such as Standard Least Squares. One primary advantage of DOE is to AVOID this problem...but historical data doesn't usually come from a designed experiment...you get what you get...and multicollinarity/correlation among predictor variables is often present. All is not lost if you have substantial amounts of multicollinearity...there are still modeling personalities in JMP (like partial least squares) and JMP Pro (the penalized regression methods in the Generalized Regression personality) which are useful in this eventuality.