- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Report Inappropriate Content
Predictor Screening Question
Just a general question - for the Predictor Screening Platform, if you're trying to find variables related to a particular Response Output (say product Yield) - would you ever bring in another response variable & add it to the "X" box - like "impurity" or something along those lines? Or should you only be bringing in explanatory variables in that "X" box?
1 REPLY 1
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Get Direct Link
- Report Inappropriate Content
Re: Predictor Screening Question
Created:
Feb 15, 2024 03:23 AM
| Last Modified: Feb 15, 2024 12:51 AM
(763 views)
| Posted in reply to message from mjz5448 02-14-2024
Hi @mjz5448,
I will provide you 2 answers, a formal one and a "philosophical/conceptual/methodological" one.
- Formal answer: If you read the JMP Help section about Predictor Screening, you can read that "The Predictor Screening platform provides a method of screening many predictors for their ability to predict an outcome." So to answer directly your question, your factors/explanatory variables go into the X panel, and your responses go into the "Y, Response" panel.
- Conceptual/methodological answer: Predictor Screening is a tool designed to help identification of important predictors for different responses. It is based on Bootstrap Forest (you can read the documentation and many ressources on the web to better learn how Random Forests work).
What is the purpose of predictive models ? There are a lot of possible definitions, but generally, they tend to align on the fact that they are used in order to make predictions about future events or outcomes. In the context of experimental data, predictive models are used in order to predict the most probable outcomes of an imaginary (= not yet practically realized) experiment, based on a model. So if you're using any variable as "X" in the Predictor screening platform, that means this information should be available to be used for the prediction without having done the experiment. It isn't the case for your response, since it's a measurement of a specific experiment done in practice, so this information is not available at the time of the prediction.
It can still be interesting to see patterns in your data and correlations between responses thanks to other platforms like Graph Builder and Multivariate.
Hope this response makes sense to you,
Victor GUILLER
"It is not unusual for a well-designed experiment to analyze itself" (Box, Hunter and Hunter)
"It is not unusual for a well-designed experiment to analyze itself" (Box, Hunter and Hunter)