i am dealing with a full-factorial design with two response variables and four independent variables,i want to fit two model with different terms at the same time but i dont find a way to realize it.the reason i want to do it is that the resulting contour plot have two response variable so that i can find design space that meet two srandard simultaneously. i also tried some other methods to construct a contour plot with two response variables but none of these methods is satisfactory
Hi @ZHANDOUJI,
As explained by @statman, Stepwise method is not a default model building method recommended with DOE data, as you already have assumed a model that guide your experimentation and data collection strategies. You're not a in a situation where the data collection is imposed to you, and you have to "figure out"/determine a possible likely model for this data.
Starting with your assumed model with the simple Least Squares method should be a good starting point.
You can check the "Fit separately" option.
In order to refine your model (based on p-values, R² and R² adjusted, RMSE, and/or Information criterion like AICc), you can then for each of your response remove the least important/active terms from your model :
You can read more about model building and selection in other related discussions :
Analysis of split plot design with full factorial vs RSM
Fit Definitive Screening vs. Stepwise (min. AICC) for model selection
Hope this answer will help you,
Thanks for taking the time to help me.I have retried to fit model with your advice and I have achieved my goal.I refited my model by stepwise method which allow you to select the terms twice before the model fitted instead of the standard least square method. pick the terms I want then click “run model” I can get two model with different terms in one report,and by the method you mentioned using the contour profile in "fit group” I get the contour plot with two response variable.
is this protocol i did right? is that can be realized directly by fit the least square method?
Just a note, you shouldn't use stepwise for analysis of DOE. You should know the model before running the experiment. It is the model you select that decides how to run the experiment.
yes,I known.most time design works for model.in this problem I used the least square method find which effect is significant then I choose these significant terms by stepwise method.the reason i use it is it give me a second chance to pick the terms for different models,and they can be presented in one report
Hi @ZHANDOUJI,
As explained by @statman, Stepwise method is not a default model building method recommended with DOE data, as you already have assumed a model that guide your experimentation and data collection strategies. You're not a in a situation where the data collection is imposed to you, and you have to "figure out"/determine a possible likely model for this data.
Starting with your assumed model with the simple Least Squares method should be a good starting point.
You can check the "Fit separately" option.
In order to refine your model (based on p-values, R² and R² adjusted, RMSE, and/or Information criterion like AICc), you can then for each of your response remove the least important/active terms from your model :
You can read more about model building and selection in other related discussions :
Analysis of split plot design with full factorial vs RSM
Fit Definitive Screening vs. Stepwise (min. AICC) for model selection
Hope this answer will help you,
OH,Thank you very much! Idont find the "remove" button before.I know these basic knowledge about fitting model and model selecting.I used to use Design expert,I am just not familiar with this software.
I have tried to click the option “fit separately” and I get two model with same terms.I cant refine my model directly in the report,so I use the stepwise method and this allow me to re-select the terms I want for each model.
Hi @ZHANDOUJI : Is this another homework problem?
yes