I'm working on multiple linear regression again. I have to deal with many variables about 10 qualitatives and 20 quantitatives. I would like to know the basis in order to assess the interaction between all these variables. I tried to do something: For each variables -> I performed a simple linear regression with one of my 30 explanatory variable and my response variable and then under JMP i did a "group by". With this manipulation I thought that i would be able to see visually a positive or negative effect of the interaction. But now that I have performed over 30 simple linear regressions, I think that this process doesn't work that way. I think it only works when people want to test a model with only 2 variables in order to know if they have to add the interaction term in the model.
My final goal is to build a model of regression with my response variable (Y) and my explanatory variables (Xi) [all my 30 variables PLUS all the "effective interactions" that I would have found out among my simple regressions].
I hope that somebody will rescue me ! I thank you very much for reading...and I'm sorry for my english again...
You need to use the Fit Model command available under the Analyze menu. This will allow you to have multiple X variables in your model.
The dialog window for Fit Model will allow you to specify the model structure. Common model structures can be generated using the Macro button. The model type you are describing includes all main terms and two-factor interactions and can be generated using the "factorial to degree" option.
These are the steps:
1. Analyze> Fit Model
2. Select the response column(s) and click the Y button
3. Select the column names of the explanatory variables, click the Macro button and select "factorial to degree"