Hi! I'm fairly new to JMP and would appreciate any help you could give. I've read tutorials and scoured the JMP community but am still having trouble. I have a set of market research survey data and I want to do a drivers analysis on it. I understand linear regression is the way to do a drivers analysis, i.e. determine which variables have the most effect on a dependent variable. In this case the independent variables are rating questions (e.g. "how do you rate the taste, on a scale of 1-5?") and the dependent variable is purchase intent ("how likely would you be to purchase this product, on a scale of 1-5?"). All the variables are nominal. Each of the variables has 5 possible responses -- at least for now until I understand this and expand the analysis to include more variables.
Is there anyone who has run Drivers Analysis on market research data who can help?? Or if not for that specific application, could I get help understanding regression models?
Issues I'm having as I go about this:
1) I'm choosing Nominal logistic as the personality when running the Fit Model. I have 1 Y and 12 Construct Model Effects variables. I was expecting to get a beta coefficient for each of the 12 model effects variables, plus a p value to determine if it is meaningful. But I don't see a single beta coefficient estimate. Instead, under the Parameter Estimates section, there are estimates for every one of the possible responses (e.g. Taste [Excellent], Taste [Very Good], Taste [Good], Taste [Fair], Taste [Poor]), and each of those variable sets is repeated 4 times. Thus, the list of Terms for Parameter Estimates is very long. I was expecting it to be 13 items, one for each variable plus one for the intercept, and instead it is several screens long. How do I interpret these estimates?
2) The Effect Likelihood Ratio Tests section lists each variable and a p value, so here I can clearly see which variables are important and fit in the regression to predict the dependent variable. But it doesn't tell me relative importance.
3) Does the Effect Summary section show relative importance? How do I interpret the LogScale column? I would like to say the most important variables are these 5 (or whatever has a p value of 0.1 or less), with Variable A the most important and twice as important as variable B, etc. I expected to get that from beta coefficients, e.g. a standardized coefficient of 0.2 for Taste would mean an increase of 1 in Taste rating would lead to an increase of 0.2 in Purchase Intent. And if the coefficient for Aroma were 0.1, then it would be half as important because an increase of 1 in Aroma rating would lead to an increase of 0.1 in Purchase Intent.
Once I figure this out, I'll need to know if I can run the regression using variables that have different nominal scales. But now I need to understand the basics!
Please help! Thanks!
One of the first things you can try is determining if all your independent variables are active and important to the drivers analysis. Set up your regression as you have already then go to the Personality drop down and select Stepwise instead if Nominal. Select OK. This will take you another window for the Stepwise process. You will see several options there to change things. For now accept the defaults and select Go in the middle left. It may take a few seconds to process, but once the algorithm completes select Make Model in the upper right. This will bring up a new window where you will see which factors are important to your drivers analysis. Again select OK and you will be taken to the same model report you are currently seeing when you build your model.
There are also a couple of Mastering JMP on demand videos that you should watch that will show more about regression both with continuous and nominal data.
This isn't the answer to all (or maybe any) of your questions, but many of them will be answered by the videos.
Try the video in the link below to start.