I need to do a key driver analysis and identify the top 10 drivers for customer satisfaction. The data set that I have consists of 1 ordinal dependent variable and about 20-30 independent variable (ordinal, nominal, binary).
I tried to analysis the data set by Fit Y by X and Fit Model. However, I do not know what to look out for to identify the most important drivers.
For numerical data, I am aware that I can use the beta coefficients. However, I do not know what to do/ where to find similar information for this case.
Please help me!
Both the Fit Y by X and Fit Model platforms can provide useful insights. Have you looked through the JMP documentation on logistic regression? There are lots of model fitting ouputs to look at to help you decide if you have a useful model for your key driver analysis. Here's a link to the landing page in the documentation for logistic regression in general, with sub topics for ordinal logistic regression.
You may also want to take a look at the partition platform and if you are running JMP Pro some of the penalized regression methods as well as the other tree based modeling methods like boosted tree and bootstrap forest.
Thanks for the reply. I've looked through all the web pages in the link given. However, it doesnt contain any information on how I can find the relative importance.
I am using JMP Pro now. However, I dont know which columns/ information to look at so that I can find the top drivers/ importance. I am having a big problem now because I have never worked with ordinal variables before.
I suggest taking a look at the worked example in the cheese.jmp data table supplied with your JMP Sample Data Directory. The thinking and workflow for that example might be similar to what you are trying to accomplish with your problem?
Make sure to take a look at the two embedded scripts as well. The Contingency script does a nice job of showing the mosaic plots...and you'll get additional plots for your continuous variable predictors as well. The Fit Model script uses the Fit Model -> Ordinal Logistic personality to build and evaluate the model. Once you fit your model to your data, I suggest taking a look at various model fit evaluation statistics and visualizations such as the Effect Summary window, the Whole Model test, the Lack of Fit test, parameter estimates, ROC curves, confusion matrix, and the JMP Pro prediction profiler. In short there is no ONE thing to look at to decide if your model does a satisfactory job of helping you identify your 'key drivers'.
And last but not least...compare what the model is telling you to your knowledge of the process, rational human behavior, and expectations...if the model if flying in the face of process knowledge then maybe there is something amiss? For example suppose you are evaluating customer satisfaction with a telephone customer contact center and one of the predictor variables is 'wait time to talk to an agent' and the model suggests that satisfaction increases with increasing wait time...well something is fishy.