Stepwise tool throwing error after Count Regression

anishgrandhi

Hi All,

I get the following error after using Stepwise tool -

The workflow is attached the post. The data is basically a survey data where participants answer yes or no. Yes is marked as 1 or else it is 0. We want to predict the variable Case. We use SAS code for this analysis but I am trying to replicate the process in Alteryx.

Is there a better way to do this/ what mistake am I doing here?

NBRegression.yxzp

CharlieS

Hi @anishgrandhi 

I can help with what's causing the error and how to resolve it, but for suggestions beyond that, we might need to call in some support.

Troubleshooting steps:

- What's sending the error? You can see in the error message that Tool #5 inside the Stepwise tool is the source. If we right click > open the Stepwise tool (read: macro), we can find the lines of R code that control this specific message:

} else if (!(any(class(orig.model)[1] %in% c("glm", "svyglm", "lm", "rxLogit", "rxGlm", "rxLinMod")))) {
  stop.Alteryx("Stepwise variable selection is only relevant for models created using the Linear, Logistic, and Count Regression tools.")

- Now that we know that that's the list of model objects that are accepted, we need to find out what is being sent by the Count Regression tool. This Community post shows how you can use an R tool to display the model object properties. When I add this to the the workflow, this is what we see:

- Ah, the "glm.nb" is not in the approved list. What's the ".nb"? That's the negative binomial model type you selected on the Count Regression tool. Change that to Poisson and the Stepwise tool is happy. 

It's interesting that if the "Quasi-Poisson" option is selected, there's a specific error message in the Stepwise tool saying "Stepwise methods do not work for models in the quasipoisson family". Which tells me that maybe the error handing in the Stepwise tool should be updated for Negative Binomial selections. 

20200709-CountStepwise.yxmd