Standard betas for logistic regression

[#1]

I would like to compare strength of effect in my logistic regression model.  In least squares models I can bring up and compare standardized beta coefficients.  Can this be done with logistic models?  (I don't see the option in "columns" option).  Or, is there another method to compare the strength of effect of each of my terms in the model?

Thanks

Fishguy

[#2]

This command in Fit Least Squares is convenient, but all it is doing is centering and scaling the continuous predictors. You can accomplish the same thing yourself with a column formula. It is easy to do:

1. Select the predictor columns in the data table.
2. Right-click the header for one of them.
3. Select New Column Formula > Distributional > Standardize.

Now use these columns in place of the original columns when you set up your modeling in Fit Model dialog. Using the Fitness data table in the sample data folder, I fit a model Oxy versus Age through MaxPulse. I created the standardized version of the predictors and used them for the linear predictor of a second fit. I exported the Parameter Estimates as a data table for each and concatenated them so I could examine them together. Here are the results:

Notice that the p-values are identical between the two fits. Then I plotted the estimates, where the importance is clearer:

Only Runtime seemed to be important when using the original predictors. Runtime, RunPulse, and MaxPulse seem to be important when using the standardized predictors.

Both models produce identical predictions, of course.

Another technique is built into the Prediction Profiler that might be helpful for your purpose: Variable Importance.

[#3]

Thanks for the reply!  So my model has a mix of predictors that are continuous and a couple that are binary.  Do I standardize all the predictor variables in the model?

Thanks again:)

[#4]

Categorical variables are always coded so no need. See help here and here for more information.

[#5]

Great, thanks!