For the latest version of JMP Help, visit JMP.com/help.


Fitting Linear Models > Stepwise Regression Models > Perform Binary and Ordinal Logistic Stepwise Regression
Publication date: 06/21/2023

Perform Binary and Ordinal Logistic Stepwise Regression

The Stepwise personality of Fit Model performs ordinal logistic stepwise regression when the response is ordinal or nominal. Nominal responses are treated as ordinal responses in the logistic stepwise regression fitting procedure. When a response has only two levels, ordinal logistic regression models are equivalent to nominal logistic regression models. To run a logistic stepwise regression, specify an ordinal or nominal response, add terms to the model as usual, and choose Stepwise from the Personality menu.

The Stepwise reports for a logistic model are similar to those provided when the response is continuous. The following elements are specific to logistic regression results:

When the response is categorical, the overall fit of the model is given by its negative log-likelihood (-LogLikelihood). This value is calculated based on the full iterative maximum likelihood fit. See Likelihood, AICc, and BIC.

The Current Estimates report shows chi-square statistics (Wald/Score ChiSq) and their p-values (Sig Prob). The test statistic column shows score statistics for parameters that are not in the current model and shows Wald statistics for parameters that are in the current model. The regression estimates (Estimate) are based on the full iterative maximum likelihood fit.

The Step History report shows the L-R ChiSquare. This value is the test statistic for the likelihood ratio test of the hypothesis that the corresponding regression parameter is zero, given the other terms in the model. The Sig Prob is the p-value for this test.

Note: If the response is nominal, you can fit the current model using the Nominal Logistic personality of Fit Model by clicking the Make Model button. In the Fit Model launch window that appears, click Run.

See Example of Logistic Stepwise Regression.

Want more information? Have questions? Get answers in the JMP User Community (community.jmp.com).