Specifying the Model in Logistic Regression
main topics 

The logistic regression procedures can fit models with:

·    up to 9 factors and up to 50 covariates

·    crossed or nested factors

·    covariates that are crossed with each other or with factors, or nested within factors

Model continuous predictors as covariates and categorical predictors as factors. Here are some examples. A is a factor and X is a covariate.

Model terms

A  X  A*X

fits a full model with a covariate crossed with a factor

A | X

an alternative way to specify the previous model

A  X  X*X

fits a model with a covariate crossed with itself making a squared term

A  X(A)

fits a model with a covariate nested within a factor

The model for logistic regression is a generalization of the model used in Minitab's general linear model (GLM) procedure. Any model fit by GLM can also be fit by the logistic regression procedures. For a discussion of specifying models in general, see Specifying the Model Terms and Specifying Reduced Models. In the logistic regression commands, Minitab assumes any variable in the model is a covariate unless the variable is specified as a factor. In contrast, GLM assumes that any variable in the model is a factor unless the variable is specified as a covariate. Be sure to specify which predictors are factors in the main dialog box.

Model restrictions

Logistic regression models in Minitab have the same restrictions as GLM models:

·    There must be enough data to estimate all the terms in your model, so that the model is full rank. Minitab will automatically determine if your model is full rank and display a message. In most cases, eliminating some unimportant high-order interactions in your model should solve your problem.

·    The model must be hierarchical. In a hierarchical model, if an interaction term is included, all lower order interactions and main effects that comprise the interaction term must appear in the model.