Where is logistic regression?
What are the available options for logistic regression in SPSS Statistics?
Resolving the problem
SPSS Statistics has a number of procedures that offer options for fitting logistic regression models. Which procedure you will want to use will depend upon the type of logistic regression model you want to fit, and the specific options you want the procedure to have.
Logistic regression models can be applied to categorical responses that are binary (two response categories) and to responses with more than two categories. When there are more than two response categories, responses may be either ordinal or nominal (not ordered). Binary (binomial) models are special cases of both ordinal and nominal response models for more than two categories, so any procedure that will fit ordinal multinomial or nominal multinomial response models will also fit binary response models.
The Base system offers the PLUM or Ordinal Regression procedure, which includes logistic models among the five types of models available.
The procedures most specifically designed for logistic regression modeling are the LOGISTIC REGRESSION (Binary Logistic Regression in the menus) and NOMREG (Multinomial Logistic Regression) procedures. LOGISTIC REGRESSION fits binary response models and includes stepwise fitting methods. NOMREG fits nominal response multinomial logistic models, and also includes stepwise modeling capabilities. These procedures are both in the Regression Models option, and are the only currently available options offering stepwise modeling. This option also includes the PROBIT (Probit Regression) procedure, which is designed for grouped dose-response data using either a probit or logit model, and would typically be used only for that specific purpose.
The GENLIN procedure for generalized linear modeling (Generalized Linear Models) and generalized estimating equations or GEE (Generalized Estimating Equations) offers binary and ordinal response logistic models. The GEE functionality is appropriate for handling large-sample problems involving correlated data such as from repeated measurements.
Beginning with Version 19.0, the GENLINMIXED procedure, which fits generalized linear mixed models, is available in the Advanced Statistics module. GENLINMIXED will fit binary logistic regression models with or without random or repeated effects. In Version 19, only nominal multinomial response models are supported. In versions beginning with Version 20, ordinal multinomial response models are also supported. Random effects are allowed in all of these models. However, repeated effects are not allowed with multinomial response models.
If data are from complex samples, the CSLOGISTIC (Complex Samples Logistic Regression) procedure is available for nominal multinomial modeling, and the CSORDINAL (Complex Samples Ordinal Regression) procedure is available for ordinal multinomial modeling.
Translate this page: