1. Create a data table

From the Welcome or New Table dialog, choose to create a multiple variables data table. If you are just getting started, you can choose to use the sample data for multiple logistic regression

2. Enter data

Each row represents a different observational unit, for example an individual, animal, or experimental replicate

Each column represents a different variable. All variables must be entered as numbers. Usually these will be continuous variables. Note that the exception is that for multiple logistic regression, the dependent (Y) variable must contain values of only 0 or 1 (read more about why).

Other variables (predictor variables) can also be categorical. If a categorical variable only has two possible values, we recommend that you encode the values using a 1 or 0, the same approach used when encoding the dependent variable. For example, if studying mice, you may want to include the mouse sex as a variable, and could enter “0” for male and “1” for female.

If a categorical variable has three or more possible values, you'll have to do some extra work. The simplest approach is called dummy coding (also called indicator coding or reference coding), but there are other alternative methods that could be used such as effects coding. One good source to learn about these coding methods is Glantz and Slinker, cited below.

Note that there is no need to code interactions manually. Prism will allow you to add interactions automatically in the parameters dialog

3. Run the multiple logistic regression

Click Analyze, and then choose multiple logistic regression from the list of analyses for multiple variables tables. The multiple logistic regression dialog has five tabs:

•Model. Choose which variable is the dependent (Y) variable. Add labels for the dependent variable outcomes 1 and 0. Specify which other variables to include as dependent variables, and choose any interactions or transforms you wish to include in the model.

•Compare. Choose a second model and specify how the fit of the two models should be compared.

•Options. Specify which results Prism should report (note, additional results options on Goodness-of-fit tab).

•Goodness-of-fit. Select from a range of various metrics that provide some insight into how well the model fits the entered data.

•Graphs. Select which visualizations you would like Prism to generate from the fit.

Reference

Glantz and Slinker, Primer of Applied Regression and Analysis of Variance, 3rd edition, Chapter “Using linear regression to do one-way analysis of variance with any number of treatments”, page 391