About this Resource

Using Statistical Regression Methods in Education Research

SRME

4.8 Methods of Logistic Regression

Entry Methods

As with linear regression we need to think about how we enter explanatory variables into the model. The process is very similar to that for multiple linear regression so if you’re unsure about what we’re referring to please check the section entitled ‘methods of regression’ on Page 3.2. The control panel for the method of logistic regression in SPSS is shown below.

 Methods of Logistic Regression

As you can see it is still possible to group the explanatory variables in blocks and to enter these blocks in to the model in order of importance. Thus the above screen shot show we are at ‘Block 1 of 1’, but we can use the ‘Next’ button to set up a second block if we want to. The ‘Enter’ option should also be familiar - when selected, all explanatory variables (here labeled ‘covariates’ by SPSS – just to add an extra little challenge!) in the specific block are forced into the model simultaneously.

The main difference for logistic regression is that the automated ‘stepwise’ entry methods are different. Once again the forward and backward methods are present. They differ in how they construct the regression model, with the forward method adding explanatory variables to a basic model (which includes only the constant, B0) and the backwards method removing explanatory variables from the full model (one including all the specified explanatory variables). SPSS makes these decisions based on whether the explanatory variables meet certain criteria. You can choose three different types of criteria for both forward and backward stepwise entry methods: ‘Conditional’, ‘LR’ and ‘Wald’. ‘LR’ stands for Likelihood Ratio which is considered the criterion least prone to error.

We haven’t gone into too much detail here partly because stepwise methods confuse us but mainly because they are not generally recommended. They take important decisions away from the researcher and base them on mathematical criteria rather than sound theoretical logic. Stepwise methods are only really recommended if you are developing a theory from scratch and have no empirical evidence or sensible theories about which explanatory variables are most important. Most of the time we have some idea about which explanatory variables are important and the relative importance of each one, which allows us to specify the entry method for the regression analysis ourselves.

Page contact: Feedback to ReStore team Last revised: Fri 22 Jul 2011
Back to top of page