What logistic regression does

Logistic regression helps with predicting which of two categories a case is likely to belong to given certain other information (Field, 2000, p 163). The term ‘predict’ is used in the statistical sense that ‘within [a] particular model the variable is strongly, significantly and independently associated with the outcome, and may therefore be viewed as influential in the “pathway” to that outcome’ (Ghate and Hazel, 2002, p 293). It does not mean that the presence of a predictor variable automatically leads to the outcome in question, or that the former causes the latter. Logistic regression is suitable when the outcome or ‘dependent’ variable is categorical, and the predictor or ‘independent’ variables are categorical or continuous.

Why logistic regression is useful

Although several variables may be significantly associated with a target variable, any one may not have independent predictive power once other factors have been controlled for. In other words, the association may be the product of their mutual association with a third factor. For example, since income tends to increase with age, so anything else that generally increases with age, such as dental decay, can be used with some success to predict income. A possible but wrong conclusion would be to say that not brushing one's teeth might be a good career move (more decay equals higher income). Multivariate analysis reduces the chances of making such an error; it is unlikely that dental decay has any predictive power that cannot be explained in terms of age (Bullock et al, 1998). Logistic regression is a multivariate technique that assists with disentangling which are the key factors that predict a target variable that is categorical, and which are statistically associated with the target variable primarily due to a shared underlying factor (Ghate and Hazel, 2002).

How a logistic regression is done

First, correlations are carried out to identify factors that are significantly associated with the outcome variable. Second, those variables are fed into the logistic regression model in SPSS (they can be added all at once or individually). There are various techniques for doing this, but a common one is the ‘forward stepwise’ approach.