I’ve created a binary logistic regression model with two continuous independent variables (age and annual salary) to predict car purchase (0-not purchased and 1 for puchased) using SPSS.
Testing for the linearity of the logit (using a logistic model with interaction terms consisting of the variables x the natural logarithm of the variable, as e.g. described by Andy Field’s IBM SPSS Statistics) reveals, that the interaction term has a significance of p=0.001. Therefore, as I understand it, the assumption of the linearity of the logit is not met and the model is not valid and cannot be used this way.
What options do I have to deal with this/to transform the model?
In suc cases, Will the model be considered as statistically significant?
Is the model robust?
Linearity of the logit means that linear data goes to linear data under the logit transformation ..it has nothing to do with the predictors side of reg model. See the attached for some references. Best wishes David Booth
Right. As it necessary to explore alternative models or further investigate the relationship between the variables to ensure that the model is robust.
If the assumption of linearity of the logit is not met for one predictor in a binary logistic regression model, there are several options that can be considered to address this issue. Here are some possible solutions: