The steps of carrying out a logistic regression are similar to MLR (Section 5.27), but with the following differences.
- The outcome in a logistic regression is categorical, not continuous, and should not be transformed.
- Check for separation and resolve any issues (see Section 6.10).
- Make sure the outcome is coded as a factor and you know which probability
glm()is modeling (see Section 6.6.2).
- If any step leads to a change in sample size or in the included categorical predictors then re-evaluate separation.
- Evaluate linearity (Section 6.12), outliers (Section 6.13), and influential observations (Section 6.14). In logistic regression, there are no normality or constant variance assumptions
- Test goodness-of-fit using the Hosmer-Lemeshow test and assess goodness-of-fit visually using a calibration plot (see Section 6.16).