Pseudo R-squared is a family of goodness-of-fit measures for logistic regression. Unlike linear-regression R-squared, it does not represent the proportion of variance explained; it evaluates how well the model explains a binary outcome. section 5.2.3
Statsmodels reports McFadden’s pseudo R-squared. It is calculated as 1 minus the model log-likelihood divided by the intercept-only model’s log-likelihood. The null model supplies a baseline, so a larger value indicates greater added predictive value from the independent variables and better fit. section 5.2.3
Interpret pseudo R-squared by:
- Comparing the fitted model’s log-likelihood with the intercept-only model’s log-likelihood. section 5.2.3
- Treating higher McFadden’s pseudo R-squared as evidence of better fit within comparable models. section 5.2.3
- Using the metric to compare models fitted to the same data, while avoiding isolated interpretation of raw log-likelihoods. section 5.2.3
