Practical Predictive Analytics

McFadden statistic

The McFadden statistic approximates an R-Square type measure by looking at the ratio of the log-likelihood of the full model (analogous to the sum of squared errors in the OLS R-square model formula previously defined), divided by the log-likelihood of an intercept-only model (similar to the R-square total sum of squares denominator). This approximates the improvement of a logistical model over the intercept-only null model. The McFadden statistic can be considered a pseudo R-square since it will try to measure the same thing, and produce similar R-square results within the same 0-1 scale.

Here is the McFadden formula from the UCLA Institute for Digital Research and Education. (FAQ: WHAT ARE PSEUDO R-SQUAREDS?):

In the ...