What's The Weight Of Evidence?

How does using weight-of-evidence to transform categorical data in a logistic regression differ from Bayesian logistic regression?

• Conceptually, it seems like a weight-of-evidence transformation is inherently Bayesian, but can be used in a "regular" logistic regression. Is there a difference between a logistic regression with weight-of-evidence transformed variables and Bayesian logistic regression?

• Answer:

  Given that log(Bayes Factor) = weight of evidence (source: https://en.wikipedia.org/wiki/Bayes_factor#Interpretation), your intuition is largely correct. However, based on how you set up the regression, Bayes method for regression might be more sensitive to small perturbations (noisy data). I would extrapolate that guess and say that when Bayes regression start to run into the risk of over-fitting,  weight-of-evidence methods might make for a more generalizable model. Please take this response with a good pinch of salt. I am willing to be corrected.