Thomas Lumley • 1/26/2026

Do predictive models need to be causal?

The article examines the necessity of causal relationships in predictive modeling. It argues that while models don't need to be causal to be predictive, their reliability depends on coefficient stability between training and production. The discussion uses examples like predicting drinking behavior and credit scores to illustrate how predictive power can stem from underlying causal chains, even without direct causation.

0 comments

#Machine Learning #statistics #Data Science