Should preface all methodologic teaching: "Formal machinery is valuable for making assumptions visible, but it gives false comfort when the assumptions cannot be verified or when it quietly substitutes a different question for the one that was asked."
New blog article documenting shortcomings in a causal inferential method commonly used when standard analyses do not inherit randomization (truncation by death). It demonstrates the value of statistical models that restore intent-to-treat, vs. causal formalism. fharrell.com/post/trunc/ #Statistics @ildiazmView quoted post on X