IMPACT Design and Statistics Core member Keith Goldfeld, DrPH, MS, MPA, continues his four-part blog series exploring Targeted Minimum Loss Estimation (TMLE)—a causal inference approach developed for observational studies that is increasingly relevant for randomized trials as well, including cluster randomized and stepped-wedge designs.

In the latest post, Dr. Goldfeld recaps key ideas from earlier installments, including how TMLE goes beyond simply improving “nuisance” models by making a targeted update so the empirical mean of the estimated efficient influence function is brought back to zero.

He then turns to simulation to examine what the targeting step changes in practice. By comparing two estimators of the average treatment effect (ATE) with TMLE, the post aims to clarify what the targeted step is doing mechanically and how it shapes the final estimate.

Getting to the bottom of TMLE: targeting in action | March 18, 2026