Leah
summary
Implements the Bayesian g‑formula in Stan, fitting logistic outcome and mediator models and using posterior predictive draws in the generated quantities block to produce population counterfactual means and average treatment effects. Uses bootstraped confounder samples and shows mediation via natural direct effects, handling a binary unmeasured confounder U by marginalizing and by injecting informative priors from external fits (point estimates plus covariances) for data‑driven sensitivity analysis; code and notebooks live on GitHub.
AI-written summary · about
chapters (14)
- 0:00Causal inference with g-formula
- 1:13Correlation vs causation policy
- 2:37Potential outcomes crash course
- 4:01G-formula standardizes data
- 7:32Causal assumptions limitations
- 7:53Stan for causal inference
- 9:16Observational data setup
- 11:43Bayesian g-formula generated
- 13:26Average treatment effect nuances
- 13:46Mediation natural effects
- 16:55Unmeasured confounding priors
- 19:46Data integration confounders
- 20:26Summary Stan’s advantages
- 20:47Q&A g-formula matching
related talks
31mBias Meets Bayes: A Bayesian Perspective on Improving Model Fairness
25mScaling Bayesian Experimentation
01h 30mBayBE: A Bayesian Back End for Experimental Planning in the Low-To-No-Data Regime
17mMaking sense of marginal effects
31mBayesian Data Analysis of Nonparametric Models in Clojure
27mI have no clue what my model is doing...
43mToward Human-Centered Machine Learning
27mUsing Causal thinking to make Media Mix Modeling