Hazards on the Causal Path
summary
A Bayesian implementation of Dynamic Path Analysis in PyMC models discretised time hazards with smooth spline priors and a Poisson discrete‑time likelihood, benchmarks against canonical dpasurv examples, and recovers the original decomposition of direct and indirect hazard effects while providing full posterior uncertainty. The model samples and predicts, shows the indirect harm peaking around ~100 days after intervention, and supports G-computation to produce interpretable time‑varying causal estimands.
AI-written summary · about
chapters (6)
- 0:00Survival and causal inference intro
- 1:25Mediation puzzle in causal inference
- 7:19Dynamic path analysis in PyMC
- 13:17Model results and interpretation
- 16:47Mediation structures in survival analysis
- 18:30Summary and recommendations
related talks
01h 24mBayesianism and Survival Analysis
42mBayesian Non-parametric Models for Data Science using PyMC3
30mClimbing the causal ladder for fun and profit
01h 27mBayesian Time Series
36mWhat-if- Causal reasoning meets Bayesian Inference
27mUsing Causal thinking to make Media Mix Modeling
31mCounterfactual Analysis with Bayesian Models
31mIntroduction to Causal Inference using pgmpy