2011 HSR&D National Meeting Abstract
2008 — Modern Approach to Causal Inference
Sauer B (SLC IDEAS Center), Greene T
(SLC IDEAS Center), Redd A
(SLC IDEAS Center)
Introduce the audience to modern theories and methods for causal inference:
A. Define the counterfactual framework for causal effects;
B. Present assumptions required for causal inference in randomized and non-randomized studies;
C. Present the use of directed acyclic graphs and background knowledge to select a minimal set of covariates to produce conditional exchangeability required for observational studies;
D. Describe the use of statistics developed specifically to address the counterfactual missing data problem.
1. Present probability notation for counterfactual framework and rationale for studying probabilistic causal effects;
2. Define and discuss the crucial difference between causal and associational effect estimates;
3. Discuss how ideal randomized trials can be used to estimate counterfactual causal effects;
4. Describe the assumptions required to estimate average causal effects from observational data;
5. Describe the use of Directed Acyclic Graphs to define variable structural types and identify the minimal subset of variables required for conditional exchangeability;
6. Present the unified therapy of bias based on causal structures;
7. Present statistical techniques specifically developed to estimate counterfactual causal effects;
8. Use simulated example that has both counterfactual and observed data to demonstrate covariate selection issues and statistical modeling;
9. Share the R code with attendees that we used to simulate data for pedagogical purposes.
Clinical researchers, methodologists, statisticians.
Assumed Audience Familiarity with Topic:
The audience should have a basic understanding of randomized controlled trials, observational research, and regression modeling.