Causal Methods
Design
How the study population and the comparison are assembled, before any modeling begins.
- Randomized designs (RCTs, cluster, stepped-wedge)
- Self-controlled designs (case-crossover, SCCS)
- Target trial emulation
Identification
The assumption that licenses reading a causal effect out of observational data.
- Back-door Adjustment (conditional exchangeability)
- Difference-in-Differences
- Front-door Adjustment
- Instrumental Variables
- Interrupted Time Series
- Regression Discontinuity
- Synthetic Controls
Estimation
How the identified quantity is computed once those assumptions are in place.
- Augmented Inverse Probability Weighting (AIPW)
- Bayesian Structural Time Series
- G-computation
- Inverse Probability of Treatment Weighting (IPTW)
- Marginal Structural Models
- Matching
- Propensity Scores (the shared ingredient in matching, IPTW, and AIPW)
- Targeted Maximum Likelihood Estimate (TMLE)
Sensitivity
What happens to the conclusion when those assumptions fail.
- E-value
- Negative Controls
- Positive Controls
- Quantitative Bias Analysis