Dynamic causal inference
Synopsis
An intervention can change the level, persistence, or dynamics of an outcome, with effects that evolve after treatment. A causal analysis of time series therefore needs to describe entire potential outcome trajectories and state the assumptions connecting them to observed data. The dynamic average treatment effect records how the difference between treatment and control mean trajectories develops over time. Statistical models then support estimation and uncertainty assessment under those assumptions. Sequential intervention studies use dynamic Bayesian models to assess intervention effects across time and cases.
Specific topics
- Potential outcome trajectories
- Potential outcomes are represented as entire stochastic processes under alternative interventions.
- Dynamic treatment effects
- Treatment effects describe changes in outcome levels and in how those outcomes evolve through time.
- Identification and estimation
- Causal interpretation requires assumptions that connect the unobserved potential outcomes to the observed time series.
- Sequential intervention studies
- Dynamic Bayesian models support the sequential analysis of intervention effects in single-case experimental designs.
Selected papers
Sequential Dynamic Bayesian Modeling for Single-Case Experimental Designs: A Novel Approach to Many Labs Research
Multivariate Dynamic Treatment Effects: A Model-Based Causal Estimator, Decomposition, and Marketing Applications
Dynamic Interrupted Time Series Analysis