In Chaps. 4–8, we showed how multi-predictor regression can be used to control for confounding in observational data, with the purpose of estimating the independent association of an exposure with an outcome. The cautious language of associations notwithstanding, the underlying purpose is often to quantify causal relationships. In this chapter, we explain what is meant by the average causal effect of an exposure, and discuss the conditions under which regression might be able to estimate it. We also show the extra steps that are needed to estimate marginal effects, which sometimes differ from the conditional effects that regression models estimate by default.
Keywords
- Propensity Score
- Causal Effect
- Potential Outcome
- Propensity Score Analysis
- Inverse Probability Weight
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.