Introduction to Causal Inference

Causal inference is one of the most important intellectual frameworks in modern science and medicine. At its core, causal inference is about answering a deceptively simple question: "What would happen if we intervened?"

This question lies at the heart of virtually every important decision we make in healthcare, public policy, and scientific research. When a physician considers prescribing a medication, they are not merely asking whether patients who take the drug tend to have better outcomes—they need to know whether the drug causes those better outcomes. When a policy maker considers implementing a new public health program, they need to know whether the program will actually improve population health, not just whether communities with such programs happen to be healthier.

The Two Fundamental Questions

Consider how different these two questions are:

1. Associational question: "Among patients who take statins, what is their risk of heart attack?"
2. Causal question: "If we were to give a patient statins, how would this change their risk of heart attack?"

The first question can be answered by simple observation and statistical analysis. The second question requires a fundamentally different approach—one that accounts for all the reasons why patients who take statins might differ from those who don't.

"Causal inference can be viewed as a missing data problem: we can never observe the same individual under both treatment and control conditions." — What If, Chapter 1

This insight transforms how we think about research design and data analysis. We move from simply describing patterns in data to reasoning carefully about what those patterns tell us about cause and effect.

Why Traditional Statistics Falls Short

Traditional statistical methods excel at describing associations—correlations, regression coefficients, odds ratios calculated from observed data. But these methods, on their own, cannot distinguish between:

• A treatment that genuinely improves outcomes
• A treatment that is preferentially given to patients who were already likely to improve
• A treatment associated with outcomes due to a common cause affecting both

The methods you will learn in this course provide the conceptual and practical tools to make this distinction when certain assumptions are met.