Observational Studies Fundamentals

In an observational study, the researcher observes treatments that were assigned by nature, circumstance, or individual choice—rather than by random assignment. This fundamental difference from randomized trials creates both opportunities and challenges for causal inference.

The Defining Characteristic

The essence of an observational study is that the researcher does not control treatment assignment. Instead, we observe what naturally occurred and try to draw causal conclusions from these observations.

Aspect	Randomized Trial	Observational Study
Treatment Assignment	Controlled by researcher	Determined by nature, patients, or clinicians
Exchangeability	Achieved by design	Must be assumed (conditional)
Confounding	Eliminated by randomization	Present and must be addressed
Ethical Limits	Cannot randomize harmful exposures	Can study any exposure that occurs
Generalizability	May be limited by strict eligibility	Often reflects real-world populations

Types of Observational Studies

Observational studies come in several forms, each with its own strengths:

Cohort Studies: Follow a group of individuals over time, comparing those exposed to those unexposed.

• Prospective: Enroll before outcomes occur, follow forward
• Retrospective: Use existing data, look backward

Case-Control Studies: Start with cases (those with the outcome) and controls (those without), compare their exposure history.

Cross-Sectional Studies: Measure exposure and outcome at the same time point.

Modern Data Sources

Today, observational data come from diverse sources:

Source	Description	Example Use
Electronic Health Records (EHR)	Clinical data from healthcare encounters	Drug safety surveillance
Claims Databases	Insurance billing records	Comparative effectiveness
Registries	Disease or procedure-specific databases	Cancer treatment outcomes
Cohort Studies	Research-grade longitudinal data	Epidemiologic research

"In observational studies, we must assume that treatment groups are conditionally exchangeable given measured confounders—an assumption that cannot be verified from data alone." — What If, Chapter 3