Observational studies use data on choices and events that happened without any intervention by the researcher.
They exist because many important questions cannot be answered with a trial.
Most data a company holds is observational, from purchases and clicks to customer service records.
Cohort studies follow groups over time, comparing those with and without an exposure.
Case-control studies start from the outcome, comparing people with a condition to similar people without it.
Cross-sectional studies measure exposure and outcome at one moment, which makes the order of cause and effect hard to tell.
The main threats are confounding, selection bias and reverse causation.
Strong observational studies tackle these with careful design, not only with statistics.
Comparing groups that are as similar as possible, such as siblings, removes many confounders at once.
Measuring many likely confounders, and adjusting for them, reduces bias when the list is good.
Sensitivity analysis asks how strong an unmeasured confounder would need to be to explain the result away.
Several studies with different designs pointing the same way give more confidence than one large study.
The link between smoking and lung cancer was established mainly through observational studies.
In practice, observational findings are best treated as strong leads, confirmed by an experiment where possible.
The usual mistake is reporting an observational link in causal words, such as cycling improves health, without saying how it was studied.