"Causal Inference in Statistics: A Primer" Chapter 3より引用（...は中略を示す）
As you have undoubtedly heard many times in statistics classes, "correlation is not causation." (...) For this reason, the randomized controlled experiment is considered the golden standard of statistics. In a properly randomized controlled experiment, all factors that influence the outcome variable are either static, or vary at random, except for one -- so any change in the outcome variable must be due to that one input variable.
Unfortunately, many questions do not lend themselves to randomized controlled experiments. We cannot control the weather, so we can't randomized the variables that affect wildfires. (...) Even randomized drug trials can run into problems when participants drop out, fail to take their medication, or misreport their usage.
In cases where randomized controlled experiments are not practical, researchers instead perform observational studies, in which they merely record data, rather than controlling it. The problem of such studies is that it is difficult to untangle the causal from the merely correlative. (...)