What correlation and causation mean
Two things moving together can be a clue, a coincidence, or a consequence of something else. The difference is the beginning of good reasoning, not a technical footnote.
If umbrella sales and traffic accidents rise on the same days, umbrellas do not cause crashes. Rain changes both. The example is simple, but the same mistake appears in dashboards, news reports, academic studies, and product analytics whenever a relationship is treated as an explanation.
Correlation is a description
Correlation tells us that two measured variables change together in a pattern. The pattern may be positive, negative, or close to zero. It can be useful evidence: a clue that deserves investigation.
But correlation does not tell us which variable came first, whether the relationship is meaningful, or whether a third factor explains both. It also does not tell us what would happen if we deliberately changed one variable.
Causation is a counterfactual claim
To say that X causes Y is to make a statement about what would happen if X were different while the relevant alternatives stayed comparable. That counterfactual is difficult to observe directly because the same person, firm, or society cannot occupy both worlds at once.
Researchers use experiments, natural experiments, statistical controls, time ordering, and causal models to make the comparison more credible. None is a magic guarantee. Each method works by making some alternative explanation less plausible and stating the assumptions that remain.
The confounder problem
A confounder is a factor connected to both variables of interest. Education and income may be related, but geography, family background, health, industry, and opportunity can affect both. Controlling for a variable can help, but controlling for the wrong variable—or for a consequence of the treatment—can create a new distortion.
This is why a good chart is not automatically a good study. The chart may accurately display the data while the headline overstates what the design can support.
How to read a causal claim
Ask four questions: What exactly was measured? Which people or cases were included? What other explanation could produce the pattern? What changed, and how was that change compared with what would otherwise have happened?
The goal is not to distrust every statistic. It is to match the strength of the conclusion to the strength of the evidence. Correlation is often where a useful question begins. Causation is what must be earned before the question becomes a policy or personal decision.
Sources & methodology
The sources below anchor the explanation. They are starting points for verification, not decoration.
- 01 Royal Society Open Science — Correlation, causation, and causal inference
Peer-reviewed discussion of how causal questions differ from associations in data.
- 02 U.S. National Academies — Reproducibility and replicability in science
Institutional context for testing, uncertainty, and the reliability of scientific findings.
- 03 OECD — Statistics and data
Public data context for definitions, comparability, and evidence-based analysis.
