What Is Correlation vs. Causation?

The definition, precisely

Two variables are correlated when they move together — ice cream sales and drowning deaths rise in the same months. One variable causes another when manipulating the first reliably changes the second. The gap between those sentences is where the classic fallacies live: the ice cream effect is a confounder (summer drives both), reverse causation explains why low self-esteem tracks poor grades (grades shape esteem as much as the reverse), and selection effects put better outcomes in programs that attracted better-resourced participants to begin with.

The evidence hierarchy that separates the two is short. Randomized experiments assign participants to conditions by chance, which balances confounders across groups and licenses the verb causes. Quasi-experiments — natural experiments, difference-in-differences, regression discontinuity — approximate that balance with design logic and support careful causal claims with stated assumptions. Everything else, including the cross-sectional surveys and cohort analyses that fill health and social-science journals, produces associations: real, useful, and unable on their own to rule out the rival explanation that something upstream moves both variables.

Writing with the distinction is a mechanical skill that graders check sentence by sentence. Match verbs to designs: randomized trial earns reduced or caused; observational work earns associated with, linked to, or predicted. Name the mechanism when you argue causality from association — the best papers explain the pathway, note which confounders the authors controlled, and state what would remain unmeasured. And quote the study's own hedged abstract rather than the press release, because the gap between them is where student papers inherit their overclaims.

The four mistakes graders see most

Every one of these shows up in a stack of sixty papers — usually more than once. Each pairs the error with the fix, so the table doubles as a pre-submission check.

A worked example

A policy paper cites a finding that cities with more bike lanes have fewer pedestrian injuries. The overclaiming version: adding bike lanes saves pedestrians. The calibrated version: cities that expanded bike networks saw pedestrian injuries fall 21 percent relative to matched cities — an observational, quasi-experimental design that adjusted for traffic volume and population, though unmeasured safety culture could still contribute. Same numbers, same citation, but the second version survives a methods-savvy grader and a hostile seminar, because every clause of the claim is bought and paid for by the design.

Pattern-match your own draft against this example rather than against the definition: concrete cases transfer better than abstract rules, and the example shows what the rule looks like when it is done right.

How correlation vs. causation differs from its neighbors

Correlation is not the same as coincidence — a correlation is a systematic pattern across observations, and its strength is measured, not guessed. It is also not causation weakened; association and causation are different claim types, not points on one dial. And the distinction is not a reason to dismiss observational research: most epidemiology, economics and education evidence is associational, and the field's answer is triangulation — many studies, different designs, same direction. Your job as a writer is to report which claim type the evidence supports, not to upgrade it.

Boundary confusion is where most grading deductions live: the term is understood roughly but applied against the wrong neighbor. Pin the boundary above and the term stops being fuzzy in practice, which is what the rubric is actually testing.