Correlation Versus Causation in an Argument

When two things happen together, it is tempting to say one caused the other. Correlation is evidence that variables are associated; causation is a claim about what would happen to one variable if the other were changed. The second claim is stronger. On a reading or argument question, ask whether the evidence supports the relationship the author actually asserts.

Three explanations for an observed relationship

What a causal claim needs

• Temporal order: the proposed cause must occur before the effect.
• Association: the variables must actually be related in the data.
• A plausible mechanism: there should be a reason the cause could produce the effect.
• Alternative explanations considered: confounders, selection effects, measurement problems, and reverse causation should be examined.
• Evidence proportional to the conclusion: a descriptive pattern usually supports a cautious association claim, not a definitive causal conclusion.

Study design changes how much the evidence can establish

A randomized controlled experiment can strengthen causal inference because random assignment helps balance known and unknown confounders between groups. It does not guarantee perfect measurement or eliminate every source of bias, and the results may not generalize beyond the participants and conditions studied. Observational studies can establish useful associations and may support causal reasoning when they use strong designs and carefully address confounding, but a large sample alone does not turn an association into proof of cause.

In an argument passage, identify what kind of evidence is actually presented. A before-and-after comparison, a survey, a correlation coefficient, an anecdote, and a randomized trial answer different questions. The key is not to dismiss observational evidence; it is to keep the conclusion within what the method supports.

Worked example

A city finds that neighborhoods with more trees have lower summer electricity bills. The data show a relationship. They do not prove that trees alone caused the lower bills. Shaded buildings may use less air conditioning, but income, building age, density, and local utility rates might also differ. A useful follow-up would measure those factors and define the outcome; an intervention or natural experiment could provide stronger evidence about the effect of planting trees.

How to answer a test question

1. Underline the author’s causal verb: causes, reduces, leads to, prevents, or produces.
2. Name the evidence: experiment, comparison, survey, trend, anecdote, or association.
3. Ask whether the cause came first and whether another variable could explain both outcomes.
4. Choose the critique that identifies the actual gap without claiming more than the evidence shows.
5. Prefer a conclusion such as “the evidence shows an association but does not establish causation” when that is the precise limitation.

Common reasoning mistakes

• Post hoc reasoning: assuming that because Y followed X, X caused Y.
• Ignoring a confounder that affects both variables.
• Assuming a correlation is meaningless; it can be valuable evidence even when it is not conclusive.
• Treating a larger sample as a cure for biased measurement or poor study design.
• Rejecting an argument by inventing an unsupported alternative cause rather than identifying a real evidentiary gap.

Key takeaway

Correlation describes a pattern; causation explains how a change produces an outcome. Test the direction, timing, confounders, mechanism, and study design, then match the strength of the conclusion to the strength of the evidence.

Association does not identify the cause

Correlation describes how two variables vary together. A positive association means higher values of one tend to occur with higher values of the other; a negative association means higher values of one tend to occur with lower values of the other. Correlation alone does not show that changing one variable causes the other to change.

A third variable may influence both. Ice-cream sales and swimming incidents can both rise during warmer weather; that association does not mean ice cream causes incidents. The unmeasured factor—temperature—may help explain the pattern. Reverse causation is another possibility: a health condition might lead people to change a habit rather than the habit causing the condition.

Look at how the evidence was gathered

Randomized experiments can support causal conclusions because random assignment helps balance other factors between groups, though execution and attrition still matter. Observational studies measure existing behavior and outcomes without assigning the exposure; they can reveal patterns and support careful causal analysis, but confounding must be addressed. A before-and-after comparison without a control group is especially vulnerable to other changes over time.

Timing can support a causal story—causes generally precede effects—but sequence alone is not enough. If scores rise after a curriculum change, ask whether the test, student group, instructional time, or other supports also changed. A comparison group measured during the same period helps estimate what might have occurred without the intervention.

Match causal language to the design

Use wording proportional to the evidence. A scatterplot may show that two measures are associated. A carefully controlled experiment may show that an intervention caused an average change under the tested conditions. Neither result necessarily generalizes to every person or setting. Consider effect size, uncertainty, population, and replication before making a broad statement.

A strong argument acknowledges alternative explanations and explains why they are unlikely or controlled. It distinguishes the observed relationship from the proposed mechanism. Evidence that a program improved scores does not automatically show why it worked; additional measures or experiments may be needed to test the mechanism.

• Describe the observed association first.
• Consider confounding variables and reverse causation.
• Check whether the design includes random assignment or a valid comparison.
• Confirm the cause precedes the outcome, while recognizing timing alone is insufficient.
• Limit causal conclusions to the population and conditions actually studied.

Apply the distinction carefully

A correlation describes how two variables vary together; it does not, by itself, show that one causes the other. A third variable may influence both, the direction may run the opposite way, or the association may be coincidental. For example, sunscreen use and sunburns can rise together because both are more common on sunny days; sunscreen is not thereby shown to cause sunburn. To evaluate a causal claim, ask whether the study establishes that the proposed cause came first, addresses plausible confounders, uses a suitable comparison, and measures outcomes consistently. Random assignment can strengthen causal inference by balancing many confounders, though implementation and attrition still matter. In passage questions, state the observed association at the strength the evidence supports.