Correlation Does Not Show Cause and Effect

No, correlation does not show cause and effect. A correlation is a statistical measure that describes the size and direction of a relationship between two variables. It tells you that two things move together, but it cannot tell you that one thing caused the other. The distinction matters because mistaking correlation for causation leads to wrong conclusions in medicine, policy, and everyday decisions.

What Correlation Actually Measures

When two variables are correlated, their values change in a predictable pattern. If one goes up and the other goes up too, that’s a positive correlation. If one goes up while the other goes down, that’s a negative correlation. The correlation coefficient, expressed as a number between -1 and +1, captures how strong and consistent that pattern is.

But here’s the critical point: it’s possible to find correlations between many variables where the relationships have nothing to do with one causing the other. The correlation coefficient should not be used to make any claim about a cause and effect relationship. It’s a starting point, not an endpoint. Two things can rise and fall together for years without either one driving the other.

Three Reasons Correlation Misleads

Confounding Variables

A confounding variable is a separate factor connected to both things you’re measuring. It creates the illusion of a direct link between them. Consider the relationship between birth order and Down syndrome: studies show that Down syndrome becomes more common with higher birth order (second child, third child, and so on). But birth order isn’t the cause. Maternal age is. Parents who have more children tend to be older by the time later children arrive, and maternal age is the actual factor driving Down syndrome risk. Birth order just happens to track alongside it.

Similarly, if you studied car exhaust exposure and asthma rates, you might find a strong correlation. But people living near heavy traffic are also more likely to be exposed to factory pollution and secondhand cigarette smoke. Those confounders can strengthen, weaken, or even completely create the apparent relationship between the two variables you’re looking at.

Reverse Causality

Sometimes the relationship between two variables is real, but the direction is backwards from what you’d assume. In studies of patients with advanced kidney disease, researchers observed a U-shaped relationship between blood pressure and heart problems: very low blood pressure seemed to predict worse outcomes. The intuitive reading is that low blood pressure somehow causes heart damage. But the evidence suggests the opposite. Long-standing high blood pressure causes changes in heart structure and function that eventually lower blood pressure while also increasing cardiovascular risk. The effect was being mistaken for the cause.

Pure Coincidence

With enough variables to compare, some will line up by chance. The per-capita consumption of mozzarella cheese in the United States correlates almost perfectly with the number of civil engineering doctorates awarded each year. Nobody thinks cheese consumption drives doctoral research. But if you only saw the numbers without context, the correlation would look convincing. This is why researchers don’t treat a single correlation, no matter how strong, as evidence of anything causal.

What It Takes to Prove Cause and Effect

The gold standard for establishing causation is the randomized controlled trial. In an RCT, participants are randomly assigned to different groups, so that both known and unknown characteristics are balanced across groups. Any difference in outcomes can then be attributed to the intervention rather than to some hidden variable. No other study design can do this. The tradeoff is that RCTs are expensive, time-consuming, and sometimes ethically impossible (you can’t randomly assign people to smoke for 20 years).

When trials aren’t feasible, researchers use a framework developed by the epidemiologist Austin Bradford Hill. His nine viewpoints help evaluate whether a correlation is likely causal:

  • Strength: A larger effect size makes a causal link more plausible.
  • Consistency: The same finding appears across different populations and settings.
  • Specificity: The exposure leads to a particular outcome, not a broad range of unrelated ones.
  • Temporality: The cause must come before the effect. This is the only absolute requirement.
  • Dose-response: More exposure produces more of the effect.
  • Plausibility: A biological or logical mechanism can explain the link.
  • Coherence: The finding doesn’t contradict what’s already known.
  • Experiment: Removing the exposure reduces or eliminates the effect.
  • Analogy: Similar exposures produce similar effects.

No single criterion is sufficient on its own, and apart from temporality, none is strictly required. Researchers weigh them collectively. A correlation that satisfies most of these viewpoints carries far more causal weight than one that only shows a strong statistical association.

How Genetics Helps Solve the Problem

One increasingly important tool for untangling correlation from causation is called Mendelian randomization. It takes advantage of a basic fact about genetics: during reproduction, gene variants are randomly assigned to offspring, much like participants are randomly assigned in a clinical trial. Because your genes are fixed at conception, they can’t be affected by confounders you encounter later in life, and they can’t be subject to reverse causation.

Here’s how it works in practice. Say you want to know whether a specific risk factor (like high cholesterol) truly causes a disease, rather than just appearing alongside it. Researchers identify a gene variant known to raise cholesterol levels and then check whether people carrying that variant also have higher rates of the disease. If they do, it strengthens the case that cholesterol itself is causal, because the gene variant was assigned randomly and isn’t tangled up with lifestyle, income, or other confounders. This approach is especially valuable when running a controlled trial would be impractical or unethical.

Statistical Significance Is Not the Same as Real-World Impact

Even when a study moves beyond correlation and finds a statistically significant result, that doesn’t automatically mean the finding matters in practice. Statistical significance tells you the result is unlikely to be due to chance. It says nothing about how large or meaningful the effect is.

A useful example: imagine two cancer drugs are tested in separate trials, and both produce statistically significant improvements in survival compared to no treatment. Drug A extends survival by five years. Drug B extends it by five months. Both results are statistically significant, but only one represents a clinically meaningful improvement. For decades, the emphasis on statistical significance led researchers and journalists to treat mathematically notable findings as if they were always important, blurring the line between “real pattern” and “pattern worth caring about.”

How to Read Health Headlines More Carefully

Most health headlines that seem to announce a cause and effect relationship are actually based on observational studies that can only show correlation. A study analyzing news coverage on a major German medical news site found that correlational findings were routinely framed in causal language. The headline “Fresh fruit prevents cardiovascular diseases and death,” for example, was based on a cohort study, a type of observational research that can only identify associations. A more accurate headline would have been: “Cohort study examines the association between fruit consumption and cardiovascular events.”

You can decode the strength of a claim by watching for specific language. Words like “associated with” or “linked to” signal correlation. “Might,” “could,” or “may” suggest a possible but unproven causal connection. “Can cause” is stronger but still hedged. Only a claim with no qualifying language (“causes,” “prevents,” “leads to”) asserts full causation, and that claim should be backed by experimental evidence, not an observational study. When a headline drops all qualifiers, check whether the underlying study was a randomized controlled trial. If it wasn’t, the headline is almost certainly overstating what the research found.