Key points#
- Two things moving together can mean one causes the other, the reverse, a shared third cause, or coincidence. Correlation alone does not tell you which.
- Confounding is the big trap: a hidden factor that influences both the thing being studied and the outcome, faking a cause-and-effect link.
- Researchers can adjust for factors they measured, but not for ones they missed, so observational findings keep some uncertainty.
- Randomized trials reduce confounding because chance assignment tends to balance known and unknown factors.
"Correlation is not causation" is one of those phrases everyone has heard and few get to use well. It is genuinely one of the most important ideas in reading evidence, and it deserves more than a slogan. Understanding why two things moving together is only a clue, and what specifically can fool us, makes you far harder to mislead and far better at judging the studies behind the headlines.
Four reasons two things move together#
When a study reports that A and B rise and fall together, there are several possible explanations, and only one of them is the exciting one.
- A causes B. The interesting case, and the one headlines assume.
- B causes A. The relationship runs the other way. People who are already unwell may take up a behavior, making it look as if the behavior caused the illness.
- Something else causes both. A third factor drives A and B together, so they move in step without either causing the other.
- Coincidence. With enough data, some things line up by chance alone.
The skill is not to assume the first explanation. It is to keep all four in mind and ask which the evidence actually supports.
Confounding, the big one#
Of those explanations, the third has a name and a starring role in epidemiology: confounding. A confounder is a hidden third factor that influences both the thing being studied and the outcome, creating an apparent link that is not genuine cause and effect.
A familiar pattern makes it concrete. Suppose people who do a certain healthy-seeming activity also tend to be wealthier, more health-conscious, and better connected to medical care. If those people also live longer, is it the activity, or is it everything else that tends to travel with it? The activity and the longer life are correlated, but the wealth and health-consciousness may be doing the real work. Mistake the confounder for the cause, and you draw the wrong conclusion.
Confounding is everywhere in observational research, not because researchers are careless, but because people are not randomly sorted into their habits and circumstances. Those who choose one path differ from those who choose another in countless ways, and any of those ways might be the true driver.
What researchers can and cannot do#
Good researchers take confounding seriously and work to reduce it. The main tool in observational studies is statistical adjustment: measuring the factors they think might confound the relationship and accounting for them in the analysis. This genuinely helps, and a well-adjusted study is more trustworthy than a raw correlation.
But adjustment has a hard limit. You can only adjust for factors you measured and thought to include. The confounder you did not measure, or did not imagine, remains in the background, untouched. This is why even careful observational studies are usually described as showing an association rather than proving a cause, and why honest authors say so plainly.
Why randomized trials are different#
This is where randomized trials earn their reputation. When people are assigned to groups purely by chance, both the known and the unknown factors tend to balance out between the groups. Wealth, health-consciousness, the things nobody measured, all of it scatters roughly evenly. So when the groups differ in the outcome, it is far more likely to be the treatment itself, not some hidden confounder, that explains the difference. That is the whole logic of randomization, and it is why trials sit higher than observational studies when the question is whether something causes an effect.
This does not make observational studies worthless. They are essential, especially where trials are impractical or unethical, and they generate the hypotheses that trials then test. The point is to read them for what they are: strong clues, not final verdicts.
Reading cause and effect with care#
Next time you see a claim that one thing is linked to another, run the quick checklist. Could the relationship run the other way? Could a third factor explain both? Could it be chance? Was it an observational study, where confounding lurks, or a randomized trial, where it is reduced? You will not always reach certainty, but you will stop taking every correlation at face value, which is exactly the goal. Treat associations as the start of a conversation, and let the strength of the evidence, not the strength of the headline, decide how much weight to give them.
Sources and further reading
Questions and answers
Does correlation prove causation?
No. Two things can move together for many reasons: one might cause the other, the relationship might run the opposite way, a third factor might drive both, or it might be coincidence. Correlation is a starting point that needs more to confirm a cause.
What is confounding?
Confounding is when a hidden third factor influences both the thing you are studying and the outcome, creating an apparent link that is not really cause and effect. A classic pattern is when people who choose a habit also differ in other ways that affect health.
Can researchers fix confounding?
They can adjust for factors they know about and measure, which helps. But they cannot adjust for factors they did not measure or did not think of, so some uncertainty usually remains in observational studies.
Why are randomized trials better at showing cause?
Because assigning people to groups by chance tends to balance both known and unknown factors between the groups, so differences in the outcome are more likely to be due to the treatment itself rather than to confounding.