An association means two things tend to occur together in a set of data. It does not, by itself, prove that one causes the other. Turning an observed link into a causal claim requires reasoning beyond the link itself, which is why the classic writing on this question treats association as a starting point rather than a conclusion the paper on association or causation. For a treatment claim, the practical difference is large: "people who did this tended to see that" is a very different statement from "doing this produces that".
What does an association actually tell you?
An association tells you that two things were observed together in the data — nothing more about direction or cause. It could mean the first caused the second, the second caused the first, a third factor drove both, or the pattern appeared by chance in this particular sample. All of those produce the same observed link.
That is why a reported association is a finding about the study's data rather than a verdict about the world.
Why can an association appear without causation?
Because a shared cause, a reversed direction, or ordinary variation can all produce a link that looks meaningful. Three everyday patterns cover most cases, and the first is what researchers call confounding:
- Confounding — a shared cause. Two things move together because something else moves both.
- Reversed direction. The link exists, but it runs the opposite way from the claim.
- Chance or selection. The pattern appeared in this sample without reflecting a general rule.
None of these is exotic. They are the ordinary reasons a link needs more than itself before it becomes an explanation.
How do you move from a link to a cause?
By asking a series of questions about the link rather than accepting it: how strong is it, does it repeat, does the cause come before the effect, and is there a plausible mechanism? This is the logic behind the classic considerations for causal interpretation — strength, consistency, temporality, biological plausibility and coherence among them.
Two of those carry most of the weight for a consumer reading a claim. Temporality asks whether the proposed cause actually came first; without that, direction is unknown. Consistency asks whether the link appears across different studies and populations; a single appearance is weaker than a repeated one. A link that is strong, consistent, correctly ordered and mechanistically plausible is a serious candidate for causation — and it is still not proof on its own.
What should you do when a claim turns a link into a cause?
Ask whether the claim is describing an observation or asserting a mechanism, and check which one the evidence supports. Three quick checks:
| The claim says | What to ask | Why it matters |
|---|---|---|
| "X is linked to Y" | Is that all the evidence shows? | A link is not a cause |
| "X causes Y" | Did the design establish order and rule out alternatives? | Causation needs more than co-occurrence |
| "X works by doing Z" | Is Z a measured mechanism or a proposed one? | A plausible mechanism is not a demonstrated one |
Applied to treatment content, the third row is the one to watch. A description of how a technique is thought to work is a mechanism statement; it is not the same as a study result showing a benefit.
Our view
We make causal claims only where the design supports them, and we say which kind of evidence gave us the confidence. Where a statement rests on a measured comparison, we describe it as a finding. Where it rests on a plausible mechanism, we describe it as the mechanism — because the difference between those two is exactly what a buyer needs in order to weigh the claim.
Frequently asked questions
If two things are associated, does one cause the other?
Not automatically — an association on its own is not proof that one thing causes the other. The link is a finding about the data, and cause has to be established separately.
Can an association still be useful?
Yes — it identifies a question worth investigating properly. It is a lead, not a conclusion.
What is the single strongest clue to causation?
Order in time. If the proposed cause did not come before the effect, a causal reading is hard to sustain.
Does a plausible mechanism prove causation?
A mechanism explains how a cause could work rather than establishing that it did in a given study. Treat it as the reason a question is worth testing, and look for the study that tested it.
Separate the link from the claim
When a claim says two things are connected, ask whether that is all the evidence shows. The rosacea study overview is a useful place to see how a specific finding is described when the design, rather than the wording, sets the limit.