| Situation in a health claim | What the evidence shows | Planning use |
|---|---|---|
| Two measures rise and fall together | The correlation coefficient runs from +1.0 to -1.0, and a correlation does not automatically mean that a change in one variable causes the other. | Treat it as a lead to investigate, not a reason to change what you eat or buy. |
| Both measures follow a third factor | Ice cream and sunscreen sales move together because of the season, not because of a direct link. | Look for the shared driver before crediting either measure. |
| The outcome may be shaping the habit | In alcohol studies, abstainers show the highest mortality risk, reflecting poorer prior health. | Ask whether illness could have changed behaviour before the study began. |
| An unmeasured factor steers who gets what | Prior spine disease influences both the choice of dual mobility hip implants and dislocation risk. | Check whether the groups differed at the start, not only at the end. |
| Participants are randomly assigned | Randomization minimises the effect of measured and unmeasured confounders. | Give trial results the most weight when a claim affects health, spending or risk. |
Figures and examples are drawn from the Australian Bureau of Statistics guide to correlation and causation and a PubMed Central methods article on causal inference.
Causation vs correlation: correlation means two measures move together; causation means one produces the other. The strongest evidence for cause comes from controlled studies, especially randomized trials, that account for third variables.
The same distinction underpins how to tell if a health study is reliable, which is the wider checklist for judging any single result.
Reading nutrition and supplement headlines
Most food, drink and supplement headlines rest on observational studies. The Association of Health Care Journalists notes that observational studies can only show correlation, and that seasoned health reporters avoid causal wording in leads and headlines for that reason. A headline that says a food “causes” or “prevents” something, built on a study of that kind, is selling a link as a cause.

That has direct consequences for what you do with a claim. The choices below fit the most common situations.
- Hold off on buying or changing habits. If a supplement claim rests on people who happened to take it, the result describes who takes supplements, not necessarily what the supplement does. A comparison in which participants are assigned to it is the stronger basis for spending money.
- Read “linked to” and “associated with” as a lead. Cautious wording usually signals that the study could not test direction or rule out other explanations.
- Check the direction of the link. Reverse causality is common in studies of lifestyle factors such as BMI, alcohol, vitamins and diet, according to the PubMed Central article. People who are already unwell often change what they eat, drink or take.
- Look at who is in each group. If the people who take or avoid something differ in other ways, the headline may be describing those differences.

Health examples where a third factor does the work
A confounder is a third variable that influences both the supposed cause and the effect. The PubMed Central article calls confounders the most common and best-known reason correlation is mistaken for causation. The examples below show how it plays out in health and nutrition.
Diet, alcohol and exercise
The amount of sodium a person eats is closely correlated with the total calories they eat, and both are linked to high blood pressure, as the Association of Health Care Journalists glossary explains. A study that finds a sodium and blood pressure link therefore has to separate sodium from overall eating before it can name salt as the driver.
The same glossary describes heavy drinkers with higher rates of pancreatic cancer. Without more information, it is impossible to know whether the cancer was caused by their drinking or by something else, such as smoking. Exercise and skin cancer offer a further case: JMP’s statistics portal points out that both are affected by a third variable, sunlight exposure. Sunny weather also explains why ice cream sales and sunburn rise together.
Clinical examples
Confounding is not limited to lifestyle questions. In a study of hip implants, an unmeasured confounder such as prior spine disease can influence which patients receive dual mobility implants and also their risk of dislocation, so the implant can look better or worse than it is. Reverse causality appears too: very low BMI is associated with a higher risk of short-term mortality in joint replacement patients, and in alcohol research abstainers show the highest mortality risk because of poorer earlier health. In both cases, a risk label can be attached to a measure that is itself a sign of illness.
Spotting a confounder
No formula flags a confounder automatically, but a few questions find most of them.
- Ask what else could move both measures. Season, sunshine, smoking, overall calorie intake and existing illness are typical candidates in nutrition and health stories.
- Be wary of a simple pairing. Correlations describe a pair of variables, so they cannot account for a possible underlying relationship with a third variable, as the JMP portal puts it.
- Ask whether the outcome could have come first. If poor health could have changed the exposure, the arrow may point the other way.
- Check what was adjusted for. Adjustment only handles the confounders the researchers thought of and measured. Randomization is the approach that also handles unmeasured ones.
What correlation and causation mean
Correlation is a statistical relationship between variables. Its numerical value, the correlation coefficient, ranges from +1.0 to -1.0, according to the Australian Bureau of Statistics. The size of the coefficient describes how closely the variables move together. It says nothing on its own about why.
Causation means a change in one variable produces a change in another. The PubMed Central article lists the conditions for a causal inference: the cause precedes the effect in time, the cause and effect are empirically related, and the relationship is not explained by a third variable. A correlation can satisfy only the second condition, which is why it is a starting point.
The same article lists three non-causal routes to a correlation: chance, reverse causality and confounding. Correlational associations are common in non-experimental, observational studies, where the direction of the relationship is often not tested. That is not a flaw in observational work. It is a limit on what a single observational result can claim.
What evidence supports a causal claim
Controlled studies are the most effective way of establishing causality between variables, according to the Australian Bureau of Statistics. A typical medical example gives one group a placebo and another a new medication, then compares outcomes. Random assignment is what makes the comparison fair: the PubMed Central article notes that the effect of third variables, measured and unmeasured, is minimised through randomization. JMP’s explanation of the principle is that random assignment lets researchers infer a direct effect without worrying about unmeasured variables. Placebo arms also matter because expectation alone can shift outcomes, as the cases in Placebo Effect Examples: 9 Real-World Cases That Reveal How Expectation Shapes Health (2026 Update) – Testpledge show.
Trials are not always possible. The Australian Bureau of Statistics points out that ethical reasons limit the use of controlled studies, so observational studies are often used to investigate correlation and causation in populations. Nobody can ethically assign people to a harmful exposure. For those questions, the choice of observational design matters, and Case-Control Vs Cohort Studies: How To Choose The Right Observational Design In 2026 – Testpledge compares the two main options.
In practice, evidence for causation typically comes not from observational data but from careful experimental design, as JMP notes. When no trial exists, treat an observational finding as a reason to look for more evidence rather than as a settled effect.
Frequently asked questions
Does a strong correlation coefficient mean one thing causes the other?
No. A coefficient near +1.0 or -1.0 only describes how closely variables move together. It cannot account for a third variable, chance or reverse causality.
Can an observational study ever support a causal claim?
It can supply information toward causality in populations, especially where trials would be unethical, but it cannot settle the question alone. It cannot match the confounder control that randomization provides.
Why do headlines turn associations into causes?
Causal verbs make a story more definite. Careful health reporters avoid them when the underlying study is observational, so a headline using them is a prompt to check the study design.
What is a quick test before I act on a health claim?
Check whether the cause came first, whether a third factor could explain both measures, and whether participants were randomly assigned. If the answer to the last is no, hold the claim loosely.



