The Ingredient Brief

Evidence

Association, Causation, and Supplement Claims

Published July 12, 2026 · The Ingredient Brief

A sales page cites a real study. The study is real. The quote is accurate. And the conclusion you walk away with is still not supported by it. This is not a rare failure. It is the normal way supplement marketing works, and it turns on one word: associated.

What “associated with” actually licenses

Most nutrition research is observational. Researchers take a large group of people, record what they eat or what they take, wait, and see who develops what. If the people who reported more of an ingredient had fewer of some outcome, the ingredient is “associated with” that outcome.

That finding is real information. It is also compatible with several explanations, and the study design cannot tell them apart:

  • The ingredient caused the difference.
  • Something else about those people caused it (confounding).
  • The outcome influenced the intake, not the other way round (reverse causation).
  • Chance, or a quirk of how intake was measured.

An association narrows the field of hypotheses. It does not pick the winner. The US Federal Trade Commission, which polices advertising claims, puts it plainly in its guidance to the industry: observational studies “can be valuable to show an association between a product or ingredient, but they don't prove a causal link.”

Confounding, in one sentence

People who take a supplement are different from people who don't — in ways that also affect their health.

In the US and UK, adults who report high intakes of vitamins and health foods tend to smoke less, exercise more, weigh less, be less deprived, and see a doctor sooner. Every one of those traits independently predicts better outcomes. So an observational study comparing “high intake” to “low intake” is never only comparing intake. It is comparing two different populations, and the ingredient is riding along with everything else that comes with being the kind of person who takes it. Researchers call this the healthy-user effect. Statisticians try to adjust for it, but you can only adjust for what you measured, and you never measured everything.

Reverse causation, in one more

Sometimes the arrow points backwards. People in early, undiagnosed decline often change what they consume — appetite drops, diets narrow, routines lapse — months or years before anything is recorded as a diagnosis. In the data, low intake appears to precede the outcome. It looks causal. It is the outcome, casting a shadow backwards.

The case that settled the argument

Antioxidant vitamins are the standing example, and worth knowing because it is the pattern, not an anecdote. Through the 1990s, observational studies consistently found that people with higher intakes and higher blood levels of antioxidant vitamins had less cardiovascular disease. The associations were strong, repeated, and biologically plausible.

Then the randomised trials ran — the design where people are assigned by coin flip, so the two groups differ only in what they were given. The trials did not reproduce the benefit. A 2004 analysis in The Lancet examined why, and found that antioxidant intake in the observational data was tangled up with socioeconomic position and health behaviour: the people with more vitamin C in their blood were, systematically, people with more of everything else that protects a heart. The vitamins had been carrying credit for their company.

Note what this does not mean. It does not mean the observational studies were fraudulent, or that the researchers were fools, or that nutrients are useless. It means that a strong, repeated, plausible association was tested against randomisation and did not survive. That is the whole point of running the trial. And it is the reason a citation to an observational study, by itself, cannot tell you what a bottle will do.

How marketing gets from association to cause without lying

Here is the mechanism, and it is worth seeing clearly, because it is legal, common, and mostly invisible.

A marketer does not have to say “this product cures X.” Saying that would convert a supplement into an unapproved drug claim, and it is exactly the sentence regulators look for. What a marketer does instead is juxtapose:

  • Sentence one is a true statement about a study. “A 2019 study found participants with higher levels of [compound] had better [measure].”
  • Sentence two is a true statement about the product. “Each capsule contains [compound].”
  • There is no sentence three. There doesn't need to be.

The inference — therefore this capsule will do that to me — is assembled by the reader. Nobody wrote it down. It exists only in your head, where it cannot be fact-checked, cited, or prosecuted. Everything on the page is technically defensible in isolation.

Three details make the trick work. First, the study is often in a different form of the compound, at a dose the product doesn't reach, in a population unlike the reader — or in cells, or in rats. Second, an association is quoted with the causal verb quietly upgraded: “supports,” “promotes,” “helps maintain.” Those are structure/function claims, which US law permits on supplements without FDA review, provided the label carries the disclaimer that the statement has not been evaluated by the FDA and that the product is not intended to diagnose, treat, cure, or prevent any disease. Third, the disclaimer sits at the bottom in grey six-point type, after the inference has already been made.

None of this requires a bad actor. It requires only a page laid out so that the reader does the illegal part.

The three questions

You do not need statistical training to defuse this. You need three questions, asked of any study a sales page cites. They take about two minutes, and you can usually answer all three from the free abstract on PubMed.

  1. Was anyone randomly assigned to take it? If yes, you are looking at a controlled trial and causal language is on the table. If no — if researchers merely observed what people already consumed — you have an association, and every conclusion drops one level, from shows to suggests. This single question sorts most citations. The FTC's own view is that randomised, controlled human trials are the most reliable form of evidence, and generally what experts require before a health benefit is asserted.
  2. What exactly was tested, and does the bottle match it? Check four things: the specific compound and form, the daily dose, how long people took it, and who they were — humans, or cells in a dish, or mice. Then compare to the label. A study using 500 mg a day for twelve weeks in adults says nothing about 50 mg for a fortnight, and a result in rodent liver cells says nothing about a person. If the citation and the panel don't match on all four, the study is about something else.
  3. What did they measure, and is it what you were promised? Trials often report a surrogate — a number in a blood test, a score on a questionnaire — rather than the outcome you actually care about. Ask how large the difference was, over how long, and whether the endpoint is the thing being sold to you. “Moved a marker by a few percent over eight weeks” and “changed how I feel” are not the same finding, and only one of them is on the page.

Ask these three, in that order, and the page stops being persuasive and starts being checkable. Some products survive it. That is useful too: a citation that answers all three cleanly — a randomised human trial, at the dose on the panel, measuring something that matters — is a genuinely strong signal, and it is rarer than the internet suggests. When you find one, you will know exactly why you believe it.

The short version

  • “Associated with” is not “causes.” Observational studies show that two things travel together; they can't tell you which one is driving.
  • Two things routinely break the inference: confounding (supplement takers differ in a dozen other ways) and reverse causation (the outcome changed the intake).
  • Marketing rarely lies outright. It puts a real study next to a real bottle and lets you supply the causal sentence — the one nobody printed.
  • Ask any cited study: (1) Was anyone randomly assigned? (2) Does the compound, form, dose, duration and population match the bottle? (3) What was measured, how big was it, and is that what I was promised?

Sources