Correlation vs Causation in Nutrition
A personal essay, not an evidence assessment. Our graded assessments of individual ingredients — written to a published standard, from full texts — are in the evidence section. Nothing here is a statement about what any product does.
Coffee drinkers live longer.
People who eat more X have less Y.
Higher blood levels of nutrient A are associated with better cognition.
People following diet B have lower disease risk.
Nutrition is full of these sentences.
Some eventually point toward real causal relationships.
Some do not.
The difficulty is knowing which is which.
Correlation is not a mistake
This is worth saying first.
An association is evidence.
It tells us two things vary together.
That can be scientifically important.
The mistake is changing the verb.
From:
is associated with
to:
causes.
That transition requires additional evidence.
People do not eat nutrients in isolation
This makes nutrition particularly difficult.
Someone who eats a lot of vegetables may also:
exercise more,
smoke less, sleep differently, have different income, receive different healthcare, consume less ultra-processed food, have different body weight, and follow dozens of other behaviours.
Researchers can adjust for many variables.
But diets and lifestyles are extremely complex.
Reverse causation is another problem
Suppose people with poorer health consume less of a particular food.
We observe:
low intake associated with poor health.
Did low intake contribute to poor health?
Or did deteriorating health change appetite and diet?
Both are possible.
The arrow is not visible in the correlation.
Nutritional biomarkers create similar traps
Imagine higher blood levels of a nutrient correlate with better cognitive performance.
Perhaps the nutrient helps cognition.
Perhaps people with healthier diets have both higher nutrient levels and better cognition.
Perhaps another biological variable influences both.
Perhaps disease changes the biomarker.
Again:
association first.
Causal interpretation later.
RCTs help, but nutrition is difficult to randomise perfectly
Long-term dietary trials are expensive and challenging.
Adherence changes.
Blinding whole diets is often impossible.
People eat outside the study environment.
The effects we care about may take years.
This is why nutritional science frequently requires several forms of evidence to converge rather than waiting for one perfect experiment.
I like triangulation
If we have:
observational associations, a plausible mechanism, controlled intervention evidence, dose-response information, consistent findings across populations, and perhaps natural experiments or genetic evidence, the causal case can become progressively stronger.
No single piece has to perform every job.
The important thing is not to pretend that the first association completed the puzzle.
Marketing often removes the uncertainty
A study reports:
Higher intake of X was associated with better scores.
A headline becomes:
X protects the brain.
A product page becomes:
X supports cognitive health.
Three linguistic steps.
A major evidentiary transformation.
No new experiment occurred between them.
Only the verbs changed.
I want MindHeaven to protect the verbs
That may sound like a strange scientific principle.
But I think it matters.
Associated with should remain associated with.
Observed in should remain observed in.
Suggests should remain suggests.
Caused should be reserved for evidence capable of supporting causality with appropriate confidence.
Language is not decoration around evidence.
Language determines what people believe the evidence means.
And in nutrition, where the systems are complex and certainty often arrives slowly, choosing the correct verb may be one of the most responsible things we can do.