The Healthy-User-Bias Audit
Separate a food's apparent effect from the lifestyle surrounding its users
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 4
- Confidence
- 99%
Lugavere uses healthy-user bias to explain why nutrition associations can mislead. People who follow culturally approved health behaviors may also exercise, avoid smoking, shop differently, and seek healthcare. People who eat more of a stigmatized food may consume it in fast food, smoke more, or be more sedentary. A measured association can therefore reflect the surrounding lifestyle rather than the food alone. His quinoa example makes the mechanism concrete: quinoa consumption might correlate with good health because its consumers are unusually health-conscious, even if the grain is not solely responsible. The audit asks what behaviors cluster with the exposure, whether a study adjusted for them, what outcome was measured, and whether randomized evidence agrees. It improves interpretation but cannot erase uncertainty or prove the opposite conclusion. The transcript itself notes that long-term randomized nutrition trials are scarce.
Origin
Lugavere introduces healthy-user bias while challenging simple conclusions drawn from observational research on meat, vegetables, and quinoa. Extracted from The Diary of a CEO.
Core principles
- 01Association does not establish that a food caused an outcome
- 02Health-conscious behaviors tend to cluster
- 03Less healthy behaviors can cluster around stigmatized foods
- 04Study design and adjustment quality matter
- 05Randomized evidence and clinical outcomes deserve separate weight
How to run it
- 1
Classify the evidence
Determine whether the claim comes from an observational association, a randomized trial, or another design. Do not assign causal certainty before this classification.
Pro tip Look for the design in the paper or study description rather than the headline.
Watch out Randomization can strengthen causal inference but does not automatically make every trial definitive.
- 2
Map the behavior cluster
List plausible behaviors that travel with the exposure, such as smoking, exercise, fast-food intake, income, healthcare use, or general diet quality. Ask which could independently influence the outcome.
Pro tip Imagine the typical context in which the food or behavior appears, then verify rather than assume it.
Watch out Stereotypes about users are hypotheses, not measured confounders.
- 3
Inspect adjustment and outcome
Check which confounders the researchers measured and adjusted for, and whether the outcome is a clinical event, biomarker, or self-report. Note important factors that remain unmeasured.
Pro tip Distinguish relative association from absolute risk when the source allows it.
Watch out Statistical adjustment can reduce confounding without eliminating it.
- 4
Triangulate the conclusion
Compare observational findings with randomized trials, mechanisms, and findings in other populations. Phrase the conclusion at the confidence level the combined evidence supports.
Pro tip Use words such as associated, suggests, or may when causality remains unresolved.
Watch out Confounding in one direction does not prove the reverse claim.
In the wild
A headline says quinoa eaters have better health. The audit asks whether those eaters also exercise more, smoke less, shop at health-focused stores, or follow higher-quality diets. It then checks whether the study adjusted for those variables and whether any randomized evidence isolates quinoa itself.
→ The conclusion is narrowed from a causal food claim to the level actually supported by the study.
Common mistakes
Promoting correlation to causation
An observed difference between groups can arise from clustered behaviors rather than the named food alone.
Using confounding to prove safety
Showing that a study is confounded weakens that inference; it does not establish the opposite conclusion.
Ignoring study design
The transcript distinguishes observational findings from the much smaller set of randomized trials.
Is it for you?
Best for
It is best for readers evaluating nutrition headlines, cohort studies, and claims built on population associations.
Not ideal for
It is not a licence to dismiss observational research or to conclude that an exposure is safe because confounding exists.
From the transcript
“this is the whole concept of healthy user bias”
“people who consume more meat especially in this country tend to be more sedentary and they tend to smoke more”
“if you look in the nutrition literature you can find a study to back up anything that you want to say”
From the episode
The No.1 Health Expert: The One Food (WE ALL EAT) That's Slowly Hurting Us!: Max Lugavere