Case Study 2 — How "A Glass of Red Wine" Became Common Knowledge

A media and evidence-history case. The studies, trials, and institutional events described are real; the four-decade reconstruction is my own reading of them.

Setup

Chapter 1 traced a pipeline from a finding to a reel in nineteen days. This one runs the same analysis over forty years, on the most successful nutrition claim of the modern era.

"A glass of red wine a day is good for your heart."

Almost everyone reading this knows that sentence. A great many believe it. A substantial number have made decisions based on it — including clinicians, who repeated it to patients.

It is, on the best current evidence, probably wrong. Here is how it got installed.

Stage 1 (1970s–80s): a genuine observation

Epidemiologists studying cardiovascular disease across countries noticed something that didn't fit. France had lower rates of coronary heart disease than its saturated fat intake predicted.

This was a real observation, honestly reported, and it acquired a name: the French paradox.

What it actually was: an ecological correlation — country-level averages compared to country-level averages, which is close to the weakest observational design there is (Chapter 2, rung 4 territory and arguably below).

What it needed: an explanation. And wine was the most conspicuous difference.

Alternative explanations existed and got less attention, including differences in how cause of death was recorded, differences in the timing of dietary change relative to disease incidence, and the rest of the French dietary pattern. The wine hypothesis was one candidate among several, and it was the most interesting one.

Stage 2 (1980s–90s): the cohort studies

Then the individual-level data arrived, and it looked extraordinary.

Study after study, in country after country, found the J-curve (§12.3). Moderate drinkers had lower all-cause mortality than abstainers. The finding replicated across populations, designs, and decades.

This was not fringe. It was one of the most reproducible findings in observational epidemiology, and by the standards of the time it looked overwhelming.

The problems were knowable and were noted — sick-quitter bias was discussed in the literature from early on. But they were treated as caveats to a robust finding rather than as candidate explanations for the finding, and that framing choice determined the next thirty years.

Stage 3 (1990s): the mechanism arrives

Observational findings become believable when a mechanism appears, and two did.

HDL. Alcohol raises HDL cholesterol, and at the time HDL was understood as straightforwardly protective. The chain looked complete: alcohol → HDL up → cardiovascular risk down.

Resveratrol. A polyphenol in red wine, doing interesting things in cell and animal models — including, later, longevity signalling (Chapter 6 §6.9). It gave the French paradox a molecule.

Both mechanisms have since substantially weakened, and it's worth noting how:

• HDL — Chapter 9 §9.10. Drugs that raise HDL failed repeatedly to reduce events; Mendelian randomization doesn't support a simple causal role. HDL is a marker, not a lever — which means a mechanism running through HDL was never as strong as it appeared, and this was knowable before the alcohol MR studies arrived.
• Resveratrol — §12.5's verdict. Wine's content is very low; matching animal doses would require implausible quantities; human supplementation trials underwhelmed.

But by the time both mechanisms weakened, the claim was already common knowledge, and common knowledge does not have a retraction mechanism.

Stage 4 (1990s–2010s): amplification, and who paid for it

Here is where Chapter 1 §1.5 becomes unavoidable.

The alcohol industry had a finding worth more than any advertising campaign could buy: a health benefit, from mainstream epidemiology, reported by neutral outlets. It funded research, funded social aspects organizations, and supported the dissemination of favourable findings — which is legal, routine across industries, and exactly the funding-effect problem Chapter 2 §2.8 describes.

And I want to be careful here, because the lazy version of this story is that the industry manufactured the finding. It didn't. The J-curve appeared in independently funded cohorts across many countries. The industry's role was in amplification, framing, and the funding of research programmes likely to produce favourable results — which is subtler, more effective, and harder to correct than fabrication would have been.

The media did the rest for free. "Wine is good for you" is a story with everything: counterintuitive, pleasant, actionable, and endlessly re-runnable. It has been re-reported every year for three decades.

Stage 5 (2010s): the evidence turns

Three things happened, roughly together.

Better handling of sick-quitter bias. Analyses restricting the comparison to lifelong abstainers found the protective dip attenuating or disappearing.

Mendelian randomization. §12.5. Genetic instruments — immune to sick-quitter and healthy-user bias by construction — found no cardioprotection, with relationships looking monotonic.

The cancer evidence strengthened. IARC Group 1, with breast cancer showing dose-response from low intakes.

And the trial that would have settled it was halted (§12.5b) amid concerns about alcohol industry involvement in its development — meaning the definitive answer will probably never come from the design that could have produced it.

Stage 6 (2016–2023): the guidelines move, quietly

Four major bodies, in seven years, all moving the same direction.

And the public messaging did not move with them. "A glass of red wine is good for your heart" remains in circulation, is still repeated by clinicians, and still appears in mainstream health content.

The asymmetry

Lay the two propagation curves side by side.

💡 Aha moment. This is Chapter 1's pipeline running in both directions — and it is not symmetric.

A finding that is pleasant, actionable, counterintuitive and commercially supported propagates extremely well. Its correction is unpleasant, less actionable, commercially unsupported, and arrives as a retraction — which is the least shareable format in existence.

The mechanism that installed the belief cannot uninstall it, because the properties that made it spread are exactly the properties the correction lacks. This is Chapter 1's Case Study 2 — the correction reaching 1/300th of the audience — extended across four decades.

And it generalizes uncomfortably: anything you believe about nutrition that is pleasant, simple, and commercially convenient deserves more scrutiny than the things you believe that are annoying, purely on propagation grounds and before you've evaluated any evidence at all.

Analysis

1. Nobody lied and it took forty years to correct. The epidemiologists reported real findings. The mechanisms were plausible. The journalists reported accurately. The industry amplified rather than fabricated, which is more effective and much harder to attack.

2. The caveats were known the whole time. Sick-quitter bias was in the literature from early on. It was framed as a limitation of a robust finding rather than as a candidate explanation for the finding — and that framing choice, not any missing data, is what cost thirty years.

3. Two independent mechanisms both weakened, and the claim survived both. HDL and resveratrol were what made the association believable. Both have substantially collapsed. Common knowledge has no retraction mechanism, so the belief outlived its supports.

4. Mendelian randomization was the tool that broke it, and it worked precisely because its weaknesses don't overlap with the cohorts' weaknesses (Chapter 2 §2.10). This is the clearest worked example in the book of why convergence is the criterion that matters.

5. And the guidelines moved without anyone noticing. Four major bodies revised in seven years. Ask five people what current alcohol guidance says and you will get 1990s answers, because guidance changes quietly and advice is given once — the same finding as Ruth's egg advice in Chapter 8 and the margarine sequence in Chapter 9.

Discussion Questions

1. The caveats were in the literature the whole time and were framed as limitations rather than explanations. What would have had to be different for the field to take them as candidate explanations in 1995? Is that a realistic ask?

2. The industry amplified rather than fabricated. Is that meaningfully better? Design a policy that would address amplification without banning industry-funded research entirely.

3. The correction has no natural constituency — nobody profits from it and it's unpleasant. Design a public health campaign that would actually shift this belief. Then estimate its budget against the forty years of free media the original claim received.

4. The chapter argues that pleasant, simple, commercially convenient beliefs deserve extra scrutiny on propagation grounds alone, before evaluating evidence. Is that a legitimate epistemic heuristic or is it prejudice dressed as rigour?

5. Compare this case to Chapter 9's margarine sequence and Chapter 8's egg advice. All three involve advice outliving its evidence. Are the mechanisms the same? If not, what distinguishes them?

Your Turn

Find a nutrition belief you hold that is pleasant, simple, and commercially convenient.

Not one you're suspicious of — one you're comfortable with. Coffee's benefits. Dark chocolate. Something being "anti-inflammatory." Olive oil. Something you'd defend at a dinner table.

Then trace it:

1. Where did you first hear it? Can you place it?
2. What is the underlying observation? Is it ecological, cohort, trial, or mechanism?
3. What is the proposed mechanism, and has it held up? Check specifically whether the mechanism has weakened since the claim became common knowledge — that's the pattern here.
4. Who profits if you believe it?
5. Has any guidance moved in the last ten years? Check the current version rather than the version you learned.

Most people find at least one belief where the mechanism has quietly weakened and they never heard about it — because mechanism collapses don't make headlines and pleasant claims don't get retracted.

That's the finding, and it's a more useful one than anything specific to alcohol.