Healthy Skepticism

Why this site

Medical information arrives every day: headlines, advertisements, guidelines, test results, diagnoses. Most of it is harder to read than it looks. This site exists to help you read it. The goal isn't to make you distrust medicine. It's to give you the tools for a better conversation with the people who practice it.

None of this is cynicism about medicine itself. The progress is real: people live longer, cancers that were death sentences are now managed, and drugs like statins and anticoagulants genuinely save lives. But the messaging around that progress has been muddled by incentives that have little to do with your health. Researchers are rewarded for publishing early and often, which fills the literature with findings that don't hold up. Drug companies are rewarded for maximizing sales, which favors the framing that makes a benefit sound largest. Neither of these requires anyone to lie. A relative risk reported without its baseline is perfectly true and still misleading. The purpose here is to help you see past the framing to what the numbers actually say.

What you'll find here

The papers are open to anyone. The kits turn a paper into something you and your doctor use together in a visit. The course materials are for students, behind the password given with enrollment. The newsletter examines one claim or recommendation at a time and sends it to you.

The six faults

The six faults are what I found when I applied my training and experience to the state of medical information reaching the public: the headlines and advertisements, the guidelines and categories, the test results and diagnoses. Across all of them, the numbers are usually computed correctly. What breaks is the reasoning wrapped around the numbers, and it breaks in the same six ways whether the package is a headline, a category line, or a diagnosis. Each fault has a plain question that exposes it.

1. Relative reporting What does a 36% improvement from taking a drug really mean?

   A drug ad says the pill cuts your risk by 36 percent. Thirty-six percent of what? If your ten-year risk falls from three in a hundred to two in a hundred, that is one person in a hundred helped, and also a 33 percent reduction; both numbers are true, and only one tells you what you need. The fault is reporting the ratio and withholding the baseline.

   The paper: Relative Reporting

2. Probability reversal Does a positive COVID test mean I have it? Does a negative mean I’m not contagious?

   A test is 96 percent accurate and you test positive, so the chance you have the disease is 96 percent. It is not. Accuracy gives the probability of a positive result in someone who has the disease; you need the reverse, the probability of the disease given your positive result, and when the condition is rare the two differ enormously. Confusing the two directions is the most consequential error in reading test results, and a negative read as the all-clear is the same fault run the other way.

   The paper: Probability Reversal

3. Arbitrary categories How concerned should you be if your systolic blood pressure goes from 119 to 120?

   A systolic pressure of 119 is normal and 120 is elevated, yet nothing happens in your body between 119 and 120. Continuous measurements get cut into categories at lines drawn by committees, and the lines move. The categories are administrative conveniences; the fault is treating them as facts about you.

   The paper: Arbitrary Categories

4. The Bernoulli fallacy Why does the published medical advice keep changing?

   A statistically significant finding sounds like a proven one. It isn't: significance measures how surprising the data would be if there were no effect at all, and says nothing about how large the effect is or how likely it is to be real. Chasing significance fills the literature with findings that fail when others repeat the work; the irreproducibility crisis is this fault operating at scale, and it is one reason the advice keeps changing.

   The paper: The Bernoulli Fallacy

5. The causal leap Does bird watching prevent dementia?

   It is reported that bird watchers get less dementia, so bird watching must protect the brain. But people who take up bird watching differ from people who don't in a dozen ways that matter, and seeing two things travel together is cheap. The gap can be closed: that smoking causes cancer was actually proven, by converging lines of evidence that eliminated the alternatives one by one. That is the standard; most headlines jump the gap instead.

   The paper: The Causal Leap

6. The syndrome trap If two people have the same diagnosis, what do they actually have in common?

   A syndrome is a checklist with a name, not a disease, yet it gets discussed as if it had one mechanism and one treatment. Two people with metabolic syndrome can share a single criterion; two with the same ADHD diagnosis can share none. Treating the label as one condition is imprecise medicine.

   The paper: The Syndrome Trap

The course teaches all six, across five sessions. To see four of them at work on one claim, read the case study: Is Nicotine Good for You?

New on the site

The papers are this site’s working library. Each one either defines one of the six faults or works a single medical question through its evidence, with every source named. They are free to read and to share. The newest:

• September 2026 The newsletter Medical claims and recommendations, examined one at a time. Free, by email.
• September 2026 Is Nicotine Good for You? A case study. One wellness claim, four faults at work, and what the trials found once each fault is removed.
• September 2026 The six fault papers One paper for each fault, with the course figures: what the claim says, why it misleads, and what to ask. Linked under each fault above.
• August 2026 The Syndrome Trap One label, many illnesses: why treating the name instead of the patient has failed, from metabolic syndrome to autism.

All papers

The course

The course distills this site’s material into a live online offering. I developed it with Joel Keenan, MD, and teach it myself. Joel is a practicing physician; I bring the mathematics and a patient’s perspective. The course is for anyone who wants to read medical claims clearly, and especially for older adults, who face the most medical decisions. It assumes no statistics background. Everything is done by counting: no formulas, just careful arithmetic you can check yourself as we go. It runs as five live sessions covering the six faults, one worked example at a time, with a take-home summary after each session and deeper papers for anyone who wants more. You leave knowing what to ask about any medical claim: out of how many, how many, and counted over whom. The course is educational and is not medical advice. Dates for the next online offering will be announced in the newsletter.

About me

I hold a PhD in mathematics from UC Berkeley and began my career as an assistant professor of mathematics at Duke University and the University of Texas at Austin. For decades I brought causal reasoning and decision support to complex programs, retiring as an IBM Distinguished Engineer. Along the way I wrote two books on leading complex development (Wiley, 1998; Addison-Wesley, 2001), earned fifteen IBM patents, and lectured often, from university seminars to industry conferences. I also write as a patient: I've lived with metabolic syndrome since childhood, and working with my doctor and semaglutide (Ozempic) I lost a hundred pounds and have kept the weight off. Both halves lead to the question this site asks: given the evidence in front of you, what can you honestly conclude?