An unusually high result is usually followed by a more ordinary one — whatever you did in between. This single statistical fact is why so many things appear to work.
Why every measurement varies, what regression to the mean is, and why it makes ineffective interventions look effective.
Measure anything about yourself twice and you get two different answers. Blood pressure, resting heart rate, almost any blood test. Nothing changed about you between the readings, and yet the numbers differ.
Two separate sources of variation are at work, and neither is error in the ordinary sense.
Where variation comes fromBiological variation is you, genuinely changing. Levels fluctuate through the day, with meals, sleep, stress, hydration, recent activity, illness and — for many measures — the time of year. Some analytes are remarkably stable within a person; others swing widely.
Analytical variation is the measurement itself. Run the identical sample twice and a good laboratory returns slightly different numbers. Small, but not zero.
Together these mean every result is really a value plus noise. Which sets up the most useful idea in this lesson.
The effect that fools everyoneIf a reading is unusually extreme, part of that is likely to be noise that happened to fall the same way. Noise does not persist. So the next measurement will probably be closer to your average — with no treatment, no change, nothing at all.
People do not usually start an intervention at a random moment. They start when a number looks bad — which is precisely when it is most likely to have been an extreme reading. Then they retest, it has improved, and the intervention gets the credit.
This is not a small effect. It is one of the main reasons ineffective treatments accumulate enthusiastic personal testimonials, and it is a large part of why the control group in a trial so often improves too. Any before-and-after with no comparison group is vulnerable to it.
This is also the honest limit on personal experimentation, and it is worth saying plainly. You can learn real things about yourself from tracking — but a single person, without a control group, changing one thing while life continues around them, cannot cleanly separate an effect from noise. That is not a reason to stop measuring. It is a reason to hold conclusions more loosely than the numbers make you feel entitled to.
Biological and analytical variation are both well characterized, and vary substantially between analytes. Standard laboratory science.
Regression to the mean is a mathematical property of repeated measurement with noise, not a hypothesis. It is a recognized source of apparent treatment effect in uncontrolled studies.
Blood pressure diagnosis requires repeated measurement, partly for this reason. Reflected in all major guidelines.
Standardizing measurement conditions reduces variability and improves the interpretability of a trend.
That an improvement between two readings, with no control comparison, demonstrates that an intervention worked.
Every measurement is a value plus noise, so an extreme reading tends to be followed by a more ordinary one regardless of what you did — which is why trends beat points and why uncontrolled before-and-afters are so persuasive and so unreliable.
Health makes more sense when more than one person is looking at it. Everything here is free to read.
Want the rest? The foundation bundles collect these lessons into one file, free.
A visual guide to how metabolism, mitochondria, hormones, sleep, muscle, the heart, the gut and the brain influence one another — the whole system on one page, with the strongest connections written out. Use it as your roadmap while you work through the lessons.
Free. The map is yours to keep. Plus occasional new lessons and evidence-based updates. No spam. Unsubscribe any time.
Free to download right now — no email needed:
The Health Map — a visual guide to how metabolism, mitochondria, hormones, sleep, muscle, the heart, the gut and the brain influence one another. Every lesson on the site stays free to read either way.
Free. No spam, unsubscribe any time.
