PHYS-43 · Principle
My FTP went from 248 to 253. Did I improve?
Maybe not. Even a lab power test carries about 1-2% noise, and two readings differ by more than that by chance, so one 2% rise is compatible with no change. Repeat the same test.
How the body works, agreed across the reviews below; a study entry shows what one experiment measured.
Practical Application
My FTP went from 248 to 253. Did I improve? Maybe not. Even a lab power test carries about 1-2% noise, and two readings differ by more than that by chance, so one 2% rise is compatible with no change. Repeat the same test.
What it does not mean: that small real changes are worthless: a trained athlete's expected gain can be the size of the noise (SYNTH-93).
Mechanism
Typical error, the spread of one person's repeats, is the standard noise measure; a high retest correlation can hide it in a varied sample (REF-1361). As a CV: about 0.9% for incremental peak power, 0.9-2.0% for constant-power tests of 1 min to 3 h, about 1.5% for lactate-threshold power, 1.3 times larger in non-athletes; the second trial beat the first by 1.2%, later ones by 0.2% (REF-1278).
A difference between two readings carries √2, about 1.4, times the single-test error, so "change ± one typical error" is only a ~50% interval, a lenient rule (REF-1362).
Whether a real change matters is a second question, the smallest worthwhile change (practice, REF-441).
Common misreading
"Up 5 W, so fitter"; "my watch VO2max fell 2 points, so I am losing it".