What your watch actually measures

0:03 · Three places a number comes from

Every number on your watch came from one of three places: a sensor, a formula, or something you typed in once. This video teaches you to tell them apart.

0:18 · A number with a colour next to it

On your wrist right now there is a number with a colour next to it, and it looks measured. Some of it was, and the rest was worked out from a value you typed in on the day you unboxed the watch.

0:33 · Three rungs

One of the values it started from moves every heart-rate zone on your watch — we will get to which one, and by how much. This is EnduranceProof, and this video is a ladder with three rungs: measured, modelled, assumed. Three kinds of number, and three levels of trust.

0:56 · What the watch can sense

Strip the software away and the watch has four ways of knowing anything: light through your skin, where the satellites say you are, how your wrist moves, and air pressure. Across a review of 158 validation papers, wrist heart rate landed within about 3 percent on average. On a heart rate of 150, that is four or five beats — smaller than anything you would feel. The light is the one that matters most, and here is how it counts. A small green lamp and a detector sit against your skin, and the light that comes back changes as the blood underneath swells with each pulse. So the watch does not see your heart beat, it sees the pulse arrive at your wrist, and counts. That is why a chest strap, which reads the heart's own electrical signal, is still the reference in a lab. So where does the watch go wrong?

2:00 · Rung one: measured

The catch is in the word "average". Over a whole run the wrist is very close, but any single reading in the middle of that run can be well off. The failure runners notice most has a rhythm to it. A study of 53 people found the error during activity was, on average, 30 percent higher than at rest. With a steady arm swing the sensor can lock onto the swing and mistake it for the pulse. If your heart rate ever sat suspiciously close to your cadence for a few minutes, that was it. Distance is the other measured number, and its weather is the sky. Eight watches were taken over courses measured to the metre. On an open track every watch was within 5 percent, and read slightly long. In forest and city streets every watch read short, by between 3.5 and 9 percent. The signal bounces off buildings and trees before it reaches the wrist. Up to 9 percent short on a 10K is up to 900 metres. So a measured number is close on average, wanders on any single reading, and has a short, physical list of things that push it around. On the next rung that list gets longer, and quieter.

3:33 · Rung two: modelled

Pace, calories, VO2max, training load, recovery time: none of these was measured. Each was worked out from the first rung by a formula, and the formula needed something about you to start from. Here is what that costs. Sixty volunteers wore seven wrist devices, a few at a time, while a chest ECG measured the truth. Heart rate came out within 2 to 7 percent across the seven devices. Calories, from exactly the same wrists, were off by between 27 and 93 percent. Same wrist, one rung up, and no device got within 20 percent. The fitness number is the most interesting case, because one manufacturer has published how it works. It logs your age, and it records heart rate and speed on a normal run. It keeps the reliable segments and reads fitness off how heart rate relates to speed. And it scales that to a maximum heart rate estimated from your age, because you never ran to your limit. The maker's own chart says what that costs, and we read it as documentation, not as proof. A maximum guessed 15 beats too low puts the estimate about 9 percent off; with the real maximum known, about 5. So read a modelled number as a trend of you against you, not as your level. And notice what every one of those models needed to start from.

5:16 · Where the models start

Training load works the same way: time multiplied by how hard. And "how hard" is your heart rate placed between a resting value and a maximum. A resting value and a maximum, both from the third rung. That was what the watch measures and what it models. Now the third rung: what it assumes about you — and that is where the biggest number in this video sits.

5:45 · Rung three: assumed

Most watches start you off with a maximum heart rate of 220 minus your age. Two exercise physiologists went looking for where that came from, and published the answer in 2002. It goes back to a 1971 review. The authors plotted about 35 data points from about eleven earlier sources, and fitted no line to them. They wrote in the caption that no single line would adequately represent the data. And yet, they added, 220 minus age "defines a line not far from many of the data points". When the 2002 authors fitted a line through those same points properly, they got roughly 215 minus 0.9 times age. Individual maxima scattered 21 beats around that line. Across the research since, real maxima scatter 7 to 11 beats either side of the age formulas. Here is what that does to you. For a forty-year-old the formula says 180. Real maxima at that age sit between 169 and 191. Now build a 70-to-80-percent zone on each of those. At 169 the zone is 118 to 135, and at 191 it is 134 to 153. The lowest and the highest zone barely touch. And every number above it moves with it: every zone, every training effect, and, as the maker's own chart showed, the fitness estimate too. The most useful value in your whole settings screen is a maximum that was actually observed on you, not estimated from your age.

7:45 · HRV and readiness

One last stop, for the numbers with a colour next to them. Heart-rate variability starts with something measured: the gap between beats, in milliseconds, usually at night while your wrist is still. Everything after that is a choice. Which statistic — most watches use one that averages how much each gap differs from the last. The baseline is your own history, so the score compares you with you. And which weights turn stress and recovery into one readiness number. One maker has published how it detects stress and recovery from heart rate, variability and breathing. It looks at a four-hour window that starts thirty minutes after you go to bed. The weights that turn that into the score on your screen are not published by anyone. Whether that score predicts anything about your next session is a different question, and it is in our queue. Here the reading rule is short: several rungs above the sensor, calibrated to your own past, and one morning inherits every wobble from the pulse underneath.

9:02 · How to read the number

Two habits fall out of all this. Read a measured number as a value and a modelled number as a trend. Your heart rate right now is a fact with a small error bar, and your VO2max this week is a direction of travel. And when a number surprises you, look down the ladder before you look at your training. An old weight, a maximum that was never measured on you, a run under trees, a loose strap. A watch is a very good sensor wearing a very confident dashboard. The full piece, with all fifteen sources and every figure, is at enduranceproof dot com. Our next video takes the zones themselves apart: what training zones are, which one does what, and how you set yours. If you would like that one when it lands, subscribing is how you get it.