What happens when the pace picks up

A step counter that is accurate walking is not automatically accurate running, and it is not obvious in advance which way it should go. Two things change at once. The signal gets easier: footstrike is harder, the arm swings through a wider arc, and the rhythm is more regular than a walk through a car park. And the signal gets faster: cadence roughly doubles, and a counter that mistakes one stride for two, or two steps for one, has twice as many chances to do it.

Taken together the figures put running ahead of walking: jogs and runs come in at 5.3% from the hand count and walks at 6.6%, which is the direction the validation literature reports too — one review found consumer devices generally more accurate running than walking, and another found wrist comparisons landing closest to a true count at a jog and drifting below it at every other speed. Read in absolute steps rather than percentages the ordering flips, because a jog session puts about half as many steps again on the counter.

The indoor-to-outdoor gap is where running stops behaving like walking. On a walk the belt is the easier place and the pavement the harder one. At a run it goes the other way: 96.7% out on the road against 94.0% on a treadmill. The second half of the first paragraph is the reason. A treadmill holds a runner at one cadence for as long as they stay on it, and it holds them at the highest cadence any of these conditions reaches — and the faster the feet, the more often a counter has the chance to read one stride where two steps were taken. The direction of the error is the tell: on a belt the ring reads low, by about six steps in every hundred, and on the road by fewer than one.

The study cited below is the nearest published comparison, and it set a separate mark for each of these two conditions — 5% on a treadmill and 10% in free motion — measuring five wrist devices against the mean of two manual counters while forty volunteers walked and jogged. On the road the ring is well inside its mark, at 3.3%. On a belt it is at 6.0%: inside the 10% to 15% an expert consensus accepts for general use, which is the line drawn on the chart, and outside the tighter of the two marks that study set, which is the one it set for the belt on the assumption that a belt is the easier place to count.