Between-unit agreement of GPS-derived mechanical work and power: time to trust the numbers?
Buchheit M, Lopez Sagarra A. Between-unit agreement of GPS-derived mechanical work and power: time to trust the numbers? Sports Performance & Science Reports. 2026 Jun;300:v1.
New SPSR paper out. GPS 3.0 series, post #2.
If GPS 3.0 gave us mechanical work and mechanical power as a new way to read locomotor demands, the obvious next question from practitioners was simple: can I trust these numbers when I swap units between players?
So we tested it. Ten WIMU PRO EVO units (Hudl), mounted on a rigid trolley so every unit followed the exact same path, pulled through 12 football-specific tasks plus a full 16 min session in Doha. We processed the raw files two ways: native WIMU (GPS 2.0) and through ADI (Athletic Data Innovations) for the GPS 3.0 mechanical metrics.
What we found for the ADI metrics:
- Total mechanical work: between-unit CV of 0.39%
- Mean mechanical power: CV of 0.38%
- Peak mechanical power: CV of 3.52%
- The multi-directional components (cuts and arcs) were noisier (CV 7.0 to 7.7%) but still small once standardized
We did not stop at CV%. A CV alone does not tell you if the noise matters. So we standardized each error against the between-player variability seen in elite matches. On that scale, mechanical work, mean power and peak power are all trivial. Cuts and arcs are small. And the error is random, not systematic, so units can be rotated between players without introducing a consistent bias.
Two honest caveats I want to be clear about:
- This is a trolley study. It isolates pure device-to-device error and removes the body-worn vibration and soft-tissue movement you get on a real player. So this tells you the units agree with each other under clean conditions. It does not by itself tell you the metric is valid on a back. Reliable is not the same as valid.
- The directional components (cuts/arcs) are likely overestimated here, because units sat 10 to 20 cm apart on the trolley and took slightly different arcs through turns. On a single body-worn unit that artifact goes away.
One detail worth repeating: mechanical work and mechanical power are computed from GPS position and velocity only. The accelerometer is not involved. The only accelerometer metric here is PlayerLoad. That matters because GPS 3.0 is built as a better proxy for internal neuromuscular load, but it is still an external measure.
