REF-1255 · Maier et al., 2017 · International Journal of Sports Medicine · Paper · PubMed

Will two power meters agree?

Not always: 54 riders' own meters (9 makers, pre-2017) read 0.9% low on average against a physics reference, varied ±3.2% (SD) between meters, and 6 of 54 were off by more than 5%.

B-limited, mixed evidenceMeasured on recreational to elite athletes

Maier et al. (Swiss Federal Institute of Sport, Magglingen) asked how true and how precise the mobile power meters that cyclists actually use are.

They calibrated 54 meters from 9 manufacturers belonging to 32 elite and recreational cyclists against a first-principles mathematical model of treadmill cycling: with the rider coasting downhill on a motorised treadmill, a back-pulling system was set to cancel the downhill force, then loaded 3 times with 4 different masses (the abstract's wording) while the rider pedalled to hold position, so the power needed is known from physics.

Trueness was the mean deviation from the model, precision the coefficient of variation. On average the meters read 0.9% low (−0.9 ± 3.2%, mean ± SD) and six deviated by more than ±5%; precision was high (CV 1.2 ± 0.9%) but differed between makers, with Stages varying more than SRM and PowerTap (both p < 0.001).

The authors conclude that power meters vary considerably in trueness and advise calibrating and adjusting every meter against a first-principles reference.

Our reading: the base's first fleet test of riders' own meters; most read within a few per cent, about one in nine was off by more than 5%, and the error is mostly a systematic per-meter deviation in trueness (a percentage of power, not the zero offset), not noise (precision) — so a meter is a consistent instrument for itself and a doubtful one for comparison with another meter.

Pre-2018 hardware; read from the abstract.

Protocol

Who and what: n = 54 mobile power meters, 9 manufacturers, belonging to 32 cyclists described as elite and recreational.
Criterion: a first-principles mathematical model of treadmill cycling. The cyclist coasted downhill on a motorised treadmill while a back-pulling system was adjusted to counter the downhill force; the system was then loaded 3 times with 4 different masses while the cyclist pedalled to keep position.
Outcomes: trueness = mean deviation from the model; precision = coefficient of variation; comparisons between manufacturers.
Not in the abstract: treadmill speed and gradient, the power range and cadence covered, the meter models and their measuring sites (crank, spider, pedal, hub), firmware, whether a zero offset was done before each test, temperature, and the statistics behind the brand comparison beyond the p-values.
Funding and interests: the abstract record states that the authors declare no conflicts of interest; funding is not given in the abstract (full text not read).

Key Results

OutcomeResult (abstract)
Trueness, all 54 metersMean deviation −0.9 ± 3.2% (mean ± SD); read as meters below the model, a sign convention the abstract does not define
Meters off by more than ±5%6 of 54
Precision, all 54 metersCV 1.2 ± 0.9% (mean ± SD)
Precision by manufacturerStages varied more than SRM (p < 0.001) and PowerTap (p < 0.001)
Our arithmetic: 6 of 54 is 11%, about one meter in nine. If deviations were roughly normal, a mean of −0.9% with SD 3.2% puts 95% of meters between about −7.2% and +5.4% of true power; at 250 W that is about −18 to +14 W.
Our arithmetic (assumes independent errors): the difference between two meters picked at random has an SD of about 3.2% × √2 ≈ 4.5%, so two meters on the same ride could disagree by up to about ±9% (95%) — the reason two head units or a trainer and pedals may show different watts for the same effort.

Mechanisms

Measured: trueness (each meter's systematic offset) varied far more between meters (SD 3.2%) than precision (repeat-to-repeat CV 1.2%) varied within them.
Authors' interpretation: current meters vary considerably in trueness while precision is generally high but differs between manufacturers; every meter should be calibrated and its trueness adjusted against a first-principles reference.
Inference: a strain-gauge meter's slope (its gain) is set at the factory and drifts with temperature, ageing, damage or reassembly, while the rider's zero offset corrects only the reading with no force applied; a meter can zero perfectly and still read a few per cent high or low under load. The abstract does not say which of these the deviations came from.

Practical Application

Will two power meters agree? Not always: 54 riders' own meters (9 makers, pre-2017) read 0.9% low on average against a physics reference, varied ±3.2% (SD) between meters, and 6 of 54 were off by more than 5%.
Keep one meter for the numbers that matter (B-, from the abstract): a meter repeated itself within about 1% but its offset from true power varied between meters, so FTP, CP and zones measured on one meter belong to that meter.When the meter changes — new pedals, a trainer indoors and a crank outdoors, a borrowed bike — expect a few per cent and up to about 5% or more, and retest on the new meter (Not from this paper; the base's own practice: or record one ride on both meters together to see the offset). This extends SYNTH-53's same-instrument rule for lactate meters to power meters.
A "known" power meter is a reference only for itself: SYNTH-56 advises checking trainer power against a known power meter before building zones on it; Maier shows that the known meter is itself off from true power by a typical ±3% (SD), so the check tells you whether trainer and meter agree, not which is right.
Small changes need the same meter: a 2-3% change in FTP is inside the spread between meters here; read it only on the same meter, same protocol (base practice in SYNTH-53; inference from these numbers).
Zero offset before riding, per the maker (Not from this paper; the base's own practice): it removes drift in the unloaded reading but not a slope error; the authors advise checking trueness against a physical reference, which few riders can do.

Limitations

From the abstract only. The power range, cadence, meter models, measuring sites, firmware, zeroing routine and per-brand trueness are not in the abstract; the full text (Thieme, paywalled; the repository copy at BFH ARBOR is access-restricted) was not read.
Model-based criterion. The reference is a first-principles model of treadmill cycling with known masses, not a calibrated laboratory instrument such as a dynamic calibration rig; its own uncertainty is not given in the abstract.
Old hardware. Meters in riders' hands before the 2017 publication; makers have since changed hardware and firmware, and pedal-based meters may be under- or over-represented (not stated). No current model was tested.
Conditions. Steady pedalling on a treadmill; low power, sprints, out-of-saddle efforts, temperature change and road vibration are not covered by the abstract.
Funding. No conflicts of interest declared; funding not stated in the abstract.
How this entry was checked and changed
Why this level

from the abstract; a fleet test of n = 54 power meters, which the rubric places at B as a validation study; held at B- because the criterion is a first-principles model of treadmill cycling, not a laboratory reference instrument, and because the abstract gives no power range, cadence, meter models or calibration routine (key methods missing)