REF-795 · Atkinson et al., 2007 · Journal of Sports Sciences, Vol. · Paper · PubMed

Variable Versus Constant Power Strategies During Cycling Time-Trials: Prediction of Time Savings Using an Up-to-Date Mathematical Model

Cweak evidence: opinion or lab dataMeasured on a mixed groupWritten before October 2026, being re-checked

Using the Martin et al. (1998) cycling power model with validated coefficients, the authors quantified the time a time-trial rider saves by varying power in parallel with gradient and wind instead of holding it constant.

For a hypothetical 70 kg rider on a 10 kg bike at a mean of 289 W, varying power by ± 10% saved 126 s on a 10-km course of alternating 1-km climbs and descents of 10%, 51 s on a 40-km course of alternating head- and tailwind sections, and 26 s on a "standard" 40-km time trial.

The savings were larger than an earlier model had predicted and largest for the lower-powered riders who could vary power the most. Whether a rider can tolerate such variation was, in the authors' words, still unclear.

Participants

No human participants. One hypothetical cyclist (mass = 70 kg) on a bicycle (mass = 10 kg), simulated at mean power outputs of 164, 289 and 394 W, chosen to mirror baseline performances studied previously.

Protocol

Model: Martin et al. (1998) power model with validated coefficients, replacing the Di Prampero et al. (1979) model that Swain (1997) had used.
Conditions: wind velocities from -10 to 10 m/s, gradients from -10 to 10%, and different pacing strategies (constant vs power varied with gradient and wind).
Scenarios: (i) 10-km time trial of alternating 1-km sections at 10% and -10% gradient; (ii) 40-km time trial of alternating 5-km sections of 4.4 and -4.4 m/s wind (Swain's course); (iii) the 40-km time trial described by Jeukendrup and Martin (2001).

Key Results

Scenario (mean power 289 W)Power variationTime saved
Hilly 10 km (Swain's course)± 10%126 s
Windy 40 km (Swain's course)± 10%51 s
"Standard" 40 km TT (Jeukendrup and Martin 2001)10%26 s
Savings for most scenarios were greater than Swain (1997) had predicted.
The largest savings were for the hypothetical riders with the lowest mean power output who could vary power to the greatest extent.
The authors judged these savings to compare favourably with the predicted benefits of altitude training or carbohydrate-electrolyte drinks.
Savings at 164 and 394 W are not given numerically in the abstract.

Practical Application

In a validated cycling power model (hypothetical 70 kg rider, 10 kg bike), varying a 289 W mean by ± 10% with the terrain saved 126 s on a 10-km course of steep climbs and descents, 51 s on a 40-km course of alternating head- and tailwinds and 26 s on a standard 40-km time trial; the slower the rider, the bigger the saving. To "should I push the climbs and headwinds in a 40-km TT?", the physics says yes: ride harder where speed is low and ease where it is high, and expect tens of seconds, more on hilly or windy courses.
The model does not know the rider's physiology: whether ± 10% swings can be held without fading was unclear to the authors, so this is a direction to practise, not a prescription.
Inference, not tested here: at the same mean power, variable power cost more oxygen and lactate than steady power in elite cyclists (REF-252), while small swings cost nothing measurable (REF-241); keep surges modest and rehearse them.
This entry puts numbers on the direction REF-126 (also a model, level C) gives, but it does not raise the level of that evidence: both are simulations.
Not from this paper; the base's own practice: on a flat, still-air course hold even power; the variable-power case applies to real gradient and wind changes.

Limitations

Simulation only: hypothetical rider, stylised courses (alternating 10% climbs and descents, alternating wind sections), no measured rides.
Constrains mean power, not physiological cost; the tolerability of the power swings was not modelled.
Only the 289 W savings are given numerically in the abstract.
An erratum exists (see the Journal line); its content was not checked. Written from the abstract and record metadata only (full text not read).
How this entry was checked and changed
Why this level

modelling: the power model's coefficients are validated, but the pacing predictions were not tested on riders