Running in the wild: Energetics explain ecological running speeds

Abstract

Human runners have long been thought to have the ability to consume a near-constant amount of energy per distance traveled, regardless of speed, allowing speed to be adapted to particular task demands with minimal energetic consequence.^1-3 However, recent and more precise laboratory measures indicate that humans may in fact have an energy-optimal running speed.^4-6 Here, we characterize runners' speeds in a free-living environment and determine if preferred speed is consistent with task- or energy-dependent objectives. We analyzed a large-scale dataset of free-living runners, which was collected via a commercial fitness tracking device, and found that individual runners preferred a particular speed that did not change across commonly run distances. We compared the data from lab experiments that measured participants' energy-optimal running speeds with the free-living preferred speeds of age- and gender-matched runners in our dataset and found the speeds to be indistinguishable. Human runners prefer a particular running speed that is independent of task distance and is consistent with the objective of minimizing energy expenditure. Our findings offer an insight into the biological objectives that shape human running preferences in the real world-an important consideration when examining human ecology or creating training strategies to improve performance and prevent injury.

Keywords: activity tracking; big data; cost of transport; energetics; fitness tracking; gait; metabolic cost; preferred speed; running; wearable sensing.

Figure 1.. Characteristics of free-living runners and runs.

We analyzed 4,645 runners and 37,201 runs. These runners spanned a broad range of ages (A), body mass indices (B) and average running speeds (C). Average running speed decreased with increasing age (D) and increasing body mass index (E). Few runners ran more than 20 runs, with 8 runs being the average (F). The most commonly run distance in the database was approximately 5 km (3 mi), with the majority of runs being under 10 km (G). Related Table S1 shows the criteria used to select runners from the database for these analyses. To further assess our runner population, related Figure S1 compares runners’ top performance paces to online race statistics.

Figure 2.. Free-living runners prefer a particular speed, largely independent of distance.

(A) Run average speed versus run time across a range of run distances. Each grey dot represents a single run. Striations in the data illustrate that runners often target integer run distances (in miles). Curved colored lines represent distances included in the comparisons presented in B and C. (B) Runners’ average speed did not change across run distance within any of the three distance comparisons (orange, teal, and purple), except for the longest, 11.27 km, distance. A comparison of speeds across orange, teal, and purple appears to reveal an increase in speed for longer distances; however, this is the result of different runner populations within each color. This across-color comparison does not control for differences in runners who choose to run longer or shorter distances, while our within-color statistical comparisons do. (C) Histograms illustrating runners’ percent change in speed relative to their average speed for each of the three run distances in the comparison. Color and shading correspond to the distances presented in A and B. Related Table S1 shows the criteria used to select runners from the database for these analyses.

Figure 3.. Free-living runners’ preferred speeds are energy optimal.

A comparison of average energy optimal speeds from lab experiments {Rathkey:2017fi, SteudelNumbers:2009dp, Willcockson:2012ed} for females (A, teal dashed line) and males (B, orange dashed line) to age and BMI matched free-living runners average preferred speeds (black dashed lines). Black and colored bars represent the 95% confidence interval in the average preferred speeds and average energy optimal speeds, respectively. For illustrative purposes, the cost of transport curves, averaged across experimental participants^–, are plotted for females (A, teal curve) and males (B, orange curve) and the grey histograms display free-living runners’ average preferred speeds. There were no differences between average preferred speed (free-living data) and average energy optimal speed (laboratory data) for either females or males. Note that the average energy optimal speeds, plotted as colored dashed lines, differ slightly from the minima of the average cost of transport curves because to calculate the average energy optimal speeds (and 95% CI in these speeds) we fit cost of transport data for individual participants, solved for the individual optimum, and then averaged these values. Related Table S1 shows the criteria used to select runners from the database for these analyses. Related Figure S2 shows the results of the free-living runner and experimental participant matching methods, while Table S2 presents a sensitivity analysis.