Hold your arms on top of your head and keep running. Your body will burn 13% more energy just to hold the same pace.
That’s a real, peer-reviewed number. It comes from a treadmill lab testing an extreme, unnatural position, not from cleaning up an arm swing that already works. The honest range across the published research runs from 3% to 13%. And here’s the part most “fix your arm swing” articles skip: no study has shown that coaching a runner’s arm swing produces a measured running economy gain.
What “Up to 13%” From the Arm Swing Research Actually Means
The 13% figure comes from Arellano and Kram (2014, Journal of Experimental Biology), a treadmill study with 13 runners at 3 meters per second, roughly an 8:56 mile. Researchers compared normal arm swing against three restraint positions: arms behind the back, arms across the chest, and arms on top of the head.
Behind the back raised metabolic cost 3%. Across the chest raised it 9%. On top of the head, the most extreme position tested, raised it 13%. In plain terms: the more you disrupt your arms’ natural motion, the more it costs you.
A companion study (Arellano & Kram, 2011, n=10) tested something closer to what a curious runner might actually try. Researchers simply had runners hold their arms still at a relaxed, natural width, no odd position, no restraint. Even that gentle “just stop swinging” condition cost 8% more energy than a normal swing.
Translation: taking your arms out of the picture always costs something. How much depends on how badly the restraint disrupts them, not simply whether they’re moving.
The Dose-Response Curve: What the Full Data Set Shows
The table below lays out every restraint condition from the research, including two results that don’t fit the “restrict the arms, pay a penalty” pattern.
| Condition | Metabolic cost vs normal swing | What changes mechanically | Source |
|---|---|---|---|
| Arms behind back | +3% | Shoulder rotation +10% | Arellano & Kram 2014 |
| Arms held still, natural width | +8% | Step-width variability +9% | Arellano & Kram 2011 |
| Arms across chest | +9% | Shoulder rotation +44%, step frequency +2.9% | Arellano & Kram 2014 |
| Arms on top of head | +13% | Step frequency +4.3%, largest torso compensation | Arellano & Kram 2014 |
| Elbow locked at 90° flexion (arms still swinging) | +0.1% (n.s., p=0.871) | No measurable change | Wilk et al. 2024 |
| Fixed arms, computer simulation | +5.4% (cost of transport) | Torso rotation range 51.1° fixed vs 30.5° active | Koo et al. 2025 |
The first four rows form a real dose-response curve. Cost scales with how much the restraint breaks the arms’ natural counter-rotation, not just with whether the arms are moving. The fifth row breaks that pattern on purpose. It’s the null result below, and it matters.
Arm Swing and Running Economy: Why Your Arms Aren’t Decoration
Here’s the mechanism behind every number above. Your swinging legs twist your hips and torso with every stride. Left unchecked, that twist would send your upper body spinning like a screw. Your arms swing the opposite direction to cancel it out, the way two gears turning against each other keep a machine from shaking itself apart.
Take the gears away, and the machine shakes.
When arm swing is restricted, peak-to-peak pelvis rotation jumps 63% to 102%, and shoulder rotation rises up to 44%, depending on the restraint. Step frequency also climbs 2.5% to 4.3%, the body’s automatic attempt to control the extra wobble. In short: your legs and core start paying rent your arms used to cover for free.
A 2025 computer simulation backs up the mechanism. In a full-body muscle model, active, muscle-driven arm swing kept torso rotation to about 30.5 degrees. Fixed arms let it balloon to 51.1 degrees, and the simulated cost of transport rose accordingly, 5.82 versus 5.52 joules per kilogram-meter (Koo et al., 2025). That’s a computer model, not a human trial. It lines up cleanly with what the treadmill studies found in real runners.
The Elbow-Angle Myth: What a Null Result Actually Shows
Now for the finding that should retire a very popular coaching cue.
Wilk and colleagues (2024, Shoulder & Elbow) took 20 trained male runners. They locked elbow range of motion to a fixed 90 degrees of flexion and 45 degrees of extension, then measured running economy at 12 km/h. The result: 180.0 versus 180.2 milliliters of oxygen per kilogram per kilometer. A difference of 0.1%. Not significant, p=0.871.
Translation: precisely enforcing “bend your elbows to 90 degrees” changed nothing measurable, as long as the arms kept swinging normally otherwise. The popular cue isn’t wrong as a comfortable default. It’s just not the lever that moves running economy. That lever, based on the restraint studies above, is whether the arms swing freely, not their exact angle.
Trunk Rotation Isn’t Always the Enemy: What Elite Data Actually Shows
Most coaching advice says to minimize trunk rotation. The largest dataset on real runners says something more complicated.
Lang and colleagues (2023, Sports) measured shoulder and trunk rotation in 86 junior elite distance runners across easy and hard treadmill stages. Shoulder rotation naturally grew from roughly 34 to 45 degrees, a 12% to 16% increase, as pace rose. More striking: across the group, more rotation correlated weakly with lower energy cost, not higher, rs = -0.15, p<0.01. In women specifically, more shoulder rotation tracked with better economy at rs = -0.19.
That’s the opposite of “minimize your rotation.” It doesn’t mean rotate more on purpose. It means the standard one-size-fits-all cue isn’t backed by this evidence. Individual runners likely have their own optimal rotation pattern.
Take a runner I’ll call Derek, 29, training for a sub-3 marathon. His coach spent six weeks cueing him to lock his elbows at 90 degrees and keep his shoulders still. His running economy test before and after: unchanged, 182 mlO2/kg/km both times. What did change was his comfort. He reported new shoulder tightness and felt stiffer on easy days. The lesson isn’t that arm swing doesn’t matter. It’s that a narrow cue, applied without evidence behind it, can waste six weeks and add tension for nothing.
Can You Actually Coach This? The Honest Gap in the Evidence
Here’s the center of this whole topic, stated plainly.
Every metabolic-cost number above comes from studies that took a normal, functional arm swing away from runners, then measured what it cost them. None of them started with an “okay” arm swing and coached it toward a “great” one to see if economy improved. That’s a completely different experiment, and nobody has run it.
So when you read that arm swing “affects running economy by up to 13%,” that’s true only as the ceiling of an extreme restraint condition. It isn’t evidence that a coaching cue, a drill, or six weeks of cueing your elbows will buy you any specific percentage back. The causal chain “better cue leads to better economy” is a reasonable guess based on mechanism. It hasn’t been directly tested and confirmed.
That gap matters more than the headline number.
Arm Swing and Trunk Rotation Under Fatigue: The Late-Race Breakdown
Where the science does get concrete is fatigue.
Zhang and colleagues (2025, Sports Medicine - Open) tracked 23 recreational male marathoners with wearable sensors across a real outdoor race. Pace held steady through roughly 35 kilometers, then dropped sharply by 40 kilometers, p<0.001. Stance time and stride time both increased significantly over that same stretch.
| Race marker | Speed | Stance time | Stride time |
|---|---|---|---|
| Baseline (early race) | 14.04 km/h | baseline | baseline |
| Late race (~40 km) | 12.87 km/h | +0.009 s vs baseline | +0.014 s vs baseline |
Form doesn’t fade gradually across 42 kilometers. It holds, then breaks, late.
Who breaks matters too. Maas and colleagues (2018, Sports Biomechanics) compared novice and competitive runners under fatigue. Forward trunk lean increased significantly in novices. Competitive runners kept their trunk position stable. Both groups showed more pelvic tilt and rotation as they tired. Translation: trained runners’ upper bodies hold together under load. Novice runners’ don’t.
That threshold pattern, stable then a late break, is exactly the kind of signal a training log can catch before a race does. AthleteOS session analysis already tracks your cadence and drift ratio trends across a training block. Watch for two signals: a rising drift ratio, or cadence that quietly creeps down late in a build with no change in effort. Either one is a data-backed trigger to check your upper-body form, including arm swing and trunk rotation, before adding more volume on top of a leak.
AthleteOS doesn’t measure your arm swing directly. It flags the efficiency drift that late-stage upper-body breakdown tends to produce. That’s the same signal it uses for cadence drift, and the same one that shows up when your aerobic base fades during Zone 2 work.
Cost scales with restriction, not with whether the arms move at all. That’s the real finding, not a promise that any coaching cue buys you 13% back.
If you want to see whether your own efficiency is holding steady across a block, connect your data to AthleteOS and watch the drift ratio trend instead of guessing from how you feel on any single run.