Your stride wobbles a little on every run, even at a steady pace. For years, running media has claimed more wobble means more injury risk. A 2025 scoping review says that’s too simple.
Researchers compared stride variability between injured and healthy runners across 21 studies. Just over half, 52%, found injured runners had higher variability. About a quarter, 24%, found lower variability. Over a third, 38%, found no difference (Iqbal & Chow, 2025). A meta-analysis wasn’t possible. There’s no standard normal range.
Translation: nobody has a clean rule you can trust yet.
That split matters, because “variability” means two different things.
What the 2025 Scoping Review on Running Stride Variability and Injury Found
Two more numbers matter. Ninety percent of the 21 studies were done in a lab, not on the road where you train. And only 9% used the more advanced “nonlinear” methods needed to test why variability might connect to injury.
Most of the field is still measuring the wrong kind of wobble.
Linear vs. Nonlinear Variability: Two Numbers, One Confusing Name
“Stride variability” usually means one of two different things.
Linear variability is the simple version: the coefficient of variation (CV) or standard deviation (SD) of stride time or length, how much the numbers spread around the average. Healthy runners show a stride-time CV around 3%, or as low as 1.0-1.6% near threshold (lactate threshold study, n=33). Translation: it shows how much your stride timing bounces around, not why.
Nonlinear variability is different. The most common version is DFA-alpha (Detrended Fluctuation Analysis), a fractal method scoring the structure of your stride pattern, not just its spread. Values run roughly 0.5 to 1.0: near 0.5 looks like random noise, near 1.0 is rigidly repetitive, and a healthy pattern sits between.
Think of it like a drummer’s groove versus a metronome. A good drummer’s beat has structure, each hit loosely connected to the last, adjusting without falling apart. A metronome just repeats the same interval, with zero give. Your stride should act like the drummer. When it starts acting like the metronome, that’s the “loss of complexity” researchers watch for.
In one study of 18 runners, injury-history runners showed a lower DFA-alpha (0.79) than healthy controls (0.96), a significant gap (p=0.01). The same two groups showed no difference in stride time, SD, or CV (Meardon et al., 2010). Two runners could log an identical “3% wobble” and sit on completely different patterns underneath. Only the fractal math caught it.
That’s the gap in the literature. Ninety percent of studies never even looked.
Too Rigid Is Also a Warning Sign: The U-Shaped Curve
Here’s where “more variability is worse” breaks down.
Variability isn’t a straight line against speed, it’s a U. A 90-person study measured contact time, stride time, and duty factor at five speeds, 80% to 120% of preferred pace. Variability was lowest near preferred pace and rose moving away from it either direction (P≤0.005) (Hein et al., 2014). Translation: your stride is most consistent at the pace you naturally choose.
The same study found something that flips the “chaos is bad” intuition: previously injured runners showed significantly higher stride-to-stride correlation of footstrike pattern than never-injured runners (P=0.046). More predictable, not messier.
A separate study of 40 runners split by experience adds a wrinkle. Runners with 10+ years of experience had a lower baseline stride-rate DFA-alpha (0.66) than novices (0.71, p<0.001), and adapted faster to forced step-frequency changes. Authors read this as flexibility, not injury risk (running experience study, n=40).
So a lower number shows up in injury-history runners in one study, and in fit, experienced runners in another. Same dial, opposite meaning.
A single number means nothing without context.
What Your Garmin, COROS, or Stryd Can (and Can’t) Tell You
This is the part most articles skip. It decides what you can act on today.
| Metric | What It Measures | Typical Healthy Value | Seen in Injured / Fatigued States | On a Consumer Watch? |
|---|---|---|---|---|
| Stride-time CV% | Spread of stride duration | ~3% (as low as 1.0-1.6% near threshold) | Often no significant difference | Not shown by name |
| Stride-interval DFA-alpha | Structure of stride timing pattern | ~0.70-0.80 (no fixed norm) | Lower (~0.66-0.79), more rigid | No. Needs raw accelerometer + offline analysis |
| Ground contact time balance | Left/right stance symmetry | Roughly 50/50 ± a few % | Asymmetry anecdotally linked to fatigue | Yes. Garmin running dynamics pod, COROS POD 2 |
| Vertical oscillation | Vertical bounce per stride | Pace- and device-dependent | No universal norm | Yes. Wrist or chest strap |
| Stride length SD | Spread of stride length | Not established; U-shaped with speed | Rises away from preferred pace | Partial. Per-lap averages, not per-stride SD |
No consumer watch, not Garmin, COROS, Stryd, or Polar, computes DFA-alpha, sample entropy, or the Lyapunov exponent of stride interval. Those need raw accelerometer data and offline math, still a lab job.
What your watch can give you: ground contact balance, cadence, stride length, and vertical oscillation. None of those are the fractal “loss of complexity” signal the research points to. They’re a different, more basic layer, worth watching, especially contact-time balance during a long effort.
Don’t Mix This Up With DFA Alpha-1 From Your HRV App
One more mix-up is worth clearing up.
Stride-interval DFA-alpha analyzes the pattern between your footsteps. DFA alpha-1, the number some HRV apps surface, analyzes the pattern between your heartbeats to estimate aerobic threshold, not injury risk (Rogers et al., 2021). An alpha-1 value around 0.75 marks the aerobic threshold zone.
Same math technique. Same Greek letter. Two different signals from two different body systems: one reads your legs, one reads your heart. If a training app shows “alpha” during a threshold session, that’s the heart-rate version, not the stride version covered here.
Same name, different organ. Don’t conflate them.
A Composite Case: How Consistency Could Drift Before the Pain Does
Picture a runner named Maya, 34, training for a fall marathon, six days a week. Her average cadence holds steady at 172 spm for six weeks. By the old 180-spm heuristic, her form looks fine (though 180 was never a validated target).
What her cadence doesn’t show: she’s mid-way through a self-directed mileage build, the kind of two-week overload block common in real training logs.
This comes from a real field study tracking 9 distance runners through an overload block, not a lab treadmill (Fenton et al., 2024).
Translation: their stride pattern got more rigid under fatigue, then loosened once they rested. A runner in Maya’s position could show the same quiet shift, weeks before anything hurts. Her average cadence never moves. The pattern underneath it does.
AthleteOS doesn’t calculate stride-interval DFA-alpha today. No consumer platform does. What it tracks is a related signal: how your heart-rate-to-pace relationship, your drift ratio, shifts within a run and trends across weeks. Same idea: don’t just log the average, watch whether the pattern holds steady or slides.
Watch the Trend, Not the Average
Here’s the honest limit of this evidence. Almost every study above is retrospective, comparing runners already injured against those who never were, or a short prospective window inside one training block. Nobody has run a large, multi-month study proving stride variability predicts injury before it happens. The research supports variability as a correlate worth watching, not yet a validated screening test.
That doesn’t make it useless. It means the smart move is tracking trends, not chasing a single “good” number that doesn’t exist.
If you’re building mileage, watch your training load and freshness so a heavy block doesn’t sneak up on you. Check your morning HRV trend, not one daily reading. And watch your drift ratio on long runs, since rising drift and a rigidifying stride often share the same fatigue.
The single number lies. The trend doesn’t.
Start tracking your training load, drift ratio, and recovery trend together with AthleteOS.
Frequently Asked Questions
Does stride variability predict running injury? Not reliably. The 2025 scoping review found higher variability in 52% of studies, lower in 24%, and no difference in 38%. There’s no established normal range.
What is a good stride time CV percentage? Around 3% is typical in healthy runners, though trained runners near threshold measured as low as 1.0-1.6%. There’s no validated cutoff, so read any number as context, not a verdict.
Can my Garmin, COROS, or Stryd watch measure stride DFA-alpha? No. No consumer watch computes DFA-alpha, sample entropy, or the Lyapunov exponent of stride interval. Those need raw accelerometer data processed offline, a lab job.
Is DFA alpha-1 the same as stride DFA-alpha? No. Stride DFA-alpha analyzes the pattern between footsteps. DFA alpha-1, seen in HRV apps, analyzes the pattern between heartbeats to estimate aerobic threshold. Same math method, different signals.
Does running cadence affect stride variability? Indirectly. Variability is lowest near your preferred pace and rises as you move away from it. A steady average cadence can still sit on top of a pattern quietly degrading underneath.
Can stride variability change within a single run? Yes. A 68-runner field study of a real 5K found hills moved stride complexity more than fatigue did, while stride-time spread dropped significantly lap to lap. Terrain and pace both matter before reading one run as a health signal.