RunningForm
Research basis

What we measure, and why.

RunningForm uses biomechanics reference ranges derived from peer-reviewed sports science research. This page documents the evidence behind each metric, our measurement methodology, and known limitations. We believe in transparency about what the science supports and where our analysis is approximate.

All biomechanics measurements are computed client-side from 2D video using MediaPipe Pose Landmarker. They are estimates, not lab-grade measurements. Reference ranges are pace-adjusted across three tiers: easy (> 6:00/km), tempo (4:30-6:00/km), and fast (< 4:30/km). Metrics tagged rear-view require an optional second clip filmed from behind the runner; without it, the analysis runs from the side view alone.

01

Vertical Oscillation

Unit: % of estimated body height (shoulder-to-ankle distance)

Thresholds

easy

< 5.5% good, < 8.0% moderate

tempo

< 6.0% good, < 8.5% moderate

fast

< 6.5% good, < 9.0% moderate

How we measure it

Computed as the total range (max minus min) of hip-to-ankle vertical distance across all analyzed frames, divided by the average shoulder-to-ankle distance. Using hip-minus-ankle cancels vertical camera panning. This total-range method yields higher values than per-stride averages reported by wearables like Garmin.

Limitations

  • ·"% of body height" is not a standard metric in the literature or wearable ecosystem. Garmin and Stryd use absolute centimeters or vertical ratio (VO / stride length). This is a novel normalization designed for camera-independence.
  • ·Body height is estimated as shoulder-to-ankle distance, which is ~75-80% of true height. This inflates percentages ~25% compared to true body height.

Sources

02

Trunk Lean

Unit: degrees forward from vertical

Thresholds

easy

2-8° good, 0-12° moderate

tempo

3-9° good, 1-13° moderate

fast

3-10° good, 1-15° moderate

How we measure it

Computed as the angle from vertical of the line connecting shoulder midpoint to hip midpoint, averaged across all frames with valid pose landmarks. Lean source (ankles vs. waist) is determined by comparing upper-body lean to lower-body lean.

Limitations

  • ·Contrary to popular coaching advice, elite runners do NOT increase lean with speed. The assumption that "faster = more lean" appears to be a recreational runner habit, not an elite characteristic.
  • ·There is a trade-off: more lean reduces knee load but increases hip extensor demand (Teng & Powers 2015). The "right" amount depends on individual injury history.

Sources

03

Foot Placement

Unit: % of estimated body height (shoulder-to-ankle distance)

Thresholds

easy

< 5% good, < 9% moderate

tempo

< 7% good, < 11% moderate

fast

< 9% good, < 13% moderate

How we measure it

Computed as the absolute horizontal distance between hip midpoint and the visible-side ankle at ground contact frames, divided by the shoulder-to-ankle body height estimate. Expressed as a percentage.

Limitations

  • ·No validated cutoff values exist in the literature (Souza 2016: "likely on a sliding scale, where lower values are generally associated with lower ground reaction forces"). Our thresholds are derived empirically.
  • ·The horizontal/vertical coordinate ratio introduces a constant scaling factor dependent on video aspect ratio (~0.56x for 16:9 landscape). Thresholds are calibrated for typical phone landscape video.
  • ·2D side-view video has known limitations for this measurement. We also compute Tibial Inclination at Contact (below) as the camera-independent shank-angle metric used in clinical settings — the two together describe the overstride pattern more robustly than either alone.

Sources

04

Cadence

Unit: steps per minute (spm)

Thresholds

easy

165–180 spm good, 155–190 spm moderate

tempo

170–185 spm good, 160–195 spm moderate

fast

175–195 spm good, 165–205 spm moderate

How we measure it

Estimated from stride times detected in the gait cycle analysis. A full gait cycle (same-foot to same-foot) covers 2 steps, so cadence = 120 / average_stride_time_seconds. Accuracy depends on detecting at least 2 clean gait cycles.

Limitations

  • ·Cadence from video has lower precision than wearables — a ±5 spm margin is expected. Do not over-interpret small deviations.
  • ·Low cadence (<160 spm) is strongly correlated with overstriding and is reported alongside foot placement — the two are treated as one finding rather than two separate issues.

Sources

05

Ground Contact Time

Unit: milliseconds (ms)

Thresholds

easy

240–300 ms good, 210–340 ms moderate

tempo

200–260 ms good, 175–290 ms moderate

fast

170–220 ms good, 150–250 ms moderate

How we measure it

Measured as the duration from initial foot contact to toe-off for the visible side, averaged across detected stance phases. Reported alongside Leg Stiffness Ratio (below), which is the dimensionless flight/contact ratio derived from GCT and stride time.

Limitations

  • ·Raw GCT is speed-dependent and should not be framed as "lower is always better." Runners with equal running economy can have very different ground times — the body uses multiple strategies.
  • ·GCT from 2D video is estimated from frame timestamps and gait event detection. Precision is approximately ±20–30 ms at our capture rate.
  • ·We previously reported duty factor (GCT ÷ stride time) here. Recent meta-analysis (Anderson 2024) found duty factor was non-significant against energy cost (r=−0.06), while leg stiffness was meaningful (r=−0.28). We moved the efficiency framing to Leg Stiffness Ratio.

Sources

06

Knee Flexion at Initial Contact

Unit: degrees from full extension

Thresholds

easy

15–30° good, 10–35° moderate

tempo

15–30° good, 10–35° moderate

fast

15–30° good, 10–35° moderate

How we measure it

Computed as (180° − interior hip-knee-ankle angle) at each detected initial-contact frame, using MediaPipe's 3D worldLandmarks rather than image-space coordinates. Reported as the average across stance phases. Camera-independent: image-space angles bake in perspective and aspect ratio; worldLandmarks do not.

Limitations

  • ·A more flexed knee at contact dissipates impact through eccentric quad action; a stiff-legged landing transmits load to the tibia and patellofemoral joint.
  • ·Not strongly speed-dependent — the literature and our thresholds use the same range across pace tiers.

Sources

  • Wille et al. 2014 — "Ability of Sagittal Kinematic Variables to Estimate Ground Reaction Forces and Joint Kinetics in Running." IJSPT, 9(7)

    Knee flexion at initial contact was one of five 2D-video sagittal kinematic variables in a regression model predicting vertical average loading rate (R² = 0.51).

  • Souza 2016 — "An Evidence-Based Videotaped Running Biomechanics Analysis." Physical Therapy in Sport

    Trained runners typically present 20–30° knee flexion at initial contact; values below ~15° are associated with high loading rate and an overstride pattern.

07

Peak Knee Flexion at Midstance

Unit: degrees from full extension

Thresholds

easy

40–50° good, 35–55° moderate

tempo

40–50° good, 35–55° moderate

fast

40–50° good, 35–55° moderate

How we measure it

For each detected stance phase (initial contact → toe-off), we compute knee flexion at every visible frame and take the maximum, then average those peaks across stance phases. Uses worldLandmarks for camera independence.

Limitations

  • ·Reflects shock absorption capacity and quad eccentric load. Too little flexion (<40°) is associated with patellofemoral pain; too much (>55°) with quad fatigue and reduced push-off.
  • ·Speed-independent in the literature; same range across pace tiers.

Sources

08

Tibial Inclination at Initial Contact

Unit: degrees from vertical (absolute)

Thresholds

easy

0–8° good, 0–15° moderate

tempo

0–10° good, 0–17° moderate

fast

0–12° good, 0–18° moderate

How we measure it

Absolute angle of the shank (knee → ankle vector) from vertical in the sagittal plane at initial contact, using worldLandmarks. Direction-agnostic: we measure magnitude, not signed angle, so the metric is identical whether the runner moves left-to-right or right-to-left across the frame.

Limitations

  • ·A vertical shank (0°) means the foot lands directly under the knee. Forward inclination puts the ankle ahead of the knee — the clinical sagittal-plane overstride marker.
  • ·This is the camera-independent counterpart to the image-space Foot Placement metric above.

Sources

  • Souza 2016 — "An Evidence-Based Videotaped Running Biomechanics Analysis." Physical Therapy in Sport

    Vertical shank at initial contact is the clinical sagittal-plane overstride marker; deviation from vertical increases braking impulse and tibial bending moment.

  • Wille et al. 2014 — "Ability of Sagittal Kinematic Variables to Estimate Ground Reaction Forces and Joint Kinetics in Running." IJSPT, 9(7)

    Lower-leg angle at initial contact was a predictor in the 5-variable GRF/loading-rate regression (R² = 0.51 for loading rate).

09

Hip Extension at Toe-Off

Unit: degrees behind vertical (absolute)

Thresholds

easy

10–20° good, 5–25° moderate

tempo

10–22° good, 5–27° moderate

fast

12–25° good, 7–30° moderate

How we measure it

Absolute angle of the thigh (hip → knee vector) from vertical in the sagittal plane, measured at each detected toe-off frame using worldLandmarks. Direction-agnostic. Averaged across detected stance phases.

Limitations

  • ·Reflects hip-flexor mobility and push-off efficiency. Limited hip extension is the kinematic signature of compensatory overstriding — the runner reaches forward for what they cannot push backward.
  • ·Modestly speed-dependent; faster running typically produces slightly more extension.

Sources

10

Leg Stiffness Ratio

Unit: flight time / contact time (dimensionless)

Thresholds

easy

0.30–0.70 good, 0.20–0.90 moderate

tempo

0.45–0.85 good, 0.30–1.05 moderate

fast

0.60–1.05 good, 0.45–1.30 moderate

How we measure it

Derived from existing GCT and stride-time measurements: for a full gait cycle, flight time = stride/2 − contact, and the ratio is flight ÷ contact. This is the body-mass-free form of Morin's spring-mass model — when scale calibration is unavailable, t_flight/t_contact tracks the same direction of change as absolute leg stiffness in kN/m.

Limitations

  • ·Higher ratio = stiffer leg = greater elastic energy return per stride. The strongest individually-validated correlate of running economy in the most recent meta-analysis.
  • ·Duty factor (the inverse framing, t_c / (t_c + t_f)) was non-significant against energy cost in the same meta-analysis (r=−0.06). We deliberately report the ratio rather than duty factor.

Sources

  • Morin et al. 2005 — "A simple method for measuring stiffness during running." Journal of Applied Biomechanics, 21(2)

    Sine-wave approximation of vertical stiffness from contact and flight times — the foundational method we derive from.

  • Anderson et al. 2024 — Meta-analysis of biomechanical correlates of running economy. Sports Medicine

    Vertical stiffness r=−0.31 and leg stiffness r=−0.28 with energy cost — strongest individually validated correlates. Duty factor was non-significant (r=−0.06).

  • Lussiana et al. 2019 — "The implications of time on the ground on running economy: less is not always better." Journal of Experimental Biology, 222(23)

    Low duty-factor runners exhibit stiffer legs; high duty-factor runners use longer ground contact with a softer leg. Equally efficient at endurance speeds.

11

Elbow Angle

Unit: degrees (interior shoulder–elbow–wrist angle)

Thresholds

easy

80–110° good, 70–120° moderate

tempo

80–105° good, 70–115° moderate

fast

80–105° good, 70–115° moderate

How we measure it

Mean interior elbow angle on the visible (near) arm across all frames where shoulder, elbow and wrist are visible. Computed from worldLandmarks for camera independence.

Limitations

  • ·The canonical coaching target is ~90° elbow flexion. Persistently outside ~70–120° suggests either tight, high-shoulder posture (too flexed) or an energy-wasting "long" arm swing (too extended).
  • ·Single-sided — we only see the near arm in side view.

Sources

12

Arm Swing Amplitude

Unit: % of estimated body height

Thresholds

easy

25–65% good, 18–80% moderate

tempo

30–70% good, 22–85% moderate

fast

35–75% good, 25–90% moderate

How we measure it

Front-to-back excursion of the near-arm wrist across the capture window, expressed as a percentage of estimated body height. Uses image-space X relative to the shoulder midline (which cancels camera pan) since worldLandmarks are hip-centred and oscillate around zero by construction.

Limitations

  • ·No standardized clinical range exists. Thresholds are set empirically against what side-view 2D yields for trained recreational runners (~30–60% body-height excursion).
  • ·Restricted swing (<20%) commonly co-occurs with tight shoulders; exaggerated swing (>75%) often co-occurs with cross-body arm action.

Sources

13

Stride-to-Stride Variability

Unit: coefficient of variation, % (SD ÷ mean × 100)

Thresholds

all

< 8% good, < 15% moderate, > 15% significant

How we measure it

For each kinematic metric with per-stride samples (trunk lean, knee flexion at IC, peak knee flexion, tibial inclination, hip extension, elbow angle), we compute the coefficient of variation across detected strides and attach it to the metric. Requires ≥3 valid samples.

Limitations

  • ·Elevated CV indicates inconsistency in form — typically from fatigue, an unstable surface, or pose-detection noise on a short clip.
  • ·Frame as an efficiency / fatigue signal, not an injury predictor — the same hedging we apply to asymmetry. Variability research in running has mixed findings on injury association.

Sources

14

Contact Time Asymmetry

Unit: % difference between left and right ground contact time

Thresholds

all

< 3% good, < 6% moderate, > 6% significant

How we measure it

Computed from ground contact time estimates for left and right foot strikes detected via gait cycle analysis. Expressed as the percentage difference between sides.

Limitations

  • ·Asymmetry does NOT reliably predict injury. The largest prospective study on this topic (Malisoux et al. 2024, n=836) found no association between gait asymmetry and running-related injury risk. We frame asymmetry as a running economy concern only.
  • ·Our measurement uses 2D pose estimation from a single side-view camera, which limits accuracy for bilateral comparisons.

Sources

15

Contralateral Pelvic Drop

rear-view

Unit: degrees

Thresholds

easy

0–4° good, 0–7° moderate

tempo

0–4° good, 0–7° moderate

fast

0–5° good, 0–8° moderate

How we measure it

Pelvic tilt of the swing side relative to the stance side at midstance, measured from the rear-view clip when one is provided. Averaged across detected stance phases.

Limitations

  • ·The single strongest kinematic predictor of running-related injury across PFP, ITBS, MTSS and Achilles tendinopathy in Bramah's injured-runner cohort.
  • ·Speed-independent in the literature; same range across pace tiers.

Sources

16

Hip Adduction at Midstance

rear-view

Unit: degrees

Thresholds

easy

0–8° good, 0–12° moderate

tempo

0–8° good, 0–12° moderate

fast

0–10° good, 0–14° moderate

How we measure it

Peak femur tilt toward the body midline during stance, measured from the rear-view clip. Averaged across detected stance phases.

Limitations

  • ·Frontal-plane 2D-video methods correlate moderately (not perfectly) with 3D motion capture for hip adduction.
  • ·Elevated peak hip adduction prospectively predicts PFP and ITBS development. Healthy runners typically peak at 5–10°; injured cohorts 12–15°+.

Sources

17

Step Width

rear-view

Unit: % of pelvic width (hip-to-hip distance)

Thresholds

easy

8–25% good, 4–32% moderate

tempo

8–25% good, 4–32% moderate

fast

6–25% good, 3–32% moderate

How we measure it

Lateral distance between contralateral midstance foot positions, expressed as a percentage of pelvic width. Computed from the rear-view clip.

Limitations

  • ·Crossover gait (step width approaching zero) co-occurs with elevated hip adduction and is associated with ITBS.
  • ·Healthy runners typically 10–20% of pelvic width; <5% is crossover.

Sources

  • Meardon, Campbell & Derrick 2012 — "Step width alters iliotibial band strain during running." Sports Biomechanics

    Narrow step width (toward crossover gait) increased peak iliotibial band strain — a mechanical link to ITBS development.

18

Foot Progression Angle

rear-view

Unit: degrees from direction of travel (absolute)

Thresholds

easy

0–10° good, 0–15° moderate

tempo

0–10° good, 0–15° moderate

fast

0–10° good, 0–15° moderate

How we measure it

Angle between the long axis of the foot and the running direction at midstance, measured from the rear-view clip. We report absolute deviation; the LLM describes direction (toe-in vs. toe-out) qualitatively from the raw frames.

Limitations

  • ·Toe-in and toe-out are reported with equal weight numerically. Excessive deviation in either direction alters loading patterns at the knee and ankle.
  • ·~5–10° toe-out is typical and considered within normal limits.
Note

Why we don't prescribe a target stride length.

Many running coaches suggest shortening or lengthening stride length as a coaching cue. RunningForm deliberately does not prescribe a target stride length, and the science is clear on why.

Cavanagh & Williams (1982) showed that trained runners self-select a stride length that is at or very near their individual metabolic optimum — the point where oxygen uptake is minimized for their speed. Deviating from preferred stride length in either direction (too short or too long) increased VO₂ by an average of 2.6–3.4 ml/kg/min. The body finds its own optimum through training adaptation.

What we do flag is overstriding — when the foot lands significantly ahead of the centre of mass, creating a braking impulse regardless of stride length. The fix is increasing cadence slightly or shortening the reach of the foot at contact, not prescribing a specific stride length target.

Source

Cavanagh & Williams 1982 — "The effect of stride length variation on oxygen uptake during distance running." Medicine & Science in Sports & Exercise, 14(1)

All subjects showed a U-shaped relationship between stride length and VO₂ with an individual optimum. Trained runners self-selected a stride length at or near their metabolic optimum.

This analysis is AI-generated and intended for educational purposes only. It is not a substitute for advice from a qualified running coach or physiotherapist. Biomechanics metrics are estimates from 2D side-view video, not lab-grade measurements.