2016 — The Effect of Elbow Hyperextension on Ball Speed (Journal Version)

Research theme

The peer-reviewed, fully worked version of the elbow question. The ICC’s illegal-action law prohibits more than 15° of elbow extension between the upper arm reaching horizontal and ball release — but hyperextension (the joint going past straight, beyond 180°) is exempt, because it is treated as an involuntary consequence of the load on the joint. Angles above 20° of hyperextension are documented in bowling. So the sport’s own law grants naturally lax-elbowed bowlers a mechanical advantage — if it is an advantage. Nobody had separated extension from hyperextension, and the cause-and-effect cannot be established experimentally because you cannot ask a bowler to change their elbow laxity.

Method, plainly: a two-segment planar forward-dynamics simulation model of the bowling arm (upper arm; forearm + hand, ball as a point mass), customised to one elite male fast bowler and evaluated against a held-out delivery. The elbow is a damped linear torsional spring that only generates torque once the joint passes straight. A one-segment model (upper arm + forearm + hand fused) supplies the straight-arm counterfactual. Then ~7,000 simulations were generated by perturbing the elbow spring stiffness and the moment hyperextension begins, mapping the whole speed landscape.

What they measured

• How fast the ball leaves the hand (ball release speed, mph and m/s; wrist tangential velocity converted to ball speed using the bowler’s own measured wrist-flexion contribution).
• How far past straight the elbow goes (peak elbow hyperextension angle, °).
• How much the elbow springs back before release (recoil, expressed as a percentage of the theoretical maximum recovery).
• The elbow angle at the moment of release (°).
• The shoulder’s angle and how much work it does (shoulder release angle; work done at the shoulder from upper-arm-horizontal to ball release).
• How long the delivery period lasts (total time, s).
• The order of the arm segments’ peak speeds (proximal-to-distal sequencing: upper arm peak angular velocity vs forearm peak angular velocity).

Findings

All causal within the model, single bowler.

1. Model accuracy. Three deliveries matched concurrently with one common parameter set to 4.5%, 4.4% and 2.5% overall RMS difference (average 3.8%). A fourth, held-out delivery simulated with fixed parameters returned 4.4% — a proper evaluation. Component detail (Table 1): peak elbow angle differences 0.0–0.5°; elbow angle at release 0.4–2.5°; ball speed 1.1–5.0%; time 4.3–8.6%. Matched initial conditions ranged: shoulder angle 93–109°, elbow angle 0.1–4.3°, shoulder angular velocity 1187–1272 °/s, elbow angular velocity 144–379 °/s, ball speed 86.1–86.9 mph, peak elbow hyperextension 13.5–14.3°, elbow angle at release 11.3–12.9°.

2. This bowler’s hyperextension was worth 4%. Best match: 85.8 mph (38.1 m/s). Same initial conditions in the straight-arm one-segment model: 82.5 mph (36.7 m/s). Difference: 4%.

3. Optimised: 5%. Varying the two elbow spring parameters to maximise speed gave 86.6 mph (38.5 m/s) — +5% over a straight arm. The solution had 25° peak hyperextension (the imposed upper bound, based on King & Yeadon 2012) with 5° of recoil, and zero damping.

4. Any hyperextension beats a straight arm. Across ~7,000 perturbation simulations, classified as “Recoiling” (recoiling at release) or “At peak” (at maximum at release), both categories were faster than a straight arm. Proximal-to-distal sequencing was present in both (upper arm peak angular velocity before forearm peak).

5. The practically useful number — a rate. With optimal recoil, the relationship between peak hyperextension and speed gain has two regimes:

   • The first <1° of hyperextension is worth about 1% of ball speed (a steep initial exponential phase).
   • Thereafter, each additional 1° of hyperextension is worth about 0.2% of ball speed.

   This is the single most quotable result in the paper. It lets you price any bowler’s elbow.

6. Optimal recoil is an inverse-hyperbolic function of peak hyperextension (R² = 0.7). For essentially all real bowling hyperextensions (>1°), optimal recoil sits between 30% and 60% of the theoretical maximum, tending towards 60% as peak hyperextension approaches 1°.

7. Worked headline example. “A fast bowler with 20° of elbow hyperextension and an optimal level of recoil will have increased ball speeds of around 5% over a bowler without hyperextension.” At elite pace (>90 mph / 40 m/s) that is ~5 mph (2.2 m/s) — the paper calls this “a substantial increase in performance”.

8. Why it works — two competing mechanisms. (a) A more hyperextended elbow forces a larger shoulder release angle to land the ball in the same place, so the shoulder torque acts through a longer arc and does more work. (b) When the elbow recoils, elbow and shoulder angular velocities act in the same direction, adding speed directly. Recoil erodes (a) while adding (b) — hence an optimal recoil for every peak angle. The recoil mechanism (b) contributes more than the extension-to-peak mechanism (a), which agrees with Middleton et al. (2015) on that specific point.

9. A direct disagreement with prior literature. Middleton et al. (2015) found that an elbow held at a fixed hyperextension offset bowled slower than a straight arm. Felton and King find the opposite: even non-recoiling hyperextension is faster. They propose that a speed gain is always available from a flexion/extension or abduction/adduction offset provided the upper arm orientation increases the shoulder’s ability to do work, and note that whether “flexion” or “extension” is the helpful direction depends on how the elbow’s axis is oriented relative to the target — which is why the literature has produced contradictory answers (Portus et al. 2006 and Roca et al. 2006 found extension helps; Middleton et al. found flexion helps).

10. Speculative 3D extension. The authors suggest the same mechanics would hold in 3D, implying that bowlers with large abduction angles who move more towards the target using flexion or extension are likely to be faster than those with straighter arms. This is explicitly labelled speculation and was not modelled.

What a coach should look for on video

Same honest framing as the conference version: elbow laxity is not trainable. These cues are for talent identification and for explaining speed differences, not for a technical intervention.

Cue 1 — Screen for hyperextension (off-camera first)

• The cue: Whether the bowler’s elbow goes past straight at all.
• Camera view + frame: Start with a static check, not video — have the bowler fully straighten the bowling arm with the palm up and look at the profile of the elbow from side-on. Then confirm in the delivery: behind-the-arm or side-on, high frame rate, scrubbing upper arm horizontal → ball release.
• What “good” looks like: A visible backwards bow at the elbow; 20° is documented and is unmistakable on video.
• What the fault looks like: No fault — a locked elbow simply lacks this speed source. Do not attempt to create hyperextension; it is anatomical laxity, and forcing an elbow past its limit is an injury route, not a coaching route.
• Why it matters — and now you can price it: past the first degree, ~0.2% of ball speed per degree. A bowler with 20° carries about 5% (~5 mph at elite pace) over an otherwise identical bowler with none. This is why two bowlers with the same run-up, the same front leg and the same trunk flexion can differ by several mph.

Cue 2 — Recoil: is the elbow already coming back at release?

• The cue: Locate the frame of peak hyperextension, then compare with the release frame.
• Camera view + frame: Behind-the-arm or side-on, 240 fps or better; step frame by frame through the last few frames before release.
• What “good” looks like: Peak hyperextension occurring before release, with the elbow visibly recovering towards straight at the release frame. The paper’s optimum was 25° peak with 5° of recoil; more generally, optimal recoil is 30–60% of the theoretical maximum.
• What the fault looks like: The elbow at its maximum bend-back exactly at release, with no recovery — the “At peak” pattern. Still faster than a straight arm, but not extracting the recoil mechanism, which the paper identifies as the larger of the two contributions.
• Why it matters: This is the paper’s central conclusion — the speed gain is governed by peak hyperextension and recoil, not peak alone. It is the only part of the elbow story that varies between bowlers with the same laxity.

Explicitly not a legality cue. The 15° ICC limit applies to elbow extension, and hyperextension is exempt. Phone footage cannot assess action legality — that needs the formal ICC protocol.

Caveats and limits

• n = 1. One elite male bowler (19 years, 1.80 m, 82.4 kg), member of the ECB elite fast bowling group; four maximal-effort good-length deliveries captured on 18-camera Vicon at 300 Hz. The “0.2% per degree” rate is derived from this one bowler’s arm.
• Two-dimensional, two-segment arm model. The authors devote a full paragraph to this: the elbow is a complex 3D joint, generalised to 2D. Omitted degrees of freedom would shorten the effective segment lengths if the arm rotated out of plane or carried an abduction/adduction angle, creating a trade-off between reduced inertia and reduced wrist velocity towards the target. The 3D implications are speculation.
• Passive spring only. No active elbow torque generator — the model cannot represent voluntary elbow flexion or extension, only load-driven hyperextension. The authors list adding an active elbow torque generator as future work.
• No hand segment. Wrist flexion’s contribution is applied as a fixed percentage uplift measured from the bowler’s own trials, not modelled. The authors list adding a hand segment as future work.
• The 25° bound is imposed, from King & Yeadon (2012), and the optimum sat on it — so “5%” is bounded by that assumption.
• Simulation, not intervention. Nothing was tested by changing a real bowler.
• The conclusion is comparative, and carefully hedged: “Although it may be possible for bowlers who do not hyperextend to bowl faster than those who hyperextend due to other technique or strength parameters, a bowler who can hyperextend at the elbow and recoil optimally will have an increase in ball speed compared to a similar bowler who cannot hyperextend.” Do not read this as “hyperextenders are faster bowlers”.

Relationship to other Felton work

• Journal version of the 2014 ISBS conference paper. All headline numbers agree (4% actual, 5% optimised, same match/evaluation scores). The journal adds: the ~7,000-simulation perturbation study, the inverse-hyperbolic optimal-recoil relationship, the 0.2%-per-degree rate, the work-done-at-the-shoulder mechanism figures, the disagreement with Middleton et al. (2015), and a substantially more cautious conclusion.
• Uses a different model from the rest of this cluster — a two-segment arm model, not the 16-segment whole-body model. It sits alongside the whole-body work rather than feeding into it. The 2020 J Sports Sci paper cites it for the passive viscoelastic element used to restrain the bowling elbow (which stays at its anatomical limit through front foot contact and therefore has no active torque generator in the whole-body model).

Note where the conference version was softened: the 2014 conference paper asserts the limiting characteristics “were found to be” peak hyperextension and recoil, with no hedging about other bowlers. The 2016 journal version restricts the claim to a comparison between two otherwise similar bowlers and adds the explicit acknowledgement that a non-hyperextending bowler may still be faster.

TENSION: (with the 2025 strength paper) this paper frames elbow hyperextension purely as free ball speed. The 2025 J Sports Sci paper observes that when strength was increased in the whole-body model, elbow hyperextension increased to each bowler’s individual upper limit, and warns that repeated elbow hyperextension has been linked to posterior elbow impingement and bone stress injuries (McBride et al., 2021), and that shoulder-strength/arm-speed interventions could therefore push bowlers towards an injury-risk adaptation. The cluster contains both “hyperextension is worth ~5%” and “hyperextension is an injury pathway to be watched” without ever reconciling them. See the 2025 paper, The Effect of Increased Strength on Ball Release Speed.