2015 — PhD Thesis: Factors Limiting Fast Bowling Performance in Cricket

Research theme

This is the foundation document for everything else in this cluster. Felton’s argument is that the existing fast-bowling literature is almost entirely experimental and correlational — you measure 20 bowlers, correlate technique variables with ball speed, and get “on average” statements that conflict between studies and cannot tell an individual bowler what to change. His alternative is a theoretical forward-dynamics approach: build a computer model of one specific bowler (his real limb lengths, real mass distribution, real measured joint strength), prove the model can reproduce what that bowler actually did, then let an optimiser search for the technique that would make that bowler fastest. Because the only thing that changed is technique, any speed difference is a genuine cause-and-effect result, not a correlation.

The method, plainly: a 16-segment planar (2D) torque-driven computer simulation model of the front foot contact phase only (from the instant the front foot lands to the instant the ball leaves the hand — about 0.1 s), customised to one male bowler, then optimised with a genetic algorithm. Input data came from 18-camera Vicon motion capture at 300 Hz plus a Kistler force plate at 1800 Hz, with joint strength measured on an isovelocity dynamometer and 95 anthropometric measurements taken to build a subject-specific inertia model.

The four research questions are stated explicitly in Chapter 1:

• Q1. How close to optimal is the technique of the bowler in this study?
• Q2. How close to optimal is the body configuration at front foot contact?
• Q3. How does an increase in strength affect performance?
• Q4. How does run-up speed affect performance?

What they measured

The model’s outputs and the variables interrogated:

• How fast the ball leaves the hand (ball release speed, m/s and mph).
• How bent the front knee is through the delivery (front knee joint angle time history, front foot contact → ball release).
• How much the front ankle collapses (front ankle joint angle time history).
• How far the bowler folds forward over the front leg (trunk flexion / trunk orientation angle time history).
• How late the bowling arm starts to come over (bowling shoulder joint angle and extensor activation timing — “delay in arm circumduction”).
• What the front (non-bowling) arm does (front shoulder joint angle; whether it is held high and then pulled down).
• What the back leg does (rear hip flexion angle and timing).
• The pose the bowler is in at the moment the front foot lands (initial body configuration: trunk orientation, front ankle, knee, hip, back hip, front shoulder, bowling shoulder angles — Table 9.1).
• How hard the front foot hits the ground, forwards and downwards (horizontal and vertical ground reaction force time histories).
• How strong each joint actually is (subject-specific maximal voluntary torque–angle and torque–angular-velocity relationships at ankle, knee, hip, shoulder, wrist).
• How fast the bowler is travelling as the front foot lands (horizontal centre-of-mass velocity at front foot contact — the proxy for run-up speed).

Findings

All results below are causal within the model (simulation, single bowler), not correlational — with the important caveat that “causal” here means “causal inside a validated 2D model of one athlete”.

1. Model accuracy (the credibility gate). The torque-driven simulation matched the recorded delivery to an overall RMS difference of 3.99%. Component breakdown (Table 8.3): force 9.59%, centre of mass 0.06%, trunk orientation 0.67°, ball velocity 0.03%, phase duration 0.19%, joint angles 1.81°. The kinematic-only difference was 0.87%. Centre-of-mass RMS error was 2.8 cm horizontal and 4.1 cm vertical. No penalties (anatomical limits, wobbling-mass excursion, ground compression) were incurred.

2. The bowler’s actual speed was 35.3 m/s (79.4 mph).

3. Q1 — technique alone, same landing position: +9.8%. Letting the optimiser change only the muscle activation timings, while forcing the bowler to land in exactly the pose he actually landed in, produced 38.8 m/s (87.2 mph) — a 9.8% gain. The changes: front ankle held more extended, front knee held straighter throughout, front hip flexors ramped up to reduce co-contraction and let the trunk flex, bowling shoulder extension delayed, back hip flexing more. The peak horizontal ground reaction force rose slightly — Felton reads this as straighter front-leg kinematics braking the body more effectively and converting run-up linear momentum into angular momentum about the front foot.

4. Q2 — technique plus landing position: +21.5%. Letting the optimiser also choose the pose at front foot contact (115 parameters total) produced 42.9 m/s (96.5 mph) — a 21.5% gain. The changes in landing pose (Table 9.1, matched → optimised):

   • Trunk orientation 93.6° → 93.5° (essentially unchanged)
   • Front ankle 146.5° → 144.1°
   • Front knee 170.5° → 172.3° (straighter)
   • Front hip 138.9° → 140.1° (more extended → longer delivery stride)
   • Back hip 205.7° → 202.6°
   • Front shoulder 322.7° → 271.9° (dramatically more flexed — front arm held much higher)
   • Bowling shoulder 71.8° → 53.9° (dramatically more delayed — arm much further back)

   The two shoulders are where almost all the gain comes from. Note that the lower-body landing angles barely moved: Felton concludes “the initial configurations of the lower extremities of the fast bowler in this study are close to his optimal”.

5. Performance is largely pre-set at landing. In the optimal-initial-configuration solution, most torque generator activation levels were constant through the phase. Felton’s interpretation: “the performance of the fast bowler is pre-determined by the orientation of the body at front foot contact”, and the fiddly activation changes seen in the matched and first-optimisation solutions are the bowler compensating for a sub-optimal landing pose. This is one of the most coach-relevant claims in the whole cluster.

6. The optimal front leg behaves like one rigid segment. Front ankle and knee were held in co-contraction throughout, making the front leg function as a single strut — more efficient braking of the pelvis.

7. Better landing position lowers front-foot loading. The optimal-initial-configuration simulation had lower peak ground reaction force in both horizontal and vertical directions than the optimal-technique-from-actual-landing-pose simulation, which Felton links to the longer delivery stride (consistent with Worthington et al., 2013b). This matters for injury, not just speed.

8. Q3 — strength: +1.3%. Increasing maximum isometric torque at the ankle, knee, hip and front shoulder by 5% (bowling shoulder and wrist unchanged) and re-optimising gave 43.5 m/s (97.8 mph), a 1.3% gain over the optimal technique at the bowler’s real strength. The optimal technique did not change — front leg still straighter, arm still delayed, trunk flexion still increased; strength just let the model execute it slightly better (knee slightly straighter, trunk flexion delayed slightly longer, greater front-arm extension). So: strength is a modifier of technique quality, not an independent speed lever, at least for this already-elite bowler.

9. Q4 — run-up speed: an optimum exists, but the penalty for overshooting is small. Optimising at horizontal centre-of-mass velocities of 4.5, 5.0, 5.5, 6.0, 6.5 and 7.0 m/s showed a peak, then a plateau rather than a cliff. For this bowler, going from his actual 5.28 m/s to 6.0 m/s gained 0.4%. Past the optimum, +1 m/s cost about 1 mph of ball speed — because the model was forced to bend the front knee (not strong enough to hold it straight against the extra momentum), degrading the linear→angular momentum conversion. Felton’s inference: the optimal run-up speed is set by how strong the bowler is at holding the front knee straight with a long delivery stride.

What a coach should look for on video

This thesis supports five concrete cues. All of them are about the ~0.1 s between the front foot landing and the ball leaving the hand.

Cue 1 — The pose at the instant the front foot lands

• The cue: Freeze the frame the front foot first touches down and read the whole shape: front knee, stride length, where both arms are.
• Camera view + frame: Side-on, at front foot contact. This is the single most valuable frame in the delivery.
• What “good” looks like: Front knee already close to straight; long delivery stride; bowling arm still well back (in the model the optimal bowling shoulder angle was ~18° more delayed than the bowler’s own); front arm held high and flexed (the optimal front shoulder was ~51° more flexed — a big, obvious difference on video).
• What the fault looks like: Landing with the arms already unwinding — front arm dropped, bowling arm already halfway over. A short stride with a soft knee.
• Why it matters: In this model, changing only this frame’s pose was worth 21.5% of ball speed versus 9.8% for changing everything after it. Felton’s conclusion is that the delivery is largely pre-determined by this position. It is also the change that lowered peak ground reaction force in both directions — so it is a speed gain and a load reduction at the same time.

Cue 2 — Does the front knee hold?

• The cue: Watch the front knee from the moment it lands to the moment the ball is released. Is it a strut or a shock absorber?
• Camera view + frame: Side-on, scrub frame-by-frame from front foot contact to ball release.
• What “good” looks like: The knee straight or straightening; ankle not collapsing either — the whole front leg behaving like one rigid segment (the optimum used ankle and knee co-contraction to achieve exactly this).
• What the fault looks like: The knee visibly flexing after landing and the hips dropping. Watch the height of the head: if the head sinks after front foot contact, the leg is absorbing rather than braking.
• Why it matters: This was worth most of the 9.8% technique-only gain. Mechanically it converts run-up momentum into rotation about the front foot instead of dissipating it.

Cue 3 — How late does the bowling arm start?

• The cue: Find the frame where the bowling arm begins to circumduct (come over). Compare it to the frame of front foot contact and to when the trunk starts folding.
• Camera view + frame: Side-on or behind-the-arm; scrub from front foot contact.
• What “good” looks like: The arm is still back at front foot contact and only starts coming over after the trunk has begun to flex.
• What the fault looks like: The arm leading the trunk — already coming over at or before front foot contact.
• Why it matters: Delaying the arm lets the trunk flex further while still delivering the ball to the same length. It was the largest single change in the optimal landing pose.

Cue 4 — Front arm high, then pulled down

• The cue: Where is the non-bowling arm at front foot contact, and what does it do next?
• Camera view + frame: Front-on or side-on, front foot contact through to release.
• What “good” looks like: Front arm high/flexed at landing, then actively pulled down towards the torso.
• What the fault looks like: Front arm already low, loose, or drifting sideways at landing; no active pull-down.
• Why it matters: The optimiser moved the front shoulder more than any joint except the bowling shoulder. Felton reads the high front arm as aiding the delay of the bowling arm and the pull-down as aiding trunk flexion and shoulder rotation.

Cue 5 — Trunk flexion: late but large

• The cue: How far the bowler folds forward over the front leg between landing and release, and when the fold starts.
• Camera view + frame: Side-on, front foot contact through release.
• What “good” looks like: The fold starts slightly late but goes further than the bowler’s habit.
• What the fault looks like: Folding immediately at landing and running out of range before the arm arrives.
• Why it matters: More trunk flexion is linked to more ball speed across the whole literature; the model shows it emerges from the straight front leg rather than from working the abdominals harder — the optimal solution reduced front hip extensor torque.

On run-up speed — a cue with a caution. The model says a faster run-up helps up to a point and then plateaus. Do not chase run-up speed if the front knee starts collapsing: in the model, past the optimum the leg buckled and speed fell. Coach the run-up speed the bowler’s front leg can actually absorb.

Not supported by this thesis: any prescriptive strength target. Strength was worth 1.3% for a 5% gain at four joints on one already-elite bowler.

Caveats and limits

• n = 1. One male bowler, 18 years old, 85.0 kg, 1.935 m, England U19 squad, flagged as a potential England player. Every number in this thesis is that one bowler’s number. Felton is explicit that the value of the method is that it is individual-specific — which is precisely why the magnitudes (9.8%, 21.5%, 1.3%, 0.4%) should not be quoted at another bowler.
• Simulation, not intervention. Nothing here was tested by making a bowler bowl differently and measuring the result. The +21.5% is what the model says is available, not what anyone achieved.
• Planar (2D) model. Non-planar pelvis and trunk rotations were approximated with massless variable-length segments driven as functions of trunk angle. Genuine 3D rotation, side-flexion mechanics, and the shoulder’s internal/external rotation are not properly represented.
• Front foot contact phase only. The run-up, back foot contact and the whole delivery stride before landing are outside the model. When the model says “land in this pose”, it cannot say whether the bowler could physically get into that pose, or what it would cost earlier in the action. Felton acknowledges the back foot contact phase is probably what limits optimal run-up speed.
• Single-solution optimisation. The optimum is one set of activation parameters. Felton quotes Yeadon’s “pinnacle” warning: an optimum can sit on a narrow peak that a real athlete could never reliably land on. Robustness to perturbation was not tested.
• Force match is the weakest component (9.59% vs 0.87% kinematic) — a known limitation of pin-jointed models. Treat the ground-reaction-force conclusions more cautiously than the kinematic ones.
• Male, elite, adolescent. No claim to female bowlers, juniors, or club bowlers.

Relationship to other Felton work

• The 16-segment model built here is the model used, essentially unchanged, in every later paper in this cluster: the 2015 and 2017 conference papers, the 2020 J Sports Sci paper (which is the peer-reviewed publication of Chapters 4–9), the 2023 J Biomechanics paper (same model, ten bowlers), the 2024 ISBS abstract and the 2025 J Sports Sci strength paper.
• The elbow-hyperextension work (2014 ISBS / 2016 J Sports Sci) uses a separate, much simpler two-segment arm model and is listed in the thesis front matter as a conference output, not a thesis chapter.
• The 2015 and 2017 conference papers report the same optimisations as this thesis with rounded numbers (10% and 22% rather than 9.8% and 21.5%; 1% rather than 1.3%).

TENSION: The 21.5% initial-body-configuration result — the headline finding of the thesis — appears in the 2015 and 2017 conference papers (as “22%”) but does not appear in the peer-reviewed 2020 J Sports Sci paper, which reports only the 9.8% activation-only optimisation and explicitly says “varying the initial position of the bowling arm was outside the scope of this paper”. The most eye-catching number in this cluster was quietly not carried into the journal literature until the ten-bowler 2023 study re-derived it (13.5% for the group). Treat “22% is available from a better landing position” as a thesis/conference claim, not a peer-reviewed journal claim.

TENSION: On strength, the thesis states that when strength was increased “the optimal technique remained the same” and that added strength let the bowler keep the front leg straighter and delay trunk flexion. The 2025 J Sports Sci paper, running the same manipulation on ten elite bowlers, found the opposite non-significant trend — less knee extension and reduced trunk flexion — and described these adaptations as “contrary to expectations”. See Felton 2025 — the effect of increased strength on ball release speed.

TENSION: On front-foot loading, this thesis reports that the optimal technique (activation only) slightly increased peak horizontal ground reaction force, while the optimal initial configuration lowered peak force in both directions. The 2020 paper carries the first result; the 2023 paper (which optimised both) carries the second. Coaches should not read a single “optimal technique raises/lowers front-foot load” message out of this cluster — it depends on whether the landing position is allowed to change.