Cluster 04 — Fast Bowling Variations and Loading

Seven items from two PhD projects Paul Felton co-supervised at Loughborough. Two distinct sub-lines that never cite each other, but which together answer two very different coaching questions.

Sub-line (a) — Manawadu, bowling variations. 21 county-level male fast bowlers, 48 deliveries each (yorkers / bouncers / stock), 18-camera Vicon at 250 Hz, 53-marker model, force plates, 2D DLT to measure where the ball actually landed. Descriptive and correlational. Directly coachable.

Sub-line (b) — Lamb, acceleration transmission. 11 male fast bowlers with nine 1600 Hz IMUs from ankle to C7, plus a 16-segment compliance simulation matched to one bowler. Instrumentation and modelling. Relevant to workload monitoring and injury, not to technique coaching.

Coaching length control

The one-paragraph version. Bowlers change length by changing the upward tilt of the ball out of the hand (release angle) — not by changing pace. That tilt is produced almost entirely by the wrist and the hand in the last 10–20% of the delivery stride. Everything below the trunk — run-up speed, delivery stride, front knee, front hip, back knee, back hip, lower back — is identical across the three deliveries. If a bowler keeps missing their length, look at the wrist and the fingers, not the legs.

The numbers that matter:

• Release angle separates the three lengths cleanly: yorker ~2°, stock ~5°, bouncer ~13°. Effect size η² = 0.78–0.79 — by far the largest in the dataset, and the only release variable significant in all three pairwise comparisons.
• Release speed barely changes. Yorker 31.5, stock 32.1, bouncer 33.1 m/s (η² = 0.07–0.08). Horizontal ball speed did not differ at all (p = 0.07). All the difference lives in the vertical component: 1.2 / 3.0 / 7.4 m/s (η² = 0.77). Bowlers point the ball, they don’t throw it harder or softer.
• Release height is a moderate effect (η² = 0.22) and runs against intuition: the bouncer is released lowest (107% of standing height vs 112% for the yorker) because the arm has come further forward and down.

“My bowler keeps dropping short when he tries to bowl a yorker” — what to look at

In order:

1. The wrist, side-on, at ball release and through the whole delivery stride. Wrist angle at release alone explained 75.3% of yorker release-angle variance; adding hand orientation reached 86.6%. Yorkers carry a more extended (cocked back) wrist from back foot contact onwards (~193° at release vs ~189° for a bouncer), use ~6° more wrist travel through the delivery (31° vs 25° from front foot contact), and arrive at release with the fastest wrist (~16 vs ~12 rad/s). A bowler who arrives at the flexed “bouncer wrist” while intending a yorker gets a good-length half-volley.
2. The finger/ball line in the last frame before release — the one genuinely coachable-by-feel cue, and the thesis says so. ~58° for a yorker, ~52° for a stock ball, ~42° for a bouncer against horizontal. Higher finger line = flatter, fuller ball. Coaching phrase: stay behind and under the ball to the last instant. It was the only technique variable separating all three lengths (η² = 0.31).
3. Upper-back posture at release — this sets release height rather than angle. More upright for the yorker (~152°, released at 112% of height), folded further forward for the bouncer (~148°, 107%). Thoracic variables explained 55–60% of release-height variance across all three deliveries.
4. Check nothing else changed. Front knee, front hip, back knee, back hip, lumbar angle, delivery stride length, shoulder joint angle and centre-of-mass velocity showed no significant difference at any point of the action across the three deliveries. A bowler visibly running in harder or lengthening the stride to bowl a variation is compensating, not executing.

Before you blame the bowler

• Yorker success rate at county academy level was 24.8% (58/234). Stock 46.4%, bouncer 98.7%. The mean attempted yorker landed at 3.8 ± 3.3 m — about 1.8 m too short.
• A 0.005 s change in release timing — roughly one frame — moves a yorker’s landing point across a 5.68 m range, versus 3.97 m for a bouncer. The yorker is simultaneously the most timing-sensitive delivery and has the smallest target (2 m, versus open-ended for a bouncer). Much of the 25% vs 99% gap is task geometry, not technique.
• Length control is strongly individual. Best bowler in the squad averaged 0.73 m on yorkers; worst averaged 6.76 ± 4.52 m. Six of 21 averaged over 5 m. Bouncer control was uniform across everyone. The thesis argues explicitly that identifying which bowlers can repeat a yorker is a selection decision.
• Judge yorker practice by scatter, not by the best ball. SD under ~1 m is what real control looks like; over 3 m and it is a variable good-length ball with a yorker’s intention attached.

Does bowling a variation cost pace, or increase injury risk?

• Pace: essentially no. Yorker vs stock release speed did not differ (p = 0.710 in the ISBS paper). Bouncers came out ~1–1.6 m/s (≈4–6 km/h) faster, not slower.
• Injury-risk positions: no detectable increase. Front-foot loading was statistically indistinguishable across the three deliveries on every measure — peak vertical force (5.06 / 5.80 / 5.75 BW, p = 0.238), peak braking force (2.61 / 3.02 / 3.00 BW, p = 0.305), vertical and horizontal loading rates, impulses, times to peak, and continuous SPM of both force curves. Yorkers showed the lowest mean forces (non-significantly). Upper-body joint moments were likewise flat, with one brief exception (wrist moment, bouncers, 40–60% of front-foot-contact-to- release, p = 0.03 — not significant at release).
• Read that as a null result on n = 21, not as proof of safety. With ±1.5 BW SDs the study is underpowered for modest differences.

Front leg: the speed vs shock trade-off

This is the sharpest genuine tension in Felton’s whole body of work, and neither side resolves it.

The performance side (Felton, Yeadon & King 2020, Optimising the front foot contact phase of the cricket fast bowling action; Worthington, King & Ranson 2013 — both cited throughout this cluster, both filed elsewhere in this collection): a straighter, more braced front leg that resists collapsing at front foot contact produces more ball release speed. The leg acts as a rigid fulcrum for the trunk to rotate over.

The injury side (Lamb, all three papers in this folder): the reason the body survives front foot contact at all is compliance — the front ankle, knee and hip rotating and the tissues deforming, which stretches the impulse out in time and cuts the peak. That is measured:

• Peak acceleration at the ankle at front foot contact: ~150 g. At L5: ~29 g. At C7: ~9.5 g. Attenuation ankle→L5 is 80–91%; ankle→C7 is 93–95%.
• It attenuates at every step of the chain — ankle→knee→hip→L5→L1→C7, every adjacent pair significantly different (p < 0.01), with a progressive time delay at each level.
• The body attenuates proportionally harder when it is hit harder (position × phase interaction F = 10.1, p < 0.001).

And Lamb’s simulation tests the trade-off directly. Making the front ankle, knee and hip springs stiffer in a 16-segment model matched to one bowler:

A stiffer front leg transmits more shock upward. That is the same configuration the performance literature says makes the ball faster. Same joint, same 100 ms of the action, opposite advice.

What to be honest about:

• The plausible manipulation is small. At 10× stiffer — already far beyond anything a human can adopt — L5 rose only 0.6 g (+5%). The dramatic numbers come from the physiologically impossible 100× case. Do not tell a coach that bracing the front leg raises low-back shock 43%. The direction is established; the magnitude at realistic stiffness is not.
• The distal segments take the hit disproportionately. At 100× the shank rose +158% while L5 rose only +43%. Stiffening the leg damages the leg’s protection far more than the back’s.
• n = 1, 2D planar model, spring values borrowed from a drop-landing study and never optimised (the optimisation was promised at the ISCSB conference and never published — it is presumably in Lamb’s embargoed thesis).
• No paper in Felton’s output that I could access resolves the trade-off, quantifies the exchange rate, or says where the balance should sit. This is the single most important open question in this cluster.
• Manawadu’s data adds one relevant fact: front knee angle is irrelevant to length (p = 0.315 at release; p > 0.05 across the whole delivery stride). So a coach adjusting front-leg bracing is trading speed against shock only — length is not in the exchange.

Is accelerometer data a practical proxy for workload monitoring?

Partly — and less than the marketing suggests.

• Yes as impact exposure. A tibia/ankle-mounted IMU sees the largest, cleanest signal and reliably distinguishes the three impacts. Front foot contact (~150 g) and the follow-through (~110 g) are the loading events; back foot contact (~56 g) is significantly milder. The follow-through in particular is nearly as severe at the knee as front foot contact (87.7 vs 90.8 g) and the knee attenuates it least there (18 ± 20%) — and it is almost never filmed.
• No as a lumbar load measure. 80–91% of the tibial signal never reaches L5, the attenuation fraction varies bowler to bowler (±5–10% SD), and the group’s own conclusion is that the high-frequency elastic-wave component — the part most associated with bone microdamage — is largely gone before L5 and “may be unlikely to contribute to lumbar stress fracture risk.” An ankle load score measures what the leg absorbed, not what the back received.
• Surface accelerometry is not bone load. All three Lamb papers say so; muscle forces, a large part of spinal loading, are absent entirely.
• No published component tests fatigue, spell length, season accumulation, or actual injury outcomes. All of it is six maximal deliveries in a lab on 11 healthy, injury-free young men. The applied workload question the ECB funded this to answer may be in the embargoed thesis; I could not check.

Every contradiction and tension flagged in this cluster

TENSIONS (two well-supported findings pulling in opposite directions):

• Front leg — speed vs shock. (flagged in Lamb 2022 — surface measured accelerations, Lamb 2023 — acceleration transmission, Lamb 2023 — lower limb joint compliance and acceleration transmission, Lamb 2024 — PhD thesis: acceleration transmission) A straight, braced front leg produces more ball speed (Felton, Yeadon & King 2020; Worthington, King & Ranson 2013). A compliant front leg attenuates impact shock (all three Lamb papers, confirmed causally in the ISCSB simulation). Same joint, same instant, opposite advice. Unresolved anywhere in Felton’s published work. Magnitude at realistic stiffness is small (10× stiffer → L5 +5%); direction is solid.

• Does impact shock even cause lumbar stress fracture? (flagged in Lamb 2023 — acceleration transmission and Lamb 2024 — PhD thesis: acceleration transmission) Every paper in the Lamb line opens by framing high ground reaction forces as a stress-fracture risk factor (Ranson et al. 2008) and impact shock as the thing worth measuring. The 2023 ISB paper then concludes the post-impact elastic wave is “unlikely to contribute to lumbar stress fracture risk” because so little reaches L5. If that is right, the case for accelerometer-based lumbar workload monitoring — the applied motivation for the ECB funding — weakens considerably. The papers do not resolve it.

• Thoracic flexion does two jobs at once. (flagged in Manawadu 2023 — PhD thesis: fast bowling variations; raised by the thesis itself in Chapter 9) More upper-back flexion is associated with higher ball speed (Worthington 2013) and with the bouncer (Manawadu 2023 — bouncers had the most thoracic flexion and the lowest release height). A coach training more trunk flexion for pace may find the bowler’s natural length creeping shorter. The thesis calls for this to be investigated; it has not been.

• Front knee matters for speed but not for length. (flagged in Manawadu 2023 — PhD thesis: fast bowling variations) Worthington et al. (2013) and Felton, Yeadon & King (2020) put ball speed substantially in the front leg. Manawadu finds front knee angle irrelevant to length — p = 0.315 at release, p > 0.05 across the whole delivery stride, and front hip, back knee, back hip and lumbar likewise. This is a refinement, not a conflict: front leg ⇒ speed, wrist and hand ⇒ length. Worth flagging because coaches routinely treat the front leg as the answer to everything.

CONTRADICTIONS (one result directly disagreeing with another):

• Attenuation beyond L5. (flagged in Lamb 2022 — surface measured accelerations, Lamb 2023 — acceleration transmission, Lamb 2024 — PhD thesis: acceleration transmission) Lamb finds acceleration continues attenuating past L5 up to C7 (L5→L1→C7 all significantly different, p < 0.01). McErlain-Naylor, King & Allen (2021) found no attenuation beyond L5 in drop-jump landings. McErlain-Naylor co-authors both. Proposed resolution: fast bowling delivers a far larger ankle input, so more residual acceleration remains to attenuate higher up. Not tested.

• Lamb’s line criticises Felton’s own models. (flagged in all four Lamb files) All three Lamb papers name Felton, Yeadon & King (2020) as a cricket simulation model that “focused solely on performance” and has “not yet explored the relationship between performance and injury”, and argue via Allen, King & Yeadon (2012) that rigid pin-jointed models are unsuitable for estimating internal loading. This is an internal methodological critique of the supervisor’s own work by his student, and it is the explicit motivation for the compliance model.

Reporting inconsistencies (errors, not disagreements — recorded so nobody chases them):

• Lamb 2022 ISBS states “significant differences between BFC and FT and FFC and FT, however there were no significant differences observed between FFC and the FT phase” — self-contradictory. The 2023 ISB paper resolves it: BFC was significantly less than both FFC and FT; FFC vs FT did not differ. Read the 2022 sentence as a typo.
• Manawadu yorker release angle printed as 2.2 ± 0.9° in thesis Table 6.1 but 1.9 ± 1.8° in the thesis text (p. 69) and in the 2022 ISBS paper.
• Bouncer success rate given as 98.7% (thesis Ch. 5, ISBS paper) and 98.3% (thesis Ch. 9).
• Shoulder-angle regression explained variance given as 45.6% (Ch. 8) and 45.3% (Ch. 9).
• Manawadu release height, stock ball: 110.7 ± 7.9% (ISBS 2022, all successful trials) vs 111.3 ± 3.9% (thesis, one representative trial per bowler) — different analyses, not an error.
• Attenuation-to-L5 percentages differ slightly between Lamb 2022 (88/80/82%, ankle reference) and Lamb 2023 ISB (91/86/88%, distal tibia reference) for BFC/FFC/FT. Same site described differently; direction and magnitude agree.

What is missing from this cluster

• Lamb’s PhD thesis is embargoed until 2027-10-01 with no published abstract. The simulation spring optimisation — the piece that would make the compliance model quantitatively usable and might quantify the speed/shock trade-off — was promised at ISCSB 2023 and never published. It is presumably in there.
• Manawadu’s thesis title over-promises. “Fast Bowling Variations” covers pitch length only. There is nothing on slower balls, cutters, swing, seam position or spin despite the brief. If you need slower-ball biomechanics, it is not here.
• Action type (front-on / side-on / midway / mixed) was recorded but never analysed in Manawadu’s data (1 side-on, 7 front-on, 8 semi-on, 5 mixed). Named as the thesis’s own leading piece of future work.
• No women, no juniors, no club bowlers, no injured bowlers anywhere in this cluster. All male, all elite or near-elite, all healthy.
• No intervention study. Nothing in this folder shows that coaching a wrist position actually changes a bowler’s length. Every technique–outcome link here is a correlation across bowlers, on n = 21, with confidence intervals the thesis itself prints as “<1% – 97%”.