Felton’s core research line, 2014–2025: athlete-specific forward-dynamics computer simulation models of the fast bowling action, used to predict what technique change would make a given bowler faster. Ten items, all first-authored by Felton, all funded or supported by the England and Wales Cricket Board.
The arc, 2014–2025
2014–2017: build the machine, prove it on one bowler. Felton’s opening argument is that experimental fast-bowling research — measure twenty bowlers, correlate technique with ball speed — produces “on average” statements that contradict each other and cannot tell any individual bowler what to change. His alternative is forward-dynamics simulation: build a model of one specific athlete (their limb lengths, their mass distribution, their dynamometer-measured joint strength), prove it reproduces what that athlete actually did, then let an optimiser find the technique that would make them fastest. Because only technique changes, the result is genuinely causal.
Two models emerge. A small two-segment arm model answers the elbow-hyperextension question (2014 ISBS, 2016 J Sports Sci). A 16-segment whole-body planar model of the front foot contact phase — the 0.1 s between the front foot landing and the ball leaving the hand — carries everything else. Applied to one England U19 bowler, it says: optimising his movement alone gains 9.8%; letting the optimiser also choose the position he lands in takes the combined gain to 21.5% — the landing pose is worth roughly as much again on top of the movement optimisation. Strength is worth 1.3%. The prescription is a straighter front leg, a delayed bowling arm, a delayed front arm, and more trunk flexion. In 2017 Felton states the position at its sharpest: this bowler is limited by technique, not strength, and should be coached accordingly.
2020: the peer-reviewed foundation. The Journal of Sports Sciences paper publishes the model with a proper kinematic and kinetic evaluation (4.0% overall, 0.9% kinematic) — a standard much of the simulation literature does not meet. Notably, the eye-catching 21.5%/22% combined result — optimising landing position and movement together — does not make it into the journal; only the 9.8% movement-only figure does.
2023: the pivot. Critics (Glazier and Mehdizadeh, 2019) had made the obvious counter-argument: a one-bowler optimum tells you nothing about anyone else, because you cannot know whether it is suitable or attainable for them. The answer this cluster arrives at is to run the individual-specific method ten times and look for trends — computationally expensive, and never done before with forward-dynamics models. That design finds what the single-bowler design in 2017 could not show either way: the same optimal characteristics emerged for every one of ten elite bowlers. Mean gain 13.5%. The paper claims it “has resolved the controversy on whether individual and group optimisation studies of fast bowling reflect underlying commonalities.”
What survives of the individual-specific argument, and it matters: the optimal target is common; whether a given bowler can reach it is individual, depending on their range of motion, strength and anthropometry.
2024–2025: strength gets tested, and the expected answer does not appear. The same ten models, with strength raised 5%. Ball speed rises 1.5–2.0% — real, statistically significant, and tiny next to the 13.5% available from technique. But the technique changes that did show up ran the wrong way — non-significant trends towards less knee extension, less trunk flexion, more shoulder extension at release. The journal version — a larger design than the conference note that preceded it, with three strength conditions and ANOVA post-hoc tests — calls this “contrary to expectations” and closes with a caution against using strength interventions to alter front foot contact technique. That caution is the recommended reading for anyone citing this line of work.
The 5–8 coaching cues this cluster supports
Everything here is inside the front foot contact phase. Film side-on at the highest frame rate available (240 fps minimum), with the bowler filling the frame, and learn to find two frames: front foot contact and ball release.
Read the frame the front foot lands — it decides most of the delivery. Front knee more extended, bowling arm still back, front arm still high. In the thesis, optimising this frame’s pose on top of the movement pattern took the gain from 9.8% to 21.5% — the landing pose is worth roughly as much again as everything after it — and most muscle activations in the optimal solution were constant — the delivery is largely pre-set at landing. Common to all ten bowlers in 2023. Sources: thesis, 2015, 2017, 2023 (both).
Front knee straighter at landing and held through the phase — but “straighter than you are”, not “locked”. Watch the ankle too; the optimum makes the front leg behave as one strut (ankle–knee co-contraction). The tell for failure: head and hip height sinking after contact. The 2023 paper is explicit that not all optimised techniques used a fully braced knee. Sources: every whole-body paper in the cluster. The most robust finding here.
Bowling arm delayed at landing, and starting after the trunk begins to fold. Find the first frame of trunk flexion and the first frame of arm circumduction — trunk should come first. The mechanism: a delayed arm lets the trunk keep folding while the ball still goes to the right length. It is paid for with greater bowling shoulder extensor torque. Sources: thesis, 2015, 2017, 2020, 2023 (both).
Front arm high at landing, then pulled down late but hard. The 2023 J Biomech paper is the first study to identify this mechanism: a high front arm stabilises the shoulder girdle so the bowling arm can be delayed, but raises upper-body rotational inertia — so it must then accelerate down and into the torso to let the trunk flex. In the thesis optimisation the front shoulder moved ~51°, more than any joint but the bowling shoulder. Sources: thesis, 2015, 2017, 2023 (both).
Trunk flexion: delayed onset, larger total. Staying up briefly after landing, then folding further than habit. It emerges from the straight front leg rather than from muscling the torso — the optimum used lower front hip extensor torque. Sources: thesis, 2015, 2017, 2020, 2023 J Biomech.
Bowling wrist held back longer. Common to all ten bowlers in 2023. The weakest of the kinematic cues — the wrist is the joint most constrained in the model. Sources: 2023 (both).
Screen the bowling elbow for hyperextension — and then monitor it. Two separate uses. Talent ID: does the forearm pass beyond the line of the upper arm during delivery? Past the first degree it is worth ~0.2% of ball speed per degree — about 5% (≈5 mph at elite pace) for a bowler with 20°, and it is ICC-legal because hyperextension is exempt from the 15° extension limit. It cannot be coached; it is anatomy. Also check whether the elbow is already recoiling at release (optimal recoil is 30–60% of maximum) — two bowlers with identical laxity extract different amounts. Injury monitoring: the 2025 paper found strength increases drove hyperextension to every bowler’s ceiling and warns of posterior elbow impingement and bone stress injury. Sources: 2014 ISBS, 2016 J Sports Sci, 2025 J Sports Sci.
For an elite male bowler already in a professional S&C programme, spend coaching hours on technique, not on adding strength. 5% more strength = 1.5–2.0% ball speed. Optimising landing position and movement = 13.5%. Strength remains essential for injury resilience and for the earlier phases of the action, which this model does not cover — but it is not the speed lever at this level. And per 2025, do not expect strength training to fix front foot contact technique. Sources: thesis, 2015, 2017, 2024, 2025.
A cue this cluster explicitly does NOT support: rear leg timing. The 2020 paper wants it earlier; the 2023 paper wants it later. Do not cue it in either direction.
Run-up speed, with a caution: the thesis found an optimum exists but the curve plateaus rather than falls off a cliff — 1 m/s past the optimum cost about 1 mph. Past the optimum the model was forced to bend the front knee. Coach the run-up speed the bowler’s front leg can actually absorb.
Reading the record: where papers differ
The overarching caveat, which applies to every number below: every performance figure in this cluster — 9.8%, 21.5%, 13.5%, 0.8%, 1.5%, 2.0%, 4%, 5% — is a computer simulation. No paper in this cluster reports a technique intervention or a strength intervention on real bowlers with re-measured ball speed. The 2017 paper proposed that the recommendations “will be used to shape the future coaching of this individual” and that the results would be analysed; no follow-up appears in the record. The real-world strength intervention literature this cluster cites is itself mixed (Callaghan et al. 2021 and Hislen et al. 2023 found no gain; Taliep & Maker 2021 and Feros et al. 2020 found gains, with Feros also finding reduced accuracy). The simulation’s answer — a small real gain, easily lost in underpowered field studies — is plausible and untested.
Major, cluster-defining:
Superseded — The individual-specific message from 2017 (group findings cannot tell an individual what to change) and the 2023 finding (the same optimal characteristics turn up in all ten bowlers) look opposed until the designs are compared: 2017 is a single-bowler study that could not test for commonality either way, and 2023 is the first ten-bowler run of the same method — a larger, more capable design superseding the smaller one on this question. What survives from 2017 is that attainability is individual, not the target.
Superseded — The strength conclusion differs between 2024 ISBS (strength “facilitates a greater delay in bowling arm circumduction” — reinforcing the known optimum) and 2025 J Sports Sci (same ten models, same manipulation, but no significant joint angle differences, with non-significant trends towards less knee extension, reduced trunk flexion and greater shoulder extension). The 2025 paper is the larger design — three strength conditions, ANOVA with post-hoc tests — against a single-condition conference note, and its own authors call the result “contrary to expectations”, closing with a caution against using strength to alter technique.
Superseded — 2024 tested a single condition — 5% isometric strength increase at ankle, knee, hip and shoulder — giving 40.7 → 41.1 m/s (+0.8%). 2025 tested three conditions, the third of which is the same manipulation — 5% at ankle, knee, hip and shoulder, elbow and wrist excluded in both, same ten models — giving 41.5 m/s, plus an added intermediate condition of 5% lower body only (41.3 m/s). The figures differ because the optimisation scope differs, not the manipulation: 2024 re-optimised only the joint torque activation parameters, inheriting the optimal landing position from the 2023 study; 2025 re-optimised all 112 parameters, including the initial joint angles, angular velocities and trunk orientation. The wider scope gives a different ceiling. The journal is the fuller, peer-reviewed, wider-scope account of the two.
Superseded — The thesis and 2015/2017 reach a different conclusion about what strength buys than 2025 does. Early work (+5% strength, single bowler): the front leg stays straighter, trunk flexion delays, the front arm extends more — “the optimal technique remained the same”. 2025 (ten bowlers, larger design): the opposite trend.
Superseded — 2020 found optimised technique had earlier and faster rear hip flexion (reduces moment of inertia about the front hip); 2023 J Biomech found optimised technique had delayed rear hip flexion (delays trunk flexion for more efficient momentum transfer) — opposite directions, same lab, same model family. The 2023 paper cites the 2020 mechanism and proposes that delaying rear hip flexion delays upper trunk flexion for a more efficient transfer of momentum. Both papers state the function of rear hip flexion is uncertain.
Reporting and reproducibility:
Superseded — The thesis and the 2015/2017 conference papers report a 21.5–22% gain from optimising the movement pattern and the initial landing position together, from a single-bowler design. The peer-reviewed 2020 paper reports 9.8% from the movement pattern alone, and states the initial bowling arm position was “outside the scope” of that analysis. The ten-bowler group study (2023) reports 13.5%. These are three different optimisation scopes, not three estimates of one quantity.
Open question — The 2023 J Biomech abstract states optimal technique produces lower peak ground reaction forces and loading rates. Neither the discrete peaks in Table 1 (p = 0.394 horizontal braking, p = 0.093 vertical) nor the continuous SPM1D analysis in Fig. 1 (the paper’s own Results section: “no significantly different periods between the ground reaction time histories were observed”) found a significant peak-force difference. Only the loading rates reached significance (p = 0.008, p = 0.022). The abstract’s claim is supported for loading rates and not for peak forces.
Unreconciled — Does optimal technique raise or lower front-foot load? 2020 (movement only): peak horizontal GRF increased. Thesis (landing position changed) and 2023: peak forces and loading rates decreased. A plausible resolution — changing the landing position lowers load, changing only the movement raises it — does not appear written down anywhere in the cluster.
Elbow hyperextension:
Open question — 2016 frames hyperextension as an unambiguous ICC-legal ball-speed advantage (~0.2%/degree, ~5% at 20°). 2025 observes that strength increases pushed hyperextension to every bowler’s ceiling and warns of posterior elbow impingement and bone stress injury. These address different questions — ball speed versus injury risk under strength training — nine years apart, and neither cites the other on this point. How to weigh the speed benefit against the injury risk it can carry is not answered anywhere in the record.
Superseded — The 2014 conference version states the hyperextension finding without qualification; the peer-reviewed 2016 journal version restricts the claim to two otherwise similar bowlers and concedes a non-hyperextending bowler may still be faster — a normal narrowing of scope at peer-review stage.
Reading order for a coach
- Investigating Commonalities of Optimal Technique in Front Foot Contact (2023) — best evidence, ten bowlers, all the primary cues.
- The Effect of Elbow Hyperextension on Ball Speed (2016) — the elbow, for talent ID.
- The Effect of Increased Strength on Ball Release Speed (2025) — what strength does and does not buy.
- PhD Thesis: Factors Limiting Fast Bowling Performance (2015) — the mechanism and the joint-angle detail behind everything else.