Dr Paul Felton is a sports biomechanist at Nottingham Trent University (PhD Loughborough, funded by the England & Wales Cricket Board). He sits on the ICC Suspect Bowling Actions Panel (since 2014), works with the ECB Coach Association as a coach developer (since 2021), and runs The Cricket Biomechanist. He is one of the most prominent applied researchers in cricket biomechanics.
This catalogue covers his complete published output — 25 journal articles, 27 conference papers, 6 theses he authored or co-supervised, plus 2 items found outside his own publication list. Every cricket item gets a full breakdown of theme, findings, and what a coach should look for on video.
How to use this
| If you are… | Start here |
|---|---|
| A fast bowling coach worried about backs | The Coach’s Playbook, then Fast Bowling Injury & Lumbar Spine |
| A fast bowling coach chasing pace | The Coach’s Playbook, then Fast Bowling Technique & Context |
| Coaching women | Fast Bowling Technique & Context — see the “Coaching female cricketers” cues |
| Coaching spin | Spin Bowling & Batting |
| Coaching batting / power hitting | Practitioner & Reference Outputs — written for coaches |
| Working on length control | Fast Bowling Variations & Loading |
| Building or trusting a video analysis system | Methods & Modelling |
| Looking for where the papers differ | Reading the Record |
Two documents sit above the folders:
- The Coach’s Playbook — every video-checkable cue in the corpus, organised by camera angle and by the frame you scrub to. This is the practical output.
- Reading the Record — where the published papers point different ways: work that has been superseded, disagreements no published work reconciles, and questions no study in the record resolves.
On scope and vantage. This catalogue is organised around one publication list because that is the only tractable boundary for a corpus this dispersed — not because it represents a single research agenda. Roughly half of these papers are student-led, with Felton in a supervisory or third-author role. Nothing here comes from re-analysing data; it is what shows up when one reader takes a whole published record in sequence, which is a thing no reviewer, examiner or funder is ever asked to do. The observations are about the corpus as a body, not about the quality of any individual study — several of which are the best-validated work in the field.
What that boundary leaves out. A January 2024 systematic review of fast bowling technique against performance outcomes (Hands & Coventry-Searle, Sports Medicine Australia conference abstract) found 23 eligible studies — and its exclusion criteria were “optimised biomechanical model methodologies, fluid mechanics, injury, and illegal actions”. That rules out this catalogue’s entire simulation cluster and its entire injury cluster. The two bodies of work barely intersect: a review of the correlational fast-bowling literature and a catalogue built around a simulation programme are looking at the same sport through different windows. Worth knowing before treating either as the field.
The same review sizes two gaps that are real and not this catalogue’s to fill. Swing was the outcome in one study out of 23, and accuracy in two. That makes the length-control work in Cluster 04 a larger share of the world’s fast-bowling accuracy literature than its modest presentation here suggests. (It is a conference abstract rather than a peer-reviewed review, and it names none of its 23 studies, so it is cited here for scope and for those two counts — nothing is built on it.)
Five findings that carried across the corpus
1. The mixed action does not predict back injury. Fifty elite bowlers, prospective, MRI-confirmed. Shoulder counter-rotation was 43 ± 14° in bowlers who subsequently fractured versus 40 ± 20° in those who didn’t — not significant, and the injured group averaged above the famous 30° threshold. What does predict injury is a collapsed back leg at back foot contact (rear hip 146° vs 156°) and lumbopelvic extension at front foot contact; together they classify 88% of bowlers. The paper tells coach education to move on. Thirty years of orthodoxy, contradicted.
2. Optimal technique is individual — but less individual than the founding work proposed. The 2017 case-study work argued that each bowler has their own optimum, so blanket technical models are wrong. The 2023 ten-bowler studies then found genuine commonalities across elite bowlers. What survives is subtler and more useful: the direction of the target is shared; how far a given bowler can move toward it is individual. A single-bowler result refined by a ten-bowler result is the programme working as intended.
3. None of the simulation work has been tested in the field. Every ball-speed figure in the simulation cluster is a computer prediction. The 2017 paper stated that the model’s recommendations “will be used to shape the future coaching of this individual” and that the results would be analysed; that follow-up does not appear in any accessible published source. Ask “was this measured or simulated?” of every number, and the answer for that cluster is always simulated.
4. Planar models reproduce angles and speeds well, and forces poorly. Felton, Yeadon & King (2019) compared three variants of a planar simulation model against Vicon and force-plate data: orientation error ~0.9°, ball speed error 1.7–3.8%, but ground reaction force error 11–18%. That is a model-versus-measurement comparison, not a test of what a camera can recover — do not infer loading from a planar model’s force output. Fast bowling is the worst case for 2D modelling, because it is a side-on action with real out-of-plane hip and shoulder motion a planar model cannot represent.
5. Men’s findings do not transfer to women. Trunk flexion is a pillar of the male ball-speed model and predicts nothing in women (r = −0.19, p = 0.57). In batting, all 15 male batters extended the lead elbow through the downswing; 8 of 15 female batters flexed it. Coaching the men’s model into the women’s game is not conservative — it is unsupported.
How much to trust a given claim
A confidence ladder, applied across the catalogue:
| Confidence | What | Examples |
|---|---|---|
| High | Joint angles and orientations from motion capture | Trunk angle ~0.9° error; the injury group means in the lumbar-spine cluster |
| High | Ball/bat speeds, timings, frame counts | Release speeds; BFC→FFC 192 ms vs FFC→BR 103 ms |
| Medium | Between-player correlations in decent samples | Length control η² = 0.78 (n = 21); the 88% injury model (n = 50) |
| Medium-low | Simulated effects of a technique change | Everything in the simulation cluster |
| Low-medium | Ground reaction forces from models | 11–18% error |
| Low | Internal tissue loading | Not measured anywhere; inferred |
| Indicative only | “Optimal technique” | n = 1 in the founding work; never field-validated |
Two caveats attach to almost everything here. Nearly all of it is correlational — bowlers who had X also had Y; almost no one was coached to change X and re-measured. And nearly all of it is male, elite, and English pathway — the injury work compares 39 injured to 11 uninjured bowlers, mean age 18.9.