2019 — PhD thesis: Factors affecting performance in elite finger spin bowling

Research theme

This is the parent document for the two Sanders/Felton journal papers, plus two substantial studies that were never published as papers and are therefore only available here:

• Chapter 7 — a 69,552-delivery Hawk-eye analysis of finger spin in test cricket. By far the most match-relevant thing in this cluster.
• Appendix A — a kinematic study of ten elite WRIST spin bowlers, which reaches conclusions in the opposite direction to the finger spin work.

Also here: Chapter 4, a purely descriptive walkthrough of what the elite finger spin action actually looks like, angle by angle, phase by phase — a coach’s reference table for “what is normal”; and Chapter 8, a constraints-led coaching argument.

The thesis’ three data sources: 3D kinematics + PROM on 23 elite finger spinners (18-camera Vicon, 300 Hz, 56 markers, Trackman); Hawk-eye ball tracking from 60 test matches 2006–2015; and 10 elite wrist spinners on the same lab set-up.

What they measured

Chapter 5 (kinematics) and Chapter 6 (PROM) — see Sanders 2018 — kinematic parameters and Sanders 2019 — passive ROM. Not repeated here.

Chapter 7 (match play, Hawk-eye) — six ball-trajectory parameters per delivery, plus per-delivery metadata (runs, over, batter, batter handedness):

• Where it pitched, long/short (bowling length, m from the batter’s stumps; bounds 0–8 m, bucketed 1 m)
• Where it pitched, left/right (bowling line, m from centre stump; bounds ±0.6 m, bucketed 0.15 m)
• How fast it left the hand (release velocity, mph; bounds 40–60)
• How much it turned off the pitch (deviation angle, °; bounds ±9)
• How steeply it landed (incidence angle to the ground, °; bounds 14–20)
• How high it looped (zenith delivery height, m; bounds 1.8–2.5)
• Outcomes: bowling average (runs per wicket) and bowling economy (runs per over).

Appendix A (wrist spin) — 26 kinematic parameters on the same convention as the finger spin study, plus delivery stride length and normalised release height.

Findings

Chapter 4 — what the elite finger spin action actually looks like (descriptive, n = 23)

Useful as a “what’s normal” reference. Note these are group descriptives, not targets.

1. Run-up approach speed 1.4–3.8 m/s (2.55 ± 0.42 m/s) — a jog, not a run.
2. Only 14.3% of these elite bowlers used a genuinely side-on action; 64.3% used a “mixed” action (classified on shoulder alignment at BFC: side-on <210°, mid-way 210–240°, front-on >240°). Shoulder at BFC ranged 180–260.6° (218 ± 19.3°).
3. Back foot at BFC 297.7 ± 16.8° — i.e. already ~28° open of parallel-to-the-crease, not parallel as coached.
4. Pelvis-shoulder separation at BFC 16.1 ± 14.1° (range −9.0 to 46.4°). Only 7% of bowlers had a pelvis orientation >180° at BFC; 23.3% counter-rotated the pelvis below 180°.
5. At FFC: front foot 336.9 ± 17.1°, pelvis 206 ± 10.8°, shoulders 186.4 ± 7.4°, separation 19.9 ± 10.4°. Peak pelvis angular velocity 593.2 ± 100.1 °/s, reached as the bowling arm passes upper-arm-horizontal.
6. At ball release: pelvis 262 ± 13.2° — 77% of bowlers had the pelvis SHORT of square-on (<270°), which is exactly why the ones who got past it stood out on spin. Shoulders 279.1 ± 26.1°; only 36.7% were <270°. Trunk flexion 32.3 ± 5.7°.
7. Elbow extension UAH→BR ranged 0–15° (10.6 ± 6.1°) in this Chapter 4 reporting. (Note: Chapter 5 / the journal paper reports the same variable as 3.65 ± 3.24°, max 10.5°. The two numbers are not reconcilable from the text — see TENSION below.)
8. Follow-through: pelvis and shoulders rotate a further 78° and 72° after release; total BFC→FT rotation 158.0 ± 23.7° (pelvis) and 171.3 ± 23.5° (shoulders) — neither segment completes the full 180° the coaching literature calls for. Front foot pivots 107 ± 32.4° (range 16–158.5°).

Chapter 5 — kinematics and spin rate

As per Sanders 2018 — kinematic parameters. Headline: pelvis orientation explains ~43% of spin-rate variance; high-spin bowlers are mid-way (≈225°) at FFC and past front-on (>270°) at BR, with bigger hip-shoulder separation and shoulders short of side-on at FFC.

Chapter 6 — passive range of motion and spin rate

As per Sanders 2019 — passive ROM. Headline: front-hip total rotation arc is the only significant predictor (r = 0.552, 25.5% of variance); rear-hip internal rotation and bowling-shoulder internal rotation correlate at p < 0.10.

Chapter 7 — what actually works in test match cricket (the unpublished gem)

34 elite finger spinners (right-arm off break, ROB, and slow left arm, SLA), 60 test matches, Nov 2006 – Sep 2015, 69,552 deliveries, 11,592 overs, 1,021 wickets, 34,075 runs. Cumulative average 33.37, economy 2.94 rpo. Hawk-eye’s pitching-point error is quoted at 2.6 mm. Bowlers with suspect (>15°) actions were removed. 59.33% of deliveries were ROB, 40.67% SLA. (The thesis abstract says 36 bowlers; Chapter 7 says 34 — see TENSION.)

1. Three parameters together predicted 54.4% of the variance in bowling economy: length, line, and release velocity.
   • Length alone explained 22.9%. Significant quadratic relationship, r = 0.954, p < 0.001. 4–5 m from the batter’s stumps was the cheapest length — 40.96% of all deliveries landed there. Runs conceded rose exponentially both shorter and fuller. Deliveries pitching 0–1 m (i.e. right up at the batter’s feet) conceded 209 runs for 6 wickets, average 34.83.
   • Adding line took it to 52.8%. r = 0.992, p = 0.008. Landing within 0.15 m of the middle-stump centre line was cheapest — the −0.15 to 0 m bucket produced the lowest average (22.73) and lowest economy (2.30 rpo). Costs rise quadratically the further from the stumps you pitch.
   • Adding release velocity took it to 54.4%. r = 0.850, p < 0.001. Faster is cheaper, up to about 56–57 mph, beyond which economy worsens again. Group mean release velocity 52.54 ± 3.94 mph.
2. The optimised delivery (simulated annealing on the regression): 57.4 mph, 4.40 m length, 0.11 m line → predicted 0.279 runs per ball = 1.67 runs per over.
3. The speed–accuracy trade-off is real and quantified. Release velocity correlated with the combined standard deviation of line and length at r = 0.934, p < 0.001 — the faster the bowler bowls, the more scattered their line and length. The standardised residual plot shows a sharp break above 57 mph. This is the mechanism behind the economy optimum: speed helps until control costs more than speed gains.
4. Turn: away-swinging beats in-swinging, decisively. Deliveries deviating away from the batter (51.87% of all deliveries) produced a 24.3% lower bowling average and 17.3% lower economy than deliveries deviating toward. Significant effect of deviation angle on average (r = 0.565, p < 0.001; F(2,16) = 10.831, p = 0.001); no significant effect on economy (p = 0.087). Mean deviation 0.23 ± 4.41°. SLA bowlers turned it more (3.73 ± 2.63°) than ROB bowlers (2.92 ± 3.22°).
5. A tiny amount of deviation is enough. The thesis calculates that at 56–57 mph pitching in line with the stumps, 2.2° of deviation is sufficient to beat the bat or find the edge — a lateral displacement of just 0.092 m over 2.5 m. Batters cannot alter a shot within ~200 ms of ball arrival, and visual-motor delays are 55–130 ms.
6. Matchups: ROB bowling to left-handers produced the lowest average (26.52); SLA bowling to right-handers produced the lowest economy (2.64 rpo). Both are the “turning away from the batter” case. Bowlers bowling to a batter of the same handedness bowled wider lines (±0.32 ± 0.25 m) than to the opposite hand (SLA→RHB −0.12 ± 0.22 m; ROB→LHB 0.18 ± 0.23 m).
7. Incidence angle mattered for economy: 15° was optimal (2.59 rpo); 20° was worst (4.02 rpo). r = 0.966, p = 0.001. Mean 17.21 ± 1.17°. (A shallower angle of arrival is cheaper — i.e. skidding through beats dropping in steeply, in economy terms.)
8. Zenith (flight) height did nothing. No significant effect on average (p = 0.335) or economy (p = 0.547). Mean 2.19 ± 0.14 m. Lowest average at 2.2–2.25 m, lowest economy at 2.25–2.3 m (2.79 rpo) — differences not significant.
9. No parameter predicted bowling AVERAGE. Only deviation angle showed a significant effect. Wickets are much less explicable from ball trajectory than runs are.

Chapter 9.6 — three further considerations

10. The legality trade-off. Spratford et al. (2018) recommend that finger spinners exploit the 15° elbow-extension allowance, and that bowlers wanting to reduce extension should be more side-on at BFC and keep rotating the trunk to BR. Sanders points out this directly contradicts Chapter 5, which found more open pelvis orientations at BFC and FFC correlated with more spin. His words: the published literature “illustrates a mixed message to the coaching community”, and the coaching community needs a safe space “by which coaches do not fear coaching instruction with a sole purpose on improving aspects associated with performance (e.g. ball spin rate) verse safety”.
11. Spin rate alone is the wrong metric. Magnus force depends on both spin rate and release speed, so neither in isolation predicts how much the ball will drift, dip and deviate. Sanders proposes adopting Spratford’s velocity/revolution index as the process metric for finger spin, rather than assessing spin rate and ball speed separately. This is the single most important conceptual point in the thesis for anyone measuring spinners with a radar.
12. Lab spin bowlers bowl ~14% slower than test bowlers. Elite finger spinners in test cricket release at 52.12 ± 1.39 mph; those measured in labs across six studies at 44.95 ± 1.12 mph (t(3) = −12.855, p = 0.001; Δ = 13.76%). Within Chapter 5’s own sample, the four bowlers who had actually played test cricket released at 53.34 ± 1.31 mph — right at the test-match value. Release speed therefore looks like a discriminator between “elite domestic” and “international”, not just between elite and sub-elite.

Appendix A — elite WRIST spin (n = 10, unpublished)

13. Wrist spinners spin it much harder than finger spinners: 1791–2553 rpm (2209 ± 232 rpm) vs 1685 ± 170 rpm for finger spin. Release speeds 19.11–21.76 m/s (21.33 ± 1.14 m/s) — also slightly faster. Run-up 2.85 ± 0.57 m/s vs 2.53 ± 0.48 m/s for finger spin.
14. The best single predictor was shoulder rotation from BFC to BR — and the correlation is NEGATIVE: r = −0.837, p = 0.003, explaining 80.8% of spin-rate variance. Less shoulder rotation = more spin. Bowlers rotating the shoulders below 45° BFC→BR were the big spinners (group range 0.1–83.4°, mean 45.3 ± 23.7°).
15. Shoulder orientation at ball release: r = −0.875, p = 0.001 — the high-spin wrist spinners released with the shoulders at a “mid-way” orientation (>220° but short of 270°), i.e. not chest-on. Group mean 269.8 ± 20.8°.
16. Other correlates: shoulder orientation at FT (r = −0.660, p = 0.038), shoulder rotation BFC→FT (r = −0.666, p = 0.036), pelvis orientation at BR (r = −0.644, p = 0.044), pelvis rotation BFC→FT (r = −0.665, p = 0.036), pelvis-shoulder separation at BR (r = 0.589, p = 0.037).
17. Nothing else mattered: run-up velocity r = −0.512 (p = 0.130), stride length r = −0.422, back foot orientation at BFC r = 0.168, normalised release height r = −0.366, elbow extension r = −0.157.
18. Elite wrist spinners here counter-rotated the shoulders in excess of 40° from a semi-open BFC orientation (224.5 ± 10.8°) to side-on later — the signature of a mixed action, and contrary to the coached “arrive side-on at BFC” instruction.
19. Proposed mechanism: a mid-way shoulder orientation at release puts the shoulder in a position to exert a large internal rotation moment, driving the long-axis rotation of the arm that a wrist spinner (supinating) relies on. Also permits a lower spin-axis and seam azimuth angle → more drift and dip.

What a coach should look for on video

The thesis’ distinctive contribution is that Chapter 7’s cues are visible from the boundary with no motion capture at all — they are ball-flight cues, not body cues. Combined with the Chapter 4/5 body cues, that gives a coach two independent layers.

Ball-flight cues (Hawk-eye based, watchable from behind the arm or from square)

Cue 1 — Length: 4–5 m from the stumps

• Camera + frame: Side-on from square of the wicket, or behind-the-arm with a marked pitch. Scrub to the frame the ball pitches.
• What “good” looks like: Pitching 4–5 m from the batter’s stumps; optimised value 4.40 m.
• What the fault looks like: Both directions cost runs, and the cost curve is a quadratic — the further from 4.5 m either way, the faster runs accumulate. Full (0–1 m) is expensive and not compensated by wickets (average 34.83 there).
• Why it matters: The single biggest predictor of economy, 22.9% of variance on its own.

Cue 2 — Line: within 0.15 m of middle stump

• Camera + frame: Behind-the-arm, frame of pitching. A cone or a chalk mark at middle stump makes this trivially judgeable.
• What “good” looks like: Ball landing inside a 30 cm-wide corridor centred on middle stump. The −0.15 to 0 m bucket returned average 22.73 and economy 2.30 rpo against a group baseline of 33.37 and 2.94.
• What the fault looks like: Drifting wide of the corridor. Watch particularly for a bowler bowling to a same-handed batter — this cohort’s bowlers widened to ±0.32 m in that matchup and paid for it.
• Why it matters: Line takes the economy model from 22.9% to 52.8% explained.

Cue 3 — Release speed: push toward 56–57 mph, but stop there

• Camera + frame: Not a video cue — needs a radar or a broadcast speed readout. Include it because it is the third leg of the model.
• What “good” looks like: Around 56–57 mph. Group mean was 52.5 mph, so most elite finger spinners are below the optimum and have room.
• What the fault looks like: Above ~57 mph, line-and-length scatter increases sharply (r = 0.934 between speed and combined line/length SD) and economy worsens. The speed-accuracy trade-off has a visible knee point.
• Why it matters: Takes the model to 54.4%. And the crucial caveat from Chapter 9.6.2: extra speed must not come at the cost of spin rate, because drift and deviation come from Magnus force, which needs both.

Cue 4 — Turn it away from the batter

• Camera + frame: Behind-the-arm, from pitching to arrival at the bat.
• What “good” looks like: Deviation away from the batter. Off-spinner to a left-hander; slow left-armer to a right-hander. 24.3% better average and 17.3% better economy than turning the ball into the batter.
• What the fault looks like: The stock ball turning into the pads of a same-handed batter.
• Why it matters: The only parameter with a significant effect on bowling average. Also note the reassuring number: 2.2° of deviation is enough — 9 cm of lateral movement. Coaches chasing enormous turn are chasing something the batter can actually see and adjust to; the thesis found low averages at both small (±1–2°) and very large (>±7°) deviations, and the batter’s 200 ms adjustment window is what makes the small deviations work.

Cue 5 — Flight height is not the lever you think it is

• Zenith height had no significant effect on either average or economy across 69,552 deliveries. If you are coaching “give it more air” as an end in itself, this dataset does not support it. (Incidence angle did matter — 15° optimal, 20° worst — and incidence angle is coupled to trajectory shape, so “loop it more” is not neutral either; it just isn’t the height of the arc per se.)

Body cues (from Chapter 4/5 — see Sanders 2018 — kinematic parameters for detail)

Hips mid-way (≈225°) at front foot contact; shoulders slightly open of side-on and lagging the hips by ~20°; hips past square (>270°) at release; both feet pointing more down the wicket than “parallel to the crease”. Chapter 4 gives the reassuring context that only 14.3% of these elite bowlers were genuinely side-on, and neither hips nor shoulders complete the 180° rotation the textbooks call for.

The honest limit: what video CANNOT give you for spin

The single most useful methodological fact in this thesis for a video coach is buried in Chapter 3 and Chapter 9.1:

Four extra reflective markers were placed on the index and middle finger knuckles of the bowling hand specifically to quantify finger motion during the delivery — and they had to be discarded from all analysis because the markers sat too close together to track, leaving large gaps in the joint-centre trajectories.

An 18-camera Vicon system at 300 Hz in a controlled indoor volume could not resolve the fingers. A phone at 240 fps handheld from 15 m away has no chance. Consequently:

• You cannot coach finger action from video. Not the spread angle between index and middle finger, not the moment of release from the finger pads, not the “rip”.
• Forearm pronation (the one distal variable that did correlate with spin, r = 0.409, p = 0.053) is borderline. It is a ~11° rotation happening in the last few hundredths of a second before release. On a 240 fps side-on phone video you may see whether the forearm pronates, but you will not measure it. Treat it as a yes/no observation.
• Wrist flexion, upper-arm internal rotation, and elbow extension are all invisible or unreliable on phone video — and the first two didn’t correlate with spin anyway, so nothing is lost.
• What you CAN resolve is exactly what predicted spin: pelvis line, shoulder line, foot orientations, and the separation between hip and shoulder lines. These are large, slow, high-contrast segment orientations. The good news of this research programme is that the strongest predictors of spin rate happen to be the things a camera sees best. The bad news is that they explain 43% of the variance and the remaining 57% lives largely in a hand you cannot film.
• Buy a radar, not a better camera. Trackman/equivalent gives you spin rate and release speed directly. Chapter 9.6.2’s argument is that you should be tracking both together (the velocity/revolution index), because Magnus force — drift, dip, deviation — depends on their combination, not either alone.

Wrist spin: do NOT transfer the finger-spin cues

CONTRADICTION (finger vs wrist spin, within this thesis): the cues invert.

• Finger spin: more pelvis rotation past front-on at BR = more spin (r = +0.676).
• Wrist spin: pelvis orientation at BR correlates negatively (r = −0.644), shoulder orientation at BR negatively (r = −0.875), and less shoulder rotation BFC→BR (below 45°) is the single best predictor (r = −0.837, 80.8% of variance).

For a wrist spinner, the video cue is: at ball release the shoulders should be mid-way, short of chest-on, and the shoulders should have turned relatively little from back foot contact. Sanders states this explicitly — the mechanistic underpinnings differ between finger and wrist spin “and should therefore be coached accordingly, as opposed to the previously reported generic model of coaching spin bowling”.

Caveats and limits

• Chapters 5, 6 and Appendix A: n = 23, 16 and 10 respectively. All elite males. All correlational. Only one predictor entered each regression.
• Appendix A (wrist spin) is unpublished and unreviewed, n = 10, and the 80.8% figure comes from a single-predictor regression on ten data points — that is a very high explained-variance figure from a very small sample and should be treated as a strong hypothesis, not an established result. (The appendix’s methods section also says “Ten elite male finger spin bowlers”, evidently a copy-paste error from Chapter 5.)
• Chapter 7 is observational match data, not an experiment. Length, line and speed are chosen by bowlers who are also choosing them in response to conditions, batter, and match situation — the regression cannot separate “bowling there is cheap” from “good bowlers bowl there”. The optimised 57.4 mph / 4.40 m / 0.11 m delivery is an extrapolation from a regression surface, not an observed delivery.
• Chapter 7 aggregates across 34 bowlers and nine years of pitches, conditions and opponents. Buckets with <120 deliveries were dropped; ANOVA degrees of freedom are on bucket counts (e.g. F(2,6) for length), not delivery counts, so the effective sample for the significance tests is small.
• Spin rate was NOT measurable in match play — Chapter 7 has no spin data at all. The link from Chapter 5’s spin rate to Chapter 7’s match outcomes is entirely inferential, via Magnus force theory.
• Bowlers with suspect actions were removed from Chapter 7, which removes exactly the population where the legality/performance trade-off would show up.
• Chapter 8 is a theoretical/review chapter (ecological dynamics, constraints-led approach). It contains no new data.
• Several key tables (7.3, 7.4) and all figures did not extract from the PDF; findings dependent on them are quoted from the surrounding prose.

TENSION (internal, participant count): thesis abstract says 36 bowlers in the Hawk-eye study; Chapter 7 methods and results both say 34.

TENSION (internal, elbow extension): Chapter 4 reports elbow extension UAH→BR as 0–15° (10.6 ± 6.1°); Chapter 5 / the journal paper report the same variable on the same 23 bowlers as 0–10.5° (3.65 ± 3.24°). These cannot both be right. It matters, because 10.6° mean sits close to the 15° legality limit while 3.65° does not.

TENSION (internal, deviation angle): §7.5.4 states “the lowest bowling average and economies were at large angles of ball deviation (> ±7°), taking 18.9% (n = 209) of wickets at an average of 11.95 and 2.69 rpo”, while the Chapter 9 research-question summary states “deliveries with small degrees of deviation e.g. ±2°, form the lowest bowling average” and §7.6 notes low averages at −2° and ±1°. The likely reconciliation is that the average-vs-deviation curve is U-shaped with low averages at both extremes, but the thesis never says so explicitly, and a coach should not read either statement as “the optimum amount of turn”.

TENSION (internal, pelvis 43.1%): the thesis abstract attributes 43.1% of spin-rate variance to pelvis orientation at FFC; the Chapter 9 research-question summary attributes 43.1% to pelvis orientation at ball release. Same 0.2-point discrepancy carried through from the journal paper’s text-vs-Table 3 mismatch.

Relationship to other Felton work

• Chapter 5 = Sanders 2018 — kinematic parameters (J Sports Sci) and = Felton 2019 — spin technique and spin (conference, Felton first author).
• Chapter 6 = Sanders 2019 — passive ROM (J Sci Med Sport). The thesis lists it as “in review”, confirming the thesis predates publication.
• Chapter 7 and Appendix A appear to be unpublished — this thesis is the only accessible source for the 69,552-delivery match analysis and for the elite wrist spin kinematics.
• CONTRADICTION (with coaching orthodoxy): rejects Woolmer & Noakes (2008) side-on model, for both finger and wrist spin.
• CONTRADICTION (finger vs wrist spin, internal): pelvis and shoulder orientation at release predict spin in opposite directions for the two spin types. Detailed above.
• CONTRADICTION (with Beach et al. 2017/2018): Beach found club-level finger spin to be a push-like action with maximum internal shoulder rotation velocity correlating with spin; Chapter 5 finds a sequential action with no upper-arm internal rotation relationship. Also, Beach proposed that wrist spinners’ faster run-ups feed higher spin; Appendix A found no run-up/spin relationship in elite wrist spinners (r = −0.512, p = 0.130).
• TENSION (with Spratford et al. 2018): the legality/performance conflict, raised by Sanders himself in §9.6.1 — Spratford’s advice for reducing elbow extension (be more side-on at BFC) is the opposite of what Chapter 5 associates with high spin.
• The Chapter 9.6.2 argument that spin rate and release speed must be assessed together rather than in isolation is the most transferable idea in the thesis and is not made in either published paper.