2023 — PhD Thesis: A Biomechanical Analysis of Fast Bowling Variations

Research theme

This is the single richest coaching source in Felton’s variations line. It asks three questions in sequence: (1) how well can elite bowlers actually control length?, (2) which body movements produce a different length?, and (3) do the forces and torques change when you bowl a variation? — the last being the injury-relevant one.

Method: 21 county-level male fast bowlers (age 19 ± 2.2 y, height 1.87 ± 0.05 m, mass 81.4 ± 9.7 kg), each bowling 48 full-run-up deliveries — 12 yorkers, 12 bouncers, 24 stock (12 to an imaginary right-hander, 12 to an imaginary left-hander) — in randomised order, indoors on an artificial surface with no batter present. 18-camera Vicon MX13 at 250 Hz, a 53-marker full-body model built for fast bowling, two reflective ball markers, and two Kistler force plates at 1000 Hz at the bowling crease. Landing point measured by 2D direct linear transformation from a rear camera. Joint moments by top-down inverse dynamics from the bowling arm. Continuous curves analysed with one-dimensional statistical parametric mapping (SPM) from back foot contact to ball release.

Delivery zones (distance from the batter’s stumps): yorker 0–2 m, stock 4–7 m, bouncer >7 m.

Everything is measurement and correlation. There is no simulation and no intervention in this thesis, so nothing here proves that changing a technique parameter causes a length change.

What they measured

Outcome (ball) variables

• Where the ball actually landed, metres from the batter’s stumps (pitch length).
• How fast the ball left the hand, and split into its forward and upward parts (resultant / horizontal / vertical ball release speed).
• The upward tilt of the ball’s path at release (ball release angle, degrees above horizontal).
• Release height as a percentage of the bowler’s own standing height.

Body variables at ball release (discrete)

• How straight the front knee is (front knee angle at ball release).
• How much the upper back is bent forward (thoracic angle, 3D joint angle; and 2D thoracic segment orientation, i.e. how the upper back leans against the horizon side-on).
• The bowling arm’s angle at the shoulder (shoulder angle) and where the upper arm points in space (2D upper arm orientation).
• How cocked-back or flexed the bowling wrist is (wrist angle), and where the hand points (2D hand orientation).
• Which way the fingers/ball line points just before release (2D finger orientation — the line from the hand marker to the ball, against horizontal). This was a new measurement in this thesis.
• How long the delivery stride was as a % of standing height (delivery stride length %).
• How much each joint moved between front foot contact and release (flexion FFC→BR).

Continuous curves, back foot contact → ball release (SPM)

• Back hip, back knee, front hip, front knee, lumbar, thoracic, shoulder, wrist joint angles; 2D thoracic, 2D upper arm, 2D hand orientations; vertical and horizontal velocity of the whole body’s centre of mass.

Force and torque variables

• Braking (backward) and vertical ground reaction force at the front foot, in bodyweights, at front foot contact, at ball release, and at their peaks; loading rates; impulses.
• Joint moments at the wrist, shoulder and thoracic joints, plus their angular velocities.

Findings

A. How well can they control length? (Chapter 5)

1. Success rates: bouncer 98.7% (233/236), stock 46.4% (110/237), yorker 24.8% (58/234). Three out of four attempted yorkers were not yorkers.
2. Mean landing points: yorker 3.8 ± 3.3 m, stock 7.0 ± 3.7 m, bouncer 10.8 ± 1.4 m. The mean yorker landed ~1.8 m further from the batter than the longest legal yorker — i.e. the characteristic miss is not full enough. The mean stock ball sat exactly on the shortest edge of its band.
3. Scatter is the real difference. Bouncer SD 1.4 m versus 3.3 m and 3.7 m — roughly 2.5× tighter, despite the bouncer having an open-ended target.
4. Individual variation is huge (Table 5.3). Per-bowler mean yorker length ranged from 0.73 m (bowler 12) — an excellent yorker bowler — to 6.76 ± 4.52 m (bowler 18), who was essentially bowling good length while trying to bowl yorkers. Six of 21 bowlers had a mean yorker length over 5 m, i.e. more than 3 m past the target. Per-bowler yorker SD ranged from 0.77 m (bowler 15) to 6.44 m (bowler 16). Bouncer SDs, by contrast, were 0.72–2.05 m for everyone. Length control is a bowler-specific skill, and it is only visible on the full ball.
5. A projectile-motion sensitivity model explains why (Table 5.4). Taking each length’s representative trial and perturbing the release timing by ±1 SD (0.005 s, i.e. about one frame at 200 fps), the resulting spread in landing point was: yorker 5.68 m, stock 4.21 m, bouncer 3.97 m. The yorker is both the most sensitive to a timing error and has the smallest target (2 m vs open-ended). That combination, not a technique flaw, is the core difficulty.
6. Action type was not a selection criterion. Of the 21: 1 side-on, 7 front-on, 8 semi-on/midway, 5 mixed. No analysis was run by action type — the thesis names this as future work.

B. Which body movements produce a different length? (Chapters 6 and 7)

7. Release angle again dominates: yorker 2.2 ± 0.9°, stock 5.4 ± 1.2°, bouncer 13.0 ± 2.7°, p < 0.001, η² = 0.78 — significant in all three pairwise comparisons. It is the only ball variable that cleanly separates all three lengths. (Note: the thesis text on p.69 quotes the yorker as 1.9 ± 1.8°, matching the ISBS 2022 paper, while Table 6.1 prints 2.2 ± 0.9°. Minor internal inconsistency; the direction is unaffected.)

8. Release height: yorker 112.3 ± 4.0%, stock 111.3 ± 3.9%, bouncer 107.4 ± 4.0% of standing height, p < 0.001, η² = 0.22. The bouncer is released lowest.

9. Resultant ball speed hardly changes: yorker 31.5 ± 2.3, stock 32.1 ± 1.9, bouncer 33.1 ± 2.2 m/s, η² = 0.08; only the bouncer differs. Horizontal ball speed did not differ at all (p = 0.07). All the difference between lengths lives in the vertical component: yorker 1.2 ± 0.5, stock 3.0 ± 0.8, bouncer 7.4 ± 1.6 m/s, η² = 0.77. Bowlers change length by changing where they point the ball, not how hard they throw it.

10. Front knee angle at release did not differ at all: yorker 153.6 ± 24.6°, bouncer 156.8 ± 22.8°, stock 152.9 ± 21.8°, p = 0.315. Nor did delivery stride length (p = 0.363), shoulder joint angle (p = 0.124), or shoulder flexion FFC→BR (p = 0.882).

11. The parameters that DO differ are all upper-body and hand:

    • 2D finger orientation (ball–hand line vs horizontal): yorker 58.4 ± 9.3°, stock 52.0 ± 9.1°, bouncer 42.0 ± 9.1°; p < 0.001, η² = 0.31 — significant across all three pairs, the largest technique effect size in the thesis.
    • 2D hand orientation at release: yorker −13.7 ± 10.8°, stock −10.1 ± 11.6°, bouncer +2.1 ± 14.5°; p < 0.001, η² = 0.17.
    • 2D shoulder orientation: yorker 21.5 ± 8.0°, stock 20.9 ± 6.5°, bouncer 28.4 ± 8.1°; p < 0.001, η² = 0.14.
    • Wrist angle at release: yorker 192.6 ± 8.0° (most extended), stock 191.9 ± 10.6°, bouncer 188.6 ± 7.9° (most flexed); p < 0.001 but small η² = 0.02.
    • Thoracic angle at release: yorker 152.1 ± 10.5° (most upright), bouncer 148.2 ± 10.5° (most flexed forward); p = 0.035, η² = 0.02.
    • Thoracic flexion FFC→BR: yorker 27.7 ± 6.7°, bouncer 31.2 ± 7.0°, stock 29.6 ± 7.7°; p = 0.001. The bouncer involves ~3.5° more trunk flexion through the delivery.
    • Wrist flexion FFC→BR: yorker 31.0 ± 16.0°, bouncer 24.7 ± 18.7°, stock 25.6 ± 18.0°; p = 0.027. The yorker uses ~6° more wrist travel.

12. Regression models (one trial per bowler, n = 21 per model):

    Delivery	Predicts release ANGLE	Predicts release HEIGHT %
    Yorker	wrist angle at BR alone = 75.3%; + 2D hand orientation = 86.6%	thoracic angle alone = 38.2%; + 2D thoracic orientation = 55.2%
    Bouncer	no predictor found at all	wrist angle alone = 23.3%; + thoracic angle + 2D hand orientation = 56.9%
    Stock	2D hand orientation alone = 31.2%	2D thoracic orientation alone = 49.8%; + thoracic angle = 60.2%

    Reading across: release angle (= length) is a wrist-and-hand variable; release height is a thoracic (upper-back posture) variable. The bouncer’s release angle is unpredictable because its successful window is so wide that many techniques land inside it.

13. When in the action do the differences appear? (SPM, Chapter 7)

    • No significant differences at any time for: back hip, back knee, front hip, front knee, lumbar angle, shoulder angle, and vertical and horizontal velocity of the centre of mass (all p > 0.05). The run-up and the lower body are essentially identical across the three deliveries.
    • Thoracic angle: differs p = 0.006 in the first 0–30% of back-foot-contact-to-release (stock more extended than bouncer/yorker), converges mid-stride, then differs again p = 0.025 in the last 20% (bouncer most flexed).
    • Wrist angle: differs p < 0.001 from 40% to 80% and p = 0.035 in the last 10%. Yorkers carry a more extended (cocked) wrist all the way from back foot contact, and that gap closes into release — i.e. the yorker uses the biggest late wrist snap.
    • 2D thoracic orientation: p < 0.001 in the last 30%.
    • 2D upper arm orientation: p < 0.001 in the last 20%.
    • 2D hand orientation: p < 0.001 in the last 10% — the latest-appearing difference of all.

    The whole action is the same until roughly the last fifth of the delivery stride. The variation is a late, small, upper-body event.

C. Do the forces change? (Chapter 8) — the injury question

14. No significant difference between delivery types in ANY front-foot loading variable (all p > 0.05, all η² ≤ 0.03):
    • Peak vertical force: yorker 5.06 ± 1.52 BW, bouncer 5.80 ± 1.44 BW, stock 5.75 ± 1.56 BW (p = 0.238).
    • Peak braking force: yorker 2.61 ± 0.98 BW, bouncer 3.02 ± 0.86 BW, stock 3.00 ± 0.91 BW (p = 0.305).
    • Vertical loading rate: 224.6 ± 154.7 / 273.8 ± 157.5 / 247.7 ± 134.6 BW·s⁻¹ (p = 0.583).
    • Horizontal loading rate: 86.6 ± 49.8 / 104.2 ± 56.4 / 101.6 ± 56.3 BW·s⁻¹ (p = 0.554).
    • Vertical impulse (p = 0.12), horizontal impulse (p = 0.49), time to peak forces (p ≥ 0.63).
    • Continuous SPM of braking force and vertical GRF from front foot contact to release: no significant differences anywhere in the curve.
    • Yorkers showed consistently the lowest mean forces, though non-significantly. The thesis speculates this may be why yorker release speeds trended lowest.
15. Upper-body joint moments: almost nothing. Wrist, shoulder and thoracic moments at front foot contact and at ball release all p > 0.05. The only significant continuous finding was wrist moment, p = 0.03 (F = 5.72), in the 40–60% window of front-foot-contact-to-release, with the bouncer carrying a greater flexion moment — and it was not significant at release.
16. Wrist angular velocity at ball release did differ: yorker 15.9 ± 4.0 rad/s, stock 13.9 ± 4.1, bouncer 12.1 ± 3.4; p = 0.009, η² = 0.146. The yorker has the fastest wrist at release, the bouncer the slowest — the bouncer holds the ball marginally longer in a flexed position to build vertical velocity.
17. Kinetics explain the kinematics poorly. Wrist moment at FFC correlated r = 0.566 with wrist angle at release (30.9% explained). Shoulder angular velocity at release plus shoulder moment at FFC explained 45.6% of shoulder angle. Thoracic angular velocity explained only 17.3% of thoracic angle. The thesis’ own conclusion: “the forces / torques are minimal for the kinematic changes.”

What a coach should look for on video

This is the richest section in the whole cluster. Cues are ordered by how strongly the thesis supports them.

1. Wrist position, from back foot contact all the way to release (the yorker cue)

• Camera view + frame: Side-on, slow motion, tight on the arm. Look at three moments: back foot contact, front foot contact, and ball release.
• What “good” looks like: for a yorker, the wrist sits more extended (cocked back) than the same bowler’s bouncer from back foot contact onwards, then snaps through the biggest range (~31° of wrist flexion FFC→BR vs ~25° for a bouncer) and arrives at release still slightly more extended (~193° vs ~189°) and moving fastest (~16 vs ~12 rad/s).
• What the fault looks like: a bowler who sets the wrist in the same position for every ball, or who arrives at release with the flexed, “bouncer” wrist while intending a yorker. That ball comes out at good-length tilt and lands 3–4 m from the batter as a half-volley.
• Why it matters: wrist angle at release alone explained 75.3% of yorker release-angle variance — the strongest single technique–outcome link in the thesis. This is the first thing to check when a bowler keeps missing the yorker.

2. Where the fingers/ball point in the last frame before release

• Camera view + frame: Side-on, highest frame rate the phone allows, freeze on the last frame the ball is in contact with the fingers. Read the line from wrist through the ball against horizontal.
• What “good” looks like: ~58° for a yorker, ~52° for a stock ball, ~42° for a bouncer. A higher finger line = a flatter, fuller ball. It is the only technique variable in the thesis that separated all three lengths (η² = 0.31).
• What the fault looks like: the finger line falling toward the bouncer value on an intended yorker — the fingers coming off the back of the ball early.
• Why it matters: the thesis’ own coaching recommendation. Quoting Chapter 9: finger orientation “is one thing which is coachable where the player can be asked to feel the ball till the last moment.” It is a feel cue — “stay behind and under the ball to the last instant.”

3. Upper-back (thoracic) posture at release — this sets release HEIGHT

• Camera view + frame: Side-on, freeze at ball release. Read the lean of the upper back against vertical, and note ball height relative to the top of the head.
• What “good” looks like: a more upright upper back for the fuller ball (yorker ~152° at release, released at 112% of standing height) and a more folded-forward upper back for the bouncer (~148°, released at 107%, and ~3.5° more trunk flexion accumulated from front foot contact). The difference is small in degrees but drives ~9–10 cm of release height on a 1.87 m bowler.
• What the fault looks like: identical trunk position across variations, or (for the bouncer) a bowler staying tall and trying to “bang it in from height” — the data say the short ball comes from folding over the front leg, not from reaching up.
• Why it matters: thoracic angle plus 2D thoracic orientation explained 55.2% (yorker), 56.9% (bouncer) and 60.2% (stock) of release-height variance.

4. Check that NOTHING ELSE changed

• Camera view + frame: Side-on and front-on, run the whole delivery stride.
• What “good” looks like: run-up speed, delivery stride length, front knee, front hip, back knee, back hip and lower-back (lumbar) angle identical across the three deliveries. In this thesis none of them differed at any moment (all p > 0.05), and centre-of-mass velocity did not change either.
• What the fault looks like: a bowler who visibly runs in harder for a bouncer, lengthens the stride for a yorker, or drops the front knee to “get under it”. Those are compensations, not the mechanism, and they will cost speed and consistency.
• Why it matters: the variation is a last-20%, upper-body-and-hand event. Everything before that should be the bowler’s normal action. This is the most reassuring finding in the file for a coach: you do not need a second action to bowl a variation.

5. Judge length practice by scatter, per bowler, not by the best ball

• Camera view + frame: Behind the arm or high overhead, with the pitch chalked in 2 m bands from the batting stumps.
• What “good” looks like: in this squad the best yorker bowler averaged 0.73 m and the tightest had an SD of 0.77 m. That is what a genuine yorker bowler looks like.
• What the fault looks like: a mean over 5 m (six of 21 bowlers) or an SD over 3 m — the ball is not a yorker, it is a variable good-length ball.
• Why it matters: the thesis argues directly that identifying which of your bowlers can actually repeat a yorker is a selection decision, because a bowler who misses the yorker length repeatedly “will be attacked during the match.”

6. Understand why the yorker misses before blaming the bowler

• Camera view + frame: n/a — this is a framing cue.
• What “good” looks like: accepting that a 0.005 s (one-frame) shift in release timing moves a yorker’s landing point over a 5.68 m range, versus 3.97 m for a bouncer, while the yorker’s target is 2 m wide and the bouncer’s is open-ended.
• What the fault looks like: treating a 25% yorker success rate as a technique failure.
• Why it matters: the honest coaching implication is volume and timing consistency on the yorker, plus the wrist/finger cues above — not a rebuild.

7. Injury: bowling variations does NOT appear to raise front-foot loading

• Camera view + frame: not a video cue — this is a force-plate result you should know.
• What “good” looks like: peak vertical force ~5.1–5.8 BW and peak braking force ~2.6–3.0 BW regardless of whether the bowler is bowling yorkers, bouncers or stock balls (all p > 0.05).
• What the fault looks like: n/a.
• Why it matters: a coach can practise variations without assuming a workload penalty per ball from the variation itself. See the caveats — this is a null result on n = 21, not proof of safety.

Cues this thesis does NOT support: anything about action type (front-on/side-on/mixed) and length — it was recorded but never analysed. Anything about slower balls, cutters, or spin/seam variations — despite the thesis title, only pitch length was studied. Anything about swing.

Caveats and limits

• n = 21, all male, all county academy/MCCU level, age 19 ± 2.2. Not internationals, not women, not club or junior bowlers.
• No batter present, indoors, on an artificial surface. The thesis flags all three: (a) without a physical target the psychological task differs and yorker accuracy could go either way in a match; (b) the artificial pitch’s lower bounce may have pushed bowlers to change length.
• The regressions use one representative trial per bowler — n = 21 per model. The thesis prints 95% confidence intervals on the explained percentages that are enormous (e.g. 86.6% with a CI of “<1% – 97%”, 31.2% with “<1% – 98%”). The point estimates are far less precise than they look. Do not quote “86.6%” as if it were tight. Stepwise regression on n = 21 also over-fits by construction.
• All correlational. No simulation, no intervention, no before/after coaching study. Nothing here demonstrates that changing a bowler’s wrist angle causes a length change — only that the two travel together across bowlers.
• The kinetic chapter is largely a set of null results. With n = 21 and SDs on peak vertical force of ±1.5 BW, the study is not powered to detect modest loading differences between delivery types. “No significant difference” here means “no difference detectable in this sample”, not “no difference exists”.
• Capture rate limits the release instant. The thesis explicitly notes that 250 Hz may have been too slow given that a 0.005 s timing shift moves the ball metres, and that higher frame rates “may have helped identify additional factors linked to pitch length.”
• Action type was never analysed despite the sample containing all four types — named as the thesis’s own leading piece of future work.
• The thesis title over-promises. “Fast Bowling Variations” covers pitch length only; slower balls, cutters, swing and seam position are not studied.
• Minor internal inconsistencies: yorker release angle printed as 2.2 ± 0.9° (Table 6.1) vs 1.9 ± 1.8° (p.69 text); bouncer success rate given as 98.7% (Ch. 5) and 98.3% (Ch. 9); shoulder regression explained variance given as 45.6% (Ch. 8) and 45.3% (Ch. 9).

Relationship to other Felton work

• Chapter 5 is the 2022 ISBS paper (Manawadu, Felton, Hiley & King) — see Manawadu 2022 — pitch length variations and control — with per-bowler tables and the projectile sensitivity model added.
• Chapter 6 is the 2023 World Congress abstract (Manawadu, King, McErlain-Naylor, Felton & Hiley) — see Manawadu 2023 — intended pitch length and release angle — with the full regression tables and the ball-release-height models added. This file supersedes that abstract.
• Builds on Worthington, King & Ranson (2013) and the Felton/King speed line by measuring the same body, differently. It confirms Worthington’s finding that greater thoracic flexion goes with faster balls — here the bouncer has both the most thoracic flexion and the highest speed — and Chapter 9 names untangling thoracic flexion’s dual role in speed and in length as future work.
• Notably contradicts the assumption that front knee behaviour matters for everything. Worthington et al. (2013) and Felton, Yeadon & King (2020) locate ball speed substantially in the front leg. This thesis finds front knee angle at release (p = 0.315) and front knee angle across the whole delivery stride (SPM, p > 0.05) irrelevant to length. That is a refinement, not a conflict: front leg ⇒ speed, wrist and hand ⇒ length.
• Sits directly against the Lamb sub-line in this same folder. Manawadu finds front-foot loading unchanged by delivery type; Lamb finds the loading itself is enormous but heavily attenuated. Together they say: the variation does not add load, but the underlying stock delivery already carries it.

No CONTRADICTION with other Felton work. One TENSION worth noting for coaches, which the thesis itself raises in Chapter 9: greater thoracic flexion is associated with higher ball speed (Worthington 2013) and with the bouncer (this thesis). A coach who trains more trunk flexion to add pace may find the bowler’s natural length creeping shorter, and vice versa. The thesis calls for this connection to be investigated and it has not been.