Pith. sign in

REVIEW 3 major objections 4 minor 62 references

World Cup hydration breaks don't alter match momentum, a within-match case-crossover analysis of 99 matches finds.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 11:42 UTC pith:WW66L74Q

load-bearing objection A careful, honest case-crossover analysis that credibly shows no large average hydration-break effect, with a genuinely useful methodological point about contrast-identified interactions. the 3 major comments →

arxiv 2607.19783 v2 pith:WW66L74Q submitted 2026-07-22 stat.AP stat.ME

Do In-Match Hydration Breaks Alter Match Momentum? A Within-Match Case-Crossover Analysis of the 2026 FIFA World Cup

Debangan Dey This is my paper
classification stat.AP stat.ME
keywords hydration breakmatch momentumcase-crossover designcausal inferenceassociation football2026 FIFA World CupAttack Momentum indexexpected goals
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests the widely repeated claim that the mandatory hydration breaks at the 2026 FIFA World Cup blunt the momentum of whichever team is dominant. Using a within-match case-crossover design, where each break event is compared to non-break minutes of the same match, it finds the dominant side's momentum decays at essentially the same rate whether or not a break occurs, so the apparent post-break fade is normal mean reversion, not an effect of the stoppage. The break only removes minutes of continued pressure, and this removal leaves no trace in net expected goals. The average effect is small and near zero; state-dependent signals for leading teams and hot matches are reported as exploratory.

Core claim

The central discovery is that the dominant side's momentum reverts through the hydration stoppage as if play had continued. Under the clock-aligned counterfactual, break events and continuous-play controls decay in near-parallel, from about +20 to about +8 momentum points, so the break neither creates nor prevents the fade. What the break removes is the continuation of the dominant side's pressure during the stopped minutes, and this leaves no trace in net expected goals (break effect on net xG ≈ 0.011, interval crossing zero). The average break effect on momentum is small (+0.26 clock-aligned, −0.63 play-aligned) with wide confidence intervals, and the paper reads the results by effect size

What carries the argument

The within-match case-crossover design: each break event serves as its own control by comparing post-break momentum to non-break minutes of the same match, differencing out all match-level characteristics. The outcome is a sign-adjusted (dominance-oriented) version of a proprietary minute-resolved Attack Momentum index, with the three in-play break minutes excluded to avoid algorithmic decay artifacts. Identification of the average effect requires extrapolating a smooth game-time trend into the break window, where no untreated minutes exist, so the main effect is fragile; break-by-covariate interactions are identified from cross-sectional variation among break events and are stable across tr

Load-bearing premise

That the proprietary Attack Momentum index is an unbiased measure of real on-field momentum immediately after a stoppage; the paper itself notes the index produces algorithmic decay artifacts during breaks, and if the observed mean reversion is a property of the index rather than of the teams, the central 'no break effect' conclusion collapses.

What would settle it

A concrete check: if a hydration break produced no dip in the momentum index (or if the dip were shown to be a pure index artifact by replacing it with an event-based measure such as shot quality), the 'as if play continued' claim would fail. Conversely, the paper's placebo break at non-break minutes, which returns no effect, is evidence the estimator does not manufacture breaks; replicating that placebo on an independent dataset would further test the design.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • The strong form of the popular claim—that a break reliably halts a dominant side's momentum—is not supported; the fade occurs anyway.
  • The break's measurable cost is the removal of minutes of pressure, not a change in post-break play, so its competitive impact on chance creation is negligible.
  • State-dependent effects on leading teams and in hot matches cannot be ruled out; the data are more compatible with the break eroding a leading side's momentum than enhancing it.
  • A policy intended to be competitively neutral shows no detectable average effect, but the exploratory interactions warrant monitoring and replication in future tournaments.
  • Methodologically, when an intervention recurs at fixed times, its average effect is only weakly identified while its interactions with state are well identified—a lesson for similar designs in sport.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the momentum index's decay during stoppages is partly an algorithmic artifact (as the paper acknowledges for the play-aligned dip), the 'as if play continued' reading may overstate the true continuity of team momentum; an independent event-based momentum measure would test this.
  • The same case-crossover logic could be applied to other fixed-time stoppages—timeouts, quarter breaks—with the same caveat that the main effect is an extrapolation, while interactions are reliable.
  • The Elo interaction suggests breaks might systematically favour stronger teams; if replicated, this could have competitive-balance implications for mandated stoppages in tournaments.
  • The lack of historical no-break control minutes in extreme heat means the effect of breaks in the hottest conditions is essentially unidentifiable; only randomized or staggered implementation could resolve that specific question.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper tests the popular claim that the mandatory hydration breaks at the 2026 FIFA World Cup blunt the momentum of the dominant team. Using a within-match case-crossover design, the author compares post-break outcomes at the two break minutes (~23' and ~68') with non-break minutes of the same match, adjusting for a smooth game-time trend, pre-break momentum level/slope, and scoreline. The outcome is a sign-adjusted proprietary Attack Momentum index, with a supplementary net expected-goals (xG) outcome. The paper reports a small average break effect that is only weakly identified (the main effect is an extrapolation into a window with no untreated minutes), stable break-by-covariate interactions (notably a positive Elo interaction and a negative lead interaction), and agreement between the within-match and external-control designs. The central interpretation is that the dominant side's momentum decays through the stoppage as if play had continued, so the break neither creates nor prevents the fade, although it removes unobserved minutes of pressure.

Significance. If the result holds, it is a timely, policy-relevant challenge to the 'momentum break' narrative surrounding the 2026 World Cup. The within-match design is well suited to removing match-level confounding, and the paper is exemplary in its transparency: the fragility of the main effect (Table 5), the positivity limitation of the external design (Section 5.2), and the exploratory status of the interaction findings are all stated openly. The availability of a full reproducible codebase is a clear strength. However, the central causal claim rests on a proprietary outcome and on an identification assumption about carryover that is not stated, and the independent xG falsification is underreported. With these gaps addressed, the paper could be an important contribution to the momentum literature.

major comments (3)
  1. [§4.2 (A3), §4.4, §6.1] The SUTVA assumption is stated only as no cross-match interference and a single treatment version, but the referent minutes used as no-break controls are observed after a break has already occurred in the same match (e.g., minutes 24–67 for the first-half break). If the break has any effect that persists beyond the 12-minute outcome window, these referents are not realizations of Y(0), and the case-crossover contrast is contaminated. This is especially acute for the second-half break, where every minute after 23' is post-treatment if carryover exists. The paper should state and defend a no-carryover assumption, restrict referents to minutes before the first break as a sensitivity analysis, or model a carryover term. As written, the 'as if play had continued' interpretation in §6.1 relies on an untested assumption that is load-bearing for the central claim.
  2. [§6.1] The net-xG check is load-bearing: it is used to dismiss the play-aligned dip as 'the index rather than the team.' Yet it is reported in a single sentence with only point estimates and intervals. No estimating equation, outcome orientation, window definition, sample size, or estimation details are given. If the xG analysis uses the same fixed-effects model (3), it should be presented in a table; if it uses a different specification, its comparability is unclear. Without this information, the independent-outcome support for the central claim cannot be evaluated.
  3. [Abstract; §1.2; §6.1] The affirmative wording 'the apparent post-break fade is not caused by the break' and 'reverting through the stoppage as if play had continued' is stronger than the identification discussion in §4.4 supports. Table 5 shows the main effect β swings from +1.1 to −5.1 depending on referent band, and per-minute dummies give −1.7±7; the average effect is identified only under a smoothness extrapolation into a window with no untreated minutes. The claims should be qualified as 'consistent with no large average effect under the clock-aligned extrapolation' rather than as a causal no-effect finding, unless the carryover analysis (first major comment) changes the evidential base.
minor comments (4)
  1. [Throughout] The abbreviation 'FIFA' is typeset with a spurious space in many places (e.g., Abstract, Section 1); please correct globally.
  2. [§6.1] If Deb and Dey (2019) is the basis for the xG values used, clarify the connection; if it is only background citation, say so explicitly to avoid the appearance of circularity.
  3. [Figure 4 caption] State whether the ±1.96 bands are pointwise or simultaneous confidence bands.
  4. [§4.5] The random-intercept corroborating check is described as giving 'nearly identical' estimates, but no numbers are reported. Please include the estimates in a supplementary table or appendix.

Circularity Check

0 steps flagged

No significant circularity: the central break-effect contrast is estimated from data, not defined into existence; the lone self-citation (Deb & Dey, 2019) is non-load-bearing.

full rationale

The paper's central claim—that the break neither creates nor prevents the dominant side's momentum fade—is an empirical contrast, not a construction. The outcome Y_i(t)=o_i M_i(t) orients the proprietary momentum series by a sign fixed in the pre-break window, and the same sign and same pre/post window structure are applied symmetrically to break events and to within-match control anchors, so the treated-versus-control contrast is not defined in terms of its own estimate. Equation (3) estimates the break coefficient β jointly with match fixed effects, a game-time trend, and pre-break state; nothing in the specification forces β to zero. Table 5 shows β swings from +1.07 to −5.14 across referent bands, and a placebo break at 38'/82' returns no effect, evidence that the estimator can detect deviations from the counterfactual. The fragility of the average effect is explicitly self-flagged ('The average effect is only weakly identified, resting on a trend extrapolation into a window with no untreated minutes'; 'The main effect is thus identified only under a smoothness assumption on g(c)', §4.4)—an honest identification limitation, not a circular step, since the smoothness assumption does not assert 'no break effect.' The event-based xG outcome is SofaScore's proprietary shot-quality measure, not a value produced by the author's own model; the sole self-citation (Deb & Dey, 2019) appears in a list of shot-conversion modeling references and is not load-bearing for any estimate. No uniqueness theorem is imported from the authors, no ansatz is smuggled in via citation, and mean reversion is analyzed rather than renamed. The manuscript's own caveats—that the play-aligned dip 'reflects the index rather than the team' and that the external design has a positivity void—are measurement and identification concerns, not reductions of the result to its inputs. I therefore find no significant circularity; the score of 2 reflects only the presence of a minor, non-load-bearing self-citation in the literature review.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central claim rests on a handful of modeling choices (exclusion window, outcome horizon, heat split, trend specification) and on the validity of a proprietary external index. No new physical or theoretical entities are introduced; the analysis transforms a commercial metric. The main-effect extrapolation is the most fragile part: it depends on the smoothness of g(c) at a minute where no untreated observation exists.

free parameters (5)
  • In-play exclusion window width = 3 minutes
    Uniform 3-minute blanking at each break; justified by 96% of breaks being 2-3 min, but applied identically to controls.
  • Post-break outcome window = 10 minutes ([s+3, s+12])
    Choice of outcome horizon; sensitivity not fully explored.
  • WBGT hot/cold split = 28 °C
    Ad hoc threshold for heat modification; identified as a modifier after data inspection.
  • Quadratic game-time trend g(c) = coefficients estimated
    Smoothness assumption on which the main effect extrapolation rests; main effect swings under per-minute dummies.
  • Propensity-score caliper width = 0.2 SD
    Standard recommendation (Austin 2011); still a choice affecting external match.
axioms (5)
  • domain assumption A1: Conditional exchangeability — within a match, whether a minute is a break minute is independent of potential outcomes given game time, scoreline, and pre-break momentum.
    Stated in §4.2; the placebo break at 38'/82' is a weak check, but the assumption is untestable for the real break minutes.
  • domain assumption A2: No anticipation — play immediately before the break is not altered by the upcoming stoppage.
    Stated in §4.2; players and coaches know the break is coming, so this is questionable and untested.
  • standard math A3: SUTVA — no interference between matches and a single version of treatment.
    Standard causal assumption; plausible in this setting.
  • domain assumption A4: Within-match positivity — referent minutes exist with comparable scoreline and momentum state.
    Stated in §4.2; fails for game time itself, forcing the main effect to rely on extrapolation (§4.4).
  • domain assumption SofaScore Attack Momentum index measures match momentum faithfully.
    Used throughout; validated only indirectly via goal alignment and decay signature (§2.1), and the paper acknowledges index artefacts (§1.2, §3).

pith-pipeline@v1.3.0-alltime-deepseek · 18598 in / 13911 out tokens · 135267 ms · 2026-08-01T11:42:59.312304+00:00 · methodology

0 comments
read the original abstract

The 2026 FIFA World Cup was the first to mandate hydration breaks, brief stoppages near the twenty-second minute of each half, in every match regardless of weather. A widespread claim holds that these breaks blunt the momentum of whichever side is dominant. We test the claim with a within-match case-crossover design, using non-break minutes of the same match as self-matched controls. This differences out every match-level characteristic and adjusts only for game time, scoreline, and pre-break momentum state. The apparent post-break fade is not caused by the break: the dominant side's momentum decays at essentially the same rate whether or not a break occurs, reverting through the stoppage as if play had continued. What the break removes instead is the continuation of that side's pressure during the stopped minutes, and this leaves no trace in net expected goals. The average effect is only weakly identified, resting on a trend extrapolation into a window with no untreated minutes, whereas break-by-covariate interactions are stable; that average is small and near zero. With only 99 matches the design lacks power to reject a small effect, so we read the results by effect size and confidence interval. The estimates are not uniform across match states. The clearest modifier is team strength: the break effect shifts by about +1.6 momentum points per 100 Elo points of rating advantage, favouring a stronger in-form side over a weaker side that is momentarily dominant, and it is the only statistically significant interaction. A weaker pattern appears by scoreline, where a comfortably-leading side of average strength shows an implied loss near -4 points on a pre-break level of about +20. We report these state-dependent signals as exploratory and in need of replication.

Figures

Figures reproduced from arXiv: 2607.19783 by Debangan Dey.

Figure 1
Figure 1. Figure 1: Constructing the sign-adjusted outcome, with the break shown as a stoppage. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Breaks occur only in two narrow windows (≈ 23′ , ≈ 68′ ). The estimand is local to these minutes. and cannot, in the absence of an untestable homogeneity assumption, speak to a break at some other minute. Second, the policy breaks in every match, so that, unlike historical tournaments in which a stoppage of this kind occurred only in hot conditions, heat and the administration of a break are not separable … view at source ↗
Figure 3
Figure 3. Figure 3: The two counterfactuals for the break window. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: What the two designs show. Event-time average of sign-adjusted momentum for break events (red) and controls (blue), oriented toward the pre-break dominant side. (a) Clock-aligned: both series are aligned on the game clock and the in-break minutes are blanked; the two decay alike, so the break neither creates nor prevents the fade. (b) Play-aligned: the break is removed and the resume minute is play +1; the… view at source ↗
Figure 5
Figure 5. Figure 5: Implied break effect on the in-form side, by match state. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Positivity void for the external-control design. [PITH_FULL_IMAGE:figures/full_fig_p022_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

62 extracted references

  1. [1]

    and Brito, Jo

    Nassis, George P. and Brito, Jo. The association of environmental heat stress with performance: analysis of the 2014. British Journal of Sports Medicine , year =

    2014

  2. [2]

    Consensus recommendations on training and competing in the heat , journal =

    Racinais, S. Consensus recommendations on training and competing in the heat , journal =. 2015 , volume =

    2015

  3. [3]

    , title =

    Budd, Grahame M. , title =. Journal of Science and Medicine in Sport , year =

  4. [4]

    , title =

    Gouttebarge, Vincent and Duffield, Rob and den Hollander, Steve and Maughan, Ron J. , title =. BMJ Open Sport & Exercise Medicine , year =

  5. [5]

    Cognitive Psychology , year =

    Gilovich, Thomas and Vallone, Robert and Tversky, Amos , title =. Cognitive Psychology , year =

  6. [6]

    Journal of Sport and Exercise Psychology , year =

    Gernigon, Christophe and Briki, Walid and Eykens, Katie , title =. Journal of Sport and Exercise Psychology , year =

  7. [7]

    and Colavecchio, Paul G

    Vallerand, Robert J. and Colavecchio, Paul G. and Pelletier, Luc G. , title =. Journal of Sport and Exercise Psychology , year =

  8. [8]

    American Journal of Epidemiology , year =

    Maclure, Malcolm , title =. American Journal of Epidemiology , year =

  9. [9]

    and Maclure, Malcolm and Robins, James M

    Mittleman, Murray A. and Maclure, Malcolm and Robins, James M. , title =. American Journal of Epidemiology , year =

  10. [10]

    and Jackson, Rodney T

    Marshall, Roger J. and Jackson, Rodney T. , title =. Statistics in Medicine , year =

  11. [11]

    Environmetrics , year =

    Lumley, Thomas and Levy, Drew , title =. Environmetrics , year =

  12. [12]

    Epidemiology , year =

    Janes, Holly and Sheppard, Lianne and Lumley, Thomas , title =. Epidemiology , year =

  13. [13]

    , title =

    Mittleman, Murray A. , title =. Epidemiology , year =

  14. [14]

    Paddy , title =

    Farrington, C. Paddy , title =. Biometrics , year =

  15. [15]

    and Farrington, C

    Whitaker, Heather J. and Farrington, C. Paddy and Spiessens, Bart and Musonda, Patrick , title =. Statistics in Medicine , year =

  16. [16]

    International Journal of Epidemiology , year =

    Lopez Bernal, James and Cummins, Steven and Gasparrini, Antonio , title =. International Journal of Epidemiology , year =

  17. [17]

    and Soumerai, Stephen B

    Wagner, Anita K. and Soumerai, Stephen B. and Zhang, Fang and Ross-Degnan, Dennis , title =. Journal of Clinical Pharmacy and Therapeutics , year =

  18. [18]

    , title =

    Rubin, Donald B. , title =. Journal of Educational Psychology , year =

  19. [19]

    , title =

    Holland, Paul W. , title =. Journal of the American Statistical Association , year =

  20. [20]

    Causal Inference: What If , publisher =

    Hern. Causal Inference: What If , publisher =

  21. [21]

    Econometrica , year =

    Mundlak, Yair , title =. Econometrica , year =

  22. [22]

    , title =

    Liang, Kung-Yee and Zeger, Scott L. , title =. Biometrika , year =

  23. [23]

    Colin and Miller, Douglas L

    Cameron, A. Colin and Miller, Douglas L. , title =. Journal of Human Resources , year =

  24. [24]

    , title =

    Morris, Jeffrey S. , title =. Annual Review of Statistics and Its Application , year =

  25. [25]

    and Caffo, Brian and Reich, Daniel , title =

    Goldsmith, Jeff and Bobb, Jennifer and Crainiceanu, Ciprian M. and Caffo, Brian and Reich, Daniel , title =. Journal of Computational and Graphical Statistics , year =

  26. [26]

    and Lazar, Nicole A

    Wasserstein, Ronald L. and Lazar, Nicole A. , title =. The American Statistician , year =

  27. [27]

    Nature , year =

    Amrhein, Valentin and Greenland, Sander and McShane, Blake , title =. Nature , year =

  28. [28]

    and Rothman, Kenneth J

    Greenland, Sander and Senn, Stephen J. and Rothman, Kenneth J. and Carlin, John B. and Poole, Charles and Goodman, Steven N. and Altman, Douglas G. , title =. European Journal of Epidemiology , year =

  29. [29]

    Journal of Science and Medicine in Sport , year =

    Chalmers, Samuel and Siegler, Jason and Lovell, Ric and Lynch, Grant and Gregson, Warren and Marshall, Paul and Jay, Ollie , title =. Journal of Science and Medicine in Sport , year =

  30. [30]

    and Chalmers, Samuel and Topham, Thomas H

    Brown, Harry A. and Chalmers, Samuel and Topham, Thomas H. and Clark, Brad and Meyer, Tim and Jowett, Andrew and Jay, Ollie and P. Efficacy of the. Journal of Science and Medicine in Sport , year =

  31. [31]

    Biology of Sport , year =

    Almeida, Carlos Humberto and Leite, Werlayne Stuart Soares , title =. Biology of Sport , year =

  32. [32]

    International Journal of Sports Medicine , year =

    Link, Daniel and Anzer, Gabriel , title =. International Journal of Sports Medicine , year =

  33. [33]

    International Journal of Environmental Research and Public Health , year =

    Garc. International Journal of Environmental Research and Public Health , year =

  34. [34]

    Journal of Sports Economics , year =

    Fischer, Kai and Haucap, Justus , title =. Journal of Sports Economics , year =

  35. [35]

    International Journal of Performance Analysis in Sport , year =

    Siegle, Martin and Lames, Martin , title =. International Journal of Performance Analysis in Sport , year =

  36. [36]

    Science and Medicine in Football , year =

    Li, Yuchen and Weber, Hendrik and Link, Daniel , title =. Science and Medicine in Football , year =

  37. [37]

    Den Hartigh, Ruud J. R. and Van Yperen, Nico W. and Gernigon, Christophe , title =. Science and Medicine in Football , year =

  38. [38]

    and Sunderland, Caroline A

    Redwood-Brown, Alison J. and Sunderland, Caroline A. and Minniti, Antonio M. and O'Donoghue, Peter G. , title =. International Journal of Sport and Exercise Psychology , year =

  39. [39]

    Journal of Economic Behavior & Organization , year =

    Gauriot, Romain and Page, Lionel , title =. Journal of Economic Behavior & Organization , year =

  40. [40]

    The advantage of scoring just before the half-time break---pure myth?

    Meier, Philippe and Flepp, Raphael and R. The advantage of scoring just before the half-time break---pure myth?. Journal of Sports Economics , year =

  41. [41]

    PLOS ONE , year =

    Baert, Stijn and Amez, Simon , title =. PLOS ONE , year =

  42. [42]

    CHANCE , year =

    Kent, Jonathan and Brandes, Ulrik , title =. CHANCE , year =

  43. [43]

    Match running performance preceding scoring and conceding a goal in men's professional soccer , journal =

    Konefa. Match running performance preceding scoring and conceding a goal in men's professional soccer , journal =. 2024 , volume =

    2024

  44. [44]

    The Economic Journal , year =

    Gauriot, Romain and Page, Lionel , title =. The Economic Journal , year =

  45. [45]

    and Sanjurjo, Adam , title =

    Miller, Joshua B. and Sanjurjo, Adam , title =. Econometrica , year =

  46. [46]

    Psychology of Sport and Exercise , year =

    Bar-Eli, Michael and Avugos, Simcha and Raab, Markus , title =. Psychology of Sport and Exercise , year =

  47. [47]

    and Elmore, Ryan and Fosdick, Bailey K

    Gibbs, Connor P. and Elmore, Ryan and Fosdick, Bailey K. , title =. The Annals of Applied Statistics , year =

  48. [48]

    Separating psychological momentum from strategic momentum: Evidence from men's professional tennis , journal =

    Meier, Philippe and Flepp, Raphael and R. Separating psychological momentum from strategic momentum: Evidence from men's professional tennis , journal =. 2020 , volume =

    2020

  49. [49]

    Den Hartigh, Ruud J. R. and Gernigon, Christophe , title =. Journal of Sports Sciences , year =

  50. [50]

    Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining (KDD '21) , year =

    Robberechts, Pieter and Van Haaren, Jan and Davis, Jesse , title =. Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining (KDD '21) , year =

  51. [51]

    2026 , month = jun, day =

    Why are there hydration breaks at the 2026. 2026 , month = jun, day =

    2026

  52. [52]

    2026 , month = jun, day =

    England's. 2026 , month = jun, day =

    2026

  53. [53]

    2026 , month = mar, day =

    Vertelney, Seth , title =. 2026 , month = mar, day =

    2026

  54. [54]

    and Rubin, Donald B

    Rosenbaum, Paul R. and Rubin, Donald B. , title =. Biometrika , year =

  55. [55]

    , title =

    Stuart, Elizabeth A. , title =. Statistical Science , year =

  56. [56]

    , title =

    Austin, Peter C. , title =. Pharmaceutical Statistics , year =

  57. [57]

    2026 , howpublished =

    2026

  58. [58]

    Journal of Sports Analytics , year =

    Deb, Soudeep and Dey, Debangan , title =. Journal of Sports Analytics , year =

  59. [59]

    Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD '19) , year =

    Decroos, Tom and Bransen, Lotte and Van Haaren, Jan and Davis, Jesse , title =. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD '19) , year =

  60. [60]

    Journal of Human Sport and Exercise , year =

    Rathke, Alex , title =. Journal of Human Sport and Exercise , year =

  61. [61]

    Proceedings of the Workshop on Machine Learning and Data Mining for Sports Analytics (MLSA), ECML/PKDD , year =

    Eggels, Harm and van Elk, Ruud and Pechenizkiy, Mykola , title =. Proceedings of the Workshop on Machine Learning and Data Mining for Sports Analytics (MLSA), ECML/PKDD , year =

  62. [62]

    PLOS ONE , year =

    Mead, James and O'Hare, Anthony and McMenemy, Paul , title =. PLOS ONE , year =