Stop Guessing: Optimizing Goalkeeper Policies for Soccer Penalty Kicks

Jesse Davis; Lotte Bransen; Tim Janssen

REVIEW 3 major objections 5 minor 36 references

Stop Guessing: Optimizing Goalkeeper Policies for Soccer Penalty Kicks

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper argues that goalkeeper penalty strategy should be chosen per goalkeeper, not per generic game-theoretic optimum, because a keeper's physical dive range and ability to guess the right corner determine which policy works.

desk verdict A genuinely useful simulation framework for goalkeeper penalty policies, with one headline claim that needs an 'upper bound under non-reactive takers' asterisk. read the letter →

arxiv 2505.24629 v1 pith:NGIPTATR submitted 2025-05-30 cs.LG cs.GT

Lotte Bransen , Tim Janssen , Jesse Davis This is my paper

classification cs.LGcs.GT

keywords penaltykicksgoalkeeperpolicysportsanalyticsactioncapacitiessimulationframeworkdivetimingdirectionpredictiongametheory

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that goalkeeper penalty strategy should be chosen per goalkeeper, not per generic game-theoretic optimum, because a keeper's physical dive range and ability to guess the right corner determine which policy works. It builds a simulation that replays 7,872 annotated historical penalties and estimates save probability as the product of guessing the right corner and reaching the ball from that corner. The main results are that the optimal policy among early, late, early-educated, and mixed-educated dives shifts with the keeper's early and late dive ranges, and that standing slightly off center, toward the kicker's natural corner, raises expected save percentage by 0.5 to 3 percentage points. The point of the framework is to give coaches actionable, capacity-aware advice, including whom to put in for a shootout.

What carries the argument

The central object is the goalkeeper policy π, a rule that fixes initial lineup, dive timing, and dive direction. The argument is carried by Equation 2, which converts a Euclidean distance between the keeper's starting point and the ball's crossing point into a save probability given a correct-corner dive: zero beyond reach plus a tolerance µ, ρ within reach minus µ, and linearly interpolated in between, with ρ=0.7 and µ=0.7 m fitted to data. Around this sit two learned models — a direction classifier (natural corner, non-natural corner, center) and a distance regressor — that let a policy be 'educated' by predicting where the kick will go and whether a late dive can reach it. This machinery lets the same historical kicks be replayed under any keeper's dive ranges, which is what makes per-keeper policy comparison possible.

What would settle it

Collect data with the goalkeeper's starting position and ball end coordinates. If, at fixed within-reach distance, save rates differ by ball speed or kick height, Equation 2 is falsified. Or, if the fraction of kicks aimed at the shaded corner drops when the keeper moves off center, the estimated 0.5 to 3 percentage point benefit is overestimated.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the best policy for a specific goalkeeper depends on their action capacities, and aligning slightly off center increases the chance of saving a kick. The claim is established through a player-agnostic simulation framework: for each historical kick, the probability of a save is the probability of diving to the correct corner times the probability of stopping the kick given a correct-corner dive, where the second factor is a piecewise-linear function of the Euclidean distance between the goalkeeper's starting point and the ball's crossing point, controlled by a reach radius, a tolerance, and a within-reach save probability. Evaluated over 7,872 penalties, the framework finds no universally best policy: a keeper who can dive late should usually use the mixed educated policy, a keeper who cannot should use the early educated policy, and the aggregate game-theoretic optimum is beaten by policies that respect the keeper's actual range.

Load-bearing premise

The framework assumes both that save probability given a correct-corner dive depends only on the distance between the keeper's starting point and the ball's crossing point through a fitted curve (Equation 2), and that takers do not re-aim when the keeper stands visibly off center; if either fails, the policy rankings and the shading benefit change.

Editorial extensions

If this is right

For a keeper who can dive late, the mixed educated policy is the best of the four considered; for a keeper who cannot dive late, early educated beats plain early.
Shading 0.1 to 0.3 m toward the kicker's natural corner raises expected save percentage by 0.5 to 3 percentage points on on-target kicks, assuming the taker does not change aim in response.
The aggregate game-theoretic optimum, which says to dive late 87.1% of the time, is not a reliable individual recommendation because many keepers lack the late-dive range to execute it.
Shootout selection differs from in-game selection: a keeper with a strong early dive but no late dive looks better in shootouts because shootout takers use keeper-dependent kicks less often (15.1% versus 21.2%).

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Editorial inference: if ball speed and kick height were recorded, the distance-only save curve in Equation 2 could be re-estimated as a conditional surface; a testable prediction is that the fitted tolerance µ shrinks once those covariates are included.
Editorial inference: the shading result could be coupled to the taker-response studies the paper cites, yielding an equilibrium offset at which the keeper gains from better reach without the taker shifting aim; the paper stops at the no-response case.
Editorial inference: the framework can be run in reverse as a training diagnostic, identifying which dive range (early or late) the recommended policy is most sensitive to, so coaches can target that range in drills.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Short version: this paper deserves a real referee slot. It offers a genuinely new simulation framework for comparing goalkeeper penalty policies, with a large expert-annotated dataset and a sensible focus on action capacities. The headline off-center positioning result is real but conditional: it assumes takers do not change their aim, and the paper itself flags that in Section 5.2 even though the abstract and conclusion state it more flatly.

What is actually new: prior game-theoretic models (Palacios-Huerta, Tuyls et al.) treat penalty kicks as zero-sum with independent pre-selected actions. This paper explicitly models early vs late dives, dive ranges, initial alignment, and uses learned player-agnostic direction/distance models. The policy comparison and the practical use cases (goalkeeper substitution, shootout advice) are the kind of thing a coaching staff could actually use. The nested cross-validation and calibration checks for the learned models are done properly, and the data-merging appendix is unusually detailed.

Soft spots, in order of real softness. First, the off-center benefit (0.5–3pp in Figure 2) is computed by shifting the goalkeeper’s starting position while holding the historical ball end-coordinates fixed. The only mechanism is the distance term in Equation 2. The authors cite work showing visible off-center alignment can change taker direction, so this number is best read as an upper bound under a zero-response assumption. They do include the caveat and suggest moving during the run-up, but the paper’s main conclusion would be more honest if it led with that qualification. Second, the four parameters in Equation 2 (ρ, μ, early/late ranges) are grid-searched on the same dataset used for evaluation. The absolute expected save percentages are therefore in-sample; the relative policy rankings are more trustworthy than the point estimates. Third, the mixed educated policy is defined so that a late dive is chosen whenever the predicted distance is within late range, so its advantage over early-educated is somewhat built in rather than discovered.

None of these are fatal. The framework is coherent, the limitations are mostly acknowledged somewhere, and the paper is a legitimate contribution to sports analytics. If I were refereeing it, I’d ask for (1) a sensitivity analysis with a simple reactive-taker model, (2) bootstrap or prior-based uncertainty on the save percentages, and (3) an explicit restatement of the off-center claim as conditional on taker non-response. That would make it a solid accept. For a general ML audience it is a good example of simulation-based policy evaluation in a low-data domain.

Referee Report

3 major / 5 minor

Summary. The paper proposes a simulation framework for evaluating goalkeeper policies in soccer penalty kicks, using a merged dataset of 7,872 penalties annotated with kicker/goalkeeper strategies. The framework learns a direction model and a distance model, defines four policies (late, early, early educated, mixed educated), and evaluates them under varying goalkeeper dive ranges and initial positions. The main claims are that the optimal policy depends on a goalkeeper's action capacities and that aligning slightly off-center increases expected save percentage. The paper also demonstrates practical use cases such as selecting a goalkeeper for shootouts and generating taker-specific advice.

Significance. The work is valuable for sports analytics because it addresses the interaction between kickers and goalkeepers, incorporates action capacities, and provides a player-agnostic simulation framework with nested cross-validation for the learned models. The use of expert-annotated strategies and the practical use cases are strengths. However, the headline off-center claim rests on an unmodeled taker-response assumption, and the physical save model's parameters are fitted in-sample, which limits the strength of the quantitative conclusions until these are addressed.

major comments (3)

[Section 5.2, Figure 2] The off-center benefit is computed by evaluating historical end coordinates of kicks taken when goalkeepers stood centered, with only the distance term in Eq. (2) changing. The manuscript itself cites evidence [21,25,31] that takers adjust their aim when the goalkeeper is visibly off-center, so the 0.5–3 percentage point gain is an upper bound under a zero-response assumption. A sensitivity analysis that models taker re-aiming (or at least varies the assumed response rate) is needed to substantiate the claim that aligning off-center increases save percentage; otherwise the claim should be explicitly reframed as an upper bound.
[Appendix C and Section 5] The parameters ρ, μ, r_early, and r_late in Eq. (2) are fitted by grid search on the same dataset used for policy evaluation in Sections 5.1 and 5.2. Because these parameters directly determine p(s|k,c,π,gk), the reported expected save percentages are in-sample and may be optimistically biased. The calibration plot in Figure 6 is also on the same data. A nested or held-out calibration procedure, or at least a report of how sensitive the policy rankings are to variations in these parameters, is required to support the quantitative comparisons.
[Section 4.1, Section 5.1, Figure 1] The mixed educated policy is defined to dive late exactly when the predicted kick distance is within the goalkeeper's late range. Its superiority over always-late or always-early policies, while expected under this decision rule, is partly by construction rather than an empirical discovery. The paper should benchmark the mixed educated policy against an oracle policy that uses true distances (or a policy that uses the distance model with an alternative threshold) to quantify the cost of distance prediction error, especially given that the distance model's accuracy is only slightly above baselines (Figure 3).

minor comments (5)

[Section 5.4, Table 2] The text for Use Case 2 is inconsistent with the table: it says GK2 is expected to stop 18.4% but the table reports 0.187, and it says GK1 and GK3 save '17.4% (GK1) and 17.5% (GK2)' under the mixed policy, whereas the table shows GK3, not GK2, at 0.178; the corresponding percentages in the text should be corrected.
[Section 4.2] The word 'emperical' should be 'empirical'.
[Appendix C] The phrase 'proabily estimation task' should be 'probability estimation task'.
[Appendix A.3] The similarity measure is referred to as 'Ratfcliff-Obershelp' but should be 'Ratcliff-Obershelp'.
[General notation] The notation d(locs, ke)(π) in and around Eq. (2) is confusing because the policy argument appears as a function on the distance; consider using r_π(gk) or an explicit subscript to denote the dive range for policy π.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's claims are explicit simulation outputs of a stated model, with the key limitation acknowledged by the authors.

full rationale

The paper is an explicit simulation study. Eq. (2) defines p(s|k,c,π,gk) as a piecewise-linear decreasing function of the distance between the goalkeeper's starting location and the ball's crossing point; the off-center evaluation in §5.2 re-evaluates this function with a shifted starting location. The reported 0.5–3 percentage point benefit is therefore a consequence of the model assumption, but it is not a circular derivation: Eq. (2) is not defined in terms of the policy rankings or the off-center conclusion, and the paper does not present the result as an independent empirical measurement. The parameters ρ=0.7, μ=0.7 and the dive ranges are fitted in Appendix C on the same dataset, which limits external validity and makes the counterfactual dependent on the fitted distance–save relationship; this is a correctness/robustness concern, not a self-referential reduction. Similarly, the mixed-educated policy's definition — dive late when the predicted distance is within the late range — makes its strong performance under Eq. (2) unsurprising, but §5.1 presents this as a simulation comparison rather than an independent discovery. The authors explicitly flag the most important threat to the off-center claim in §5.2: 'players may change their penalty direction based on the positioning of the goalkeeper,' citing [21,31,25]; this is a stated limitation rather than a hidden circular step. The only self-citation ([6]) appears in related work and is not load-bearing. No step in the derivation chain reduces to its own input by construction.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The framework's conclusions depend on a fitted distance-based save model and on behavioral assumptions about takers. The free parameters (ρ, µ, dive ranges) are all fitted in-sample on the same dataset used for policy evaluation. The most fragile assumption for the off-center recommendation is that taker direction is independent of goalkeeper position, which is acknowledged as questionable for visible shifts.

free parameters (4)

ρ (within-reach save probability) = 0.7
Fitted in Appendix C by grid search (range 0.5-1.0) to minimize Brier score for Eq. 2 on the merged penalty dataset; used to compute expected save percentages in Section 5.
µ (distance tolerance/uncertainty) = 0.7 m
Fitted in Appendix C alongside ρ via grid search; defines the linear falloff region in Eq. 2.
r_early (early dive range) = 3.1 m
Fitted in Appendix C as the value minimizing Brier score for Eq. 2; used as the default early range in experiments and use cases.
r_late (late dive range) = 2.8 m
Fitted in Appendix C as the value minimizing Brier score; used as the default late range and varied in Section 5.1.

assumptions (6)

standard math Minimax theorem for finite zero-sum games
Used in Section 3 to derive optimal mixed strategies from the empirical payoff matrix.
domain assumption Penalty kick can be modeled as a zero-sum game with the specified action sets (kicker: N, C, NN, Dep; keeper: N, Late, NN)
Basis for the augmented game in Section 3; the Minimax solution is computed on the empirical payoff matrix.
domain assumption Save probability given correct corner depends only on Euclidean distance between goalkeeper start and ball crossing point, via Eq. 2 with fitted parameters
Core of the simulation (Section 4.3); no dependence on ball speed, height, spin, or individual technique beyond distance.
domain assumption Goalkeeper starts at the center of the goal when computing distances
Stated in Appendix C; off-center policies are then modeled as shifts in start position.
ad hoc to paper Taker direction distribution is unaffected by goalkeeper initial position in the off-center analysis
Assumed in Section 5.2; the paper acknowledges takers may adjust to clearly off-center keepers, and suggests dynamic alignment instead, but the Figure 2 result uses the fixed-distribution assumption.
domain assumption Keeper-dependent takers' outcomes can be summarized by the empirical probability of guessing the correct corner
Used in Section 4.2 for dependent kicks; the interaction between keeper movement and dependent taker choice is not explicitly modeled.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Stop Guessing: Optimizing Goalkeeper Policies for Soccer Penalty Kicks." pith.science (2026). https://pith.science/paper/NGIPTATR

@misc{pith2026250524629,
  author       = {Pith},
  title        = {Pith review of: Stop Guessing: Optimizing Goalkeeper Policies for Soccer Penalty Kicks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NGIPTATR}},
  note         = {Machine review of arXiv:2505.24629}
}

read the original abstract

Penalties are fraught and game-changing moments in soccer games that teams explicitly prepare for. Consequently, there has been substantial interest in analyzing them in order to provide advice to practitioners. From a data science perspective, such analyses suffer from a significant limitation: they make the unrealistic simplifying assumption that goalkeepers and takers select their action -- where to dive and where to the place the kick -- independently of each other. In reality, the choices that some goalkeepers make depend on the taker's movements and vice-versa. This adds substantial complexity to the problem because not all players have the same action capacities, that is, only some players are capable of basing their decisions on their opponent's movements. However, the small sample sizes on the player level mean that one may have limited insights into a specific opponent's capacities. We address these challenges by developing a player-agnostic simulation framework that can evaluate the efficacy of different goalkeeper strategies. It considers a rich set of choices and incorporates information about a goalkeeper's skills. Our work is grounded in a large dataset of penalties that were annotated by penalty experts and include aspects of both kicker and goalkeeper strategies. We show how our framework can be used to optimize goalkeeper policies in real-world situations.

Figures

Figures reproduced from arXiv: 2505.24629 by the authors.

**Figure 1.** Each plot shows how the expected save percentage varies as function of the early dive range for a fixed late dive range. from 2.6m to 2.9m and the early range from 3.0m to 3.2m in increments of 0.1m [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗

**Figure 2.** Expected save percentage for on-target penalties as function of moving x meters toward the kicker’s natural corner. tigate a policy where the goalkeeper’s initial alignment is slightly closer to one corner. Research has indicated that player’s are able to shoot more forcefully (i.e., higher ball velocity) when aiming toward their natural corner [20]. Moreover, the natural corner is the most common placement locatio… view at source ↗

**Figure 3.** Accuracy per range threshold for our Penalty Distance Model and baselines. Most shots are placed within 2.5 and 3 meters from the center, and this is also the area in which our model slightly outperforms the baselines. probability estimates are, we report the logloss. On all penalties, our model obtains a logloss of 0.954 ± 0.011 vs. 0.969 ± 0.014 for the base model. When only considering shootout penalties, our mod… view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: The distribution of penalties (both shootout and non-shootout) taken by players prior to a shootout, per order in the shootout. We observe that the further we progress into the shootout, the less experienced players get. B Features for the Penalty Distance and Directio…

**Figure 5.** Figure 5: The influence of within-reach save probability ρ and uncertainty factor µ on the save probability. the fact that player’s might be biased by their most recent penalty’s result. Therefore we add the same information (goal, save or miss, and corner) for the player’s most…

**Figure 6.** Figure 6: Calibration plot for predicted probabilities of saving a penalty given that you are in the correct corner using r(πe, gk) = 3.1, r(πl, gk) = 2.8, µ = 0.7, ρ = 0.7. show that this Equation makes sense and learn the uncertainty values from our dataset. We evaluate this E…

**Figure 7.** Figure 7: Calibration plots for the predictions by the Penalty Direction Model for each of our classes (non-natural, center, natural). corner, with this effect being stronger when a goal will result in a shootout win. Finally, the first kick for each team in a shootout is aimed …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

36 extracted references · 35 canonical work pages

[1]

International Journal of Performance Analysis in Sport16, 508–522 (08 2016)

Almeida, C., Volossovitch, A., Duarte, R.: Penalty kick outcomes in UEFA club competitions (2010-2015): The roles of situational, individual and performance factors. International Journal of Performance Analysis in Sport16, 508–522 (08 2016)

work page 2010
[2]

American Economic Review 100(5), 2548–64 (December 2010)

Apesteguia, J., Palacios-Huerta, I.: Psychological pressure in competitive envi- ronments: Evidence from a randomized natural experiment. American Economic Review 100(5), 2548–64 (December 2010)

work page 2010
[3]

Azar, O.H., Bar-Eli, M.: Do soccer players play the mixed-strategy Nash equilib- rium? Applied Economics43(25), 3591–3601 (2011)

work page 2011
[4]

Soccer and Society10 (03 2009)

Bar-Eli, M., Azar, O.: Penalty kicks in soccer: An empirical analysis of shooting strategies and goalkeepers’ preferences. Soccer and Society10 (03 2009)

work page 2009
[5]

Journal of Economic Psychology 28, 606–621 (10 2007)

Bar-Eli, M., Azar, O., Ritov, I., Keidar-Levin, Y., Schein, G.: Action bias among elite soccer goalkeepers: The case of penalty kicks. Journal of Economic Psychology 28, 606–621 (10 2007)

work page 2007
[6]

In: Proceedings of the Sports Tomorrow Congress 2021 (2021)

Bransen, L., Davis, J.: Predicting football penalty directions using in-match per- formance indicators. In: Proceedings of the Sports Tomorrow Congress 2021 (2021)

work page 2021
[7]

International Journal of Performance Analysis in Sport22, 571 – 582 (2022)

Buscà, B., Hileno, R., Nadal, B., Serna, J.: Prediction of the penalty kick direction in men’s soccer. International Journal of Performance Analysis in Sport22, 571 – 582 (2022)

work page 2022
[8]

In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

Chen, T., Guestrin, C.: XGBoost: A scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. pp. 785–794 (2016)

work page 2016

Show all 36 references

[9]

American Economic Review 92(4), 1138–1151 (2002)

Chiappori, P.A., Levitt, S., Groseclose, T.: Testing mixed-strategy equilibria when players are heterogeneous: The case of penalty kicks in soccer. American Economic Review 92(4), 1138–1151 (2002). https://doi.org/10.1257/00028280260344678

2002 doi
[10]

Percep- tion 39, 1111–24 (08 2010)

Dicks, M., Button, C., Davids, K.: Availability of advance visual information con- strains association-football goalkeeping performance during penalty kicks. Percep- tion 39, 1111–24 (08 2010)

2010
[11]

Attention, Perception & Psychophysics72, 706–20 (04 2010)

Dicks, M., Button, C., Davids, K.: Examination of gaze behaviors under in situ and video simulation task constraints reveals differences in information pickup for perception and action. Attention, Perception & Psychophysics72, 706–20 (04 2010)

2010
[12]

Ecological Psy- chology 19, 383 – 410 (2007)

Fajen, B.R.: Affordance-based control of visually guided action. Ecological Psy- chology 19, 383 – 410 (2007)

2007
[13]

FIFA: Laws of the game 2022/2023 (2022), https://www.theifab.com/laws-of-the- game-documents/

2022
[14]

International Journal of Perfor- mance Analysis in Sport21, 139 – 155 (2020)

Horn, M., de Waal, S.J., Kraak, W.: In-match penalty kick analysis of the 2009/10 to 2018/19 english premier league competition. International Journal of Perfor- mance Analysis in Sport21, 139 – 155 (2020)

2020
[15]

Scandinavian Journal of Medicine & Science in Sports28 (08 2018)

Hunter,A.,Angilletta,M.,Wilson,R.:Behaviorsofshooterandgoalkeeperinteract to determine the outcome of soccer penalties. Scandinavian Journal of Medicine & Science in Sports28 (08 2018)

2018
[16]

Journal of Biomechanics141, 111208 (06 2022) 16 L

Hunter, A., Pavlic, T., Angilletta, M., Wilson, R.: Identifying the best strategy for soccer penalty success: A predictive model for optimising behavioural and biome- chanical trade-offs. Journal of Biomechanics141, 111208 (06 2022) 16 L. Bransen et al

2022
[17]

Journal of Sports Sciences 37, 1–9 (07 2018)

Ibrahim, R., Kingma, I., de Boode, V., Faber, G., Van Dieen, J.: Kinematic and ki- netic analysis of the goalkeeper’s diving save in football. Journal of Sports Sciences 37, 1–9 (07 2018)

2018
[18]

Journal of Applied Sport Psychology24, 73–91 (01 2012)

Jordet, G., Elferink-Gemser, M.: Stress, coping, and emotions on the world stage: The experience of participating in a major soccer tournament penalty shootout. Journal of Applied Sport Psychology24, 73–91 (01 2012)

2012
[19]

Science and foot- ball pp

Kuhn, W.: Penalty-kick strategies for shooters and goalkeepers. Science and foot- ball pp. 489–492 (1988)

1988
[20]

Journal of sports sciences 16 3, 211–34 (1998)

Lees, A., Nolan, L.: The biomechanics of soccer: a review. Journal of sports sciences 16 3, 211–34 (1998)

1998
[21]

Psychological Science18, 222–3 (04 2007)

Masters, R., Kamp, J., Jackson, R.: Imperceptibly off-center goalkeepers influence penalty-kick direction in soccer. Psychological Science18, 222–3 (04 2007)

2007
[22]

Princeton University Press (1947)

von Neumann, J., Morgenstern, O.: Theory of Games and Economic Behavior. Princeton University Press (1947)

1947
[23]

Journal of Sports Sciences (06 2014)

Noël, B., Furley, P., Kamp, J., Dicks, M., Memmert, D., Francis, T.: The devel- opment of a method for identifying penalty kick strategies in association football. Journal of Sports Sciences (06 2014)

2014
[24]

Frontiers in Psychology12 (2021)

Noël, B., van der Kamp, J., Klatt, S.: The interplay of goalkeepers and penalty takers affects their chances of success. Frontiers in Psychology12 (2021)

2021
[25]

PLOS ONE10 (08 2015)

Noël, B., van der Kamp, J., Memmert, D.: Implicit goalkeeper influences on goal side selection in representative penalty kicking tasks. PLOS ONE10 (08 2015)

2015
[26]

Review of Economic Studies70, 395–415 (02 2003)

Palacios-Huerta, I.: Professionals play minimax. Review of Economic Studies70, 395–415 (02 2003)

2003
[27]

Ergonomics48, 1686–97 (2005)

Savelsbergh, G., van der Kamp, J., Williams, A., Ward, P.: Anticipation and visual search behaviour in expert soccer goalkeepers. Ergonomics48, 1686–97 (2005)

2005
[28]

Journal of Sports Sciences41(7), 668–676 (2023)

Secco Faquin, B., Teixeira, L.A., Coelho Candido, C.R., Boari Coelho, D., Bayeux Dascal, J., Alves Okazaki, V.H.: Prediction of ball direction in soccer penalty through kinematic analysis of the kicker. Journal of Sports Sciences41(7), 668–676 (2023)

2023
[29]

Tuyls, K.e.a.: Game plan: What AI can do for football, and what football can do for AI. J. Artif. Int. Res.71, 41–88 (sep 2021)

2021
[30]

Journal of Sports Sciences24(05), 467–477 (2006)

Van Der Kamp, J.: A field simulation study of the effectiveness of penalty kick strategies in soccer: late alterations of kick direction increase errors and reduce accuracy. Journal of Sports Sciences24(05), 467–477 (2006)

2006
[31]

Journal of Cognitive Psychology24 (12 2012)

Weigelt, M., Memmert, D., Schack, T.: Kick it like Ballack: The effects of goal- keeping gestures on goal-side selection in experienced soccer players and soccer novices. Journal of Cognitive Psychology24 (12 2012)

2012
[32]

Wood, G., Jordet, G., Wilson, M.R.: On winning the “lottery”: Psychological prepa- rationforfootballpenaltyshoot-outs.JournalofSportsSciences 33(17),1758–1765 (2015)

2015
[33]

Zheng, R., Reus, C., Kamp, J.: Goalkeeping in the soccer penalty kick: The dive is coordinated to the kicker’s non-kicking leg placement, irrespective of time con- straints. Human Movement Science76, 102763 (04 2021) Stop Guessing: Optimizing Goalkeeper Policies for Soccer Pen...

2021
[34]

kicker strategy (independent versus dependent kick)
[35]

goalkeeper dive direction (left, center, or right)
[36]

However, Sections 2 and 3 contain useful aggregate infor- mation about kicker strategy and dive timing

goalkeeper timing (early versus late dive) Annotating this requires trained analysts who can identify the differences be- tween the strategies. However, Sections 2 and 3 contain useful aggregate infor- mation about kicker strategy and dive timing

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

International Journal of Performance Analysis in Sport16, 508–522 (08 2016)

Almeida, C., Volossovitch, A., Duarte, R.: Penalty kick outcomes in UEFA club competitions (2010-2015): The roles of situational, individual and performance factors. International Journal of Performance Analysis in Sport16, 508–522 (08 2016)

work page 2010

[2] [2]

American Economic Review 100(5), 2548–64 (December 2010)

Apesteguia, J., Palacios-Huerta, I.: Psychological pressure in competitive envi- ronments: Evidence from a randomized natural experiment. American Economic Review 100(5), 2548–64 (December 2010)

work page 2010

[3] [3]

Azar, O.H., Bar-Eli, M.: Do soccer players play the mixed-strategy Nash equilib- rium? Applied Economics43(25), 3591–3601 (2011)

work page 2011

[4] [4]

Soccer and Society10 (03 2009)

Bar-Eli, M., Azar, O.: Penalty kicks in soccer: An empirical analysis of shooting strategies and goalkeepers’ preferences. Soccer and Society10 (03 2009)

work page 2009

[5] [5]

Journal of Economic Psychology 28, 606–621 (10 2007)

Bar-Eli, M., Azar, O., Ritov, I., Keidar-Levin, Y., Schein, G.: Action bias among elite soccer goalkeepers: The case of penalty kicks. Journal of Economic Psychology 28, 606–621 (10 2007)

work page 2007

[6] [6]

In: Proceedings of the Sports Tomorrow Congress 2021 (2021)

Bransen, L., Davis, J.: Predicting football penalty directions using in-match per- formance indicators. In: Proceedings of the Sports Tomorrow Congress 2021 (2021)

work page 2021

[7] [7]

International Journal of Performance Analysis in Sport22, 571 – 582 (2022)

Buscà, B., Hileno, R., Nadal, B., Serna, J.: Prediction of the penalty kick direction in men’s soccer. International Journal of Performance Analysis in Sport22, 571 – 582 (2022)

work page 2022

[8] [8]

In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

Chen, T., Guestrin, C.: XGBoost: A scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. pp. 785–794 (2016)

work page 2016