A Stochastic Analysis of Table Tennis

Yves Dominicy, Christophe Ley, Yvik Swan

Abstract: We establish a general formula for the distribution of the score in table tennis. We use this formula to derive the probability distribution (and hence the expectation and variance) of the number of rallies necessary to achieve any given score. We use these findings to investigate the dependence of these quantities on the different parameters involved (number of points needed to win a set, number of consecutive serves, etc.), with particular focus on the rule change imposed in 2001 by the International Table Tennis Federation (ITTF). Finally we briefly indicate how our results can lead to more efficient estimation techniques of individual players' abilities.

Introduction

In 2001, the International Table Tennis Federation (ITTF) implemented a major rule change, shifting sets from 21 points to 11 points and reducing the service rotation from five serves to two. While sports like tennis have extensive statistical literature, table tennis has been less studied due to its unique deterministic service rule. This paper establishes a generalized mathematical framework to calculate the exact probability distributions of scores and match durations, providing a rigorous tool for evaluating rule configurations.

Modeling the Game

The researchers developed a general $(m, n, G)$-scoring model, where $m$ is the number of serves per turn, $n$ is the points needed to win a set, and $G$ is the sets needed to win a match. They employed two stochastic models: a "No-Server Model" with constant rally-win probability, and a more realistic "Server Model" where win probabilities $p_a$ and $p_b$ depend on who is serving.

Using Euclidean division to account for the deterministic serve rotation, the authors decomposed the game into independent binomial experiments. This allowed for the derivation of general formulas for score distributions and the number of rallies required to conclude a set or match.

Results and Impact of Rule Changes

The analysis reveals that the number of consecutive serves ($m$) has negligible impact on the overall winning probability. However, the shift from 21 to 11 points significantly altered the game's dynamics. The 11-point system is more "balanced," slightly increasing the underdog's chances by reducing the sample size of rallies where the law of large numbers would favor the stronger player.

Expected number of rallies (a) and variance (b) in a 21-point set as a function of player strengths.

As shown in Figure 3, the expected duration and variance of a set are highly sensitive to player skill. For equally matched players ($p_a = p_b = 0.5$), a 21-point set averages approximately 37 rallies. The expected duration scales linearly with $n$; thus, the transition to 11 points effectively halved the time spent per set.

Predictability and Match Duration

A primary motivation for the ITTF's rule change was to make match durations more predictable for broadcasters. The study confirms that the standard deviation of match length is generally lower in the 11-point system, particularly when the number of sets is adjusted to maintain competitive balance.

Ratios of expected duration (left) and standard deviation (right) between the new (11-point) and old (21-point) systems.

Figure 6 illustrates the ratios of expectation and standard deviation between the two systems. While the expected duration can be kept similar by increasing the number of sets ($G$), the relative standard deviation often decreases, particularly for players of unequal strength, enhancing predictability.

Conclusion and Future Directions

The current model assumes independent and identically distributed (i.i.d.) rallies, omitting psychological factors like momentum. Future work aims to utilize these formulas for Maximum Likelihood Estimation (MLE) of player abilities and to develop dynamic win-probability models for real-time match analysis.

Summary updated

Abstract

We establish a general formula for the distribution of the score in table tennis. We use this formula to derive the probability distribution (and hence the expectation and variance) of the number of rallies necessary to achieve any given score. We use these findings to investigate the dependence of these quantities on the different parameters involved (number of points needed to win a set, number of consecutive serves, etc.), with particular focus on the rule change imposed in 2001 by the International Table Tennis Federation (ITTF). Finally we briefly indicate how our results can lead to more efficient estimation techniques of individual players' abilities.
BibTeX
@inproceedings{dominicy2013stochastic,
  author = {Yves Dominicy and Christophe Ley and Yvik Swan},
  title = {{A Stochastic Analysis of Table Tennis}},
  booktitle = {Brazilian},
  year = {2013},
  doi = {10.1214/11-bjps177},
}
Figure