1 Introduction
One of the most popular problems when it comes to select an alternative from a set of available options is probably that of selecting a winner in a sports competition. In this paper we present an axiomatic approach to the fairness aspect of such selection problems.
Any precise definition of a competition system entails answering the following basic questions (cf. Haigh 2009): Who the players in a match are, at what stage of the competition this match takes place, and how the final winner is decided. In general, competition rules may be set with many different objectives in mind, such as match intensity, attracting the interest of the spectators (cf. Dagaev and Suzdaltsev 2018), minimizing organizational costs, and so on. Fairness is, or should be, certainly one the main goals of any competition designer. Moreover, as outlined in Bartsch et al. (2006), in real world-sport leagues, fairness issues are quite important.
Anzeige
In a league competition, every player participates in a given number of matches against other players. A certain number of points is assigned to the winner of each match, and the final winner is the player with the most points.
The most widely used leagues are “single round-robin tournaments”, where each player plays against every other player once, and “double round-robin tournaments”, where each player plays against every other player twice (cf. Rasmussen and Trick 2008). Regular seasons in most sports have a league structure. However, there are well-known examples of leagues that are not round-robin tournaments, e.g. the US National Football League (NFL), where each team plays 16 matches in the regular season but does not play the same number of times against every opponent and does not play at all against some teams. In the US National Basketball Association (NBA) each team plays 82 matches in the regular season and plays against every other team, but does not play the same number of matches against every opponent.
Leagues where different players play different number of matches are more difficult to find. However we believe that this possibility should not be disregarded, at least at the theoretical stage. The same principle that supports giving byes in knockout tournaments (favoring more meritorious participants) could be applied to leagues, so that certain players gain an advantage by playing more matches. It should be noted that such an advantage is given if points are assigned only for winning matches, though there are exceptions to this rule. For instance, a bye can be awarded the same number of points as for winning a match in Swiss-system tournaments with an odd number of players.1
We nevertheless include the mentioned theoretical possibility of gaining by playing more matches in our formal model, if only to investigate its implications. Recent proposals to change standard round-robin domestic leagues in football also support the pertinence of such a more general analysis (cf. Feehely 2021). We refer the reader to Lasek and Gagolewski (2018) for an overview of tournament formats used in the majority of European top-tier association football competitions as well as to Csató (2020) for the relationship between the existence of ground-robin groups of different strength in tournaments and the quality of matches in the corresponding competitions.
Anzeige
Many of the discussions about fairness in competition systems are based on rather informal arguments. In this paper we offer a formal setup within which such discussions can be framed and precisely define two fairness principles that are commonly pursued in real practice. The first principle requires a competition system not to favor weaker players (we refer to this as “monotonicity in strength”), and the second requires that all players should have the same chance of being the final winner if all players are equally strong (“equal treatment”). Our main purpose is thus to study the extent to which league-type competition systems can be considered fair according to these principles.
Literature overview
There are several papers on modeling and studying fairness in league-type competition systems. One possibility is to focus on the influence of scheduling in sequential round-robin tournaments (cf. Durán et al. 2017, Krumer and Lechner 2017, Sahm 2019), including carry-over effects (cf. Russell 1980, Lambrechts et al. 2018) or on the requirement that a team should not play against extremely weak or extremely strong teams in consecutive periods (cf. Briskorn 2009, Briskorn and Knust 2010, Nemhauser and Trick 1998). Moon and Pullman (1970) concentrate on “equalizing” handicapping methods. Rubinstein (1980) shows axiomatically that the points system used in round-robin competitions is the only one that satisfies three axioms inspired by social choice theory. Levin and Nalebuff (1995) draw an analogy between round-robin points systems and voting systems.
Fairness in sports has also been analyzed with respect to tie-breaking mechanisms within the field of economics (cf. Apesteguía and Palacios-Huerta 2010and Che and Hendershott 2008). Most studies that compare competition systems apply simulation techniques to check that particular properties are satisfied (cf. Appleton 1995, McGarry and Schutz 1997, Scarf et al. 2009, Ryvkin and Ortmann 2008 and Ryvkin 2010). There is also a sizeable number of papers in operations research related to sports that focus on topics other than fairness and thus lie outside the scope of this study. Readers are referred to Csató (2021), Kendall and Lenten (2017), Kendall et al. (2010), Lenten and Kendall (2021),and Wright (2014) for surveys of this literature.
Our contribution
We start our analysis by formulating two basic fairness axioms. The first is an “equal treatment” requirement which states that “if all players are equally strong then all should have the same probability of being the final winner”. The second is a “monotonicity in strength” condition requiring that “a weaker player should not have a higher probability of being the final winner than a stronger player”.
Each of these axioms is presented in a strong form and in a weak form. The strong versions impose that a league-type competition should fulfill the property for every possible assignment of the players, while the weak forms only require that the property be fulfilled for at least one such assignment. The effect of these weak versions of the axioms was also studied in Arlegi and Dimitrov (2020) and Arlegi (2022). However, the analysis in these works is set within the framework of elimination-type competitions (knockout tournaments) which is completely different from the model of league-type competitions we develop and study in the current paper. In particular, the corresponding characterizations for the case of elimination-type competitions are heavily based on both the tree structure of the knockout tournaments and on the specific way in which the sequential structure of the competition determines players’ winning probabilities.
In contrast, we study in the current paper the other main class of competitions (league-type competitions) and focus exclusively on the effect of the assignment of the players, i.e. on who is playing against whom and how many times, seeking to isolate this effect from others related to scheduling when the competition is structured sequentially. To that end, we investigate what type of league-type competitions satisfies the said fairness properties. Generally speaking, equal treatment leads to competitions where every player plays the same number of matches (Theorems 1 and 2), while monotonicity in strength drastically restricts the number of participants to two (Theorem 3). When weak monotonicity in strength is considered, the class of competitions that fulfills the property increases (Theorem 4).
2 Framework
Let N be a finite set of at least two players who compete with each other by playing a total number s of matches. An assignment \({\textbf{A}}\) is a symmetric matrix \({\textbf{A}}=\left( a_{ij}\right) _{i,j\in N}\) with \(a_{ij}\) being the number of matches played between players i and j. Thus, \( {\textbf{A}}\) determines the participants in each of the s matches, with \( (\Sigma _{i,j\in N}a_{ij})/2=s\). For example, for \(N=\{1,2,3,4\}\), the assignment \({\textbf{A}}\) defined by \(a_{ii}=0\) and \(a_{ij}=1\) for all \(i,j\in N\), \(i\ne j\), describes a single round-robin tournament between four players with the set of \(s=6\) matches being played \( \{(1,2),(1,3),(1,4),(2,3),(2,4),(3,4)\}\). As shown below, our definition of a league allows (and the corresponding results account) for the possibility of players playing different numbers of matches.
The rest of this section describes the main ingredients of our model. These include the ordinal information about each player’s relative strengths represented by a binary relation R over the player set N and the set of “winning probability matrices” that are consistent with R (Subsection 2.1), plus the notion of an assignment and the probability of each player being a final winner of a league competition (Subsection 2.2). Subsection 2.3 presents the fairness axioms, making use of all the previous formal elements.
2.1 Players’ strength and winning probabilities
We follow Arlegi and Dimitrov (2020) and assume that the members of the player set N are completely ordered according to a binary relation R of strength so that, for all \(i,j\in N\), iRj is interpreted as “player i is at least as strong as player j”. The corresponding asymmetric and symmetric factors of R are denoted by P and I respectively, so iPj reads “i is strictly stronger than j ” and iIj reads “i and j are equally strong”.2
Now assume that information is available about the probabilities of one player defeating another in a match. For \(i,j\in N\), denote by \(p_{ij}\) the corresponding probability and note the following natural connection between R and these probabilities: A player i is at least as strong as another player j if and only if \(p_{ij}\ge 0.5\). Since the binary relation of strength is complete, iPj means \(p_{ij}>0.5\) and iIj implies \(p_{ij}=0.5\). For convenience, in what follows we order the players in N according to R; that is, if iPj then \(i<j\) (if iIj then either \(i<j\) or \(j<i\)).
We denote by \({\mathcal {P}}_{R}\) the set of all probability matrices such that, for \({\textbf{P}}\in {\mathcal {P}}_{R}\), \(p_{ij}\ge 0.5\) if and only if iRj. Moreover, we assume that each probability matrix in \({\mathcal {P}}_{R}\) satisfies the following two conditions:These conditions are commonly used in the literature (cf. David 1963, Hwang 1982, Horen and Riezman 1985, Schwenk 2000) and taken together they are equivalent to the “strong stochastic transitivity” of the corresponding probability matrix (cf. David 1963). The crucial condition here is Condition (2 ), which states that any player is more likely to beat a weaker player than a stronger one. It is worth mentioning that, for a fixed binary relation of strength R, our fairness properties are imposed to hold for each \({\textbf{P}}\in {\mathcal {P}}_{R}\), so that no further details on these probability matrices are needed for our results.
$$\begin{aligned}{} & {} \forall i,j\in N,\text { }p_{ij}+p_{ji}=1. \end{aligned}$$
(1)
$$\begin{aligned}{} & {} \forall i,j\in N,\text { }p_{ij}\ge 0.5\text { implies }p_{ik}\ge p_{jk} \text { for each }k\in N\setminus \left\{ i,j\right\} . \end{aligned}$$
(2)
Finally, it is easy to see that the strong stochastic transitivity of the probability matrices in \({\mathcal {P}}_{R}\) implies that the binary relation of strength R is transitive. Moreover, for \(i,j,k,\ell \in N\) the following useful interval property connecting the strength of these players to their corresponding winning probabilities emerges:
$$\begin{aligned} iRjRkR\ell \text { implies }p_{i\ell }\ge p_{jk}. \end{aligned}$$
(3)
2.2 Assignments and final winners
Given a league competition \(\left( s,N\right) \), an assignment of size \(s\in {\mathbb {Z}} ^{+}\) is a symmetric matrix \({\textbf{A}}=\left( a_{ij}\right) _{i,j\in N}\) with \(a_{ij}\in {\mathbb {Z}} ^{+}\cup \left\{ 0\right\} \) for all \(i,j\in N\) being the number of matches played between players i and j such that:The set of all assignments for N of size s satisfying the above conditions is denoted by \({\mathcal {A}}^{(s,N)}\). We assume that player \(i\in N \) wins the competition \(\left( s,N\right) \) at assignment \({\textbf{A}}\in {\mathcal {A}}^{(s,N)}\) if no other player wins more matches than i from the matches induced by \({\textbf{A}}\). Given a probability matrix \({\textbf{P}}\) containing the probabilities of a player winning a match, we denote by \( \varphi _{i}({\textbf{A}},{\textbf{P}})\) the probability that player \(i\in N\) wins the competition. The exact mathematical expression of \(\varphi _{i}( {\textbf{A}},{\textbf{P}})\) is provided in the Appendix and it should be noted that it does not use tie-breaking criteria (cf. Berker 2014 and Csató 2021), i.e., multiple final winners are allowed.
-
\(a_{ii}=0\) for each \(i\in N\) (no player plays against herself);
-
\(\Sigma _{j\in N}a_{ij}>0\) for each \(i\in N\) (each player plays at least one match);
-
\(2\le \left| N\right| \le \Sigma _{i,j\in N}a_{ij}=2s\) (the total number of matches is s).
2.3 Fairness axioms
We now formally introduce the two fairness principles (cf. Arlegi and Dimitrov 2020). Each of these ideas is presented in a strong form and in a weak form. To state them, we assume that a binary relation R of strength is defined on the player set N.
Equal Treatment (ET) A league competition (s, N) satisfies ET if for all \({\textbf{A}}\in {\mathcal {A}}^{(s,N)}\) we have \(\varphi _{i}({\textbf{A}}, {\textbf{P}})=\varphi _{j}({\textbf{A}},{\textbf{P}})\) holding for all \(i,j\in N\) whenever \({\textbf{P}}\in {\mathcal {P}}_{R}\) is such that \(p_{ij}=0.5\) for all \( i,j\in N\).
Weak Equal Treatment (WET) A league competition (s, N) satisfies WET if there exists \({\textbf{A}}\in {\mathcal {A}}^{(s,N)}\) with \(\varphi _{i}( {\textbf{A}},{\textbf{P}})=\varphi _{j}({\textbf{A}},{\textbf{P}})\) holding for all \( i,j\in N\) whenever \({\textbf{P}}\in {\mathcal {P}}_{R}\) is such that \(p_{ij}=0.5\) for all \(i,j\in N\).
Monotonicity in Strength (MS) A league competition (s, N) satisfies MS if for all \({\textbf{A}}\in {\mathcal {A}}^{(s,N)}\), for all \(i,j\in N\), and for all \({\textbf{P}}\in {\mathcal {P}}_{R}\) such that \(p_{ij}>0.5\), \( \varphi _{i}({\textbf{A}},{\textbf{P}})\ge \varphi _{j}({\textbf{A}},{\textbf{P}})\) holds.
Weak Monotonicity in Strength (WMS) A league competition (s, N) satisfies WMS if there exists \({\textbf{A}}\in {\mathcal {A}}^{(s,N)}\) such that, for all \(i,j\in N\) and for all \({\textbf{P}}\in {\mathcal {P}}_{R}\) such that \( p_{ij}>0.5\), \(\varphi _{i}({\textbf{A}},{\textbf{P}})\ge \varphi _{j}({\textbf{A}},{\textbf{P}})\) holds.
ET and WET express the idea that in regard to the final probability of winning, the competition system should not be biased towards any particular player if all of them are equally skilled.
MS and WMS require the competition system not to reward weaker players under any of the possible probability matrices compatible with the strength of the players. In fact, many tournaments are precisely designed to minimize the role of luck in winning the competition: For example, round-robin tournaments and even double round-robin tournaments minimize that effect by staging a large number of matches, and in knockout tournaments the strongest player is often matched with the weakest, the second strongest with the second weakest, and so on.
Apart from the fact that WET is logically weaker than ET and WMS is logically weaker than MS, the normative power of the weaker versions compared to the strong versions may depend on the specific application intended, and in particular on the conjectures as to the benevolence of the competition designer. On the one hand, ET and MS prevent manipulation by a potentially corrupted competition designer because they ensure that there exists no assignment rule benefiting a particular player in relation to another one who is more or equally skilled. On the other hand, WET and WMS rely on confidence in the benevolence of the competition designer, in the sense that the focus is on competition systems where it is always possible to find an assignment rule that is fair, independently of the values in the probability matrices supporting the strength relation.
In formulating the monotonicity properties we were guided by the general view that the information on the exact values of the probabilities of winning might be rather imprecise. A consistent way of incorporating such a view into the corresponding requirements is to require them to hold for all probability matrices. Clearly, MS and WMS would then guarantee fairness for all possible probability values which are compatible with the binary relation of strength R. This is also important from a practical perspective as the rules used for assignment in most professional sports playoffs are usually proxies of R obtained on the basis of end-of-season standings. Hence, it is important to know whether the corresponding competition structure is fair independently of the precise numerical scores that describe the strength of the teams at the end of the regular season.
3 Results
3.1 Leagues and equal treatment
Whether the two equal treatment axioms (ET and WET) are fulfilled by league competitions is closely related to the fact that players should play the same number of matches. The first result in this section shows that a league satisfies ET if and only if either each player plays a single match or only two players play all matches. Clearly, both cases imply that there must be an even number of players participating in the competition.
Theorem 1
A league competition (s, N) satisfies ET if and only if either \(2s=\left| N\right| \) or \( 2s>\left| N\right| =2\)
It is sometimes suggested that leagues are fair competition systems. However, this is not true in our framework because our definition of league is much broader than the sense in which the term is popularly used, which is usually identified with round-robin tournaments. In fact, the theorem given above excludes leagues as fair structures unless every player plays exactly one match or there are only two players. We now show that weakening ET to WET extends the class of fair league competitions by including those competitions where each player plays the same number of matches against every other player.
Theorem 2
Any league competition (s, N) with either \(2\,s=\left| N\right| \) or \(2\,s=k\left( \left| N\right| -1\right) \left| N\right| \) for some integer \( k\ge 1\) satisfies WET.
It should be noted that Theorem 2 is not superfluous in the sense that not every league competition satisfies WET. For example, it can easily be checked that there is no way to assign three players to a two-match competition so that WET is fulfilled.
3.2 Leagues and monotonicity in strength
Our next result shows that MS restricts the number of players in a competition to only two.
Theorem 3
A league competition
\(\left( s,N\right) \)
satisfies MS if and only if
\(\left| N\right| =2\)
When MS is weakened to WMS, the class of fair competition systems is considerably enlarged because any league competition turns out to satisfy this axiom provided that either each player participates in exactly one match or the total number of matches is at least \(\left( \left| N\right| -1\right) \). In particular, this implies that any league competition covered by Theorem 2 satisfies the weak versions of both the equal treatment and monotonicity in strength properties.
Theorem 4
Any league competition (s, N) with \( 2\,s=\left| N\right| \) or \(s\ge \left| N\right| -1\) satisfies WMS.
It is not difficult to prove that the assignment used in the proof of Theorem 4 (see the Appendix) when \(2s=\left| N\right| \) is the only one for which (s, N) satisfies WMS. However, there are assignments other than those used when \(s\ge \left| N\right| -1\) holds for which (s, N) satisfies WMS.
3.3 Round-robin tournaments
Let \(k\ge 1\) be an integer. A k-round-robin tournament is a league competition \(\left( s,N\right) \) together with an assignment \({\textbf{A}}\in {\mathcal {A}}^{\left( s,N\right) }\) satisfying the additional condition \(a_{ij}=k\) for all \(i,j\in N\), \(i\ne j\). That is, according to \( {\textbf{A}}\), each player in N plays exactly k matches against every other player in N. Clearly then, a single round-robin tournament requires \( k=1\), while a double round-robin competition implies \(k=2\). Notice finally that, by \({\textbf{A}}\in {\mathcal {A}}^{\left( s,N\right) }\), \(2\,s=k\left( \left| N\right| -1\right) \left| N\right| \) follows.
Given the way we introduced our axioms and the fact that a round-robin tournament is defined with respect to a specific assignment, it makes sense to check whether such a competition system satisfies WET and WMS only i.e. the relevant results are to be derived from Theorem 2 and Theorem 4.
It follows directly from Theorem 2 that the first condition in the above definition of a k-round-robin tournament is satisfied. In the proof of Theorem 2, we show further that a league competition \(\left( s,N\right) \) with \(2\,s=k\left( \left| N\right| -1\right) \) satisfies WET with respect to \({\textbf{A}}\in {\mathcal {A}}^{(s,N)}\) letting each player participate in exactly k number of matches against every other player. Thus, we conclude that every k-round-robin tournament satisfies WET.
As for WMS, Theorem 4 first tells us that any single round-robin tournament with two players playing exactly one match fulfills WMS. This follows from \( 2s=\left| N\right| \) which, in order to be satisfied for a k -round-robin tournament (with \(2\,s=k\left( \left| N\right| -1\right) \left| N\right| \)), requires \(k=1\) and \(\left| N\right| =2\). Further recall that, according to Theorem 4, any league competition where the total number of matches is at least \(\left( \left| N\right| -1\right) \) also satisfies WMS. Since the latter condition is necessary for a league competition to be a round-robin tournament, one might be tempted to conclude that round-robin tournaments also satisfy WMS. It should be noted though, that the assignment used here to prove the fulfillment of WMS in this case does not match the one in the definition of a round-robin tournament. It is highly likely that league competitions \(\left( s,N\right) \) with \(2\,s=k\left( \left| N\right| -1\right) \left| N\right| \) for some integer \(k\ge 1\) do generally satisfy WMS with respect to the round-robin tournament assignment as defined above. However, providing a formal proof is much more complex than what it may appear at first sight.3
4 Conclusion
The results reported in this paper enable different league-type competition systems to be compared in terms of their fairness. In particular, we have shown that there are not many league-type competitions that are fair in the sense of simultaneously satisfying both types of fairness in their corresponding strong or weak forms. MS combined with either ET or WET produces a degenerate competition consisting of a two-player league (Theorem 3). Weakening MS to WMS and imposing it together with ET slightly enlarges the class of admissible leagues to those where each participant plays a single match. In contrast, the combination of WMS and WET results in an expansion of the class of admissible leagues to include leagues that allow players to play the same number of times in at least \(\left| N\right| -1\) matches (Theorems 2 and 4). An interesting direction for future research in that respect could be the quantification of the extent to which these properties are violated in certain competition formats. A first attempt in that general research direction was made in Csató (2022), where simulation techniques are used for an investigation of monotonicity and incentive compatibility properties of sports tournaments.
Theorem 2 proves that k-round-robin tournaments (which include single and double round-robin tournaments as special cases) satisfy WET. It seems quite likely that also other league-type competitions do satisfy WET; for instance, one could consider the assignment recently adopted by the UEFA Champions League from the 2024/25 season (cf. UEFA 2022), where each player plays the same number of matches against the same number of other players (not necessarily against every other player).4 Most likely the mentioned assignments also satisfy WMS, although the formal proof remains an open question even if one restricts probabilities to take two possible values; that is, if \(p_{ij}=\alpha \in \left[ 0.5,1\right] \) for all \(i,j\in N\) with \(i<j\).
Our analysis in this paper does not cover Swiss-system tournaments as the assignment in the later class of tournaments, in contrast to our definition of league-type competitions, is not fixed a priori but it rather depends on the results of the matches already played. The possibility for such assignment path-dependence leads to very interesting operational research problems as how to pair the players (cf. Biró et al. 2017, Führlich et al. 2021, Kujansuu et al. 1999, Ólafsson 1990) or how to generate a players’ ranking at the end of the tournament (cf. Csató (2013, 2017)).
Finally, it is worth mentioning that the strong stochastic transitivity of the probability matrices plays a crucial role in the results obtained. Any potential weakening of this condition would clearly enlarge the set of matrices that satisfy it and thus leads to a narrower class of fair competition systems.
Declarations
Conflict of interest
The authors have no relevant financial or non-financial interests to disclose.
Open AccessThis article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.