Journal of Sports Economics http://jse.sagepub.com Competitive Balance and Match Uncertainty in Grand-Slam Tennis: Effects of Seeding System, Gender, and Court Surface Julio del Corral Journal of Sports Economics 2009; 10; 563 originally published online May 6, 2009; DOI: 10.1177/1527002509334650 The online version of this article can be found at: http://jse.sagepub.com/cgi/content/abstract/10/6/563 Published by: http://www.sagepublications.com On behalf of: The North American Association of Sports Economists Additional services and information for Journal of Sports Economics can be found at: Email Alerts: http://jse.sagepub.com/cgi/alerts Subscriptions: http://jse.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav Citations http://jse.sagepub.com/cgi/content/refs/10/6/563 Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 Articles Competitive Balance and Match Uncertainty in Grand-Slam Tennis Journal of Sports Economics Volume 10 Number 6 December 2009 563-581 # 2009 SAGE Publications 10.1177/1527002509334650 http://jse.sagepub.com hosted at http://online.sagepub.com Effects of Seeding System, Gender, and Court Surface Julio del Corral Fundación Observatorio Económico del Deporte University of Oviedo This article tests whether the increase in seeded players in tennis grand-slam tournaments from 16 to 32 in 2001 led to decreased competitive balance. In doing so, two alternative measures of competitive balance for single-elimination tournaments are proposed based on the performance of seeded players. Likewise, the differences in competitive balance due to gender and court surface are also analyzed. Additionally, using data from tennis grand-slam matches from 2005 to 2008, this article analyzed the determinants of ex post match level uncertainty with probit models. Keywords: competitive balance; seeding system; match uncertainty; tennis Introduction Competitive balance in sports has attracted a lot of research in the past. The considerable interest from sports economists on this issue is driven by the fact that participants need each other to produce at all requires a basic interdependence. In other words, sports firms have to avoid monopoly (Neale, 1964). This is in sharp contrast to other types of business firms, for which the monopoly is the preferred market Author’s Note: Correspondence concerning this article should be addressed to Julio del Corral, Department of Economics/Fundación Observatorio Económico del Deporte, University of Oviedo, Avda del Cristo s/n, 33006 Oviedo, Spain; e-mail: [email protected]. ‘‘I have received helpful comments from Paloma Corrales, Kiran Gajwani, Juan Prieto, Daniel Solı́s, graduate student Cornell seminar participants, X IASE Meeting placed in Gijón participants and two anonymous referees. The author would also like to thank Julio del Corral Sr. and Pelayo González for their research assistance. Lastly, the author acknowledges the Spanish Ministry of Science and Technology for his FPU fellowship. Any remaining errors are the sole responsibility of the author’’. 563 Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 564 Journal of Sports Economics structure. Neale explained this peculiar fact of sport economics with the well-known Louis-Schmelling paradox, in which the heavyweight champion of the world needed a strong contender to maximize profits. Hence, several studies analyzing the demand for sports have focused on the uncertainty of outcome hypothesis.1 Szymanski (2003) stated that while the evidence is far from unambiguous, there is some support for the belief that attendance is positively affected by competitive balance. For this reason, there are several studies that have analyzed what happened to competitive balance over time (e.g., Garcı́a & Rodrı́guez, 2007; Horowitz, 1997; Koning, 2000). Manuscripts that analyze changes in competitive balance as a result of changes in business practice are more interesting than those that merely analyze competitive balance over time (Fort & Maxcy, 2003). For instance, Balfour and Porter (1991) tested the effect of free agency in the National Football League (NFL) and in Major League Baseball (MLB). Vrooman (1995) analyzed whether the salary cap and free agency policies altered competitive balance in MLB, the NFL, and the National Basketball Association (NBA). Schmidt (2001) tested the effects of expansion in the MLB. Haugen (2008) tested whether the three-point victory rule changed competitive balance in some European soccer leagues. La Croix and Kawaura (1999) investigated how the change from an open-bidding regime to a player draft regime affected competitive balance in the professional Japanese baseball leagues. All the above-mentioned articles analyzed the effect of some policy on competitive balance in team sports. However, as Neale pointed out in his Louis-Schmelling paradox, competitive balance is also important in individual sports. Likewise, organizers of individual sports events are also able to alter competitive balance within a competition. For instance, they can change the scoring system or some rules of the game.2 In addition, they can provide advantages to stronger competitors such as seedings (tennis) or the assignment of lanes (swimming). These practices tend to decrease competitive balance (Sanderson, 2002). This article aims to analyze the impact of a change in an individual sport policy on competitive balance. In particular, in 2001, the tennis grand-slam tournaments (i.e., the Australian Open, the French Open, Wimbledon, and the U.S. Open) changed the number of seeded players from 16 to 32. The expected outcome of this change was to decrease competitive balance. This article tests this hypothesis using data obtained from the grand-slam tournaments from 1994 to 2008. Thus, the sample is split into two periods: 1994-2000 and 2002-2008. In the first sampling period, the number of seeded players was 16, whereas in the second period the number of seeded players was 32. Given the nature of single elimination tournaments, traditional measures of competitive balance, such as the Gini coefficient or the Herfindahl-Hirschman index, cannot be used. Hence, to measure the competitive balance in these tournaments are proposed two alternative measures based on the performance of the seeded players. It is important to note that this article is not the first to analyze competitive balance in individual sports. Previously, Rohm, Chatterjee, and Habibullah (2004) Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 del Corral / Competitive Balance and Match Uncertainty in Grand-Slam Tennis 565 investigated the degree of competition at Wimbledon from 1968 to 2001, using the degree of dominance among the top four seeds. Their results support the idea that competitiveness at Wimbledon has been extremely high. Also using Wimbledon data, Magnus and Klaassen (1999) analyzed the advantage for the seeded players. They found no evidence for differences between men and women with regard to the difficulty of achieving an upset. Du Bois and Heyndels (2007) studied match-specific uncertainty, inter-seasonal uncertainty, and indicators for long-term uncertainty in men’s and women’s tennis. To do so, they analyzed the tightness of several grandslam finals, the coefficient of variation of ranking points for the top 10 players, and the Spearman rank correlation coefficient between two consecutive seasons. They found that the share of new top 10 players was larger in men (Association of Tennis Professionals [ATP] Tour) than in women (Women’s Tennis Association [WTA] Tour). However, they claim that there is no difference in competitive balance between WTA and ATP, using indicators for match-specific and seasonal uncertainty. Boulier and Stekler (1999) found that 53% of male players and 63% of female players were seeded players in the 4th round of tennis grand slams. Therefore, the objective of the article is threefold. First, two alternative measures of competitive balance in single-elimination tournaments, based on the performance of seeded players, are proposed. Second, those measures are used to analyze competitive balance in tennis grand-slams tournaments. Specifically, whether the increase in the number of seeded players from 16 to 32 in 2001 led to a decrease in competitive balance is tested. Likewise, the effects of gender and court surface differences are investigated. Third, the determinants of upset are tested using discrete choice models in grand-slam matches.3 The data consist of more than 4,000 tennis matches from the tennis grand-slam tournaments played in 2005, 2006, 2007, and 2008. The rest of the article is organized as follows. The following section describes the context of professional tennis and the tennis seeding system. This is followed by an analysis of competitive balance in elimination tournaments. Subsequently, the data are presented. Then the results of competitive balance and match uncertainty are shown. Finally, we draw some conclusions from our analysis. Professional Tennis and Tennis Seeding System Professional tennis is split into the ATP (men) and WTA (women) tours. The most important and prestigious tournaments in the tennis circuits are the grand-slam tournaments (i.e., the Australian Open, the French Open, Wimbledon, and the U.S. Open). Competition in those tournaments begins with 128 players in its main draws4 and its prize money and prestige is much higher in grand-slam tournaments than in other tournaments. Hence, the incentive to perform well in those tournaments is so Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 566 Journal of Sports Economics high that it is very unlikely that players treat the event as a practice or underperforming because the player does not want to risk aggravating an injury. A tennis match is comprised of an odd number of sets (three or five). A set consists of a number of games (a sequence of points played with the same player serving), and games, in turn, consist of points (Paserman, 2007). Men play to the best of five sets, while women play to the best of three sets in grand-slam tournaments. Assuming that the probability of winning a set by the favorite player is the same throughout the match,5 probability theory tells us that the larger the number of sets played, the higher the probability that the favorite wins.6 On the contrary, Magnus and Klaassen (1999) argued that it may be that men are more equal in quality than women. Therefore, it will be interesting to test the differences in competitive balance and match uncertainty among gender. As stated in Du Bois and Heyndels (2007), a critical dimension of tennis’ structure is the playing surface because it determines the speed of the game and thus the strategy of the game. The Australian Open and the U.S. Open are played on hard courts, whereas the French Open and Wimbledon are played on clay and grass courts, respectively.7 From a theoretical point of view, underdogs are believed to use riskier strategies than favorites (Dixit, 1987). Moreover, given that, in general, fewer ‘‘winners’’ hits are needed to win a point on quick surfaces than on less quick surfaces, the quicker the court, the more effective risky strategies should be. Thus, we expect that outcome uncertainty will be higher on grass than on hard courts and clay. Public preferences, especially in the final rounds of elimination tournaments, are assumed to be to watch both contests between popular contestants8 and even, rather than uneven, matches. Hence, to protect top contestants from early elimination and ensure that strong will play the strong in later round matches (Sanderson & Siegfried, 2003), organizers of elimination tournaments produce seeding positions based on previous performances.9 Thus, the seeding system can be viewed as a tool for organizers to decrease competitive balance. The seeding system in tennis’ grand slams has evolved over time. The number of seeded players has been 8, 12, 16, and currently 32, since the 2001 French Open. However, in 1924 a seeding system was introduced in which up to four representatives of a nation were drawn in the four different quarters of the draw. In 1927, full seeding was carried out and competitors were selected according to ability, irrespective of nationality. Since 1975, the seeding system has been based on computer rankings.4 Moreover, Wimbledon is the only tournament that has traditionally used the power to ignore these rankings and select seeds using their own criteria. In particular, a committee adjusted seedings subjectively. As a result, top clay court specialist players protested the unfairness of Wimbledon’s seeding system by boycotting Wimbledon in 2000. However, other professionals like Brad Gilbert (Andre Agassi’s coach) claimed to not understand why other grand-slam tournaments did not follow the All England Club’s example with regard to assigning seeds. To end this argument among tennis professionals, in 2001, four grand slams agreed that the number of seeded players would be increased to 32 and the seed order Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 del Corral / Competitive Balance and Match Uncertainty in Grand-Slam Tennis 567 could be determined using a formula that assesses not only actual ranking but also past performance on the event’s playing surface. In particular, Wimbledon developed a new seeding system for men in which the 32 highest ranked players would be selected for seeding. However, the seeds would be changed by constructing a new indicator that summed the total ranking points, plus a 100% increase for points earned for all grass court tournaments in the past 12 months, as well as an additional 75% increase in points earned for the best grass court tournaments in the prior 12 months. However, for women, the seeding order followed the ranking list, except where, in the opinion of the committee, the grass court credentials of a particular player necessitated a change, in the interest of achieving a balanced draw. Still, the other grand-slam tournaments preferred to stick to the ATP/WTA rankings. The ATP welcomed the decision to seed 32 players at the grand-slams tournaments instead of 16 players, but it was opposed to determining the order based on a surface-based system. Given that, by its very nature, a surface-specific system is highly technical and likely to be understood by very few fans, the ATP preferred to produce the seeds with a surface-biased base rather than to use a more complex seeding system. Competitive Balance in Elimination Tournaments Most individual sports events are organized on an elimination tournament basis.11 Given that the tournament design of a single elimination tournament is far from that of a league, measures of a leagues’ competitive balance, such as the dispersion of winning percentage, Herfindahl-Hirschman Indexes, and so on (see Humphreys, 2002) cannot be adopted for individual sports tournaments. Therefore, it is necessary to look for alternative measures. However, very few articles have attempted to measure competitive balance in single elimination tournaments. Boulier and Stekler (1999) proposed using the relative frequency with which 16 seeded players reached the 4th round in tennis grand-slam tournaments. In this article, we propose to extend this idea to all rounds. Thus, a competitive balance measure for each round can be computed as the percentage of the seeded contestants who achieved a round, out of the seeded contestants who should achieve12 such rounds—if seeds were perfect predictors. For instance, if the players who achieve the final round are the first and third seeded, the measure will be 50%. An advantage of this measure is that it is bounded between 0 and 100, so it is very easy to interpret: the closer to 100, the poorer the competitive balance. The main drawback is that it does not account for the quality of the players who ‘‘surprisingly’’ advance to some rounds. For instance, it provides the same measure of competitive balance if the final is played by the first and third seeded players and by the first seeded and an unseeded contestant. Obviously, an effective measure should indicate Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 568 Journal of Sports Economics higher competitive balance in the latter situation. An alternative measure that overcomes such problem is given by P ðn þ 1Þ seedp ij i CBj ¼ P 100; ððn þ 1Þ seedt Þij ð1Þ i where i denotes seeded players, j denotes round, p denotes players who have achieved round j, t denotes players who should achieve round j according to their seed number, seed is the seed position, and n is the total number of seeded players. Therefore, if the number of seeded players is 16, the CB will be 96.77% 16þ14 if the final is achieved by the first and third seeded players and 16þ15 100 51.61% 16 16þ15 100 if the final is achieved by the first seeded and an unseeded player. Data Two kinds of data were gathered. To analyze competitive balance, we collected the round reached by seeded players in tennis grand-slam tournaments from 1994 to 2008. To do so, draws were consulted or the information was obtained from Wikipedia. However, to study match-specific uncertainty, data from the matches in grandslam tournaments since 2005 to 2008 were used. In particular, we collected the results of the matches and whether the player was either qualified from the qualification round or obtained a wild card. Additionally, we collected some individual characteristics such as the ranking at the time of the tournament, the best previous ranking, country, height, weight, and date of birth of the players that were involved in the tournaments. Most data were gathered from the Web sites of the ATP (www.atptennis.com) and WTA (www.sonyericssonwtatour.com), but some data were gathered from the Web site www.tennis-data.co.uk.13 Each draw is composed of 128 players. Thus, 127 matches for both men and women are played in each tournament. In total, we gathered data on 4,064 matches. Tournaments’ Competitive Balance Results In this section, we tested whether the increase in the number of seeded players from 16 to 32 led to decreased competitive balance in tennis grand-slam tournaments. To do so, the performance of the top 16 seeded players is analyzed during two Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 del Corral / Competitive Balance and Match Uncertainty in Grand-Slam Tennis 569 Figure 1 Evolution of the Percentage of the Seeded Players (16) that Achieve the Expected Round Using a 3-Year Moving Average 100 Men 35 35 55 55 % % 75 75 100 Women 1994 1998 2002 Year Semifinal Third 2006 Fourth Second 1994 1998 2002 Year Semifinal Third 2006 Fourth Second different periods: 1994-2000 and 2002-2008. In the first period, there were 16 seeded players, while in the second period there were 32. A way to compare competitive balance between both periods is by calculating the competitive balance measures proposed above, using the performance of the top 16 seeded players. Therefore, to compute the percentage of seeded players who achieved the expected round, only the performance of the top 16 seeded players should be analyzed. Likewise, n in equation (1) is 16 to compute the CB measure. Figures 1 and 2 show the evolution over time of the average across tournaments of the percentage of seeded players who achieved the expected round and the CB measure, in some rounds.14 To avoid that the inherent noise of the data cause some problems, both figures have been made using a 3-year moving average. It is noteworthy to indicate that all points in the figures are averages of 3 years of data using the same number of seeded players. Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 570 Journal of Sports Economics Figure 2 Evolution of the CB Measure Using a 3-Year Moving Average 75 100 Men 55 35 35 55 % % 75 100 Women 1994 1998 2002 Year Semifinal Third 2006 Fourth Second 1994 1998 2002 Year Semifinal Third 2006 Fourth Second From both figures, it seems that there is a structural break point, especially for men, in the year 2001, when the 32-player seeding system was adopted. In particular, both measures take higher values in the second period of analysis. Therefore, competitive balance has been lower since the new seeding system was adopted. Moreover, it can be seen that the measures take higher values especially for men. Thus, it seems that competitive balance is lower in the 32-player seeding system. Still, regression analysis should be used to get a clearer picture of the effects of the change in the number of seeded players. Before doing that, however, Table 1 provides the average of the percentage of expected seeded players who get some rounds and Table 2 provides the average of the CB measure. Wimbledon carried the lowest percentages of seeded players who advanced to the 2nd and 3rd rounds; however, Wimbledon is also the tournament with the highest percentage of the four top seeded players in the semifinals. Thus, lower seeded players were more likely to lose in the preliminary rounds in Wimbledon than in other tournaments. Likewise, the French Open is characterized by the lowest predictive Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 571 Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 88% (7%) 78% (9%) 66% (9%) 62% (20%) 55% (17%) 54% (37%) 36% (50%) Women 82% (9%) 66% (15%) 52% (16%) 49% (19%) 48% (33%) 36% (36%) 36% (50%) Men 85% (9%) 72% (13%) 59% (14%) 55% (20%) 52% (26%) 45% (37%) 36% (49%) Both Australian Open 86% (10%) 79% (12%) 63% (11%) 58% (15%) 54% (29%) 36% (31%) 14% (36%) Women 82% (5%) 64% (10%) 48% (11%) 38% (17%) 30% (33%) 21% (43%) 0% (0%) Men French Open Note: Standard deviations are shown in parentheses. Winner Final Semifinal Quarterfinal 1/8 Final 3rd Round 2nd Round Tournament Gender 84% (8%) 72% (13%) 55% (13%) 48% (19%) 42% (33%) 29% (37%) 7% (26%) Both 89% (10%) 72% (11%) 58% (9%) 60% (11%) 57% (27%) 50% (34%) 43% (51%) Women 81% (11%) 60% (15%) 45% (14%) 42% (17%) 50% (26%) 61% (35%) 71% (47%) Men Wimbledon 85% (11%) 66% (14%) 52% (13%) 51% (17%) 54% (26%) 55% (34%) 57% (50%) Both 90% (9%) 81% (11%) 64% (12%) 63% (11%) 57% (25%) 64% (36%) 36% (50%) Women 83% (9%) 66% (12%) 50% (9%) 42% (18%) 48% (21%) 36% (36%) 36% (50%) Men US Open 86% (10%) 73% (14%) 57% (13%) 52% (18%) 53% (23%) 50% (38%) 36% (49%) Both 88% (9%) 78% (11%) 63% (11%) 60% (14%) 56% (24%) 51% (35%) 32% (47%) Women Men All 82% (9%) 64% (13%) 49% (13%) 43% (18%) 44% (29%) 38% (39%) 36% (48%) Table 1 Percentage of Seeded Players Who Achieve Each Round With 16 Seeded Players (1994-2008) 85% (9%) 71% (14%) 56% (14%) 52% (18%) 50% (27%) 45% (38%) 34% (48%) Both 572 Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 91% (6%) 83% (8%) 73% (8%) 71% (17%) 72% (17%) 81% (20%) 77% (34%) Women 85% (9%) 73% (15%) 60% (17%) 60% (20%) 63% (25%) 63% (23%) 86% (17%) Men 88% (8%) 78% (13%) 67% (14%) 65% (19%) 68% (21%) 72% (23%) 81% (27%) Both Australian Open 89% (10%) 84% (10%) 71% (10%) 69% (13%) 77% (17%) 75% (20%) 79% (18%) Women 83% (8%) 65% (13%) 50% (14%) 49% (16%) 50% (30%) 52% (34%) 57% (37%) Men French Open Note: Standard deviations are shown in parentheses. Winner Final Semifinal Quarterfinal 1/8 Final 3rd Round 2nd Round Tournament Gender 86% (9%) 75% (14%) 61% (16%) 59% (18%) 64% (28%) 64% (30%) 68% (31%) Both 92% (7%) 78% (9%) 64% (8%) 68% (15%) 73% (18%) 85% (17%) 81% (28%) Women 84% (13%) 66% (18%) 52% (16%) 53% (18%) 62% (25%) 76% (30%) 91% (25%) Men Wimbledon 88% (11%) 72% (15%) 58% (14%) 60% (18%) 68% (22%) 80% (24%) 86% (26%) Both 92% (8%) 86% (11%) 73% (13%) 75% (12%) 81% (12%) 85% (20%) 82% (28%) Women 85% (8%) 71% (12%) 57% (10%) 53% (15%) 65% (19%) 74% (26%) 69% (40%) Men U.S. Open 89% (8%) 78% (13%) 65% (14%) 64% (17%) 73% (18%) 79% (24%) 76% (34%) Both 91% (8%) 83% (10%) 70% (10%) 71% (14%) 76% (16%) 82% (19%) 80% (27%) Women Table 2 The CB Measure of Competitive Balance With 16 Seeded Players (1994-2008) 84% (9%) 69% (14%) 55% (15%) 54% (17%) 60% (25%) 66% (29%) 76% (33%) Men All 88% (9%) 76% (14%) 63% (15%) 62% (18%) 68% (22%) 74% (26%) 78% (30%) Both del Corral / Competitive Balance and Match Uncertainty in Grand-Slam Tennis 573 Table 3 Number of Seeded Players Who Achieve the Predicted Round Determinants All Model 1 DSEED16 DMASC HARD CLAY R3 R4 R5 R6 R7 R8 R2HARD R3HARD R4HARD R5HARD R6HARD R7HARD R8HARD R2CLAY R3CLAY R4CLAY R5CLAY R6CLAY R7CLAY R8CLAY R2GRASS R3GRASS R4GRASS R5GRASS R6GRASS R7GRASS R8GRASS Constant Observations R2 ** 4.1 10.3*** 2.1 11.8*** 14.2*** 29.2*** 33.3*** 34.9*** 40.3*** 51.0*** 96.2*** 784 .30 Women Model 2 ** 4.1 10.3*** Model 1 0.3 Model 2 0.3 88.4*** 392 .32 Model 1 8.0 *** Model 2 8.0*** 6.6** 18.1*** 18.0*** 33.1*** 39.0*** 37.6*** 43.4*** 46.1*** 2.4 5.5 10.4** 25.3*** 27.6*** 32.3*** 37.2*** 55.9*** 28.3*** (1) 15.4** (1) 0.8 (1) 3.3 (1) 4.9 (2) 9.8 (2) 21.4*** (2) 26.8*** (3) 14.5** (2) 1.8 (2) 8.9 (3) 15.2** (3) 28.6*** (3) 50.0*** (3) 27.7*** (2) 9.2 (3) 5.6 (3) 6.3 (2) 3.6 (1) 1.8 (1) (1) 64.4*** 784 .35 Men 45.8*** (1) 36.6*** (1) 22.1*** (1) 19.2** (1) 13.4 (2) 16.1* (2) 7.1 (2) 43.3*** (2) 36.6*** (2) 20.1** (2) 15.2 (3) 10.7 (3) 7.1 (3) 28.6*** (3) 46.0*** (3) 29.5*** (3) 15.2 (3) 17.0* (2) 14.3 (1) 7.1 (1) (1) 43.0*** 392 .34 93.6*** 392 .29 10.9 (2) 5.8 (1) 20.5** (1) 25.9*** (1) 23.2*** (2) 35.7*** (1) 35.7*** (2) 10.3 (3) 7.6 (1) 23.7** (2) 33.0*** (3) 41.1*** (3) 50.0*** (3) 71.4*** (3) 9.4 (1) 11.2 (3) 26.3*** (3) 29.5*** (2) 21.4** (1) 107 (2) (1) 75.4*** 392 .38 Notes: In parentheses, the ranking of the coefficients relative to the other two surfaces in the same round and model are shown. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 574 Journal of Sports Economics Table 4 CB Determinants All Model 1 DSEED16 DMASC HARD CLAY R3 R4 R5 R6 R7 R8 R2HARD R3HARD R4HARD R5HARD R6HARD R7HARD R8HARD R2CLAY R3CLAY R4CLAY R5CLAY R6CLAY R7CLAY R8CLAY R2GRASS R3GRASS R4GRASS R5GRASS R6GRASS R7GRASS R8GRASS Constant Observations R2 *** 3.6 12.6*** 1.3 5.3*** 12.0*** 25.1*** 25.6*** 19.7*** 13.9*** 10.0*** 96.5*** 784 .25 Women Model 2 *** 3.6 12.6*** Model 1 0.7 Model 2 0.7 89.5*** 392 .16 Model 1 7.8 *** Model 2 7.8*** 0.1 11.0*** 15.5*** 29.2*** 30.7*** 24.0*** 18.0*** 8.5** 2.7 0.4 8.5*** 20.9*** 20.5*** 15.4*** 9.7*** 11.5*** 2.3 (1) 8.0* (1) 20.2*** (1) 21.4*** (1) 15.7*** (1) 10.5** (2) 7.6* (2) 0.0 (3) 11.6** (2) 25.4*** (2) 27.2*** (3) 22.6*** (3) 22.5*** (3) 18.3*** (3) 1.8 (2) 14.2*** (3) 27.9*** (3) 25.8*** (2) 18.5*** (2) 5.7 (1) (1) 94.3*** 784 .27 Men 10.6** (1) 3.2 (1) 8.1 (1) 8.5 (1) 4.5 (1) 1.5 (2) 1.8 (2) 7.7 (3) 2.4 (3) 10.0 (3) 11.8* (3) 4.0 (3) 5.9 (3) 2.2 (3) 11.0* (2) 2.9 (2) 17.3*** (2) 13.0** (2) 8.3 (2) 4.0 (1) (1) 80.9*** 392 .17 91.0*** 392 .25 6.0 (2) 19.2*** (1) 32.4*** (1) 34.4*** (1) 26.8*** (2) 22.4*** (2) 13.4* (2) 7.8 (3) 25.6*** (2) 40.9*** (2) 42.6*** (2) 41.2*** (1) 39.0*** (3) 34.4*** (3) 7.4 (1) 25.4*** (3) 38.6*** (3) 38.5*** (3) 28.6*** (3) 15.5* (1) (1) 95.0*** 392 .28 Notes: In parentheses, the ranking of the coefficients relative to the other two surfaces in the same round and model are shown. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. power of seeds, especially from the quarterfinals. When comparing the difference in competitive balance between men and women, we found that competitive balance is lower for women than for men (the only exception is the winner round of Table 1). Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 del Corral / Competitive Balance and Match Uncertainty in Grand-Slam Tennis 575 Tables 3 and 4 show the determinants of the competitive balance measures using regression analysis. In doing so, a different observation is considered for each round of each tournament for men and women. Thus, there are 784 observations,15 of which 392 belong to women and 392 to men. Those measures are evaluated using ordinary least squares (OLS) regression with a number of dummy variables. The first dummy variable takes the value of one if the tournament had 16 seeded players and zero otherwise (DSEED16). A negative coefficient is expected for this variable. Another dummy variable represents gender (DMASC) and takes the value of one for men’s matches. Likewise, a set of dummies for surface (HARD, CLAY, and GRASS) and dummies for round were also used. Those sets of dummies were interacted in Model 2. The coefficient of DSEED16 is negative and significant in the models for all players.16 Thus, the increase in the number of seeded players led to decreased competitive balance in tennis grand-slam tournaments. However, if the sample is split into men and women, this coefficient is not significant for women but takes a lower and significant value for men. Thus, the change in the number of seeded players affected men’s and women’s grand-slam tournaments differently. Moreover, average competitive balance measures are lower for men than for women by more than 10%. Therefore, men’s tournaments are characterized by a much higher competitive balance than women’s tournaments. Finally, with respect to court surfaces, clay turned out to be the type of court surface in which the competitive balance is highest, if all rounds are aggregated. This result could be qualified using the results from Model 2. Thus, in general, clay is the surface that produces the most competitive balance, especially in the final rounds. In the preliminary rounds, Wimbledon is the tournament generally characterized by the highest competitive balance. Match-Specific Uncertainty Results The biggest ex post match uncertainty outcome happens when the underdog beats the favorite. Following Abrevaya (2002), this outcome is labeled as upset. To analyze upset determinants, we estimated probit models in which the dependent variable takes the value of one when there is an upset and zero otherwise. Two kinds of covariates are used. The first refers to the characteristics of the match, while the second refers to players’ characteristics. To control for match characteristics, we used dummies for the court surface (i.e., HARD, CLAY, and GRASS) where GRASS is the reference category and a set of dummies controls for the round of the match. With regard to court surface, we expected that the quicker the surface, the higher the likelihood of upsets. Thus, it is expected that the lowest coefficient would belong to clay surface and the highest to grass surface. Likewise, the round dummies are expected to be lower when advancing round because Gilsdorf and Sukhatme (2007) found that the larger the Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 576 Journal of Sports Economics difference in prizes between the winner and the loser, the less likely the upset, but upsets will be less likely on 1st rounds due to the seeding system because seeded player are going to face between them. Thus, it is unclear which effect will predominate. With respect to the player characteristics, the following variables were used. A measure of the ex ante quality differences between players (DIFRANKING) was constructed as the difference between players’ rating measures (Rating ¼ 8 log(Rank)), which was proposed by Klaassen and Magnus (2001).17 The main advantage of this measure is that it assumes that quality in tennis is a pyramid. The expected value of the coefficient of this value is negative. Moreover, we included the age of the favorite (AGEF) and the age of the underdog (AGEU). Older players are more experienced than younger ones but may also lose some of their abilities and incentives. Hence, it was not clear, which sign the coefficient of those variables would take. However, with a large data set, the coefficients of those variables are likely to demonstrate that young players are more likely to improve their ranking than old players. Therefore, a negative coefficient for AGEU and a positive coefficient for AGEF were expected. In addition, a dummy variable for players who played in the qualification round (DPREV) and a dummy variable for wild cards players18 (DWILD) were introduced. Moreover, because countries could specialize in producing players with particular surface skills (e.g., Spain with clay), dummy variables for countries with enough observations were interacted with the surface dummies.19 In tennis, the height of the players is correlated positively with service skills (i.e., the taller player is the one more likely to serve fast). As a result, we assumed that differences in height could influence upsetting. Therefore, we include the height difference (DIFHEIGHT) between the favorite and the underdog, with the coefficient expected to be negative because the higher height difference between the favorite and the underdog the less likely are upsets. Lastly, we created a dummy variable that takes the value of 1 if the player is a former top 10 player at a minimum of five 5 years ago. The model includes variables for if the former top-10 player is the underdog (EXTOP10U) or the favorite (EXTOP10F). The expected values of those coefficients are negative for EXTOP10F and positive for EXTOP10U. This is based on the expectation that former top 10 players will remember how to play well if they are motivated enough. Finally, to test whether the differences between men and women are significant, we used gender dummy variables (i.e., DMALE and DFEMALE). Likewise, those variables were interacted to test for joint differences between gender and court in an alternative specification for which GRASS*DMALE was the reference. We estimated two alternative specifications for all players, as well as a model for men’s matches and another for women’s matches. Thus, we estimated four models of upset determinants. Table 5 shows the estimates as well as the marginal effects.20 As expected, the difference in quality between contestants is negatively correlated with the probability of upset and is significant in all models. The variable DMALE has a negative coefficient but is not significant. The estimated coefficients of the court surface dummies are not significant at any confidence level. The only Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 577 Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 0.025*** 0.022*** 0.204*** 0.147 0.229*** 0.424*** 0.007** 0.401 0.404 0.285 0.252 0.389 0.005 3,952 .12 2,070 0.235 0.424*** 0.075 0.031 0.093 (0.008) (0.007) (0.068) (0.049) (0.070) (0.148) (0.002) (0.127) (0.119) (0.084) (0.074) (0.108) (0.002) (0.135) (0.024) (0.010) (0.030) ME 0.103 0.032 0.191 0.223 0.179* 0.025*** 0.022*** 0.205*** 0.151 0.231*** 0.428*** 0.007** 0.403 0.406 0.287 0.253 0.390 0.008 3,952 .12 2,069 0.107 0.424*** Coefficient Model 2 (0.033) (0.010) (0.064) (0.075) (0.060) (0.008) (0.007) (0.068) (0.050) (0.071) (0.150) (0.002) (0.128) (0.120) (0.084) (0.074) (0.108) (0.002) (0.135) ME (0.006) (0.004) (0.074) (0.076) (0.112) (0.236) (0.002) (0.308) (0.249) (0.193) (0.179) (0.196) (0.055) (0.044) (0.030) 0.141 0.100 0.020** 0.013 0.227** 0.229 0.389*** 0.657*** 0.008* 1.021** 0.988** 0.799** 0.775* 0.941** 0.192 1,930 .16 966 (0.128) ME 0.764 0.414*** Coefficient Model 1 (Women) Note: ME, marginal effect measured at the means of the other variables. *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. All models include country effects interacted with the surface for favorite and underdog, respectively. CONSTANT DIFRANKING DMALE HARD CLAY HARD*DFEMALE HARD*DMALE CLAY*DFEMALE CLAY*DMALE GRASS*DFEMALE AGEF AGEU DPREV DWILD EXTOP10F EXTOP10U DIFHEIGHT DROUND1 DROUND2 DROUND3 DROUND4 DROUND5 DROUND6 Observations Pseudo-R2 Log likelihood Coefficient Model 1 Table 5 Upset Probit Results 0.031*** 0.036*** 0.138 0.061 0.083 0.359*** 0.009** 0.022 0.014 0.070 0.101 0.007 0.054 2,022 .15 1,030 0.143 0.248 0.011 0.453*** Coefficient ME (0.010) (0.011) (0.045) (0.020) (0.026) (0.122) (0.003) (0.007) (0.004) (0.022) (0.033) (0.002) (0.017) (0.045) (0.081) (0.142) Model 1 (Men) 578 Journal of Sports Economics exception is the coefficient for GRASS*DFEMALE, which is positive in Model 2. Therefore, the hypothesis that quicker courts have higher match-specific uncertainty is only supported by the data for women. The coefficients for AGEF and AGEU are positive and negative, respectively. This suggests that younger players tend to perform better than older players, given the ranking and other covariates. Similar results were found by Gilsdorf and Sukhatme (2007), using data from men’s matches. The coefficients of EXTOP10U are positive and EXTOP10F are negative, reflecting that players who were top 10 players several years before tend to make more upsets as underdog and to lose less matches as favorites than other players. Moreover, this effect is higher for women than for men. Finally, the coefficient of DIFHEIGHT is negative in all models, reflecting that if the difference in height between the favorite and the underdog is positive, the upset likelihood will be reduced. If the difference is negative (i.e., the underdog is taller than the favorite), then the upset-likelihood will increase. Conclusions In this article, we tested whether the increase in the number of seeded tennis players in grand-slam tournaments from 16 to 32 in 2001 led to a decrease in competitive balance. To do that, we proposed two measures of competitive balance based on the performance of the seeded players. The results indicated that the outcome of the change in the seeding system differed by gender. In particular, competitive balance decreased significantly for men. For women, no significant effect on competitive balance was incurred as a result of the change. Moreover, the competitive balance measured by the performance of seeded players is much higher for men than for women. Furthermore, Wimbledon faced the highest amount of competitive balance in the preliminary rounds, while in the final rounds, the opposite was found. Additionally, we analyzed the upset determinants using probit models. We found that quality difference is the main determinant of upsets. Moreover, conclusive evidence of differences among surfaces and gender were not found. Given that many sports events use seeds in single-elimination tournaments, the methodology used in this article can be implemented to analyze the competitive balance in these events. This could be especially important for testing the implications of some rule changes on competitive balance in sports and events. Notes 1. See Borland and MacDonald (2003) or Szymanski (2003) for surveys. 2. For instance, the International Table Tennis Federation has recently banned the use of speed glue to stick rubbers to the blade. Such a measure will certainly alter the competitive balance and will benefit Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 del Corral / Competitive Balance and Match Uncertainty in Grand-Slam Tennis 579 players with high defensive skills. Table tennis has also made some previous changes that could affect the competitive balance, such as the change in the scoring system and the service rule. 3. Magnus and Klaassen (1999), Boulier and Stekler (1999), Clarke and Dyte (2000), Klaassen and Magnus (2001, 2003), Paserman (2007), and Gilsdorf and Sukhatme (2007) have previously estimated discrete choice models using tennis data. 4. Some players need to qualify to the main draw by playing the qualifying stage. If an unexpected spot becomes available, then a lucky loser of the qualifying stage will play in the main draw. 5. Magnus and Klaassen (1999) found that there is no decrease in service dominance during a match. 6. For instance, assuming that the probability of winning a set by the favorite is .6, then the probability that the favorite wins the match is .65 within a match to the best of three sets and it is .68 if the match is played to the best of five sets. 7. The fastest court is the grass court, followed by hard court. The slowest type of surface is the clay surface. 8. The most popular players are usually the best classified players in the official rankings (see Pujol & Garcı́a-del-Barrio, 2007). However, there can be some exceptions from players who garner fame because of circumstances other than being a good player. Anna Kournikova in the beginning of the 2000’s is a good example. 9. Seeds are not only used in single elimination tournaments but also in several competitions in which round robin matches are played and subsequently the classified players are pooled in a single elimination tournament such as the Olympic competitions in sports, UEFA Champions League or FIFA World Cup. See Groh, Moldovanu, Sela, and Sunde (in press) for a theoretical analysis of optimal seedings. 10. The Appendix contains an explanation of how the computer rankings are constructed. 11. See Szymanski (2003) for the economic design of individual contests. 12. The 1st and 2nd seeded players should achieve the final, the 3rd and 4th should achieve the semifinals, the 5th, 6th, 7th, and 8th should achieve the quarterfinal, and so on. 13. Data from this source have previously been used by Forrest and McHale (2007) to estimate the relationship between returns and odds. 14. Some rounds are precluded because the figure turns fuzzy. The graphs containing all rounds are available upon request. 15. 14 (years), times 4 (tournaments), times 7 (rounds), times 2 (genders). 16. The coefficient of DSEED16 is the same in all models 1 and 2 because model 1 and model 2 contains the same information. The only difference is that in model 2 the round dummies and surface dummies were interacted. 17. The absolute difference in the ranking was also used, but because we obtained better results with this variable only those results are presented. 18. These wild cards are usually given to players from the country in which the tournament is played or well-recognized players who may not have a high enough ranking. 19. The coefficients of these variables are ignored in the presentation of the results, but they are available upon request. Not surprisingly, the country effects of Spain reflect the ability of Spanish players on clay. 20. It is important to note that it is not possible to use all observations due to missing data for some variables (i.e., ranking, height, and age). Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 580 Journal of Sports Economics Appendix A How Are Tennis Rankings Computed? The ATP and WTA ranking system reflects players’ performance in tournament play using a cumulative points system in which the number of tournament results that comprise a player’s ranking is capped at 17 and 18 tournament results for women and men, respectively. Both systems have some ‘‘mandatory’’ events (i.e., grand slams, ATP Master Series, and WTA Tier I). In both cases, for each ‘‘mandatory’’ event in which a player is not accepted, he or she may count one extra event toward his or her ranking Thus, the ATP ranking is based on calculating, for each player, his total points from the 13 ‘‘mandatory’’ events—the four grand slams and the nine ATP Masters Series tournaments—and his best five results from all eligible International Series tournaments from the past 52 weeks. The Tennis Masters Cup counts as an additional 19th tournament for players who qualify for the circuit finale. However, the WTA ranking is based on the player’s highest ranking points over the past 52 weeks including her ranking points from the Mandatory Tier I Tournament and the grand-slam Tournaments, if the player qualifies. Players are ranked on the basis of their total points. In the following table, the points awarded for each round in the ATP and WTA ranking systems are presented. Table A1. Points Awarded in Various Types of Tournaments (2008) Women Grand slam ATP master series Tier I (US$3,000,000) Tier I (US$2,000,000) Tier I (US$1,340,000) Men W F S Q R16 R32 W F S q R16 R32 1,000 700 450 250 140 90 1,000 500 700 350 450 225 250 125 150 75 75 35 500 465 430 350 325 300 225 210 195 125 115 110 70 65 60 45 40 35 Notes: W ¼ winner; F ¼ finalist; S ¼ semifinal; Q ¼ quarterfinal. References Abrevaya, J. (2002). Ladder tournaments and underdogs: Lessons from professional bowling. Journal of Economic Behaviour and Organization, 47, 87-101. Balfour, A., & Porter, P. (1991). The reserve clause in professional sports: Legality and effect on competitive balance. Labor Law Journal, 42, 8-18. Borland, J., & MacDonald, R. (2003). Demand for sport. Oxford Review of Economic Policy, 19, 478-502. Boulier, B., & Stekler, H. (1999). Are sports seedings good predictors? An evaluation. International Journal of Forecasting, 15, 83-91. Clarke, S., & Dyte, D. (2000). Using official ratings to simulate major tennis tournaments. International Transactions in Operational Research, 7, 585-594. Dixit, A. (1987). Strategic behaviour in contests. American Economic Review, 77, 891-989. Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009 del Corral / Competitive Balance and Match Uncertainty in Grand-Slam Tennis 581 Du Bois, C., & Heyndels, B. (2007). It’s a different game you go to watch: Competitive balance in men’s and women’s tennis. European Sport Management Quarterly, 7, 167-185. Forrest, D., & McHale, I. (2007). Anyone for tennis (betting)? The European Journal of Finance, 13, 751-768. Fort, R., & Maxcy, J. (2003). Competitive balance in sports leagues: An introduction. Journal of Sports Economics, 4, 154-160. Garcı́a, J., & Rodrı́guez, P. (2007). Spanish football: Competitive balance and the impact of the UEFA Champions League. In P. Rodrı́guez, S. Késenne, & J. Garcı́a (Eds), Governance and competition in professional sports leagues. Spain: Universidad de Oviedo, 171-190. Gilsdorf, K., & Sukhatme, V. (2007). Testing Rosen’s sequential elimination tournament model. Incentives and player performance in professional tennis. Journal of Sports Economics, 9, 287-303. Groh, C., Moldovanu, B., Sela, A., & Sunde, U. (IN PRESS). Optimal seedings in elimination tournaments. Economic Theory. Haugen, K. (2008). Point score systems and competitive imbalance in professional soccer. Journal of Sports Economics, 9, 191-210. Horowitz, I. (1997). The increasing competitive balance in major league baseball. Review of Industrial Organization, 12, 373-387. Humphreys, B. (2002). Alternative measures of competitive balance in sports leagues. Journal of Sports Economics, 3, 133-148. Klaassen, F., & Magnus, J. (2001). Are points in tennis independent and identically distributed? Evidence from a dynamic binary panel data model. Journal of the American Statistical Association, 96, 500-509. Klaassen, F., & Magnus, J. (2003). Forecasting the winner of a tennis match. European Journal of Operational Research, 148, 257-267. Koning, R. (2000). Balance in competition in Dutch soccer. The Statistician, 49, 419-431. La Croix, S., & Kawaura, A. (1999). Rule changes and competitive balance in Japanese professional baseball. Economic Inquiry, 37, 353-368. Magnus, J., & Klaassen, F. (1999). The final set in a tennis match: Four years at Wimbledon. Journal of Applied Statistics, 26, 461-468. Neale, W. (1964). The peculiar economics of professional sports. A contribution to the theory of the firm in sporting competition and in market competition. The Quarterly Journal of Economics, 78, 1-14. Paserman, M. D. (2007). Gender differences in performance in competitive environments: Evidence from professional tennis players. 6335 CEPR Discussion paper. Pujol, F., & Garcı́a-del-Barrio, P. (2007). Press release of tennis media value annual report. Retrieved March 2009 from http://www.unav.es/econom/sport/files/resourcesmodule/@random4562ca6fb6e9a/1209313454_R06_Valor_Mediatico_Tennis_2007.pdf Rohm, A., Chatterjee, S., & Habibullah, M. (2004). Strategic measure of competitiveness for ranked data. Managerial and Decision Economics, 25, 103-108. Sanderson, A. (2002). The many dimensions of competitive balance. Journal of Sports Economics, 3, 204-228. Sanderson, A., & Siegfried, J. (2003). Thinking about competitive balance. Journal of Sports Economics, 4, 255-279. Schmidt, M. (2001). Competition in Major League Baseball: The impact expansion. Applied Economics Letters, 8, 21-26. Szymanski, S. (2003). The economic design of sporting contests. Journal of Economic Literature, 41, 1137-1187. Vrooman, J. (1995). A general theory of professional sport leagues. Southern Economic Journal, 61, 971-990. Julio del Corral is a Ph D. Student at the University of Oviedo. His fields of specialization are efficiency and productivity analysis and sports economics. He has published articles in the Journal of Sports Economics, Revista de Economia Aplicada and Journal of Dairy Science. Downloaded from http://jse.sagepub.com at Univ de Oviedo-Bib Univ on December 2, 2009
© Copyright 2026 Paperzz