Journal of Sports Analytics

This article examines the use of the PageRank algorithm to rank the teams and predict team performance in the tournament. This method has the potential to be utilized as an alternative method to choose tournament participants, as opposed to the traditional ranking and seeding process currently employed by the NCAA. PageRank allows for the consideration of all games played during the regular season (average of 5832 games per season) and for customizable performance weights in the prediction. The PageRank algorithm is a viable tool in predicting tournament outcomes due to depth and extensiveness of the data. The PageRank analysis helped to predict over half of the tournament game outcomes correctly in 2014-2018 and produced an average bracket score of 81.8 points out of 192 possible over the same 5 years.

In order to find the determinants of non-penalty goals scored per match, in association football (soccer), this paper developed a regression model consisting of 8 explanatory variables, based on observations for 98 teams playing in the top tiers of club football in England, Spain, Germany, France and Italy. We started with a framework that considered twenty-one different pitch actions that included both technical and tactical variables. Using data for the 2015-16 football season we narrowed down to the 8 variable model. The paper used a log-linear regression model in order to remove heteroscedasticity. The model estimated the number of non-penalty goals per game with error of less than |0.33| for 93 teams out of 98. For 52 teams the margin of error was less than |0.1|. Shots from penalty box per game, share of shots from goal box in total shots and long pass accuracy were found to have statistically significant positive impact on non-penalty goals scored per game. Share of long passes in total passes and crosses per game have significant negative impact.

College wrestling rankings have a major impact upon the sport, as they are one of the criteria used to determine advancement to postseason championships and seeding, thereby benchmarking individual, team, and program success. In addition to ordering athletes based upon past performance, rankings may function as predictors for future match outcomes. This research identifies potential biases in traditional ranking methodologies and offers alternatives based on the network-based PageRank and Elo ratings. Using data from the 2013-2014 NCAA Division I wrestling season, we evaluate both existing and our new rankings on the basis of predictive accuracy. Our new methods significantly improve upon a baseline of 67% accuracy and several of the existing ranking methods. These results and follow-up analyses suggest that Elo presents an especially attractive alternative to current ranking systems for college wrestling, with potential extensions to the prediction of point differentials, ensemble methods, and generalizability to other combat sports.

Every year, there is a popular debate over how many teams should take part in the NCAA’s FBS-level college football championship tournament, and especially whether it should be expanded from 4 teams to 8 or even 12. The inherent tradeoff is that the larger the tournament, the higher the probability that the true best team is included (“validity”), but the lower the probability that the true best team will avoid being upset and win the tournament (“effectiveness”). Using simulation based on empirically-derived estimates of the ability to measure true team quality and the amount of randomness inherent in each game, we show that the effect of expanding the tournament to 8 teams could be very small, an effectiveness decrease of only 2-3% while increasing validity by 1-4%, while a 7-team tournament provides slightly better tradeoffs. A 12-team tournament would decrease effectiveness by 5-6%.

Twenty-eight years ago Golden and Wasil (1987) presented the use of the Analytic Hierarchy Process (AHP) for ranking outstanding sports records. Since then much has changed with respect to sports and sports records, the application and theory of the AHP, and the availability of the internet for accessing data. In this paper we revisit the ranking of outstanding sports records and build on past work, focusing on a comprehensive set of records from the four major American professional sports. We interviewed and corresponded with two sports experts and applied an AHP-based approach that features both the traditional pairwise comparison and the AHP rating method to elicit the necessary judgments from these experts. The most outstanding sports records are presented, discussed and compared to Golden and Wasil’s results from a quarter century earlier.

Estimating probability is the very core of forecasting. Increasing computing power has enabled researchers to design highly intractable probability models, such that model results are identified through the Monte Carlo method of repeated stochastic simulation. However, confidence in the Monte Carlo identification of the model can be mistaken for accuracy in the underlying model itself. This paper describes simulations in a problem space of topical interest: basketball season forecasting. Monte Carlo simulations are widely used in sports forecasting, since the multitude of possibilities makes direct calculation of playoff probabilities infeasible. Error correlation across games requires due care, as demonstrated with a realistic multilevel basketball model, similar to some in use today. The model is built separately for each of 20 NBA seasons, modeling team strength as a composition of player strength and player allocation of minutes, while also incorporating team persistent effects. Each season is evaluated out-of-time, collectively demonstrating systematic and substantial overconfidence in playoff probabilities, which can be eliminated by incorporating error correlation. This paper focuses on clarifying the use of Monte Carlo simulations for probability calculations in sports.

Difficulty bias was found in Morgan and Rotthoff (2014), claiming that trying something harder leads to a higher overall ranking, ceteris paribus. This was measured in elite level gymnastics, which at some meets allows for a unique measurement of this form of bias. Multiple studies also find evidence of an overall order bias. I analyze if these two forms of bias still exist in elite level gymnastics. I continue to find evidence of difficulty bias with this updated dataset and find strong evidence that these biases are different across genders – females have evidence of a large and significant difficulty bias, whereas males do not. However, the evidence of an overall order bias does not exist in this dataset.

In a close game of basketball, victory or defeat can depend on a single shot. Being able to identify the best player and play scenario for a given opponent’s defense can increase the likelihood of victory. Progress in technology has resulted in an increase in the popularity of sports analytics over the last two decades, where data can be used by teams and individuals to their advantage. A popular data analytic technique in sports is deep learning. Deep learning is a branch of machine learning that finds patterns within big data and can predict future decisions. The process relies on a raw dataset for training purposes. It can be utilized in sports by using deep learning to read the data and provide a better understanding of where players can be the most successful.

In this study the data used were on division I women’s basketball games of a private university in a conference featuring top 25 teams. Deep learning was applied to optimize the best offensive play in a game scenario for a given set of features. The system is used to predict the play that would lead to the highest probability of a made shot.

George Lindsey was one of the first to present run scoring distributions of teams of Major League Baseball. One drawback of Lindsey’s approach is that his calculations represented a situation where the team is average. By use of a multinomial/multilevel modeling approach, we look more carefully how run-scoring distributions vary between teams and how run-scoring is affected by different covariates such as ballpark, pitcher quality, and clutch situations. By use of exchangeable models over ordinal regression coefficients, one gets a better understanding which covariates represent meaningful differences between run-scoring of teams.

We present a new, simple knockout format for sports tournaments, that we call “Choose Your Opponent”, where the teams that have performed best during a preliminary group stage can choose their opponents during the subsequent knockout stage. The main benefit of this format is that it essentially solves a recently identified incentive compatibility problem when more than one teams from a group advance to the knockout stage, by effectively canceling the risk of tanking. This new design also makes the group stage more exciting, by giving teams a strong incentive to perform at their best level, and more fair, by limiting the risk of collusion and making sure that the best group winners are fairly rewarded in the knockout round. The choosing procedure would add a new, exciting strategic component to the competition. Advancing teams would choose their opponent during new, much anticipated TV shows which would attract a lot of media attention. We illustrate how this new format would work for the round of 16 of the UEFA Champions League, the most popular soccer club competition in the world.

Data from the pool rounds of three HSBC World Rugby Sevens competitions (2016–17, 2017–18, and 2018–19) are used to investigate the number of tries required to win in international rugby sevens. The data consist of 4,391 tries scored in 720 matches (1,440 team performances) and are used to calculate the probability of winning a match given that T tries are scored (P[W|T]). The distribution of the number of tries scored by each team ranges from zero to nine and is shown to be well-represented by a Poisson distribution computed from the mean value of tries scored in that competition. The number of tries scored by the winning team in each match within a competition is well-described by a Gamma function evaluated at the integer number of tries scored with parameters derived from the data set. This appears to be a novel result not previously reported in the literature. Generalizing within each competition, teams scoring either zero tries or one try have less than a 2% chance of winning; those scoring two tries win 10% to 20% of the time; three tries result in nearly a 50% chance of winning; teams scoring four tries are almost sure to win (around 90%); and that for teams scoring five or more tries winning is virtually assured. Based upon the results from these three tournaments we conclude that competitive teams should strive to score three or more tries per match and that there is no winning advantage accrued by scoring more than five tries.

In football, many people argue that in knock-out competitions the team that plays the second game in its stadium would have an advantage, a greater probability of victory in the final outcome of a two leg knock-out match. The purpose of this study is to verify the veracity of this statement using data from the Brazil Cup. We find evidence that the ability spread between the teams participating in a match is the main factor that explains the qualification of one of the teams for the next round. Until 2018, there were three criteria for break the tie in the Brazil Cup, they were used in the respectively order: goal difference, away goals rule, penalty shootout (there is no extra-time in any play-off of the championship). It is estimated that 36% of the matches end tied and need a criterion to determine the winner. Of these, 51% use the goal difference decision, 29% use the away goals rule and 20% penalty shoot-out. When considering the championship in general there is evidence that the home team wins the match in approximately 63% of the matches, a significant advantage. However, in the confrontations that were decided by the away goals rule or the penalty shoot-out, the home team wins percentage is 20% lower, indicating that these criteria level the odds of both teams.

In recent years, data analytics has been a growing phenomenon in many fields, including sport, and there has been an increased focus on how technology will impact the work of coaching and performance professionals. This paper provides a reflection of the use of data analytics within the Gaelic Athletic Association’s (GAA) football coaching practice in Ireland, and evaluates for coaches how to enhance kick-out distribution and effectiveness. Specifically, the study aimed to dissect and analyse kick-out strategies to assess from a coaching perspective, the impact on distribution and effectiveness. The research is a result of the modern day professional approach taken by most GAA Inter County teams, and intends to fill the research gap of performance analysis within the context of Gaelic Football. Video analysis of eight Men’s Inter-County Ulster Senior Gaelic Football Championship matches was completed using video analysis software. The key parameter of investigation was kick-outs, with distance and direction being the variables specifically analysed. Three factors were established to measure kick-out effectiveness: retaining possession, attack building, and score/scoring opportunity. A questionnaire was devised for data collection with 75 players and 25 coaches to gather opinions and add further depth to the statistical evidence. Kick-outs to the wings (57.6%) and 65m+ (53.8%) were the most common distribution strategy employed by goalkeepers. Kick-outs to 21–45 m produced notably greater success rates of possession retention (87%) and attack building (55%) than the longer kick-outs. Kick-outs to the wings contributed an 8% greater success rate of creating scoring opportunities as opposed to central kick-outs. Kick-outs to the centre were the only parameter to result in a negative net difference of scores between both teams (–1%). It was established that kick-outs can be successful attacking elements when used efficiently. Short kick-outs to the wings appear to be the most effective strategy to improve the likelihood of possession retention, attack building and creation of scoring opportunities. An implication of this study for sports science is to make data analytics more accessible to GAA coaches, and address the challenges faced by coaches in this area by translating knowledge of performance analysis of kick-outs into working practice on the field.

This paper introduces a probabilistic method to evaluate the tactical decisions players and coaches make at the end of NBA games. For the purposes of this research, these decisions include whether to shoot a two-point or three-point field goal for the offensive team and whether to intentionally foul for the defensive team. With a win probability model built using logistic regression and player statistics, the optimal decision for both teams in a given possession is found. The End-of-game Tactics Metric (ETM) is the difference between the win probability of the optimal decision and the win probability of the actual decision. This research extends beyond current applications of win probability models to evaluate the actual on-court decision as opposed to evaluating the result of a possession.

To evaluate the usefulness of ETM, the winning percentage of teams in games decided by a margin of five points or fewer can be compared with the mean ETM difference between a team and its opponent. The correlation coefficient of the relationship is -0.64. When combined with other variables that affect winning percentage in close games, a linear regression on those explanatory variables has an adjusted R² value of 0.79. This analysis shows that the ETM difference has a significant effect on winning close games, despite having little reliance on player performance.

In this manuscript, different approaches for modeling and prediction of tennis matches in Grand Slam tournaments are proposed. The data used here contain information on 5,013 matches in men’s Grand Slam tournaments from the years 2011–2022. All regarded approaches are based on regression models, modeling the probability of the first-named player winning. Several potential covariates are considered including the players’ age, the ATP ranking and points, odds, elo rating as well as two additional age variables, which take into account that the optimal age of a tennis player is between 28 and 32 years. We compare the different regression model approaches with respect to three performance measures, namely classification rate, predictive Bernoulli likelihood, and Brier score in a 43-fold cross-validation-type approach for the matches of the years 2011 to 2021. The top five optimal models with highest average ranks are then selected. In order to predict and compare the results of the tournaments in 2022 with the actual results, a comparison over a continuously updating data set via a “rolling window” strategy is used. Also, again the previously mentioned performance measures are calculated. Additionally, we examine whether the assumption of non-linear effects or additional court- and player-specific abilities is reasonable.

In this study, we demonstrate that men overestimate their marathon abilities compared to women and propose that this may result from men’s relative overconfidence as well as lead men to execute suboptimal race strategy relative to women. We present a novel dataset of marathoners’ predicted finish times from the 2013 Houston Marathon. Using the runners ex-ante predictions of their own marathon finish times, we show that men consistently overestimate their abilities relative to women. Further, we find that men slow more than women in the later stages of the race and find that less even pacing is correlated with worse marathon performance, results in-line with previous studies. Finally, we argue that this poor pacing is in part explained by men’s overconfidence.

This paper introduces a new metric for player evaluation in Twenty20 cricket. The proposed metric of “expected run differential” measures the proposed additional runs that a player contributes to his team when compared to a standard player. Of course, the definition of a standard player depends on their role and therefore the metric is useful for comparing players that belong to the same positional cohort. We provide methodology to investigate both career performances and current form. Our metrics do not correlate highly with conventional measures such as batting average, strike rate, bowling average, economy rate and the Reliance ICC ratings. Consequently, our analyses of individual players based on results from international competitions provide some insights that differ from widely held beliefs. We supplement our analysis of player evaluation by investigating those players who may be overpaid or underpaid in the Indian Premier League.

We develop matchup models for the probability of a ground ball and a ground ball hit using twelve years of major league baseball play-by-play data. The models are based on player descriptors that can be estimated reliably from small samples which facilitates the use of the models for prediction. The model for ground ball probability is obtained by generalizing the log5 model to include both ground ball and strikeout rates for the batter and pitcher. A strikeout rate cross term is shown to be significant in this model which leads to regions of the matchup space, termed matched and mismatched Krate configurations, where either the batter or pitcher is favored relative to the log5 prediction. We also build a model for the probability that a ground ball becomes a hit which separates the contributions of the batter, pitcher, and defense. We show that this probability has a strong dependence on the pitcher’s ground ball and strikeout rates and that the structure of this dependence changes with the platoon configuration. We give a physical justification for the model and provide examples of pitchers with characteristics that significantly lower or raise their expected ground ball hit rates. The new models for the probability of a ground ball and a ground ball hit are tested on out-of-sample data and shown to provide more accurate predictions than alternative models.