Volume 10, Issue 3

Every basketball player takes and makes a unique spatial array of shots. In recent years, technology to measure the coordinates of these constellations has made analysis of them possible, and the possibility exists for distinguishing between different shooters at a level of spatial detail finer than the entire basketball court. This paper addresses the challenge of characterizing and visualizing relative spatial shooting effectiveness in basketball by developing metrics to assess spatial variability in shooting. Several global and local measures are introduced and formal tests are proposed to enable the comparison of shooting effectiveness between players, groups of players, or other collections of shots. We propose an empirical Bayesian smoothing rate estimate that uses a novel local spatial neighborhood tailored for basketball shooting. These measures are evaluated using data from the 2011 to 2012 NBA basketball season in three distinct ways. First we contrast nonspatial and spatial shooting metrics for two players from that season and then extend the comparison to all players attempting at least 250 shots in that season, rating them in terms of shooting effectiveness. Second, we identify players shooting significantly better than the NBA average for their shot constellation, and formally compare shooting effectiveness of different players. Third, we demonstrate an approach to map spatial shooting effectiveness. In general, we conclude that these measures are relatively straightforward to calculate with the right input data, and they provide distinctive and useful information about relative shooting ability in basketball. We expect that spatially explicit basketball metrics will be useful additions to the sports analysis toolbox.

Yearly, the top National Collegiate Athletic Association (NCAA), Division I, men’s basketball teams compete for the national championship, in what is commonly referred to as “March Madness.” Teams are eligible for the tournament when they either win their conference tournament or are invited by a NCAA selection committee. Various factors influence the committee’s decision. Of primary importance is the perceived strength of a team; however, the possible measures used to assess a team’s strength can vary widely. These metrics are of particular interest for teams on the borderline or “on the bubble” of tournament inclusion. Using historical data, we assess which metrics play the most substantial roles in predicting tournament inclusion. Additionally, we determine guidelines for constructing a team’s schedule that are based on these metrics to enhance the chance of being selected for the tournament.

This paper models the TV audience for NFL games in a market without a local team. The model is estimated using all NFL games shown in the Salt Lake City market over the last 10 years. Team popularity varies season to season, with fans preferring high-scoring close games between good teams. The primary motivation of the model was to advise the local station in week-to-week selection of high TV audience games from the slate of FOX games. In 2013 the most popular team was the San Francisco 49ers and the local station broadcast more of their games than any other team. While the predictions offer modest discrimination between popular games, the predicted error precision must be reduced to compete with local station expertise. Reviewing the prediction performance in 2013 reveals insight into strengths and weaknesses of predictive analytics in business decisions.

The NHL has realigned its conferences and divisions, and starting with the 2013–2014 season the Eastern Conference features 16 teams and the Western Conference features 14. Yet because there are eight playoff spots available in both conferences, teams in the West have a 57% probability of making the playoffs, compared to just 50% for teams in the East. As a result we should expect that, on average, the last team to make the playoffs in the West will have a worse record than the last playoff team in the East. We call the difference in points earned by the 8th seed in each conference the “conference gap.” The purpose of this paper is to estimate the expected size of the conference gap under the new alignment. Using tens of thousands of simulated seasons, we demonstrate that the conference gap will be, on average, 2.74 points, meaning that Eastern Conference teams hoping to make the playoffs will have to win 1–2 games more than Western Conference playoff-hopefuls. We also show the 9th place team in the Eastern Conference has a better record than the 8th place Western team twice as often as the 9th best Western team has a better record than the East’s 8th best. Our findings inform questions about competitive balance and equity in the NHL.

Modeling and prediction of ice hockey match results are not as widely examined areas as modeling and prediction of association football match results. It is assumed that match results in football and ice hockey can be modeled by the bivariate Poisson distribution or by some modification of this distribution. The aim of this paper is to explore the possibility of using models derived for football match results also for ice hockey match results and to propose some modifications of these models. A new model based on alternative definition of the bivariate Poisson distribution is presented. The models are tested on historical data from the highest-level ice hockey league in the Czech Republic between the years 1999 and 2012.