NCAA Tournament "Dance Card"
MinV College Football Ranking
Other Sports Research

Sports Research by Jay Coleman (Available full text downloads here)

   

Minimum Violations and Predictive Meta-Rankings for College Football

B. Jay Coleman (Naval Research Logistics, 2014, DOI: 10.1002/nav.21563)

Abstract

This article presents two meta-ranking models that minimize or nearly minimize violations of past game results while predicting future game winners as well as or better than leading current systems—a combination never before offered for college football. Key to both is the development and integration of a highly predictive ensemble probability model generated from the analysis of 36 existing college football ranking systems. This ensemble model is used to determine a target ranking that is used in two versions of a hierarchical multiobjective mixed binary integer linear program (MOMBILP). When compared to 75 other systems out-of-sample, one MOMBILP was the leading predictive system while getting within 0.64% of the retrodictive optimum; the other MOMBILP minimized violations while achieving a prediction total that was 2.55% lower than the best mark. For bowls, prediction sums were not statistically significantly different from the leading value, while achieving optimum or near-optimum violation counts. This performance points to these models as potential means of reconciling the contrasting perspectives of predictiveness versus the matching of past performance when it comes to ranking fairness in college football.


Identifying the "Players" in Sports Analytics Research

B. Jay Coleman (Interfaces, 2012, Vol. 42, No. 2, pp. 109-118)

Abstract

Despite a sports analytics research history that goes back more than 50 years and a recent dramatic rise in the level of scholarly interest in sports analytics, no prior research has attempted to identify its scope, scale, and growth in terms of the body of published refereed articles in the literature. Prior research has also not identified the “players” in the field: the journals and institutions that most commonly publish sports analytics research and are most commonly cited. To answer these questions, I examined 140 journals in operations research, statistics, applied mathematics, and applied economics, and identified 1,146 articles that address the application of analytics in sports. The results provide a picture of the size and nature of sports analytics research and its purveyors, and offer some perspective on the parameters of the field.


Evidence of Bias in NCAA Tournament Selection and Seeding

B. Jay Coleman, Michael DuMond, and Allen K. Lynch (Managerial and Decision Economics, 2010, Vol. 31, No. 7, pp. 431-452)

Abstract

We investigate bias in the selection and seeding decisions of the NCAA Division I Men's Basketball Committee. Using data on 910 teams associated with the ten tournaments from 1999 to 2008, we test for bias toward teams from seven "major" conferences and six "mid-major" conferences, as well as for bias toward teams represented on the Committee. We find substantial support for the hypothesis of bias in favor of virtually all major and mid-major conferences in selection and/or seeding, as well as evidence of bias toward majors over mid-majors. We also find substantial evidence of bias toward teams with some type of Committee representation.

Note: The above research is summarized in greater detail under the "NCAA Tournament "Dance Card"" tab of this site.


Voter Bias in the Associated Press College Football Poll (free full text download available here)

B. Jay Coleman, Andres Gallo, Paul M. Mason, and Jeffrey W. Steagall (Journal of Sports Economics, 2010, Vol. 11, No. 4, pp. 397-417)

Abstract

The authors investigate multiple biases in the individual weekly ballots submitted by the 65 voters in the Associated Press college football poll in 2007. Using censored Tobit modeling, they find evidence of bias toward teams (a) from the voter's state, (b) in conferences represented in the voter’s state, (c) in selected Bowl Championship Series conferences, and (d) that played in televised games, particularly on relatively prominent networks. They also find evidence of inordinate bias toward simplistic performance measures - number of losses, and losing in the preceding week - even after controlling for performance using mean team strength derived from 16 so-called computer rankings.

Note: The above research has been featured by the Wall Street Journal, Newsday, the Florida Times-Union, WOKV (Jacksonville), and WJCT (Jacksonville).


NCAA Tournament Games: The Real Nitty-Gritty (free full text download available here)

B. Jay Coleman and Allen K. Lynch (Journal of Quantitative Analysis in Sports, 2009, Vol. 5, Issue 3, Article 8)

Abstract

The NCAA Division I Men's Basketball Committee annually selects its national championship tournament's at-large invitees, and assigns seeds to all participants. As part of its deliberations, the Committee is provided a so-called "nitty-gritty report" for each team, containing numerous team performance statistics. Many elements of this report receive a great deal of attention by the media and fans as the tournament nears, including a team's Ratings Percentage Index (or RPI), overall record, conference record, non-conference record, strength of schedule, record in its last 10 games, etc. However, few previous studies have evaluated the degree to which these factors are related to whether a team actually wins games once the tournament begins. Using nitty-gritty information for the participants in the 638 tournament games during the 10 seasons from 1999 through 2008, we use stepwise binary logit regression to build a model that includes only eight of the 32 nitty-gritty factors we examined. We find that in some cases factors that receive a great deal of attention are not related to game results, at least in the presence of the more highly related set of factors included in the model.

Note: The above research has been featured by CNBC (three prime-time spots, including two live talkback segments), CNN, ESPN The Magazine, the Associated Press, Dow Jones Newswires, ABC-Channel 25 and WJXT-Channel 4 in Jacksonville, WTOP radio in Washington, D.C., the Winnipeg Sun (CANOE news network), Network World, DM Review, WGST (Atlanta), WACV (Montgomery), KSFN (Las Vegas), BigSports590 (Omaha), WJCT (Jacksonville), WCCP (Clemson, SC), The State (S.C.), and three times by the Florida Times-Union (Jacksonville).


An Examination of NBA MVP Voting Behavior: Does Race Matter? (free full text download available here)

B. Jay Coleman, Michael DuMond, and Allen K. Lynch (Journal of Sports Economics, 2009, Vol. 9, No. 6, pp. 606-627)

Abstract

The selection process of the most valuable player (MVP) in the National Basketball Association (NBA) was recently questioned as to whether African American players were treated unfairly based on their race. Using NBA voting data from the 1995-2005 seasons, two empirical models are developed to examine the role that a player's race plays in the determination of this award. The estimates imply that after controlling for player, team, and market characteristics, there is no statistically significant effect of race on the likelihood that a player will appear on an MVP ballot or on the number of votes he will receive.


Minimizing Game Score Violations in College Football Rankings (free full text download available here)

B. Jay Coleman (Interfaces, 2005, Vol. 35, No. 6, pp. 438-497)

Abstract

One metric used to evaluate the myriad ranking systems in college football is retrodictive accuracy. Maximizing retrodictive accuracy is equivalent to minimizing game score violations: the number of times a past game's winner is ranked behind its loser. None of the roughly 100 current ranking systems achieves this objective. Using a model for minimizing violations that exploits problem characteristics found in college football, I found that all previous ranking systems generated violations that were at least 38 percent higher than the minimum. A minimum-violations criterion commonly would have affected the consensus top five and changed participants in the designated national championship game in 2000 and 2001—but not in the way most would have expected. A final regular season ranking using the model was perhaps the best prebowl ranking published online in 2004, as it maximized retrodictive accuracy and was nearly the best at predicting the 28 bowl winners.

Note: The above research is summarized in greater detail under the "MinV College Football Ranking" tab of this site. This article was also one of two published articles featured in Volume XI (Summer/Fall 2006) of ORMS Tomorrow (the INFORMS Student Magazine).


Identifying the NCAA Tournament "Dance Card" (free full text download available here)

B. Jay Coleman and Allen K. Lynch (Interfaces, 2001, Vol. 31, No. 3, pp. 76-86)

Abstract

The NCAA Basketball Tournament selection committee annually selects the Division I men's teams that should receive at-large bids to the national championship tournament. Although its deliberations are shrouded in secrecy, the committee is supposed to consider a litany of team-performance statistics, many of which outsiders can reasonably estimate. Using a probit analysis on objective team data from 1994 through 1999, we developed an equation that accurately classified nearly 90 percent of 249 "bubble" teams during that time frame and over 85 percent for the 2000 tournament. Given the NCAA Tournament's nickname of the big dance, the equation is effectively the "dance card" that determined whether a team got an invitation from past committees and is also a tool that could aid decision making for future committees. The accuracy of the dance card, and the factors and weights included in it, suggest that the committee is fairly predictable in its decisions, despite barbs from fans, teams, and the media.

Note: The above research is summarized in greater detail under the "NCAA Tournament 'Dance Card'" tab of this site.


Convergence or Divergence in Final Offer Arbitration in Professional Baseball

B. Jay Coleman, Kenneth Jennings, and Frank McLaughlin (Industrial Relations, 1993, Vol. 32, No. 2, pp. 238-247)

Abstract

Many labor relations practitioners and theorists believe that final-offer arbitration by a neutral third party encourages union and management officials to resolve their bargaining differences. However, decision scientists have found that there is no median convergence between the parties. Using professional baseball in our model, we test the assumption that major league owners tend to maximize expected monetary value (EMV), finding that claims of divergence are invalidated in dispute management contexts where there is a broad range of other motivations for settling. Decision models offer even further support for the use of final-offer arbitration in such settings.



B. Jay Coleman, Ph.D.
Richard deR. Kip Professor of Operations Management & Quantitative Methods
Department of Management | Coggin College of Business | University of North Florida | Jacksonville, FL 32224
jcoleman@unf.edu

Disclaimer