Innovative Developments in Competitor Rating Systems

Abstract Number:

1201 

Submission Type:

Invited Paper Session 

Participants:

Mark Glickman (1), Andrew Swift (3), Scott Evans (2), Victor Chan (4), Martin Ingram (5), Mark Glickman (1)

Institutions:

(1) Harvard University, Cambridge, MA, (2) George Washington University, Washington, DC, (3) University of Nebraska At Omaha, Omaha, NE, (4) Western Washington University, Bellingham, WA, (5) KONUX GmbH, Munich, Germany

Chair:

Scott Evans  
George Washington University

Discussant:

Andrew Swift  
University of Nebraska At Omaha

Reviewer:

Michael Schuckers  
St. Lawrence University

Session Organizer:

Mark Glickman  
Harvard University

Speaker(s):

Victor Chan  
Western Washington University
Martin Ingram  
KONUX GmbH
Mark Glickman  
Harvard University

Session Description:

Rating systems for head-to-head competition are essential components of organized sports and gaming. These systems, which calculate numerical measures of competitor strength, are often used for seeding competitors in tournaments, selecting top competitors for elite events, and pairing competitors of similar abilities to play evenly-matched games. Many rating systems in use, such as the ubiquitous Elo system, have a probabilistic basis; they permit calculating prediction probabilities of game outcomes. However, because these probabilistic rating systems involve algorithms that are only loosely connected to established inferential procedures, more work is needed to place these systems on firmer ground, and to make the resulting estimated competitor abilities more trustworthy.

This session involves three presentations by a geographically diverse set of speakers that solidify the foundations of probabilistic rating systems. Recognizing the difficulty of calibrating rating changes in the Elo system, Victor Chan's presentation will focus on a novel method for optimizing the magnitude of rating changes based on a mean-squared error criterion. Martin Ingram will present an important connection between the Elo system and posterior mode estimation in a Kalman filter, which allows creating extensions of the Elo model that, for example, allows incorporating margins of victory. Finally, Mark Glickman will discuss the derivation of a rating system that acknowledges that stronger competitors may tie games more often than weaker competitors, and describe implementation details of the system that has been recently adopted for use by the International Correspondence Chess Federation, the main international governing body of competitive chess by email. The three 25-minute technical presentations will be followed by a 15-minute discussion by Andrew Swift, and then by a floor discussion.

Sponsors:

No Additional Sponsor 2
No Additional Sponsor 3
Section on Statistics in Sports 1

Theme: One Community: Informing Decisions and Driving Discovery

No

Applied

Yes

Estimated Audience Size

Large (150-275)

I have read and understand that JSM participants must abide by the Participant Guidelines.

Yes

I understand and have communicated to my proposed speakers that JSM participants must register and pay the appropriate registration fee by June 1, 2023. The registration fee is nonrefundable.

I understand