Evaluating the Information in Baseball Statistics

• Activity Number: 624
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Sports
• Author(s): Richard E. Auer*+ and Alyson Lindsey Fox
• Companies: Loyola University Maryland and Loyola University Maryland
• Keywords: Logistic Regression ; Concordance ; Coefficient of Determination ; Major League Baseball ; Slugging Percentage ; On-Base Percentage

Abstract

With all of the varying measures of quality in the sport of baseball, this article attempts to find an objective way to decide which one is "the best measure." We first applied the logistic regression model to all 2430 Major League Baseball (MLB) games from the 2009 season where the identity of all of the winning teams was being predicted. The model was fit using only team identifiers as predictor variables and then fit again using final season stats from the 30 teams. A second data set was composed of all obtainable box scores statistics from a sample of 200 MLB games from the same season. With this data providing game-to-game stats, the predictive capability of well-known measures like batting average, slugging average, on-base percentage, OPS and other sabermetric measures were compared to see which single measure seems to hold the most information. Then, all of the individual box score statistics were used as predictors. Lastly, multiple linear regression models were fit using these game-to-game stats as predictors. Here, the number of runs scored by a team became the dependent variable.