Building on the research of Loeffelholz, Bednar and Bauer (2009), the current study analyzed the relationship between previously compiled team performance measures and the outcome of an “un-played” game. While past studies have relied solely on statistics traditionally found in a box score, this study included scheduling fatigue and team depth. Multiple models were constructed in which the performance statistics of the competing teams were operationalized in different ways. Absolute models consisted of performance measures as unmodified traditional box score statistics. Relative models defined performance measures as a series of ratios, which compared a team’s statistics to its opponents’ statistics. Possession models included possessions as an indicator of pace, and offensive rating and defensive rating as composite measures of efficiency. Play models were composed of offensive plays and defensive plays as measures of pace, and offensive points-per-play and defensive points-per-play as indicators of efficiency. Under each of the above general models, additional models were created to include streak variables, which averaged performance measures only over the previous five games, as well as logarithmic variables. Game outcomes were operationalized and analyzed in two distinct manners - score differential and game winner. Multiple regression analysis was used to explain the relationships between predictors and the “un-played” game’s score differential, and logistic regression analysis was used when the game winner was the dependent variable. The process of entering each model’s respective variables into the regression equations was accomplished through simultaneous entry, stepwise entry, and hierarchal entry. Statistical analyses were conducted on both the 2007-2008 and 2008-2009 National Basketball Association seasons, which served as two populations.
Taking into account goodness-of-fit measures and parsimony, superior models were identified. In regards to explained variance in score differential, the possession model with stepwise entry emerged as the best model for the 2007-2008 and 2008-2009 seasons. For predicting game winner the best models for the 2007-2008 and 2008-2009 seasons were the play stepwise entry model and the possession stepwise entry model respectively. As a whole, non-streak models were substantially more successful at explaining game outcomes than streak models. The increase in explained variance due to the entry of scheduling fatigue variables and team depth contribution factors in the second stage of the hierarchal multiple regression analysis varied among the models as well as the variables that reached statistical significance.
Overall, the findings of the present study indicate little generalizability between the two NBA seasons selected for the study. In general, the variables selected for inclusion into the regressions equations, as well as their relative importance differed from one season to the next. However, the possession models were found to be the best models in terms of predictive capabilities and parsimony, and they were the most stable over the two populations. These findings serve to support the use of composite measures of pace and efficiency in future basketball research as well as decisions made by management and members of the media.