Adjusting double Poisson models to predict the NCAA Division I Softball Championship

While its viewership has surged in recent years, college softball remains an under-researched sport in the domain of sport analytics, partially due to a lack of a longstanding major professional league. However, the postseason format of major college softball- a four-stage layout with two four-team double elimination phases and two best-of-three series- presents an intriguing challenge for predictive models. Primarily focusing on the first of these four stages, we evaluate the effectiveness of a Double Poisson model in predicting the outcome of this competition. This model, previously developed and advanced for use in predicting the outcome of soccer matches, posits that a team’s number of runs scored takes a Poisson distribution, with a mean based on its own offensive strength and its opponent’s defensive strength, with strengths expressed in terms of scoring averages. Using game-by-game results for runs scored and runs against, we construct two additional pairs of factors used in constructing the means of these Poisson random variables for each team that account for a team’s strength of schedule and conference membership. From these additional factors, we construct an adjusted model that aims to account for these factors, and assess both the unadjusted base model and the adjusted model’s ability to predict the 2024 NCAA Division I Softball Tournament through simulation, with the aim of better understanding factors associated with success in this compelling tournament format.