Having a ball: evaluating scoring streaks and game excitement using in-match trend estimation

Area of Science:

• Sports Analytics
• Statistical Modeling
• Machine Learning

Background:

• Traditional sports analysis often overlooks in-game scoring dynamics.
• A need exists for granular modeling of score differences to understand strategies and trends.

Purpose of the Study:

• To develop a novel statistical approach for modeling score differences in sports.
• To introduce interpretable indices for evaluating game trends and excitement.

Main Methods:

• Utilized a latent Gaussian process to model score differences between competing teams.
• Developed the Trend Direction Index (TDI) for probabilistic trend assessment.
• Introduced the Excitement Trend Index (ETI) to quantify game excitement.

Main Results:

• Applied the methodology to the entire 2019-2020 National Basketball Association season (1143 matches).
• Demonstrated the interpretability of TDI for analyzing individual game trajectories.
• Showcased ETI's utility in clustering teams based on their entertainment value.

Conclusions:

• The proposed latent Gaussian process model effectively captures score dynamics in sports.
• TDI and ETI offer valuable, interpretable metrics for sports analytics and fan engagement.
• This approach enhances understanding of team performance and game excitement.