Modeling of NBA Game Data and their Correlation Structure

In recent years, data analysis has become very popular and has been applied to many fields including the oil and gas industry, public health, and information technology. With the development of technology, a rapidly increasing amount of sports data, which range from numerical statistics to motion videos, becomes available and ready to explore. In this dissertation, I focus on the numerical statistics of NBA games, mainly from the 2017 - 2018 season, and attempt to build a statistical model to estimate the results of the games.Different from most research on sports analytics, which has usually been results driven without exploring the statistical structure and features, I here attempt to explain the most important factors influencing the result of a game. Unlike the ”Black Box” created by using machine learning or deep-learning techniques, I use the statistical generalized estimating equations (GEE) model.Besides the result, I also focus on the correlation structure between the games. This is important for the games, as the playoffs are held in series where two teams need to play against each other for up to seven games. Therefore, the knowledge of the corresponding correlation structure would help the teams to analyze their performance appropriately.