Dickinson Men's Basketball: Season and play-by-play analysis
We were given access to the Dickinson College Men’s Basketball account with Synergy Sports. Synergy Sports is where all season and play-by-play data is stored in their own API for most NCAA basketball games.
With this, we were able to access event level, game level, and season level data.
Areas of interest:
- How do offensive and defensive basketball statistics influence the accuracy of predicting the Points per Possession (PPP) rating and ratio?
- Are game-level statistics (fouls, points, etc.) significantly related offensive/defensive success?
- Does the winning percentage significantly change when considering event-level characteristics?
The process to get the data out of the website varied based on the type of data. The season level data for individual players and as a team was easily downloaded as a PDF and then converted to a CSV in Excel which could be read into Python. The event level data was more difficult to access. We had to inspect the elements of the page and find the login token as well as the link to the data. Then in Python, we were able to just switch game IDs to use that link and token to have requests collect the data and then we specified what elements to pull out of the mass list and into the data frames we would create.
| Data File Name | Data.Files | Link to File |
|---|---|---|
| “Seasondata” | seasondata.csv | Download Season Data CSV File |
| “All games” | all_games.csv | Download All Games CSV File |
Season-level data:
- Years 2015 through 2023, excluding 2020 (minimal data available)
- Created merged_offense and merged_defense dataframes to analyze separately, year column added
- Seasondata combines both merged datasets and creates an ID to identify offense or defense
Play-by-play (event) data:
- 27 games from the 2023-24 season
- Includes away/home teams, score, event level description, date, offensive/defensive players on the court during each event
Once the event level data was into Python in a workable state, we created dummy variables indicating what type of event occurred in each row. The variables included timeout, shot, assist, block, rebound, steal, and foul.
Season-level data:
We aim to analyze the significance of certain season metrics to predict their impact on the PPP value.
To do this, we conducted a variety of regression analyses: OLS regression with variety of control variables, separating regressions by offense and defense, and using interaction terms to determine the relationship between variables changing with one another.
Game-level data:
Our goal is to analyze the games of the 2023-24 season and estimate which factors are most impactful for a win.
Number of fouls/rebounds for Dickinson and opponent, home/away game
-
We aim to predict the PPP ratios for certain seasons while accounting for Attribute (Transition, Zone defense, last 4 sec on shot clock, etc.) and other offensive/defensive variables
-
Regression analysis using in-game statistics value to predict PPP ratios.