An All-New Hockey Analytics Prospect Scouting Tool and a Framework for Evaluating MLB Pitcher Changes
Check out what's happening in the intersection of sports and data analytics in this week's Data Points of the Week
Introducing a new hockey analytics scouting tool
This past week in sports analytics… Ruben Flam-Shepherd announced Scout, an advanced hockey analytics scouting tool offering unique visualizations to directly compare offensive production between players. These scouting visualizations use player age (down to the day) along the x-axis and display different metrics of offensive production on the y-axis, allowing the user to measure performance on an age-normalized basis. Not only does this tool incorporate NHL data, but it also accesses data from across the CHL, USHL, and BCHL to compare players coming through the junior ranks. Below we show a Scout point comparison between Maple Leaf forwards John Tavares and Mitch Marner from their days in the OHL.
Scout also offers a draft evaluation scoring system, allowing the user to break down drafting performance by team and GM. Drafting is evaluated on a cumulative “games played over expected“, a measurement of how many games a players has played in each season compared to their expected value (based on draft position). Below we focus in on the top-performing organization in the draft over the past 10 years, the Caroline Hurricanes (CAR). Specifically, we highlight Sebastian Aho, the 35th overall draft pick in 2015 currently leading all CAR prospects with nearly 350 games played over expected.
Scout is a unique offering for hockey organizations and scouting divisions to leverage moving forward! Learn more about Scout’s creator here.
An improved framework for evaluating pitching change decision making in MLB
Sean Sullivan, creator of URAM Analytics, released a publicly available dashboard serving as a framework for evaluating decision-making around pitching changes in Major League Baseball (MLB). Fans often criticize such decisions, focusing on outcomes instead of the manager's overall decision process. In this blog post, Sean explains his complete methodology used to to evaluate these decisions based on key factors like pitcher matchups, bullpen options, and game context (e.g., the Three-Batter Minimum Rule).
The framework draws from previous work, including metrics like Win Probability Added (WPA) and leverage index, alongside other models evaluating manager performance. Data is collected using Python tools like pybaseball and web scraping from ESPN for daily game roster information. The evaluation uses machine learning models to predict outcomes like on-base events or strikeouts for batter-pitcher matchups. And finally, a curated program simulates plate appearances to account for game context, passing the results from one plate appearance to the next.
Final Pitching Change Decision Making (PCDM) grades assigned to each MLB team for the 2024 season reveal an underlying relationship with two more traditional team success metrics: Bullpen ERA and Winning Percentage (WP).
The visualization of PCDM and Bullpen ERA shows a slightly negative relationship between the two metrics, meaning that as Bullpen ERA decreases, the PCDM increases on average.
And for a more direct indication into team success, the WP and PCDM show a slightly positive relationship, indicating that as one increases, so does the other on average.
This evaluation dashboard is quite an interesting tool that offers valuable insights into managerial decision-making. Hopefully, this tool can provide more context to fans around a pitching change for their favorite team and will result in less screaming at the TV.