Identify triangle plays
The project focuses on plays where multiple fielders are potential candidates to catch the same ball, making communication and positioning central to what happens next.
← Back to Projects
SMT 2026 · Sports Analytics
Triangle Tribulations
Modeling which fielder is most likely to catch a baseball that lands between several players—and turning the results into an informative, entertaining replay.
Overview
A “Bermuda Triangle” play happens when several fielders converge on a pop-up, yet the ball falls between them. Traditional catch probability can help describe how difficult a catch was, but these plays raise another question: which nearby player was most likely to make it?
I developed a Triangle Catch Probability (TCP) metric to compare fielders on the same play. I also created a proof-of-concept animation using anonymized SMT data, pairing the probabilities with player headshots to make the play easier to follow.
Methodology
Model the fielders together
I used a mixed conditional logit model to compare players within a play. The approach accommodates characteristics that differ by fielder, such as position and running direction, alongside play-level characteristics shared by all fielders, such as the ball’s exit velocity.
Make the results visual
The proof-of-concept animation attaches TCP values to anonymous player headshots, giving viewers a way to see how the model compares the fielders involved in a play.
Key Takeaways
- TCP offers a way to compare the fielders involved in the same pop-up, rather than describing only the difficulty of the catch overall.
- Combining the metric with a replay could help fans understand how positioning and miscommunication contribute to a memorable blooper.
- The current model does not estimate the probability that the ball will be dropped. That is an important limitation—not every missed catch can be explained by comparing fielders’ probabilities alone.
Impact & Next Steps
The proposed application is a customizable game-day blooper reel for Minor League Baseball teams: entertaining replays accompanied by a metric that gives fans more context about the play.
A future version could estimate the chance of the ball being dropped. The paper also proposes exploring whether a sufficiently high catch probability could help inform how errors are assigned. These are future possibilities, not features of the current model.
Skills Demonstrated
Sports Analytics
Statistical Modeling
Mixed Conditional Logit Models
Baseball Data
Data Visualization
Animation
R