Previous Data Challenges

CSAS 2026

Overview

Curling is a sport where two teams take turns sliding stones down a sheet of ice towards a target. The most common form of the game features teams with four players, with each player getting two shots per round, known in curling as an “end”. When one player shoots, two of their teammates follow the stone down the ice, sweeping its path with brooms to direct it to the desired position. Teams score by having the stone closest to the center of the target at the completion of the end, one point for each stone inside of the opponent’s closest stone. After a set number of ends (usually 8 or 10), the team with more points is the winner.

A primer on curling rules (both 4-person and mixed doubles) can be found here.

Mixed Doubles Curling presents a unique analytics opportunity described as the "Wild West of Curling" due to its distinctive format and strategic complexity. Unlike traditional curling, Mixed Doubles features:

  • 8 ends total.
  • 2 pre-placed stones.
  • Power play. Each team is allowed to opt for a “power play” once per game when they have the hammer. When a power play is being used, the pre-placed stones are moved out to one of the sides. See the primer for an image depicting the pre-placed stones during most ends, and the pre-placed stones when a power play is used.
  • Limited existing analytics - most available data comes from the Curlit website
  • High strategic variance - teams show great disparity in how they use power plays
  • Real-world application - coaches actively need these insights for upcoming competitions

The full rules of all forms of curling can be found here.

This Data Challenge focuses on power play optimization - when to use it, how to execute it, and how to defend against it - using data from international competition results that show final stone positions.

Questions to Consider

Since mixed-doubles curling is fairly new, there are a variety of topics that have not previously been studied. Below are a few example topics. However, we note that this list is far from exhaustive. Participants should feel free to study any aspect of mixed-doubles curling that interests them, provided that the analysis would be useful to a coach, curler, or some other member of the curling community.

  • Strategic Timing
    • When is the optimal time to deploy the power play? At what score differential and in which end should teams use this advantage?
    • How does power play effectiveness vary by opponent? Can strategies be customized based on the opposing team's tendencies?
  • Shot Selection & Execution
    • What are the most effective opening sequences? Should the first shot come around the guard to the open side, or are there better approaches?
    • How do the first 3 shots correlate with end outcomes?
    • What's the relationship between shot quality and scoring? How often does drawing under a guard result in zero points scored?
  • Comparative Analysis
    • Why do some teams score significantly more on power plays? For example, what makes Great Britain (Bruce Mowat) more successful than Italy on power plays?
    • How often can teams realistically expect to score 3+ points? What conditions make big scores possible?
  • Defensive Considerations
    • How should teams defend against power play attacks? What defensive setups minimize opponents' scoring opportunities?
    • Risk vs. reward analysis. When should teams "punt" an end versus fighting for a steal?

Student Deliverables Could Include:

  • Power play timing optimization model
  • Shot selection probability calculator (i.e., likelihood of scoring or giving up 0, 1, 2, 3, etc., points in the end based on the current stone positions and probability of success of upcoming shot options)
  • Team-specific strategy recommendations (based on considering the teams who will be competing at the Olympics)
  • Defensive positioning guidelines
  • Performance benchmarking across international teams

Available Data:

Shot-by-shot stone positions from Curlit are available on Github for the following competitions:

  • Beijing 2022 Olympic Winter Games
  • World Mixed Doubles Curling Championship 2023
  • World Mixed Doubles Curling Championship 2024
  • World Mixed Doubles Curling Championship 2025

More information can be found at CSAS 2026 Data Challenge.

CSAS 2025

Overview

For this data challenge, your goal is to use new baseball data on bat speed and swing length to analyze some aspect of the pitcher/batter interaction. We provide pitch-level data from Baseball Savant for 346,250 Major League Baseball plate appearances from 4/2/2024 to 6/30/2024, including relevant Statcast data along with bat speed and swing length on pitches with a swing tracked. Data from the second half of the season will be added after the conclusion of the regular season. Your analysis should involve bat speed and swing length to study any topic related to the batter, pitcher, or batter-pitcher interaction during an at bat.

Since these data are new, there are a variety of topics that have not previously been studied. Below are a few example topics. However, we note that this list is far from exhaustive. Participants should feel free to study any aspect of the batter, pitcher, or batter-pitcher interaction that interests them, provided that bat speed and swing length are used in the analysis in some meaningful way.

  • Batter Specific Plate Discipline. Can bat speed and swing length be used to measure some aspect of plate discipline?

  • Are these data useful in analyzing a batter’s decision to swing at a pitch? For example, can faster swingers be more patient?

  • Are bat speed or swing length related to a hitter’s ability to foul off 2-strike pitches, or protect the plate?

  • How is pitch location related to the above ideas? See <https://baseballsavant.mlb.com/visuals/swi ng-take?playerId=545361> which may give you some ideas.

  • Do players have different types of swings? Are these swing types related to situation (count, base runners, outs), pitcher, pitch location, pitch type, or anything else?

  • Do pitchers modify batter behavior or do batters modify pitcher behavior (or neither! Or both!!):

    • Can pitchers “dictate” swings, or is it all the batter? Are swing lengths longer against certain pitchers? To what extent is that due to pitch velocity or spin rate, as opposed to other pitcher-specific factors?

  • Do pitchers modify behavior against batters with higher/lower bat speed (e.g. throw to different locations, throw different types of pitches, etc.)?

Data

The data is in this shared folder. We recently added a file that has data for the whole season (full regular season and playoffs), and has the column arm_angle included in the data along with some additional contextual variables and other information. The new columns were added to the right of the data frame. The new columns are estimated_slg_using_speedangle through arm_angle. See the file statcast_pitch_swing_data_20240402_2024103 0_with_arm_angle.csv.

More information can be found on the CSAS 2025 Data Challenge Page.