Make prediction-based decisions with sport performance analytics
This is an excerpt from Sport Performance Analytic Methods With HKPropel Access by John R. Todorovich.
As discussed in chapter 6, the ability to make predictions about future behaviors is a very useful tool for SPA analysts and decision makers. Although each predictive SPA technique includes some degree of unreliability, it is still possible to create strong predictive models based on prior data. In short, patterns of behavior and performance tend to follow past performance. It is the discovery of these patterns and how they are applied in different situations and contexts that makes prediction-based decisions valuable to coaches and other decision makers. It should be clarified that prediction-based SPA is not based on intuition or guesses but on patterns determined through robust collection and analysis of data. Interestingly, even the inclination to change behavior often produces patterns that, given enough data, can produce strong predictive models.
As an application of the SPA model to prediction-based decision making, consider the following scenario.
Step 1: Establish What You Want to Know
An American football coach is preparing for the big game against a rival team. Every year these teams face each other, and both coaching staffs have spent a lot of time reviewing the games of their opponent to look for the best way to win. The coach asks an analyst, “Is there any way to reasonably predict when the other team will pass or run the ball?” The analyst responds that a prediction model can be created based on past performance data.
Step 2: Define the Data You Will Collect
After much discussion with the coaching staff, the analyst learns that the following variables seem to influence the other team’s decision to pass or run the football: the location of the ball on the field, the down and distance to first down, the players on the field, and the amount of time left in the game.
Step 3: Determine the Data Collection Process and Collect Data
Since the rivalry game is near the end of the season, the opponent has played several previous games. The analyst reviews each game and records data related to each variable (location of the ball on the field, down and distance to first down, players on the field, amount of time left in the game). For each offensive play, each of these variables is compared against whether the team’s next offensive play was a pass or a run.
Step 4: Analyze the Data
Because the analytic question asked to differentiate between pass or run only, the analyst performs a logistic regression procedure. To generate a strong predictive model, the analyst adds and removes variables to determine the best predictors for whether the next play would be a run or a pass.
Step 5: Interpret the Results
Using the data, the analyst determines that where the ball is on the field, the down and distance to first down, and the amount of time left in the game were the strongest combination of variables to make a reasonable prediction about whether the next play would be a pass or a run.
Step 6: Present the Results
Prior to the rivalry game, the analyst, head coach, and defensive coordinator meet and discuss the model. The coaches are comfortable using the data produced from the model. They request that, using headsets, the analyst will verbally communicate pass or run to the coaches depending on the prediction model analysis.
Step 7: Make Data-Based Decisions
The defensive coordinator decides to call defensive plays based on the data presented in the model. If the model predicts a run play, the defensive coordinator will call for a run defense. If the model predicts a pass play, a pass defensive scheme will be used.
Get the latest insights with regular newsletters, plus periodic product information and special insider offers.