Man In Motion: Classifying NFL Pre-Snap Movement
- Etienne Busnel
- Mar 25
- 4 min read
By: Etienne Busnel
Introduction
Imagine you are a cornerback in the NFL. Playing arguably one of the hardest positions in all of sports, this down you are tasked with guarding 4.29 speed Tyreek Hill. Only they hold the knowledge of what path they will take. Play off too much, and they can get easy underneath yardage. Play too close, and you could give up a massive gain deep and be the disastrous clip of the week. When a receiver makes a move in their route, you have no margin of error when deciding to commit or not. Every second the play extends on are more opportunity for you to get beat.
Then, before the ball is even snapped, he starts to move.
Pre-snap motion has exploded in popularity across the NFL. Offensive Coordinators, always looking for new ways to make defenses' lives as hard as possible, have designed players to move before the snap more than ever, with the goal of confusing defenses. The results have proven affective: when comparing how many Expected Points a play added compared to its expectation (EPA) in the first half of 2023, plays where at least one player went into motion earned an average EPA of 0.083, while plays without motion saw a -0.075 EPA, performing worse than expected.
Looking on a team-by-team basis tells the same story. Teams with higher EPAs on average used more motion than teams that did not.
While motion as a whole has proven effective at getting offenses improved results, this does not mean that any specific motion manages to confound defenses. To get a better view of specific motions, we need to look at them individually.
Categorizing Motions
The advancement of NFL player tracking technology has enabled teams to access player coordinates at every tenth of a second for every play. This level of detailed data allows analysts to study motion patterns with unprecedented precision, uncovering insights and making visualizations that were previously impossible to create. This data allows for easier classification of NFL motions.
To classify the various motions observed during weeks 1-9 of the 2022 NFL season, we utilize K-Means Clustering. This algorithm is an effective tool for organizing data when predefined labels are unavailable. The user specifies the desired number of categories, and the algorithm identifies optimal central points, grouping data based on proximity to these centers.
To categorize the NFL motions provided, the following features were selected:
Player's Initial Location - The starting position of the player before motion (relative to the football).
Player's Location at the Snap - The position of the player when the ball was snapped.
Distance Between Start and End Points - Measured both vertically and horizontally.
Movement at the Snap - Whether the player was in motion at the time of the snap. If not, the duration the player remained stationary.
Field-Side Movement - Whether the player crossed to the opposite side of the field.
Direction Changes - Whether the player changed direction during the motion.
Movement Range - Total distance traveled by the player in all four directions.
Relative Position to Tackles - The player's horizontal alignment relative to the offensive tackles.
Shifts - Whether the player registered a shift.
Speed During Motion - The player’s speed while in motion.
To improve clustering accuracy, Principal Component Analysis was applied prior to classification. By analyzing inertia and silhouette scores across different cluster counts, along with manual review, 24 distinct motion categories were identified. Below, you'll find visual representations and written descriptions of each category.
Looking at the rush percentage for every motion type reveals trends about how motion indicates the likelihood of a rush. The overall trend showed that plays where motion occurred were more likely to be passes than those that were not, with only three motion categories having higher rush percentages than plays without motion. Interestingly, the motions that led to more or similar rushing rates had a common theme: they involved a running back moving in the backfield or a wide receiver shifting closer to the line. This makes sense—bringing a receiver in tight adds an extra blocker, while running back movement often indicates direct involvement in the play. Similarly, plays where a running back moved out to a wideout or a receiver moved further outside were almost always passes.
When analyzing the EPA for each motion type, all but three produced better results than plays without motion, signifying that motion across the board, and not just a specific subset, has been effective at exploiting defenses. The most effective motions typically were by involved a wide receiver moving closer to the line, suggesting that these shifts help create more favorable offensive situations. These types of motions also tended to have higher rush percentages, signifying that while motion has been used more in the passing game, its impact on the running game is potentially more significant.
Conclusion
Motion has become a key component of NFL offenses, with teams that utilize it finding greater success. Breaking down different motion types reveals distinct trends, often indicating whether a play is more likely to be a run or a pass. It also highlighted that while motion is primarily used in the passing game, it has had arguably a more significant impact in improving play success in the running game. Overall, motion has proven to be more effective than non-motion plays, reinforcing its value in modern offensive schemes.
Future Work
Expanding the dataset to include more weeks of data would improve the accuracy of rush percentages and EPA per play, as larger sample sizes provide a clearer picture. Another key area for further analysis is how the effectiveness of motion has evolved in recent years. While motion was still gaining traction in early 2022, defensive coordinators have since had more time to adjust, potentially reducing its impact. Additionally, studying whether certain player archetypes excel in specific motion types could offer valuable insights. For example, a physical wide receiver known for blocking might be more effective when motioning closer to the line than a speedy deep threat.
Access All Code Here:
Comments