Methods

Our goal with this project was to develop a “Contested Possession Metric” — a composite statistic designed to reflect a player's in-game contributions that either create turnovers or secure possession.

In homage to club legend Brian Schmetzer, we started with a stat he frequently emphasizes: duels won. An aerial duel is defined as a challenge between two players to gain control of the ball above elbow height (typically with the head). We assigned the value for an aerial duel won at +1.

A tackle in soccer represents when a player challenges their opponent with their feet for a ball on the ground (or at least below the elbow), a game event markedly similar to an aerial duel won and often yielding the same results. As such, we also valued tackles won at +1.

To account for aerial duels lost we assigned a value of -0.75. This weighting ensures that an aerial duel lost does not fully offset an aerial duel won. Players with a 50/50 split are effectively credited for actively contesting possession, even when they don't win the ball. (Unfortunately, tackles lost are unfortunately not available in the source data. More in for below.)

After aerial duels and tackles won, the next most valued statistic for this metric is interceptions. An interception occurs when a player gains possession of the ball after an opponent passes, crosses, or shoots before the ball can reach the intended target. While this game event is not a result of a direct physical challenge for the ball, the player creates a turnover through positioning, reading game flow, and anticipating ball movement, as such interceptions were valued at +0.75.

The final statistic weighted by the algorithm is recoveries. A recovery represents any action that ends a spell of possession by the opposition and starts possession for the player's team. Again, while not a direct challenge to win the ball, recoveries were valued at +0.5 because the result of a recovery (whether the product of an error from the opposition or efforts by the player) is a turnover by the opponent and an opportunity in a new phase of play.

We chose these weights to prioritize direct challenges (duels and tackles) while still recognizing off-ball efforts (interceptions and recoveries) that affect possession dynamics. The final Schmetzer Score is the sum of the weighted values for each statistic.

In the source data a player was listed twice if they played for multiple teams in a season (this could be the result a number of scenarios including contract terms, inter-league trades or loans within the league). This could effectively lower the rank of a player as their statistics would be distributed across multiple teams. To capture the full picture of a player's performance over the season, a player's statistics are consolidated to the squad with which the player played more minutes (i.e. higher value in nineties.)

When tabulating the season average of any given statistic, we decided to exclude players with less than 90 minutes in the season. Though all players are ranked, regardless of how many minutes they played. This decision was made to ensure that the season average is a more accurate reflection of a the average for players who played in a particular season.

Data Source, Limitations, & Further Development

Our source data came from the amazing folks at FBref  (the source data set for this project) and Sports Reference. They're doing incredible work, democratizing sports data by making it publicly available. One current limitation is the granularity of available statistics. While providers like Wyscout break down duels into 4 types (offensive duels, defensive duels, loose ball duels, and aerial duels) FBRef lists only aerial duels won or lost as well as tackles won. Tackles lost, as well as duel types beyond aerial, are not available in this dataset.

While this is a limitation of the current iteration of this application, fortunately the data architecture is modular and the ETL pipelines could be pointed at a new data source in the future. This data could also be combined with field position to ascertain if a player tends to win or lose possession in specific zones on the field or during certain patterns of play. We also have under development fine tuning controls which would allow a user to manipulate the default values which weight the Schmetzer Score.

Read more about the inspiration that sparked this project on the About page.
Meet the two developers behind this project on the Team page.