In a recent-ish podcast by The Transfer Flow podcast, host Ravi Ramineni made an off hand comment that while working for the Seattle Sounders, the team had noticed that they could switch out up to three of their regular starters for bench players in a given game without causing too much of a problem. Any more than that and they began to run into trouble as a team.

This idea has been covered before - see American Soccer Analysis’ prior article on the Curtin Theorem by Eliot McKinley and Mike Imburgio - but the specific mechanism in question remains a bit of a mystery.

The connection between players and their level of familiarity with each other as a whole is a very important piece of the decision making of coaches, but it’s hard to quantify in a meaningful way. There’s also usually a tension between the coaches’ lineup choices and the desires of the fans, who often see the players on the bench as more exciting. Is the emphasis on familiarity and connection that important? Can we measure that on-field chemistry?

Ramineni’s approach was looking at starters vs depth, and in doing so requires a somewhat in-depth knowledge of the internal structures of a team. In lieu of in depth knowledge there is a metric that can be calculated quickly and compared between teams in the same league.

Adding up the running total of passes between players over the course of their careers can create a connection “score”. This is easy to calculate, and requires relatively few resources to do so. It also inherently accounts for the importance of player proximity. Players generally pass to each other more if they are close together. The level of importance of the connection between, for example, a right winger and a left full back is probably much less than the connection between a center back pairing.

In theory, the more connected a team is, the better they will perform. But is that the case? In order to test the variable, game level scores were created by taking the game level connection score of each starting lineup. To get a “connection difference”, the opponent’s connection score could be subtracted from the team’s score. The difference could then be compared against the expected goals to see what impact, if any, connection had on the outcome of the game.

Is there anything here?

To explore the relationship between pass differential and expected goals, two linear regression models were developed: one predicting xG for a team and the other predicting xG against. The models included three predictors:

The difference in the number of passes between the two teams, scaled for comparability.

The estimated difference in player quality, based on Goals Added, also scaled.

The difference in total minutes played between players on each team

Each model was fit to a filtered dataset of MLS games, using xG for or xG against as the dependent variable. By including multiple predictors, the models isolated the effect of pass differential while controlling for the influence of player quality and minutes played.

The results showed that cumulative pass differential is a statistically significant predictor in both models. In the xG for model, a higher pass differential is associated with an increase in xG, indicating that teams with better cohesion generate more scoring opportunities.

In the xG against model, a higher pass differential is associated with a decrease in xG against, suggesting that cohesion also has a role in preventing chances. This is especially interesting because passes are a solely offensive stat. (Or are they…?)

Both effects remain significant even after accounting for player quality differences and overall time on the field, both potentially confounding factors. While the overall explanatory power of the models is modest - don’t bet your house on it - there is a clear relationship between team chemistry, as measured by shared passes, and winning soccer games.

Despite best efforts to control for outside variables, there are still some potential flaws in the model. First, teams that play more possession heavy styles could potentially artificially inflate their connection totals due to sheer volume of opportunity. (Riqui Puig is flying up the charts) While the concern is valid, possession generally does correlate with team success and in MLS the difference between teams’ possession statistics are less pronounced than in other leagues around the globe.

There is also a potential for survivorship bias even controlling for player quality. Teams that are successful will continue to play the lineups that brought success, increasing the connection score. Teams that are unsuccessful over the course of a season will rebuild, sometimes midway through the year, and their connections scores will continue to remain low. Does the score predict their future success or simply reflect past success that will lead to future success? This may be a question for further research, but for the time being the metric does reveal something of value.

Areas of Interest

With the model built, the way different teams are built up can be examined in more detail. Taking a more detailed look at the data can give us a picture of how teams’ rosters have changed over time and how they have built and dismantled the connectedness of their roster over the years.

The Seattle Sounders are a good example of a team that has had an impressive level of connection over time, and managed to maintain that even through a challenging period where many club legends aged and were replaced by younger players. Since the 2020 season the average team has had roughly 7000 passes between their starting lineup in a given game. The 75th percentile of teams have hit 9200. Seattle’s average as an organization since 2013 is 9598.

When viewing these, the line colour is the coach in charge, and the dot colour is the xGD of that performance

That’s remarkable stability, and it’s notable that when they do choose to rebuild (see the 2021 season where they took a dip) it does not take long before the team is back to being well above the average for the league.The Sounders just completed big money deals for Jesus Ferreira and Paul Arriola; big name players that just happen to have played on the same team for a significant amount the last few seasons. There are more rebuilds coming for the Sounders, but if they navigate them like they have in the past they should continue to perform.

On the other end of the spectrum is the Chicago Fire. This is a team that has been through rebuild after underwhelming rebuild. They’ve made a big move to bring in former USMNT manager Gregg Berhalter to be both head of soccer operations and manager of the first team. Berhalter will have all the tools at his disposal to help fix the disaster that has been the last decade in Chicago.

Berhalter was able to stabilize the Columbus Crew and build a strong core, something that his successor Caleb Porter was not able to do at a similar level. Chicago should be able to offer him the resources, but the question is whether or not he can build a team that can have both the cohesion and quality to make a difference in the games that matter. With the USMNT, he certainly built a stable and cohesive team; they just could not win games when it counted.

If you’re curious about where your team historically stands, you can view the chart below:

Implication and uses of the model

Having modeled player connection, the question now is what can be done with that model. What are the potential edges that can be found by a coaching staff with the measure?

The most obvious would come in squad succession planning. There is no avoiding the fact that time marches on, players get older, and teams are constantly in a state of turnover. Do you replace a successful team wholesale, stripping it down and start over? Or is there a way to manage that transition carefully, never risking dropping below a certain level of experience? Both approaches are valid; if a major fixture is retiring, it might also be the time to move on from his two or three closest teammates and build a new core.

Another important question is the integration of young players into the team. It’s difficult to know exactly when to take the risk involved with playing younger players, and when a more steady hand is necessary. Understanding how to effectively ease a player in and build connection with the first team squad can prepare teams for late season fixture congestion and potential injury crises. Having confidence that a player can play their role in their squad effectively can be a comfortable position for a manager.

Finally, having an understanding of what other teams are doing can help managers determine when and how to take risks in their player selection. A team that does not have as much cohesion could provide an opportunity to bring in a wider variety of players and build cohesive depth throughout the roster, with an eye towards building to the future.