By Andrew Smith (@asmitty09)
As part of the US Soccer hackathon in Chicago last month, our team of Benjamin Harrison, Kevin Minkus, Eliot McKinley, Andrew Crago, and I developed a player decision model to evaluate final third decision making (Editor's note: Andrew is being humble and left out one important point: THEY WON THE HACKATHON. Here's a link to the project). The model uses Gaussian process classification to estimate the decision a player would make—either to pass, shoot, or attempt to take on a defender—at a specific point in the final third. Combining the players on the same team, we can extend this model to estimate the decisions the team as a whole makes in the attacking third. Then, to further understand team decision making in different game situations, the data is stratified into three categories to look at the adjustments in play when behind, tied, or ahead.
There’s not a whole lot that jumps out in an initial glance at the plots—there are no clear similarities between teams at the top or bottom of the league. There is, however, one specific team that stands out. Atlanta is extremely consistent—their attacking strategy is virtually identical regardless of the score. In the often volatile arena of professional sports consistency usually correlates with success, and Atlanta’s 2018 campaign is no exception. But ultimately it’s hard to say whether this consistency is a source or byproduct of their successes this year. Winning tends to alleviate a lot of the pressure on a team, which presumably would lead to more consistent performances, or it’s entirely possible that Atlanta is simply executing well.
One factor that affects their decision-making is that Atlanta, as the current league frontrunner, face heavily packed-in defenses in most of their matchups. Teams are effectively playing Atlanta not to lose, meaning that the defenses Atlanta sees when tied are comparable to what they see when losing. This would explain some of the similarity between the tied and behind plots.
Note: yellow means a higher probability, purple a lower probability. The dashed red line indicates that the probability is 50% at that contour.
Delving into the actual decisions, Atlanta players choose to take on defenders near the sides of the box at a slightly higher rate when up, which makes intuitive sense. Attempting to dribble through the defense is usually the riskiest of the three actions, and so would be performed more often with the lead, especially when the rest of the eleven have probably dropped back somewhat to bolster the defense and hold that lead. Then presumably, Miguel Almirón, Josef Martínez or any other attacking Atlanta player is left to their own devices, and so with less options choose to dribble. We can also note that in comparison to most other teams, Atlanta seems to pass more often when outside the box. This is probably smart, as shots from long range aren’t converted very often. Putting these factors together, Atlanta seems to have the most disciplined team in the league this year, a testament to Tata Martino and his staff.
Looking at Atlanta’s competition, NYCFC are an interesting study. In general, they have a similar distribution to Atlanta, where both dribble a bit more when ahead. However, NYCFC seems to actually shoot less when behind. Without more information, it’s hard to know if this is a result of more conservative offensive strategy, packed boxes, or just simply players cracking under pressure and refusing to shoot. Obviously this isn’t the only difference between NYCFC and Atlanta, but nonetheless it’s interesting to note.
Jumping to the Western Conference, FC Dallas appears to employ a similar strategy to NYCFC. Passing slightly more when behind and taking more risks when ahead, although Dallas seems to shoot rather than dribble.
Taking a trip to the bottom of the league, it’s clear that there isn’t a silver bullet as to why these teams are struggling this year (at least tactically). Some teams, like San Jose and DC, seem to be dribbling too much near the box, whereas Colorado looks to be blasting shots from too far away when down. The Rapids aren’t likely to be changing this anytime soon with the addition of Kellyn Acosta, who also has a propensity to shoot from deep. As for San Jose, their lack of passing from deep potentially could be remedied by Jackson Yueill, but he’d have to actually get significant playing time for that to happen. Click any of the images below to enlarge them.
At the end of the day, these plots don’t exactly take personnel into account and so it’s difficult to know with certainty whether their attacking third decisions are actively bad or simply a result of making the best of what they’ve been given. Presumably teams would like their players to play to their strengths, and so looking at these decision model plots on a player by player basis might give more insight into the overall team performance and would be an interesting follow up piece.
If there is a take home message from these plots, it’s simply that until someone makes Atlanta change up their game I suspect they will continue to ride atop the league table.