Note: This work was presented at the OptaPro Soccer Conference in Chicago earlier this January. Full video of our presentation is at the end of this post.
Michael Bradley and Wil Trapp share several obvious qualities. They are both captains for club and country. They are both smooth passing defensive midfielders, and they both possess excellent heads of hair. Another similarity is that they rarely shoot or score goals, each collecting only one goal over the last three seasons. Coincidentally, both of those goals are what we could enthusiastically describe as "wonder-goals." Bradley's long-distance chip for the US national team in a World Cup qualifier against Mexico at the Azteca (a goal not remembered as fondly as it deserves due to the rest of qualifying) and Trapp for the Crew to win a match in stoppage time against Orlando City this past summer. However, one difference between these two players was how each responded to the confidence boost that came after scoring a once-in-a-career goal.
A few weeks before Trapp's "monumental goal of the people, by the people, and for the people of Columbus, Ohio," some fellow American Soccer Analysis contributors and I traveled to Chicago to participate in the inaugural US Soccer Hackathon. Over 24 hours, we developed a framework to model player decision making in the final third. Using a couple of different methods (you can find them on Github), we were able to measure the propensity for a player or team to shoot as well as produce maps of where a pass, shot, or dribble is most likely to occur (check out Andrew Smith's article on Atlanta's ruthless attack).
Back to Bradley and Trapp. Before scoring his goal at the Azteca, Michael Bradley was overwhelmingly likely to pass the ball almost everywhere in the final third. There was a small area around the top of the 18 where he was slightly more likely to shoot, but the likelihood of a shot was still less than 10%. Trapp's shot map was similar before his own long-range goal, although his small zone of shooting was slightly farther from goal and he was still less likely to shoot from there than Bradley. The differences become much more stark following their respective goals. When Bradley returned to MLS after scoring his goal for the national team, he became about three times more likely to shoot from the top of the box when given the opportunity. Trapp however, hardly changed his patterns at all, nearly always opting to pass in the final third. This tendency becomes even more apparent when subtracting the players' post-goal maps from their pre-goal maps. Bradley's increased shooting stands out as a bright red dot, while Trapp's changes are minimal and diffused around the final third. Finally, using our expected shot model, we can see that Bradley, a player that took around 0.25 shots per game less than expected, became a player that shot around league average. This increase was equivalent to Bradley taking an extra shot every four games. Trapp, however, barely changed his behaviors, taking only an additional shot every 20 games.
In combination, these two methods can help describe what decisions players make when entering the final third and how those decisions may change over time. In this case, Michael Bradley became more likely to shoot after scoring a wonder goal, while Wil Trapp did not.
It is not difficult to imagine that following a long-range goal, a player may gain the confidence to try it again. While this may be seen this as a weakness, it is not necessarily so. Before his goal, Bradley was a lower likelihood shooter in the final third, and even when he increased his shooting it was only up to the expected level league wide. Perhaps Bradley should have been shooting more from long range when given the option, and scoring his goal at the Azteca gave him the confidence to do so. Trapp did not see a significant increase in shooting propensity, although he was also already close to the league average. Trapp also had a very defined role in Gregg Berhalter's system in Columbus, which did not include shooting from long range.
Metrics such as these can give insights into player behavior which could find potential use in several ways. An obvious example is opposition scouting. Knowing where a player is likely to shoot or pass, allows a team to plan for these events. Additionally, if a player is making decisions that a manager may want to change, these maps can bring those anomalies to light.
Watch the video of our presentation below to see a few more examples of the potential uses of decision making models in the final third.