Using Expected Threat to Find the Best Shot-Creators in MLS / by Arjun Balaraman

by Arjun Balaraman (@arjun_balaraman)

Growing up as a Manchester United fan, I have been fortunate enough to have seen some brilliant attacking players, but amidst a bevy of exciting talent one man stood out as a different kind of goalscorer – a comparatively unheralded striker from Mexico: Javier Hernandez – more commonly known as Chicharito.  

The fascinating part about Hernandez’s game was how he scored those goals. Chicharito is the ultimate goal poacher, a real throwback to the earlier decades when strikers were meant to sit in the box and score goals – nothing else. While Ronaldo et al., often created their goal scoring opportunities with their ability to move with the ball, Chicharito’s skill was moving without the ball. With his ability to, somehow, always be in the right place at the right time, Hernandez has made a career of tap-ins and intuitive finishes. In fact, all 52 of his goals in the Premier League have come from inside the box.

His prowess in the box is exemplified in this goal he scored in the Champions League against Chelsea back in 2011. On this occasion, Hernandez, is lurking in and around the six-yard box as United build the attack. As winger Ryan Giggs receives a pass from defender John O’Shea, Chicharito makes his move – a sharp dart to the back post. Giggs’ pass finds him perfectly in stride and the Mexican tapped home to give United the lead.

Chicharito Goal.gif

Contrast that goal with this marvelous individual effort scored by Houston’s Mauro Manotas during the Dynamo’s 3-1 victory over Toronto last week. Manotas picked up the ball just inside the Toronto half with multiple defenders between him and the goal. Seconds later, after a mesmerizing run, in which the Colombian took virtually four Toronto defenders out of the game on his own, he put the ball in the back of the net.  

Manotas Goal.gif

The Dynamo striker’s goal was a perfect example of a player creating a shot for himself. After receiving the ball in an area of relatively low danger near midfield, Manotas carried the ball to a much more threatening position at the top of the box before firing off his shot.

The score books would credit a goal each to Hernandez and Manotas, but they couldn’t have come in more different ways. Hernandez depended on the creativity of his teammates to score whereas Mazotas created his own opportunity.

Without necessarily taking a judgment on which is a “better” goal, I wanted to identify and credit players like Manotas that can create their own chances. As such, I decided to have a go at profiling some of the strikers/attacking players in MLS by determining what percentage of the final shot in the possession the shooter created for himself. This metric is aimed to acknowledge players for taking the ball into dangerous areas and spotlight players who are good at moving with the ball in the opposition half (for example, Manotas would have received 87% of the shot-creating credit for his goal against Toronto, according to my metric).

To measure the value of the ball at a certain time during a possession, I elected to use Karun Singh’s Expected Threat (xT) model. He originally used data from the 2017-18 English Premier League to construct the model, and while no one has yet to scientifically test whether or not xT can be generalized across leagues, I am operating under the assumption that the value of having the ball in a certain zone in MLS is very similar to the value of having the ball in the same spot in England. Here is a short excerpt from his blog on deriving the original xT model, which has since been tweaked slightly, and the link to his original post.

Since the time of his post, some slight modifications have been made to the model. Karun incorporated pass completion rates so the first expression shown above is now also multiplied by the probability that a pass is completed from zone (x,y) to zone (z,w). Including this helps us factor in pass difficulty, which was previously not accounted for. He also changed the dimensions of the zones, so field is broken up into a 12x8 grid with 96 zones rather than the original 192.  

As Karun mentions later in his post, this final metric requires a bit more work because, as noted above, the calculation for the xT value of a zone requires previously knowing xT values for all of the other zones. He solved this problem by first setting the initial xT = 0 for all of the zones and then evaluating the formula iteratively until convergence.   

Using the work Karun did to create the xT map with values for all of the zones, I first focused on finding the value of a player carrying the ball. During a possession, I attributed the change in the xT value of when a player receives the ball to when he releases the ball as the value of the ball carry for the player. Since no one else touches the ball during that time, the change in xT can only be attributed to the player with the ball at his feet, thus denoting the value of the carry.  

There are only two ways in which a player can ‘receive’ the ball. Either he has received a pass from one of his teammates, or he has won the ball back from the other team with a successful defensive action (interception or tackle won).  

When the shooter’s team already has the ball and the player is simply receiving a pass from one of his teammates before shooting, the calculation is simple. Using the ending coordinates of the pass from a player’s teammate, we are able to assign an xT value at the time the player receives the ball. Then, by subtracting that value from the xT at the time of the shot, we’re able to assign a shotcreating.xT value to each and every attempt by a player.  

The second scenario, where a player wins the ball back from the opposition and, without passing it off to one of his teammates, goes on to take a shot, is far less common – but with the increased pressing in the modern game, defenders have become more prone to trying difficult passes in dangerous areas, often surrendering possession in and around their own box.

Top 50 MLS Players in xT
Player Team Position ShotCreatingSum ShootingxTsum PercentCreated
Carlos Vela LAFC F 1.58 7.93 19.87
Josef Martinez ATL F 1.03 8.19 12.60
Aleksandar Katai CHI A 0.91 4.67 19.42
Darwin Quintero MIN A 0.75 4.62 16.26
Cristian Penilla NER A 0.72 2.55 28.13
Jefferson Savarino RSL A 0.66 2.10 31.49
Cristian Espinoza SJE A 0.66 2.78 23.73
Kacper Przybylko PHI F 0.58 5.68 10.24
David Accam PHI-CLB A 0.58 1.72 33.35
Alexandru Mitrita NYC A 0.56 2.76 20.16
Raul Ruidiaz SEA F 0.56 4.64 11.97
Fabrice-Jean Picault PHI F 0.51 5.15 9.99
Sebastian Blanco POR A 0.50 3.78 13.29
Alberth Elis HOU A 0.50 6.04 8.20
Zlatan Ibrahimovic LAG F 0.49 8.94 5.46
Romell Quioto HOU A 0.49 1.62 29.92
Victor Rodriguez SEA A 0.47 1.83 25.49
Joaquin Ardaiz VAN F 0.46 1.13 40.18
Uriel Antuna LAG A 0.45 2.59 17.39
Nicolas Mezquida COL A 0.45 3.08 14.47
Johnny Russell SKC F 0.44 3.12 13.99
Angelo Rodriguez MIN F 0.43 4.98 8.57
Christian Ramirez LAFC F 0.42 3.61 11.77
Gonzalo Martinez ATL A 0.42 2.52 16.85
Hector Villalba ATL A 0.42 1.86 22.72
Valeri Qazaishvili SJE A 0.42 4.54 9.24
Robinho CLB-ORL A 0.42 2.41 17.34
Diego Rossi LAFC F 0.40 5.14 7.81
Shea Salinas SJE A 0.40 2.32 17.16
Danny Hoesen SJE F 0.40 4.02 9.87
Chris Mueller ORL F 0.39 2.43 16.17
Brian Fernandez POR F 0.39 2.96 13.14
Tomas Martinez HOU A 0.38 2.12 17.88
Gerso SKC F 0.37 1.97 18.79
Paul Arriola DCU A 0.37 1.88 19.52
Daniel Royer NYRB A 0.37 4.05 9.01
Alexander Ring NYC M 0.37 1.45 25.24
Mauro Manotas HOU F 0.36 6.00 6.05
Latif Blessing LAFC M 0.36 2.28 15.79
Gyasi Zardes CLB F 0.36 5.74 6.23
Kekuta Manneh CIN A 0.35 1.43 24.53
Maximiliano Moralez NYC F 0.35 3.14 11.00
C.J. Sapong CHI A 0.35 6.66 5.18
Nicolas Gaitan CHI A 0.34 2.11 16.25
Teal Bunbury NER F 0.34 4.15 8.13
Darren Mattocks CIN F 0.33 1.87 17.56
Anthony Jackson-Hamel MTL F 0.32 1.66 19.15
Brian White NYRB F 0.32 3.11 10.16
Juan Fernando Caicedo NER F 0.31 3.13 9.93
Diego Valeri POR A 0.31 3.03 10.23

 When a striker or an attacking player wins the ball back, I chose to assign the value of the ball at the time of the change in possession as the value of the press since it pertains to winning the ball back. Similar to the methodology for when a team is in possession, I then took the difference in the xT at the time of the shot and at the time of the change in possession and marked it as the shotcreating.xT value.

 Some may argue that since the shooter’s team never had possession of the ball prior to him winning it back, he, for all practical purposes, ‘created’ the entirety of the xT. While that is a fair argument, I wanted to focus predominantly on the change in value when a player’s team is already in possession. Adding the impact of winning the ball back via a high press, while very important in a game context, would have diluted the main purpose of this analysis. 

By then summing up the shotcreating.xT (and calling it ShotCreatingSum) values for each player in MLS and dividing that by the sum of the xT at the time of a players’ shots (ShootingxTsum), I was able to find the percentage of a player’s shooting xT (PercentCreated) that he created for himself. To the right is a table of the top 50 players in the league so far this season based on the amount of xT they created for themselves prior to a shot. 

There is a bit of a sample size issue in the percentCreated metric as many of the high-ranking players have taken below 20 shots in the league this season which, statistically speaking, is less than you’d like for such a metric.. It is for this reason, I have preferred to sort the table by the ShotCreatingSum, to highlight players that are creating the best chances for themselves on a consistent basis.  

The one name that, as has so often been the case this season, stands out is Carlos Vela. While Zlatan may beg to differ, the LAFC striker is not only the most prolific goal scorer in the league, but also one of the most creative and leads the league in ShotCreatingSum – meaning he has fashioned more quality chances for himself than any other player.  

So far this season Vela has created over 19% of his final shooting xT, which might not seem like a lot, but compared to other players whose xG return on the season is above 3, Vela ranks in the 98th percentile. Given the high volume with which he takes shots, the Mexican is exceptional in taking on defenders and creating scoring opportunities for himself.

As evidenced by the graph above, the top goal-scorers for 2019 (or expected goal scorers, at least) from MLS thrive in different ways. At this point in his career, Wayne Rooney isn’t the kind of athlete he was back in his heyday at Manchester United, and often relies on his teammates to create a lot of his final shooting threat.  

Similarly, Chris Wondolowski has garnered a reputation as a poaching goal scorer during his time at San Jose, and that’s shown by his low percentcreated value. Wondo, much like Hernandez in a way, isn’t a modern, dynamic player and instead excels at tucking away chances that others create.  

Behind Vela on the table are Chicago’s Alexsander Katai and Atlanta’s Josef Martinez. While Katai has been relegated to the bench recently during the Fire’s dismal run, the Serbian has been a statistical bright spot for Chicago. Katai is one of the few players on the team capable of creating for himself – a fact that is further corroborated by data on his impressive dribbling skills. However, his eight combined goals and assists is nearly 4.1 short of his 12.1 xG + xA, so while Katai is getting the ball into dangerous areas on a consistent basis, he is still underperforming in terms of his final output.

In the same light, NYCFC’s Alexandru Mitrita stands out on the graph as one of the most prolific players in terms of creating for himself. As Dummy Run wonderfully detailed for The Outfield, Mitrita is a superb dribbler and shooter although he is lacking in many other areas of the game. It’s not surprising that he ranks in the top 10 for both ShotCreatingSum and percentcreated.  

While the ability to create a shot for oneself isn’t necessarily a skill that every striker needs to have, there are several strategic coaching decisions that will be informed by knowledge of the kind of strikers in each team.

Teams that often sit further back and defend have more of a need for a striker capable of creating his own opportunity since they will frequently be in the one-on-four kind of scenarios that Manotas found himself in. On the other hand, talented, attacking teams may have a greater need for pure goal poachers who are ruthless in finishing the chances that the more creative midfielders generate.  

I believe that the PercentCreated metric can also be used as a potential way of finding as well. For example, if a team sold an attacker who managed to create a lot of his chances for himself, then they would want to acquire a similar kind of player so they wouldn’t have to change systems.