Reinventing the passing wheel: What determines a good passer? / by Eliot McKinley

By Eliot McKinley (@etmckinley)

Directional Passes Over Expected: Where do players exceed passing expectations?

During the National League Wildcard playoff game, American Soccer Analysis contributor and Lamar Hunt US Open Cup champion, Sean Steffen tweeted about the baseball stat Directional Outs Above Average. This metric tells you about the defensive range of an outfielder, with positive values indicating a direction where the player is better than average at creating an out and negative where the player is below average. Obviously, this exact type of metric cannot be used in soccer, but it did inspire me to figure out how something like it could be used. Thus, Directional Passes Over Expected (DPOE) was born.

DPOE aims to visualize where a player is good or bad at passing. Expected passes (xPass) gives a global measure of whether a player is better or worse than expected at completing his or her passes. However, some players are better in one direction or another. For example, if you know Lebron James is less likely to score driving to the hoop on his left, you plan your defense to make his go left. If you know that an opponent is a less accurate passer to his right, maybe you gameplan such that that player is forced to do so.

Like its baseball inspiration, DPOE breaks up the players’ directionality into six slices of 60° each. For each sector, the players’ Per100 is shown. Per100 is defined as the number of passes a player completes compared to the number of expected passes normalized to 100 passes. So a Per100 of three indicates that a player is likely to complete three more passes in 100 than would be expected. It is derived from the xPass model developed by Matthias Kullowatz and is available on the interactive tables. xPass takes into account numerous parameters to calculate the likelihood of pass success including the spatial location, whether it is a long ball, pass types, and whether a team is up or down a player. A good passer will have a Per100 greater than zero, an average passer will have a Per100 equal to zero, and a bad passer will be below zero. DPOE contextualizes Per100 by indicating in which direction a player is particularly good, average, or bad at completing passes.

Let’s look at some examples of central defensive midfielders during the 2018 MLS season. The top row shows the DPOE plots with overall player Per100 labelled, and PassSonar plots on bottom (bars are pass angle frequencies, colors are pass distance) for Alexander Ring, Diego Chara, Michael Bradley, Tyler Adams, and Wil Trapp. DPOE plots show that Chara and Trapp are above average passers in all directions, however PassSonar shows that Trapp is a bit more likely to hit long diagonal balls than Chara. Ring and Bradley are above average in all sectors but one, with Ring below average straight ahead and Bradley towards the lower left. Perhaps if you were an opposition scout you may try to force them to play passes in these lower conversion rate directions. Finally there is Adams, who, playing for New York Red Bulls, plays much more direct than the other players depicted here in his PassSonar. Overall, his Per100 is lower than the other midfielders and his DPOE plot shows lower than average conversion rates in for forward passes.

Lastly, let’s compare Ezequiel Barco and Yamil Asad. This past offseason, Asad was swapped one United for another moving from Atlanta to D.C. for around $500,000 in allocation money while Barco was signed from Independiente for a reported MLS record fee of $15 million. By DPOE, Barco is clearly the better passer. He has positive Per100 in most sectors, and is only far below expected passing directly backwards. Asad, however has had negative Per100 in all directional sectors since 2017. PassSonar demonstrates different passing patterns for the players as well. Barco predominately passes laterally to his right, with relatively few vertical passes. Asad also mostly passes to his right, but these passes are longer than Barco’s and Asad also plays a higher proportion of vertical passes. Barco and Asad are obviously different players when it comes to passing, which shows in Barco’s xA per 96 being double than that of Asad in 2018, while both are contributing a similar amount of expected goals. Much has been said about Barco being a disappointment both on the field and off, but his underlying numbers remain relatively strong, including co-leading the league in xGC/96 with Miguel Almiron. Whether Atlanta would have been better off paying the $1 million fee for a permanent Asad transfer instead of brining in Barco is still an open question. However, given that Barco is 4.5 years younger, producing pretty good numbers while on the field, and still developing as a player, I’d hold off on answering that question definitively.

There are some inherent limitations of DPOE, such as small sample sizes on some directions (e.g. Nick Rimando has a backwards Per100 of 81 by completing 1 backwards pass this season), and perhaps there should be more directional bins. Additionally, the xPass model has some limitations which have been outlined by Matthias, which includes undetermined confounding factors due to player position, game states, and potentially playing styles. As such, when comparing players with DPOE one should be careful to take these limitations into account. But even given these, DPOE provides a means of assessing where a player passes well or poorly, and perhaps where a player should focus on in training or identifying weaknesses that an opposing team could exploit.