Expected Passing for Everyone!

Regular readers will note that I’ve been boring off about ‘expected passes’ (xP) a lot recently both on the blog and on twitter.

What is xP?

It’s basically assigning each pass a probability of it being completed successfully.

I’ve done this by gridding out the pitch in (roughly) 10yd x 10yd sections. This splits the pitch into 70-odd zones.

All the open-play passes made are then dumped into these zones. Each pass has a start and end zone.

A sideways pass (say) by a centre back to a fellow centre back in front of their own box will have a high probability of being completed (95%+).

A long punt (say) by a right back in his own channel to an attacker in the opposition half in the left channel will have a low probability of being completed (maybe 20%).

This xP model is much more granular with locations than the previous one I’ve been using. The one before had 25 zones.

My xP models do not take anything but location of the pass (start and end) into account.

I decided to include smaller zones with this latest version because centre backs with high passing volumes of simple passes were (probably) falsely crediting their account too much. The new version eases this somewhat.

Still sounds too simple, mate?

Here’s the top 10 passers this season according to the model:


Not bad, one or two dodgy looking ones. What else you got?

So, I was spending too much time trying to get the numbers in that list this top 10 was taken from, to ‘feel’ better. You know what? Sometimes (most of the time) single figure metrics don’t cut it, so eventually I just decided to let the data speak for itself. I chucked the data into Tableau and built a dashboard to visualise it. Have a play around and send a link to your pals.

Just show me mate, I haven’t got time.

Ok, example one. Compare all central midfielders with over 300 passes. Charlie Adam and Cesc Fabregas are playing a different game to everyone else. These two are going through games playing balls that on average would only be completed 75% of the time. As you can see from the graphic below, Cesc is doing it more successfully:

new-adam-v-fabYou can also just about see that the model makes a case that Claudio Yacob might be the worst central midfield distributor in the Premier League this season.

Example 2

Centre forwards. Note that the average lines move. Completing passes when you’re up front is more difficult for all sorts of reasons. It’s therefore fairer to compare them to each other:

new-forwards-comp Ighalo sticks out here. He may have been lacking goals this season but his hold up play looks to have been decent. Then there’s Sergio Aguero. He’s matching most central midfielders with the quality of his passing.

Example 3

Tottenham. It’s probably better to have the average lines league wide here rather than moving as they do in this screenshot but it’ll do:

new-spurs-compThe push ‘n’ run 61 Lilywhites are a compact, big possession, big shot volume kinda team. However, a lot of those shots are distance shots. A lot of teams graphs show more variation than the Tottenham one above. It’s the more cautious passers that boss this team at the moment. Wanyama and Dembele is an uber-safe partnership, the talented attacking mids are struggling slightly and it’s down to the full backs to take the riskier options. Still, their fans won’t be too unhappy with third even if it is a bit dull to watch.

As, I say, play with the dashboard and make your own comparisons. I love it, so ner. However, it won’t be updated for *reasons*. Another point to note is that the model was built in Excel. At a pro club and have data sat round doing nothing? Attended the OptaPro forum and think you need to be a maths n coding freak to model this shit? Fink again, and get in touch.


Posted in Sports | Tagged , , , | 1 Comment

Michael’s Keane passing

So, I’ve been mucking about making an expected pass (xP) model recently and who should turn up in it as the season’s best progressive passing centre back but reported Everton target Michael Keane.

We all know that straight passing % provides no context to the type of passes being made by players so an xP model is a way of adding some much needed value.

The basic premise with mine is to split the pitch up into zones and see how often passes from one zone to another are successful. This will give you an xP number for each pass. Compare it to how many successful passes were actually made by a team or player between those zones, et voila…a rating can be made.

It sounds (and is) very simplistic so I wanted to do a quick eye test on it. The model tells me that Keane’s passing was particularly good against Man City last month. I only have first half footage of that game so that’s what I worked with.

I took out all the passes Keane made that the model said he had more than a 70% chance of making. This left me with what the model thought his 11 ‘most difficult’ passes were in the first half.

The footage is below. I’d like you to count how many of these passes you think Keane ‘should’ have got to his man:

I count 5. Keane made 8.

The model reckons he should have completed 5.89 passes out of the 11 made here, hence it being flagged up as a good half for him passing wise. The model doesn’t know that the opponent is Man City. My pressing metrics tell me that Man City are the most effective and aggressive high pressing team in the league. You could easily feed that into the model. You could also feed into it that City were reduced to 10 men after 32 minutes.

It’s only one quick test but I’m pretty happy with that.

The model can scout every pass Keane’s ever made in the Premier League in minutes.

It’s got a better memory than your scout.

It knows how every similar pass played in the Premier League has turned out over the last 7 years.

Your scout doesn’t.

The model can tell you clearly if a player is continually overreaching himself trying passes unlikely to come off.

The time and money this saves on live or video scouting is huge.

If you’re at the OptaPro Forum this year, come and have a chat.

Follow me on twitter @footballfactman

Posted in Sports | Tagged , , ,