King Cantona pretty much sums up my reaction to the response I’ve gotten on my first post (minus his trademark imperious attitude). All I can say is wow, and a heartfelt thank you to everyone for the overwhelming reads, feedback, and support the MLS vs. NCAA Passing Styles article received. It certainly got a bit more attention than I expected this early on, to the point that my coworkers have dubbed it a Marcus Rashford-esque debut. Here’s hoping my form doesn’t dip either in subsequent appearances.
Even got some support from the MLS Armchair Analyst!
It was an exciting few days of my phone going crazy with notifications, texts, etc. Thanks again for the support, but more so for helping expand the analytics conversation to involve the college game.
Please keep an eye on this space for new articles, particularly a couple of follow-ups to that first passing style post. My rate of posting here will depend a bit on actual job obligations, but here’s hoping I get the call-up to make some appearances in other places as well… (Rashford for England at Euros?!?)
I think that perhaps the best way to look at it may be a combination of these factors. Yes, some college teams certainly lack sound tactics and can lose their way over the course of a match (so can any pro team!). I’ve also witnessed college teams who cycle multiple players in and out of a match in order to keep the energy high and apply pressure more consistently than they could when guys are going a full 90.
Interestingly enough, stats that one might expect to show an increase in pressure or recklessness (interceptions, fouls, yellow/red cards) did not differ very much between MLS and the ACC for the later stages of a game – or were actually higher in MLS in some instances. Some college games certainly do feel as though the defensive pressure is ratcheted up for longer periods… so what gives? An instinctual guess tells me that we may need to look more at the length of passes attempted in the second half, and how that would affect the overall number of passes.
If the notions are correct that some college players lack some tactics and less fatigue leads to higher pressure late in a match, then one might expect that passes will become longer in the second half. You’re placed under pressure, maybe don’t have the wherewithal to deal with it… so you dump a hopeful ball over the top. This takes a lot more time than a short pass, meaning there are more turnovers and less time left to attempt passes.
Remember that this is only a suggestion off the top of my head… stay tuned for some actual statistics and research into this idea in a future follow-up post.
Soccer Specific Presentations
Soccer Analytics Panel (video now up as of May 2)
Three big takeaways for me from this year’s soccer panel -
Expected Goals has become the standard metric for analysts everywhere… but we all calculate it differently.
That seems a to be a bit of a problem, as it becomes very difficult to compare the numbers between different analysts. Most versions have their flaws, as pointed out by Devin Pleuler – overvaluing the attack, leaving themselves open to subjectivity, and not giving enough credit to fantastic individual efforts such as Carli Lloyd’s bomb in the WWC Final. I would agree as well with the suggestion of Dan Altman and many others that perhaps the best use of xG is as a framework for other analyses. We’ve recently developed our own preliminary version of the metric here at Virginia, working around the limitations of our data in order to produce something that may not be perfect, but at least gives us a foundation to build upon for additional analysis. We’ve been able to look at individual players, season long trends, areas of the field where we created good chances or needed improvement, etc., all based off of a relatively simple model. This may be the long-term best use of xG models in soccer analysis.
Questions need to come before answers… fully understand the problem you want to solve before you do anything else.
A pretty simple but understated idea here – not groundbreaking but a good reminder to avoid jumping to conclusions and then massaging the data to back them up. Let the data tell you what it will… in fact, that’s how I came to the crux of the passing styles article. That research took several unexpected turns before it was complete. My undergrad psychology professors would be proud.
As much as the analytics world wants to be open and helpful to one another, there will always be an element of disguise.
Naturally, no one will want to give away a perceived competitive advantage by sharing every detail of the analysis that they use. This is to be expected... to an extent. Sometimes this seems to be a bit extreme though, as people become tight-lipped and overly conservative – sharing next to nothing about their methods. This was the feeling that I got at this panel – the speakers seemed a bit reserved in this sense. The panel left me wanting more information and insight than it provided by the end.
Dan Altman – How Do You Stop Leicester City?
I attended two talks by Dan Altman at SSAC – the first was a dinner he hosted, at which he presented research that he has not yet made public. It was a particularly interesting take on finding undervalued assets in the transfer market. I won’t delve too far into it though since it hasn’t been presented anywhere else. Many thanks to Dan once again for the presentation and for providing a forum to make some great connections and chat about my foray into analytics.
The work that has been seen elsewhere is Dan’s piece on opposition analysis and how teams have made adjustments to deal with Leicester’s attack. Essentially, the Foxes create their best chances when they go straight towards goal – when teams force them to pass the ball more and they become less direct, they become less likely to score. In particular, these quick attacks are most dangerous when started by Kante or Drinkwater. Opponents began to harass these two a bit more, and have given up a few higher quality chances, but much lower quantity.
This provides a perfect example of using xG as a building block, as alluded to earlier. I really like the idea of measuring directness of attack as a method for opposition analysis. If I can tell where and how your most effective attacks take place, then I can set my targets for how to disrupt you. By the same token, this analysis could be turned on your own attack – are we particularly effective in one way, or do we have some more diverse options?
Iavor Bojinov – The Pressing Game
On the surface, this work on incorporating spatial info into analysis really appealed to me as a particularly visual learner. Seeing an image of where a team is strongest/weakest in possession, or where they are most/least effective defensively, could provide an excellent basis for preparing tactics. Trends over several games or the course of a season can help as well.
The underlying mathematical formulas made it a bit less practical for me – the video analyst in me feels that a similar understanding could be gleaned from watching the game. Admittedly, this thought may be geared more towards opposition analysis, particularly in our setting of needing a quick turnaround for two games each week. I think the best use of Iavor’s model may come in terms of self-reflective or longitudinal analysis, or in profiling managers as he mentioned later in his presentation. The effect of a manager such as Pochettino or Klopp creating a very specific style of play is very easily seen in these models, creating a useful tool for owners or even pundits.
Other Notable Presentations
There were a variety of panels and presentations based around improving athletes’ performance through the use of periodization, wearable technology, and athlete surveys on RPE/sleep/nutrition. We’re in the early stages of using a variety of this information at Virginia, so it was particularly helpful to compile some insight from various sources here. Erik Korem from Kentucky’s football staff spoke during the sports science panel about the benefits of a periodization model that they adjust based on player monitoring, from which they have seen better outcomes than with a model that doesn’t take into account player readiness and training load. The Baylor athletic performance staff also spoke about the importance of monitoring players’ off-field habits in order to see what correlates with their best on-field performance.
We now use Fit For 90 player readiness surveys and rated perceived exertion (RPE) ratings before and after every session, and compare them with the objective data captured by the Catapult tracking system. Over the past year, this has allowed us to begin to hone in on the best training plan for our guys on both a team and individual level. Technology can help remind us of the importance of recovery in allowing the athletes to push themselves and then allow their bodies to adapt to the stress, in order to improve performance.
From an analytics guru’s perspective, it was pretty cool to see the reunion of the original Moneyball group, which kicked off the conference. Sabermetrics pioneer Bill James reminded everyone of the following:
Along the same lines, Jeff Van Gundy and Shane Battier spoke about the importance of coaches taking ownership of failure, allowing the players to buy in and believe in the plan.
All in all, it was a great two days of nerding out about sports analytics. I’m happy to discuss with anyone who’s interested, whether you were at SSAC and want to compare views, or didn’t attend and would like some additional insight.
As always, thanks for reading and if you enjoyed it, please pass the site along to your friends and colleagues. Sign up for updates in the box on the right side of the page in order to receive an email for each new article. Keep an eye out for the next phase of the passing styles article, and follow me on a more daily basis on Twitter @ajbarnold.