Pages

Monday, 20 November 2017

EPL Take Three : Arsenal Beat Spurs

I don't really care about these derbies. I said as much the first time I wrote one of these. So what's going on with this title? Well, simply, it's a comment on expectations.

I think most people probably would have picked Spurs to beat Arsenal or, at least, walk away with a draw. A lot of that thinking would have reflected Arsenal's previous game... their fourth loss of the season which looked pretty decisive at 3-1. That was a "fake" result.

The only real way to judge the feel of a game, i.e. which team is winning, is actually to watch the game. Goals are just proxy representations. Goals For approximates the quality of your attacking and Goals Against approximates the quality of the defending. To be clear, I don't have a problem with using goals to determine the outcomes of games. But sometimes it happens the only reason a goal was conceded was because someone slipped. That's not part of the game. Sometimes a goal is scored because it was offside. That's not part of the game. And, of course, Goals Against doesn't represent the quality of the opposition's defending (and vice versa) which also makes the goal tallies approximate representations of "winningness in the moment".

Naturally you can't watch every game. And a lot of the important things about games, such as whether or not a goal was offside, is quasi-subjective. There are judgemental calls involved. Admittedly off-sides are about the clearest cut sort of foul that regularly leads to a goal but anyone with the faintest knowledge of soccer knows how difficult it is to determine whether or not a penalty should have been given. Winningness is also a pretty subjective quality and it's difficult to use objective measures to try and assess it.

In RTS games it is possible to win a game by turtling. That means using a very defensive strategy, usually accomplished through defensive builds. The way I turtle also involves counter-attacking. Obviously this makes it harder to tell what is going on. I might not have map control but that was a conscience choice: is the opponent's control of the map providing as much information about their winningness as "usual"? Not really. This is why possession doesn't work as a measure of winningness in soccer. Apart from difficulty in converting it into goals or goal potential, plenty of teams sacrifice possession to play in a particular way. There isn't really even a clear link between possession and shots (on target).

I'm sure if you plotted it, on average the greater the possession the greater the shots taken. But every single game is unique. If we wanted to predict the number of shots taken you'd need to take into account formations, the teams playing, which players they've got and whether the team is at home as well. At least, I'd be very surprised to find these features don't help... I have not attempted such a model. After all, there's a reason why people talk about side-ways passing and the boringness of tiki taka. And taking shots doesn't really translate into shots on target properly either. Those are what really grinds an opponent down. And how to engage with different kinds of shot?

I pay a lot of attention to corners when I do watch soccer matches. This is because my mother does when she watches soccer. And this quite reasonable. In NZ following international sport is normative in a way that doesn't seem to be true anywhere else in the world (hyperbole?) and international soccer is reputed to be defensive. That is, corners are surely more likely than in domestic games... and plenty of goals come from corners in domestic games too. But corners are really just a product of possession and shots. That's an improvement on either individually. After all, to get a corner you need to use possession to make the defence do something they don't want to do. Typically the intermediate step is some kind of shot or doing something that will lead to a dangerous shooting opportunity. The problem is that certain styles of play and player can be chosen which disfavour corners. And having a lot of corners may indicate a propensity to shoot at defenders rather than the goal.

Of the usually reported statistics, the one my gut says best assesses winningness is fouls. That's a bit stupid when you think about it, right? After all, I started off talking about the difficulty in discerning fouls. And some fouls mean more than others... think about penalties and the placement of free kicks. But soccer is so into gamesmanship that "professional fouls" are a Thing. As a result, a certain number of fouls in a game are committed based on the players' perceptions of winningness, i.e. the very thing we're trying to measure. I think this is why my gut impression is tenable. Time to put this altogether.

BBC Match Report

  • Based on possession we'd be thinking City 1.5-1 Arsenal not 3-1 (so, perhaps, 3-2 or 2-1).
  • Based on shots we'd be thinking City 1.5-1 Arsenal not 3-1 (so, perhaps, 3-2 or 2-1) as well.
  • Based on shots on target we'd be thinking City 1.7-1 (so definitely 2-1 as we'd believe in at least two saved attempts each).
  • Based on corners we'd expect a draw.
  • Based on fouls we'd expect a draw too.
So what to think? Based on all this information we'd expect a tight scoreline like 2-1 or 1-0, more like 2-1 because 60% possession is quite good and these are the premier league's most possession happy sides. The point is, we would not expect a final score of 3-1. Especially remembering my beliefs about corners and fouls we'd actually prefer to believe in a 2-2 or 1-1 game. So how did the scoreline turn out to be 3-1?

Very simply this game finished at 3-1 for two reasons. Firstly, there was a penalty. I haven't seen it again (at least, I remember watching a clip of it) but all the pundits I remember reading broadly agreed with it. Secondly, there was an offside goal. In the Bundesliga it wouldn't have counted. Why? Video Ref. So, what could we expect a game without the offside to finish as? Surely there are more refined measures of performance than my gut driven inferences? Perhaps even some that take into account what I said? Well, there are expected goals. Which, as I understand them, weights all shots by the chance of it going in.

From Here.
Ah, so Manchester City were outperforming themselves by this measure as well. But one of those large circles was the offside.


What this shows is that if you disinclude the offside Manchester City look quite a lot worse than a 3 goal team. In fact, if you remove the penalty we have City 0.79 - 0.34 Arsenal. Or, 1-0. Which is consistent with our impressions from before. 2-0 isn't, however. And while it makes sense to exclude penalties as measures of winningness in an expected goals formulation as they're so likely to lead to an actual goal,* I've already talked about why penalties/fouls make sense. On to the Spurs game!

The way I have tried to get you to read the Manchester City game is that Arsenal didn't put on a bad show... and that was away from home where everyone agrees they've been poor this season. Spurs may be a very good side and easily the most consistently good team in the EPL over the past two seasons, but their record at Arsenal is a bit suspect. So why this?
Well...no one saw that coming, did they? A resolute and determined Arsenal outplaying Tottenham in all departments and securing a fully deserved victory? 
[...]
If they can do it against it Spurs when no one gave them a prayer, they should be able to do it against Burnley and Huddersfield in the next 10 days, shouldn't they? But we've been here before...
And by no-one he means his colleagues.

Based on what we've seen above, Arsenal should have been given a fighting chance. They went to clearly the best side in the division and came away with a misleading result. If we're thinking that a worse team can beat them easily when Arsenal is at home and this team usually doesn't do well away to Arsenal we should probably re-check what we're using to calibrate our expectations. Speaking of which...

2.19 to 0.59
Which suggests it should have been 2-1. But also...




  • Based on possession we'd be thinking Arsenal 1 - 1.5 Spursnot 2-0 (so, perhaps,1-2 or 3-2).
  • Based on shots we'd be thinking a draw.
  • Based on shots on target we'd be thinking Arsenal 1.25 - 1 Spurs (1-1?) or maybe 2-1 (the three shots saved thing).
  • Based on corners we'd expect Arsenal 1.75 - 1 Spurs.
  • Based on fouls we'd expect an Arsenal win... or perhaps Arsenal 1.45 - 1 Spurs (1-1?).
And Arsenal apparently had an offside goal which doesn't actually change the Expected Goals conclusion because it was a very unlikely goal (unusual, I think). But you can see why I like fouls, right?



*To clarify... Expected Goals are still goals. Sure they involve the ideas of winningness that we discussed but they really, really like goals. If we included penalties we would end up inferring teams that were very fortunate to win a penalty were doing well in terms of winningness. That's obviously wrong. You can't have a fortunate penalty if you look like you're winning. Thus penalties have to go because they're something where the goalness comes through so strongly that it crowds out the Expected Goals measure's ability to involve our ideas of winningness. Getting a shot from open play that is as likely to go in as a penalty is guaranteed to reflect a suite of traits associated with winningness. We'd have difficulty reading it over the course of the match, but because the chance isn't 1 we hope it averages out.

No comments:

Post a Comment