Defending By Numbers
Football, at it’s core, is a really simple game. So simple that the key to winning football matches can be easily summed up in two points:
1 – Score more goals than the other team.
2 – See point 1.
But that simplicity masks the complexity that makes football such an addictive sport. Yes, the goal is to outscore the opponents, but there is almost no end to the ways you can set a team up to do that, with the Kevin Keegan at Newcastle, “we’ll beat you 4-3″ approach and the George Graham school that looks to grind out 1-0s at either pole.
As the game has become more athletic and technical, teams are increasingly looking to sports science and performance analysis to give themselves some kind of advantage. This move towards a more analytical approach can be seen in the increasing investment in such areas by top clubs around the world. MLS will be rolling out the adidas miCoach system to all clubs in 2013, while OPTA and ProZone continue to carve out a lucrative niche in statistical analysis. Caleb Porter is an avowed disciple of this more modern, analytical approach and the Timbers recently announced the hiring of a performance analyst as they seek to find that edge that will turn hard-fought draws into a well-deserved wins.
Football isn’t a game that lends itself readily to statistical analysis in the way that, for example, baseball does. There are few equivalents of the 1-on-1, pitcher vs batter, battles that baseball throws up as 22 players on a 8500 sq yard pitch for over 90 minutes simply throws up too many variables to be distilled succinctly into a neat column of easily digestible figures.
Yet, such a study isn’t an entirely fruitless endeavor. Mining the numbers can still throw some light into the darkness, even though it always pays to remember that it isn’t an exact science, and much remains for individual interpretation of the numbers.
While I don’t think that analytics gives all the answers, I do believe it can, and should, be part of the discussion we have as it has much to inform our “gut” reaction. There have been plenty of times when I’ve come away from a match thinking that this player or that has had a stinker, then when I look at the numbers, they seem to tell a different story. By taking both and playing them off each other, and seeing what each can tell you about the other, you can come closer to the “truth” of it, if such thing is even possible in football.
So, accepting that finding answers in soccer analytics is more complex than simply drawing up a table and seeing who ranks top, let’s see what the numbers can tell us about how and why teams do that most crucial of things – score goals.
Well, actually, before doing that let’s take a detour by first of all looking at the problem from the other side – stopping the other team scoring.
The Upgrade Revisited
In my previous post I took a look at the Timbers goalkeepers as I tried to judge whether the front office had called their “upgrade” right. Such an approach was never going give a definitive answer, but the figures certainly seemed to suggest that Donovan Ricketts had delivered some performance improvements over Troy Perkins, though Perkins’ figures would subsequently rocket in Montreal and cast a shadow over Ricketts’.
There are too many intangibles to definitively call one way or the other. The change in style from John Spencer to Gavin Wilkinson, the fact the figures don’t take into account the quality of chances, or a goalkeeper’s distribution and communication with the defence in front of him. Respective ability of cross balls, or quickness off the line, or agility, or, well, the list goes on.
A good point was raised in the comments of that article by “Thunderbear” (I assume that’s his/her surname and not their first name) that “Perkins was traded just after Spencer was fired and after a couple of miserable team performances like the 5-0 loss in Dallas and the whipping in Utah that really affected Perkins’ statistics”.
In the five matches Perkins played under Wilkinson, the team shipped 13 goals. In the previous 17 starts, Perkins conceded 22. That’s quite a jump, I’m sure you’ll agree. The drop in shot stopping after Spencer left is really quite stark – the overall figure drops from 70.8% to 47.8%, with an In-Box Save rate (IBSv%) of 38.5% for those five games.
Even expunging those games from Perkins’ record, there is never more than a few percentage points between his save percentages and those of Ricketts in Portland, suggesting again that, at worst, the front office broke even (on the pitch at least, the salary debate is a whole other issue).
There is unlikely to be any single reason why Perkins numbers dipped so badly in these few games. The removal of a head coach is bound to unsettle some, and the change in formation and playing style wouldn’t have helped, in the short term at least.
Yet Perkins faced the same number of shots per game, with slightly more coming as In-Box Shots (IBSh) than Long Range (LRSh), but not a great leap. Certainly not in the order that might go someway to explaining why the goals just rained in.
This brings us to the part of the discussion that was missing in the previous post – the defence.
Analyzing The Defence
Analyzing defensive performance, and measuring the difference between a “good” defence and a “bad” defence is more difficult than it may seem at first glance. The bottom-line metric for any defence is the goals against column, and on that measure you could confidently, and correctly, declare that in 2012 Sporting Kansas City had a good defence, while Toronto FC had a bad one.
But it’s not that simple. Perhaps a defence was let down by poor protection from their midfield. Luck is a factor too, as in the defender who pulls off a great last-ditch sliding tackle, only to see the ball richochet off a teammate and into the path of another attacker for an easy goal, or the player who never scores hitting that one-in-a-million screamer into the top corner from 35 yards. There’s tactical issues – was the coach leaving his defenders exposed by pushing on his full-backs, or playing too wide or narrow?
However, given that a defence is ultimately there to prevent goals being conceded, it’s against that baseline they must be measured.
Taking the clearances, blocks, interceptions, recoveries and fouls as examples of “defensive action” we can see if there is any relation between how busy (and where) a team is defensively effects the number of goals they concede. Note, I haven’t included tackles as I’m not really happy with how OPTA measure them in terms on tackles won/lost, and I’d have to go through every game to judge for myself whether it was a “good” tackle or not, and then there’s the old truism that the best defenders finish the game without grass stains on their shorts because they never need to go to ground in the first place. So, yes, it’s not a comprehensive tally of every Defensive Action, by any means.
There is also the issue of defensive distribution of the ball, which is something I hope to return to in future, but for now let’s set it aside and judge a “good” defence purely on stopping the other team getting on the scoreboard.
As you can see, broadly speaking the more Defensive Actions (DfAc) a team performs, the less likely they are to concede goals, though it’s by no means definitive, especially with only 19 data points to work with. Still, it does seem to point towards some link between the two factors.
If you take Recoveries out of the equation, then the trend is more pronounced.
Backing up this link is that of the six best defences in terms of goals against (Kansas City, Seattle, Real Salt Lake, Chicago, Vancouver, Houston), three occupy spots in the top five of most Defensive Actions (minus Recoveries) – Chicago (4th GA / 3rd DfAc), Seattle (2nd / 1st), Vancouver (4th / 5th).
The big anomaly is Kansas City who are 18th in terms of DA, but are 1st in goals against. Kansas City had a high proportion of their DfAc in the back four (71.41% against the league-wide average of 66%) and, interestingly, they were 3rd in terms of fouls committed with, again, a higher than average proportion of these fouls being committed in defence (37.1% vs 30.8%). This would suggest a lot of pressing and a no-nonsense, physical approach to winning the ball back. This is largely due to keeping shots against to a minimum. No team conceded fewer shots than Sporting, and they were also good at keeping teams at distance with only 47.3% of shots conceded coming from within their own box – only Vancouver have a better record in this regard with a 46% IB:LR balance.
Those that do get through are met by a keeper in Jimmy Nielsen who performs above average in all areas of shot-stopping.
This perfect combination of tight defence and excellent shot-stopper combined to produce the fewest goals conceded in the regular season in 2012.
The Difference Makers
Sean Johnson of Chicago Fire actually posted very similar figures to Nielsen, yet whereas Nielsen let in 27 in 34 starts, the leakier Fire defence meant that Johnson conceded 38 in 31 starts as his goal was peppered by 40% more shots than Nielsen’s. So, the value of a tight defence is clear.
Yet Chicago “worked harder” in defence, with vastly more Defensive Actions than Kansas City. Where the teams also differed markedly, and perhaps this gives us an insight into the wild swing in shots against, is in possession of the ball.
I have my issues with the way OPTA measure possession, but as a metric of “ball control” it can be generally relied upon. It’s still a poor measure of which team is “better”, and terrible in terms of predicting outcomes – 17 of the 19 teams posted higher average possession rates in matches they lost than those they won – but it can tell us which team was controlling the ball, even if it can’t say where and how they were doing it.
Kansas City posted the 5th highest possession rate at 52.1%, while Chicago were 17th on 46.8%. This would seem to indicate that Sporting operated with a “keep ball” philosophy, seeking to minimise the opposition’s time on the ball. Chicago on the other hand worked best when they could spring fast attacks, giving up possession and, it would seem, inviting teams onto them, leading to an increase in shots and goals against. Only Vancouver and Toronto saw less of the ball than the Fire in 2012.
By taking Kansas City’s high possession and their high foul rate and defensive work mentioned previously, you can start to build up a blueprint for why the Sporting defence was so effective. They choked other teams of possession, and created a high number of chances (though they were actually very wasteful in converting these chances, but we’ll come to that in a future post), but when they did lose the ball, they were efficient, yet physical, in their attempts to win it back. Through this pressing and harrying they limited shots against to the minimum, but had one of the league’s best goalkeepers to take care of what did get through.
This all serves to underline how difficult it is to apply analytics to soccer in a straightforward, “the numbers don’t lie” manner. It requires a more holistic approach where a number of factors and measures can be combined and torn apart to indicate towards conclusions.
As we saw with Johnson and Nielsen, the value of a good defence cannot be under-estimated. If you were to take an “average keeper” and put him behind the Kansas City and Toronto defences, you would expect to see a difference of 20-22 goals per season, from around 29 goals to 51. Yet both keepers would perform identically.
This is all very simplicity, and ignores many of the intangibles that I’ve talked about already, but it can at least give us an idea of the importance of a good defence.
If we were to flip it around and take an “average defence” and put it in front of two keepers at either end of the scale in shot-stopping the swing would be in the order of 15-18 goals per season, from 36 to 52 goals against.
The difference between best defence & best keeper and worst/worst would be from 25 goals to 63. In actuality Kansas City conceded 27; Toronto FC conceded 62.
Looking at the other end of the table, Toronto bottomed out in both goals against and DA. As I said in the previous article on Timbers keepers, Milos Kocic actually performed reasonably well – marginally better than Ricketts across all of 2012 – but when your team allows more shots at goal than all but 2 teams in 2012, you’re going to struggle to keep them out. Indeed, even the heroic Nielsen would’ve expected to concede upwards of 40 goals when faced with the barrage Kocic did.
The two clubs that allowed more shots than Toronto were Chivas USA and Columbus Crew, however Andy Gruenebaum posted numbers than ran Nielson close in shot-stopping whereas Dan Kennedy’s were good, rather than great. Gruenebaum conceded 13 fewer than Kennedy. A very valuable keeper!
The other factor is that Chivas conceded the highest proportion of IBsh:LRSh (55.5%). Given the importance of preventing IBSh – teams are almost 4 times as likely to score in the box than out of it – each percentage point increase in IBSh:LRSh represents something in the order of an increase of 1% in goals against.
Portland were 17th in GA, and their IBSh:LRSh rate of 52%, above average, gives some idea of their problems in 2012. Despite putting in a lot of “defensive work” – they rank 8th in DA – and being “midtable” in terms of total shots against, where they really struggled was in preventing teams getting shots on target.
There would be little surprise to Timbers fans to learn that Portland were one of only 6 clubs to commit over 50% of their fouls in midfield, with Diego Chara the main culprit. Of those six clubs, only DC had a goals against record that ranked in the Top 10. Instinct may tell us that committing a foul in defence, on the edge of the box, may be more dangerous than committing a foul in midfield, but to score from 20 yards out direct from a free kick requires great skill. It is why, to me, someone like David Beckham is so feted – merely decent-to-good in most areas, where he excels is in those kind of dead ball situations.
In fact, the three goals against records in the league were from team that committed a higher than average proportion of their fouls in defence.
From 35 or 40 yards, the dynamic changes. A team can then throw forward their big guys, and a decent delivery, rather than pin-point in the case of getting a ball up-and-over or round a wall in limited space, can be good enough for someone to get on the end of it, and serve up a great opportunity to score.
There are certainly cases where the foul is preferable, and a team would rather take their chances defending a set play, but in general terms, giving away so many opportunities for a team to put the ball into the box to be attacked by numbers cannot be a good thing.
Blocking, pressuring, or generally employing sound defence can go a long way to making it difficult for the opposition to get a shot on frame. Only Colorado Rapids were worse than the Timbers at keeping opponents off target, but Matt Pickens did a better job than Perkins/Ricketts in keeping shots out and so the Rapids conceded 6 fewer over the season.
Even though the Timbers took a similar number of shots as they conceded, when you compare the accuracy rates you start to get a good idea of the Timbers problems in 2012 – 36.5% of shots against called the Timbers keeper into action but only 31.1% of the Timbers’ shots were on frame. The figures are even worse in the crucial IBSh category – 41.9% to 32.7%.
Given how long this post is already though, I’ll return to the attacking sphere in a future post.
In terms of possession, as you saw in the chart above, the Timbers came in 13th with an average of 47.85%. Under Spencer that figure was 45.38%, which would’ve dropped them to 17th. It rose to 50.16% when Wilkinson took over – around LA Galaxy level of ball control.
So what you had in Portland was a team that didn’t really control the ball particularly well, and didn’t do a good job of preventing opponents from getting shots on frame. The goalies performed adequately, but when your defence is leaking chance after chance, adequate simply isn’t going to be enough.
Well, the conclusion is that it’s difficult, nay, foolhardy to draw firm conclusions from data alone! There needs to a be a dialogue between the figures and what we see on the pitch with our own eyes. That instinct for what you’re seeing is a valuable commodity.
By drawing from both wells, we can draw a couple of fairly obvious conclusions off the bat. A good defence makes a big difference. A good goalkeeper also makes a difference, though perhaps not as much as the guys in front of him. A case could be made for seeking value in goal, while ensuring that the big bucks are reserved for shoring up the backline. In Kocic the Timbers have value at the back, but it remains to be seen whether investments in the defence and midfield will bring about marked improvements. It’s here that the leaps forward will have to be made if the Timbers are to progress in 2013.
The mid-season change wasn’t ideal. For whatever reason, it really seemed to throw off the defence and Troy Perkins, but the ship was steadied somewhat by the arrival of Donovan Ricketts. However, as a whole, the team underperformed. They let too many shots in, and from crucial areas, to expect much more than the 3rd worst defensive record.
A “better” keeper may have helped somewhat, but the key factors were repeated breakdowns in defence and midfield. There was a 71% drop between DA on the back four and DA in midfield, where in 4 of the 6 most miserly defences the drop was less than 60%, suggesting the better defences defended more as a whole rather than two distinct groups. This disjointed approach to defence led directly to a disappointing, but predictable, outcome. There is also the issue of too many fouls being committed in midfield, inviting teams to throw the ball into the box and put our defence to the test. A test the Timbers defence ultimately failed.
I’d expect to see a more cohesive approach to defence under Caleb Porter. We’re starting to see that in preseason, as the team look to press high and in numbers. I think the signing of Ryan Johnson is particularly significant in this regard as no forward player was more defensively active than Johnson was in 2012. He’s there to do more than just score goals or provide assists. The more I look at the numbers, the more his signing seems like it could be potentially the single most significant move the Front Office have made in the offseason.
Denying teams space in and around the box will be key as neither Kocic nor Ricketts have shown themselves to be especially proficient at short range.
I’d also expect to see the team climb the possession table. It is noticeable that, of the top six possession teams in 2012, five made it to the post season. Possession of the ball doesn’t guarantee goals, but the old adage that the other team can’t score if they don’t have the ball seems to hold true. Porter will seek to do more than simply keep the ball though, he’ll seek to use it with purpose in attack, and it’s to the attack that I’ll look in a future post.