Worrying Times For Liverpool At Both Ends Of The Pitch.

The recent explosion of available English Premiership data has provided a route for the curious to begin to examine in much more detail the core actions that go towards deciding the outcome of a football match. The Laws of the game award the victory to the team scoring the most goals, so it's sensible to roll the goalscoring process back one stage to try to see which teams are making the most of their scoring opportunities and equally importantly which defences are making scoring difficult for opposing strikers.

How efficiently teams turn goal attempts into shots on target and ultimately into goals often defines where a team ultimately finishes in the table. With access to x,y data points for shooting attempts it is now possible to compare team performance in this area against an average basket of hopefully, representative teams from the EPL. At the very least we should be able to highlight a side's strengths and weaknesses compared to par when dealing with goal attempts or dishing out shots of their own. This is the so called descriptive use of data, projecting suitable regressed figures for use in predicting future performance may come later.

The rolling TV soap opera that is Liverpool's seemingly constant battle with insufficient reward from copious amounts of toil in front of goal again provides an idea test case for this type of analysis. The Reds currently languish on the fringes of the relegation battle, with a negative goal difference (even after a five goal rush at Norwich) and question marks regarding their inability to adapt to a new style of play and also over the form of Pepe Reina.

Shot analysis has made great strides in recent seasons. Analysis no longer has to rely on weight of numbers to even out the quality of the shooting opportunities under investigation, but such unknowns as the position of defenders and the pressure that is being applied to the shooter are still largely absent. So with these caveats, let's look at how Liverpool have fared so far in the EPL.

Using the Opta powered Fourfourtwo app I've recorded the co ordinates of all 123 goal attempts made by Liverpool so far this season and designated each attempt as either "on target", "off target", "blocked" or "a goal"  (a goal is also recorded as an on target effort). By comparing the expected outcome for Liverpool's 123 shots, had they been attempted by an "average" EPL team, with The Red's actual outcomes, we can try to gain a better understanding of their season to date.

Shot Expectancy And Actual Outcome For Liverpool's Attack and Defence. 2012/13 To Date.

Team	Expected No. of Shots On Target.	Expected No. of Shots Blocked.	Expected No. of Goals.
"Average EPL Team"	39	31	10
Actual No. Made By L'pool Attack.	29	33	9
"Average EPL Team"	24	21	6
Actual No. Allowed By L'pool Defence.	24	21	12

Looked at over the season so far, there initially appears to be a clear pattern of average performance from the attack because their 123 shots have produced nine goals, just one less than our baseline model predicts for an average team playing similarly average opponents. However, there should be accuracy concerns because Liverpool's shooters have hit the target 29 times compared to an expected 39 and whereas it is possible that the score of 29 is the luck driven outlier, it may be that it will be the goals that begin to fall into line with the wayward shooting.

Based on goals scored Liverpool are about an average attacking side, but shots on target may suggest that currently they aren't quite that good.

If we now move onto the defensive qualities of the side. All conclusions should be tentative after even 100+ shots, so Liverpool's defensive record of dealing with opponent's shots, based as it is on 77 attempts should also be interpreted warily. But from a purely descriptive "this is how we got here" approach the expected goals column stands out.

How accurate a team can be and how quickly their attempts are closed down and blocked is down to a combination of their individual talent and the actions of the defenders. So if we accept that Liverpool have faced a fairly typical cast of opponents so far, defensively, as a unit they are around average at closing down shooters. However, they have conceded double the expected number of goals.

Video analysis would help to decide the cause of this shortfall in goal prevention. It's easy to solely implicate the keeper. But individual mistakes from players in front of him leaving opponents with clearer cut goal attempts or an unusual, but not prohibited run of (poor) outcomes occurring randomly with little actual change in Reina's ability should never be ruled out.

Liverpool's Attacking Record By Game Compared To An Average EPL Baseline.  

Team	Expected No. of Shots On Target.	Expected No. of Shots Blocked.	Expected No. of Goals.
Average Team.	5.1	4.0	1.3
Actual No. v WBA.	2	7	0
Average Team.	5.0	4.8	1.2
Actual No.v Man City	3	5	2
Average Team.	6.0	4.8	1.5
Actual No. v Arsenal.	4	4	0
Average Team.	7.2	6.0	1.9
Actual No. v Sun'land	6	4	1
Average Team	4.8	3.3	1.3
Actual No. v Man U.	6	2	1
Average Team.	5.4	3.7	1.7
Actual No. v Norwich.	6	6	5
Average Team.	5.5	4.8	1.2
Actual No. v Stoke.	2	5	0

Once we start to break down seasons into game by game slices, the numbers begin to become as much a product of Liverpool's opponents as they are of Liverpool. Liverpool may have seen a lot of their shots blocked against WBA simply because The Baggies are extremely adept at blocking shots.

Few teams are currently well known for their ability to block shots, but we can make an more educated estimation around the goal based expectancy. It's likely that Liverpool under performed in the 90 minute stretches that made up the Stoke, Sunderland, Arsenal, WBA games and possibly during the ManU one, while they over performed in the Man City game and massively over performed during the Norwich match.

This illustrates the partly random nature of scoring. Teams can hope for an ideal distribution to provide a maximum points haul, but they have to deal with the distribution of talent based rewards that actually come their way, especially in a short sequence of games. Scoring five at Norwich when an "average on average" matchup would have produced just under two, but then under performing in the majority of other matches,  coupled with accuracy issues from the season long study, may suggest again that goal scoring will be the measurement that will fall more noticeably, rather than accuracy being the one to dramatically increase.

The reality is likely to be a combination of the two, with shots becoming more accurate, goals becoming slightly less efficient and the "new" Liverpool team gravitating to a lowly position within the top ten.

Liverpool's Defensive Record By Game Compared To An Average EPL Baseline. 

Team	Expected No. of Shots On Target.	Expected No. of Shots Blocked.	Expected No. of Goals.
Average Team.	6.1	4.5	1.9
Actual No. v WBA.	6	6	3
Average Team.	3.4	3.0	0.8
Actual No.v Man City	3	3	2
Average Team.	3.5	2.8	0.8
Actual No. v Arsenal.	5	2	2
Average Team.	1.9	2.1	0.4
Actual No. v Sun'land	1	3	1
Average Team	3.0	2.0	1.2
Actual No. v Man U.	3	2	2
Average Team.	5.1	4.1	1.1
Actual No. v Norwich.	4	4	2
Average Team.	1.5	1.8	0.2
Actual No. v Stoke.	2	1	0

The same strength of schedule issues apply in a game by game look at the Red's defence, but the worrying fact is that six of their seven opponents have scored more actual goals than the expectation derived from a wider, hopefully average basket of matchps. The two Manchester clubs and Arsenal are clearly above average, but the balance is restored by the presence of Norwich, Sunderland Stoke and even the currently high flying Baggies. Once again on a game by game basis, collective or individual failings in defence appears to be a big factor in their current plight and may prove to be a vulnerability over the course of the year. Reina is certainly a keeper in the spotlight at present, along with a defence which appears uncomfortable playing the ball from the back.

The Case For Data Analysis In Football.

One persistent criticism that has been aimed at football analytics is that it hasn't overturned any existing notions that have formed around the modern game in the same way that the sabermetrics movement challenged the status quo within baseball.

I do not agree with this assertion.

Before we can address this important point, it would be helpful to give a quick (and almost certainly flawed) overview of the statistical revolution that occurred in baseball. Much has been written concerning the major differences between baseball and football. The former has discrete well defined events whereas football is a true "team on team" event where interactions are both complex and numerous. So the challenges in football are different to those found in baseball.

Secondly the timescale and resources available to either sport has been vastly different. Advanced baseball analysis was probably kickstarted in the late 70's with the formal self publication of the thoughts of Bill James, who sought to reinterpret statistical measures that had been around since what is now the last century. Ideas were then developed with the extensive self collection of data, a massive project that required huge amounts of cooperation from enthusiasts on a scale that is all but impossible to replicate within football. The gap between the birth of this fledgling movement and any acknowledged impact on the sport itself was then upwards of twenty years when "Brad Pitt" introduced these "new" ideas to MLB.

By contrast football analytics has had neither the luxury of accumulating large amounts of data, which makes the MCFC and Opta initiative so welcome, nor data laden decades in which to mature, nor so many ancient and flawed targets to demolish.

Many will have their own idea of when football analytics began to evolve, but much of the context setting work started to appear in print and on the internet in the late 90's, mainly based simply on goals scored and allowed and congregating around the many fledgling gambling sites that were and still are so prevalent. Everything from team specific win expectancy, likely final scores and expected times for a first goal to be scored were modeled using two meager team statistics. The importance of different goal scoring environments were recognised and acted upon.

So just as baseball had used models and fundamental data to describe the run and win expectancy of any game in any game state, football amateurs have already done the same for their sport, albeit on sites that are an internet backwater to many.

Football has also used this route to overturn many (journalistic) cliches that persist around the sport. The cup isn't a great leveller, it's the preserve of the Premiership particularly the Big Four or Five. It isn't harder to play against ten men, it's considerably easier. 2-0 isn't the most dangerous lead in football, it's preferable to 1-0 but not as good as 3-0. A team shooting first in a penalty shootout doesn't automatically inherit a 60% chance of winning. And more recently, raw possession isn't as important as what you actually do with it and Swansea aren't Barcelona, Britton isn't Xavi and only this week, West Ham aren't Real Madrid.

To overturn a nonsense, you first need that nonsense to exist.

Progress then stalled through lack of meaningful data, until the very recent introduction of various pay sites, resulting in a rapid familiarity with such areas of the field as "the final third". If the MCFC data dump, which in it's advanced form comprises less than 0.3% of one season's worth of games and therefore contains an even smaller amount of one year's total data, has merely confirmed perceived wisdom as of 2012, isn't that something to celebrate rather than lament.

Sabermetrics, in the view of it's supporters overturned perceived wisdom because the old time scouts got it wrong. It is hugely encouraging, but not totally unexpected to realize that the present day "traditional" football analysts, armed with superior tools and a generation or three removed from analysts in another sport have largely interpreted on field events correctly. And if number crunching can add value and quantify those conclusions, then that's surely even better for everyone involved or even mildly interested in the subject. Collaboration is always preferable to wars and perhaps there isn't a baseball like war to be fought in football. (Rather appropriately baseball is currently embroiled over which flavour of WAR to use).

The fledgling analytics movement within the NFL is probably a much more appropriate field with which to compare football's attempted leap forward. Less developed than baseball, it still has advantages over football (soccer) in terms of simplicity of on field events and access to copious amounts of data. But it's success stories are largely the same as those enjoyed by soccer. NFL number crunchers have helped to sort out the correlation and causation conflict between running the football and winning, they exposed tactical inefficiencies in fourth down decision making and they cleaned up their own self inflicted nonsenses such as the "curse" of running back overuse. They've suggested ways to project college quarterback statistics into the NFL and quantified on field events that are predictive of future wins. It apparently helps if you have a quarterback who can throw the ball........but how much does it help?

In terms of analytical progress, the two sports are neck and neck (although soccer is better placed because of it's global appeal) and with much more data input from interested parties and more than a few false starts, both should progress rapidly in the future. Although those caveats shouldn't really need to be constantly repeated.

Football analytics is in a great place at the moment.

How Passing Sequences Create Chances.

In previous posts we've looked at individual passing attempts and how the pitch position of both the passer and the intended receiver impacts on the difficulty of the pass and therefore the likelihood that the pass will be successfully completed. However, the true worth of a team's passing prowess only really becomes apparent when these individual passes are reconstructed into their original passing sequences. 

All of the data that has been used to construct these passing expectancies has been taken from the Bolton Manchester City game from the Reebok, so the usual caveats apply. The conclusions may not be entirely representative of the wider Premiership, although the contrasting styles of each team may reduce this potential problem and the relative closeness of the scoreline throughout the game probably means that both teams had a relatively attacking outlook throughout.

Below I've merely listed the the number of passing moves in each passing sequence before the passing chain is terminated through loss of possession via such events as a shot, a foul, a misplaced pass or a tackle. Every pass is intended to be completed, so I've included "chains" comprising of one misplaced or uncompleted pass and I've also initially terminated chains were a player successfully beats an opponent in a "take on" situation.

Both Bolton and Manchester City attempted very similar number of passing sequences. This is virtually inevitable because a new sequence will tend to start once possession is lost by the opponent and by including single pass chains we can get a clearer picture of the use each side made of it's possession. 

The differing passing styles on that particular match day is clearly visible in the figures. Both teams were equally careless when conceding the ball with their first attempted pass, but the visitors had more longer passing chains than Bolton. City produced nine passing sequences that lasted into double figures compared to just two for the hosts. Equally striking is the concentration of passing sequences for both teams that terminate after just a handful of passes, upwards of 84% of passing moves lasted for four passes or less

Lengths of Passing Chains from The Bolton v Manchester City Game.

No. Of Passes In The Chain.	Manchester City.	Bolton Wanderers.
1	100	93
2	24	46
3	16	15
4	17	10
5	7	5
6	7	2
7	4	4
8	4	0
9	2	5
10	3	0
11	1	0
12	1	0
13	0	0
14	0	1
15	1	0
16	1	0
17	1	0
18	1	1
Total.	190	182

Much of the previous detail could probably have been guessed at without looking at the actual data. Manchester City are an expensively assembled side who have quality players throughout with the ability to   play a possession based style, while Bolton are traditionally a much more direct team who have recognized the profit to be reaped from delivering the ball rapidly into dangerous areas of the pitch. 

It's only when we begin to look at the combined difficulty of each pass that goes to make up each chain, the input from individual players and the likelihood of chains of differing lengths and from differing starting points producing an attempt on goal can we begin to place passing as a skill into a game context.

We've seen that most passes attempted in a match have an extremely high likelihood of being completed, with the difficulty increasing, predictable with increasing pass length and in areas in and around the opponents goal. 

If we use as an example, Bolton's longest passing sequence which was started by Chris Eagles, midway inside his own half in the 50th minute, comprised 18 completed passes before the same player dragged a chance wide of City's right hand post. Over half of the 18 passes had completion expectations in excess of 80%, the most difficult pass was attempted by full back Robinson on the 12th pass of the move (58% completion expectation), and Eagles contributed six passes to the move in addition to producing the goal attempt. Based on the data and the individual pass expectations, Bolton had about a 1% chance of completing such a move and the last four passes were among the sequence's most difficult attempts. Eagles had around a 10% chance of scoring with his effort. Once we put all these numbers together it quickly becomes apparent why football is a low scoring sport!

This method allows us to pick out not only players who allow longer passing interplay to continue, but also those who may be playing the more difficult passes. Silva and Toure were frequently the most involved players in such Manchester City moves and Silva also linked together small passing sequences with successful "take ons" of opponents. Barry and Milner shared the honours for attempting the most difficult passes of these extended interchanges.

However, given the apparent difficulty in prolonging passing movements and the cash outlay required in player recruitment to build a team capable of producing such sequences, what is the payoff in chance creation? Again sample size is a big caveat, but two factors appear important when turning passing plays into goal attempts. These are, again logically, the area on the pitch where the move originates and the number of completed passes.

Below I've charted how likely it is that a chance will arise from varying lengths of passing sequences, originating from different pitch areas. Winning possession in the final third appears as significant a factor as passing ability. Based on the MCFC data, three passes originating in the final third is likely to produce a goal attempt 10% of the time. To have a similar likelihood of threatening the goal from your own penalty area over four times the number of uninterrupted completions are required. Predictably the combination of last 3rd possession and potent, prolonged passing ability gives a team a better than even money chance of a scoring opportunity arising.

Pass Sequence & Field Position & Their influence On Chance Creation.

Starting Area For Sequence.	Number Of Passes.	Chances of Producing A Goal Attempt.
Edge Of Own Box	13	10%
Edge Of Own Box.	3	0.5%
Last 3rd.	13	65%
Last 3rd.	3	10%

By pooling such factors as passing sequence, final 3rd possession and the ability to take on and beat opponents we can highlight where and why scoring opportunities arise, prioritize attainable recruitment of new players and formulate tactics that play to a team's strength and ability.

Swansea make a rare passing excursion into the final third.

We can also add context to such pass orientated teams as Swansea, who invariably begin their sequences deep in their own territory and therefore probably require much longer sequences to create comparable chances, especially with their exaggerated tendency to pass sideways or backwards. On Saturday, the Swans out passed and out possessed Stoke, but were out shot and comfortably beaten on the scoreboard.