Age profiles of sides have become a hot topic in football, particularly since Manchester City's squad of mainly peak age players began to approach 30 and potentially slip into gentle decline.
The subject is rife with many pitfalls. Survivor bias is well known in more data friendly sports, whereby only the best older players remain active later in their careers and so elevate the apparent abilities of more mature players.
An approach that attempts to account for this bias by measuring the yearly change in ability has drawbacks, most notably even the decision of which metric to chose in a fledgling data environment, such as exists for football.
Even more fundamental is how to express a side's age profile once we have settled on which parameters to measure. A sport that mixes tactical awareness, skills and physical fitness typically sees a player peaking in their late 20's.
In the case of football, different positions appear to have slightly different peaks, keepers having the most longevity and this has been covered elsewhere in this blog.
Average age, adjusted for playing time, is an obvious figure to quote to describe a team's age distribution. But this does invite misconceptions. For example, a team equally packed with youngsters and veteran players may through pure chance have a weighted average age that coincides with the accepted peak of around 28, yet without including one such player in their lineup.
A graphical representation adds more information, but the absence of numbers hinders the ability to quantify the impact of a changing age profile on a team's fortunes.
Manchester City's route to Premier League success was fuelled not only by money, but also a squad that was talented, but also played either approaching or at their peak age.
If we ignore positional variation and for simplicity assume that a Premier League player peaks at between 28 and 29, the plot above clearly shows how the age profile from City's initial Premier League winning squad moves from being predominately comprised of players approaching their peak to players at their peak when they regain the title in 2013/14 and then began to consist of players moving away from their peak during 2014/15.
As an extrapolation I've aged the squad by a year for 2015/16 and plotted how the profile would change to one consisting of a majority of players who would be likely past their prime if the share of playing time was similarly distributed in the upcoming season among the same players from the 2014/15 season.
The various graphs illustrate well the changing profile, but comparative figures would also be useful.
One simple way would be to chose a peak age and then see how far removed the weighted playing time is from this ideal peak for each age group of players. For example, an ever present 20 year old would be further away from the chosen peak age than a similarly, ever present 26 year old.
Under this scale, a side consisting of entirely peak age players would score a cumulative zero, as all minutes played would fall to peak aged players.
Manchester City's 2011/12 team score 2.5 under this method to measure the weighted proximity of their players appearances to the peak age of 28-29. Balotelli contributed 1331 minutes as a 21 year old and at the other end of the age scale, Kolo Toure played 852 minutes as a 31 year old.
As the squad converges towards their peak age by 2013/14 the arbitrary score of 2.5 has fallen to 1.1, reflecting the compression of the distribution around 28-29 years, but begins to rise to 1.7 during 2014/15, as more players move away from their peak. If 2015/16 followed the extrapolation, with no fresh blood added, City's proximity quotient (for want of a better term) will hit 3.1, compared to an average of 2.6 for the previous six Premier League champions.
Of more interest is to use this number in conjunction with a figure to describe the proportion of players approaching the peak age to those who have passed the peak age.
In 2011/12, City's weighted playing time that was outside the peak age was made up predominately of younger players. The ratio of weighted playing time for young players to those past their prime was 9.7. By 2013/14, younger players still narrowly predominated, but the ratio had fallen to 1.3, indicating near parity.
2014/15 saw City fielding a raft of 30+ year olds and relatively few minutes allocated to players aged younger than 28 and the ratio of younger non-peak players to older ones fell below one to 0.36. Older players now predominated outside of the peak age group. (Title winners average a ratio of 3.6, indicating a degree of planning from within the current playing staff).
So, as the plot for 2014/15 illustrates, City's squad has aged, moved away from the peak age and has very few replacements within the current set up.
Monday, 11 May 2015
Jose Mourinho isn’t a fan of corners, apparently. So he no doubt took the opportunity of such a 94th minute set piece at Griffin Park to slip away and beat the traffic following Middlesbrough’s first leg Championship play-off game with hosts Brentford.
Middlesbrough boss, Aitor Karanka, perhaps takes a more positive view of the much maligned corner kick than his friend, Mourinho and he was rewarded on Friday night with an injury time, deflected winner from Spanish defender Fernando Amorebieta.
The value of a corner kick is still debated, but for this particular effort we can quantify it in plain GBP. Prior to the injury time kick, Middlesbrough had around a 65% chance of making to the play-off final, reputable now worth around £130 million, for which they will be around a coin toss to reach the Premier League.
After the late winner, they’ve advanced their chances to round 82%. So in virtual, probabilistic currency, the corner, or more accurately the goal from the corner, has increased the value of Middlesbrough’s chips by around £10million.
Of course the ultimate result is an all or nothing outcome, with Brentford and either Norwich or Ipswich still very much interested parties.
More generally, corners result in much less extravagant potential swings in fortune. To take 2012/13 as a typical Premier League season, teams scored an average of 7 goals in the season from corners, compared to 46 from all other means.
So goals from corners accounted for 13% of all goals scored.
Averages inevitably hide the extremes, only 2% of Newcastle’s goals were from a corner, followed by 6% of Arsenal’s, while Stoke scored 8 such goals and that made up nearly a quarter of their total scores.
Execution also appeared to vary, Arsenal needed to take 260 corners to score precisely half the total goals scored by Stoke from almost exactly 100 fewer set pieces. The Gunners also fashioned just 61 goal attempts from their 260 corners, again compared to 60 by Stoke from only 163 flag kicks.
The style of Stoke, pre Hughes, bore little resemblance to most other Premier League sides, especially Arsenal, but we should still entertain the possibility that the spread of chance creation and goal scoring from corners, seen at its most raw between these two rivals, may be simply due to random chance, rather than differing levels of talent or intent.
Overall 3.2% of corners resulted in a goal in 2012/13, but as we’ve seen there were extremes. Newcastle scored from just 0.5% of flag kicks, Arsenal, 1.5%, Spurs and Swansea, around 2%, climbing to nearly 7% for Manchester United, 5.5% for Wigan (who won the FA Cup with a last minute goal from a corner) and 5% for WBA, Chelsea, and Stoke.
|Swansea take a corner.....Don't hold your breath.|
This spread of conversion rates does suggest we are seeing something in addition to random variation. And it is repeated if we further look at the rate at which teams muster chances from corners. Arsenal fared the worst, creating an attempt from 20% of kicks and WHU were best with a 35% rate of diverting the ball goal-wards.
Shot models also suggest that a typical team would have an average 12.2% chance of converting each opportunity created by Stoke from corners in 2012/13, but just a 9% chance from each of those fashioned by Arsenal.
It is hardly ground breaking, but the evidence suggests that in 2012/13, Stoke were better at creating better and more plentiful goal scoring chances from corner kicks than were Arsenal and that graduation of skills probably existed in the remainder of the Premier League.
Unsurprisingly, the reverse is true of open play.
There is no comparable precursor to open play chances as there is for opportunities made corners, final third possessions would probably come closest if it was readily available. The efficiency with which final third incursions are turned into open play goals may give a fairer comparison for corners to be judged against.
On this occasion, Arsenal beat Stoke hollow. They created 474 open play chances, from positions which would give an average side an 11% chance of scoring, compared to the Potters’ 236 open play attempts, each with an average generic 9% success rate.
So, in the grand scheme of an entire season, corners, as a means to create chances and ultimately score goals were a big deal to Stoke. The Potters scored a quarter of their goals from this method and they were among the best in the league in terms of creation and conversion rates. A third of the Potters’ 25 best scoring opportunities were also made from corners.So they were a rich source of big chances during the Pulis swansong.
Sides which appeared to have both a talent to over perform against the average from corners in 2012/13 and gained the largest proportion of their total goals from such a source, included Wigan, who were ranked 2nd in over performance and 3rd in terms of goals from corners as a percentage of total goals. They were followed by Chelsea, Manchester United, Sunderland and Stoke.
Sides for whom corner kicks and their outcomes were largely an irrelevance, included Arsenal, Newcastle, Spurs and Swansea.
Soccer is a low scoring sport and even at base rate conversion levels for corners, the resulting goals from an average of 11 events per match account for 13% of a game’s total goals scored and when a side, such as Stoke can increase these baseline numbers, the importance of corners to them increases.
Crudely removing Stoke’s scoring from corners in 2012/13 leaves them marooned on 36 points, the same as the final relegation spot, rather than safe in 13th spot.
Not all corners by circumstance produce £10 million shifts in fortune, but whether they are the icing on the cake, as in the case of Manchester United in 2012/13 or a means to survive, as in Stoke’s case, they are important footballing events.
Tuesday, 5 May 2015
Monday night’s Premier League game at the KC Stadium, home of Hull tigers to some, but just plain City to the majority, went largely to expectations.
Thirty three points separated the hosts from visitors, Arsenal and unlike the previous match, when Liverpool made the same trip, league form prevailed.
Ramsey scored the decisive second goal, following an obligatory injury scare, Hull slipped marginally closer to the drop, Arsenal edged nearer to consolidating second and the visiting fans once again baulked at paying £50 a ticket for the privilege of following the evening’s main attraction.
Sanchez and Ramsey, respectively claimed the first and second Arsenal goals, both scored from reasonably enticing shooting positions. But deflections on both occasions rendered Hull keeper, Steve Harper almost powerless to intervene.
Sanchez’ free kick took a 45 degree detour, via the head of Dawson and Ramsey’s goal bound shot looped in off Brady’s trailing leg. Harper had the best seat in the house for each goal and more ever was paid for the privilege.
While single incidents should not automatically validate events as the norm, the ability of deflections to turn decent chances into near cast iron, big chances, was again possibly in evidence. The impact on goalkeepers facing such deflections was looked at here, so now I’ll look at how individual teams have to cope with such unexpected events.
As you’d expect the percentage of shots from a team which take sizeable deflections vary between sides. In 2012/13, Chelsea had over 20 such efforts from their over 600 total shots, just over 3%, whereas Stoke benefited just a handful of times from their nearly 400 attempts.
However, even with Stoke’s percentage deflected shot figures around half of Chelsea’s, there is scant evidence that this is little more than natural variation within a smallish sample of shots.
Manchester City shared top billing with Chelsea, as the teams with the highest rate of deflected shots, but then followed the likes of Sunderland and relegated Reading. Manchester United and Liverpool were languishing among the “unlucky”, percentage-wise.
Chelsea may have been particularly fortunate in 2012/13 and the general trend is that to gain more deflections a side needs to take more shots or headers. Around 2% of your total shots will further trouble the keeper by taking an unexpected detour.
However, sheer weight of shots attempted and allowed by Chelsea, even using the baseline 2% figure will mean that Chelsea will turn standard chances into so called “big chances” more often than will their opponents.
In reality, during 2012/13, sixteen league games were played when Chelsea had more deflected on target shots than their opponents, the situation was reversed in just five games and one match was “tied”.
A rudimentary shot location based model will therefore fail to pick up the higher tariff save required to intercept a shot which begins its life heading in one direction before ricocheting off in a completely different one.
Judged by Harper’s enforced indifference as the ball flew feet from him last night, accounting for the increased likely goal expectation from deflected attempts will produce a different array of possible outcomes, especially for high volume shooting teams such as Chelsea were in 2012/13.
As speculated here, how goal expectation is diced per individual attempt can alter our conclusions in a sport of few scores, such as football.
Therefore, I produced two basic models, one that was aware of deflected attempts and one that wasn’t. I then simulated Chelsea’s 38 game 2012/13 season using the two models. The former greatly increased the number of “big goal expectation chances” created by Chelsea, while the latter produced a broadly similar goal expectation over the whole season, but had fewer big chances.
Disregarding penalty kicks, Chelsea had an attempt which had a likelihood of scoring in excess of 40% an average of once every other game when the potency of deflections were incorporated into the model, compared to half that number in the opinion of the model when that information was not used.
And identifying these “big chances”, even if Chelsea’s percentage was also probably luck driven, as well as volume driven, pushes higher the likelihood of the Blues gaining more than 80 league points in the simulations compared to simulations which fail to highlight this potential source of goals.
On Monday night, Arsenal had 19 attempts to Hull’s five and while Steve Bruce will no doubt feel aggrieved at the deflected nature of Arsenal’s opening two goals, there was around a 34% chance that at least one shot would be deflected during the game, requiring more from his veteran keeper.
Less likely at around 6% was that Harper would have to deal with two or more such unpleasant surprises.