Crowd sourcing Expected Goals, subjectively

Watching Spurs v Liverpool this afternoon, the commentator suggested at half-time that Liverpool could be 2 or 3 up.  However subjective, that statement in itself, is an estimate of expected goals, and got me thinking; how would crowd sourced estimates of expected goals, based on an intuitive feel for what felt like a fair scoreline, compare to the views of the array of objective expected goals models out there?

So, following the final whistle at White Hart Lane, I asked the twittersphere, and here’s what those who kindly replied thought…

Most felt Spurs were worthy of one goal, and Liverpool two goals.

The median scoreline was also 1:2.

And the average scoreline was 0.88 : 1.87

Interestingly, two responders provided fractional scorelines, these being 0.8 : 1.8 and 0.84 : 1.67. Great efforts!

And, here’s what Michael Caley’s method observed:  0.7 : 1.2 (+1pen).

Thanks to all nineteen of you who took part.  I’m keen to explore this further.  Could a subjective approach compete with the objective, could the two methods complement each other?

This approach is not without its limitations.  Collecting the opinions of a reasonable sample size being the first hurdle to overcome, and for all Premier League matches (bear in mind you need to see the game to have an opinion – perhaps something match goers or overseas followers with league-wide tv coverage can help with).

I’ll see how it goes with those games on UK TV for now, but please share, encourage as many to contribute as possible, and we will take it from there.

Please follow the links below to input your thoughts.

Next up….

WBA v Middlesbrough

Man City v West Ham


Have QPR improved under Jimmy? A look at the attack

A few weeks ago I started a series of posts looking at whether, or not, QPR have improved under the reigns of JFH.  In my first piece, I looked at the hard facts (points and goals) but given the volatility of these over short durations, I explored Jimmy’s effect on shots and the quality of the chances his teams created.  If you missed the piece, or want to remind yourself, that post can be found here.  However, the long and short of it is that, so far, the evidence doesn’t suggest improvements have been made.  But, that is in terms of outputs with a team that has been very much inherited, and on which he has been trying to impose his style of play.

But what is that style, does it differ from the style of Jimmy’s predecessors, and is it productive?

To begin to answer this I’m going to focus on attacking productivity and the creation of chances.

Inspired by David Sumpter and his book Soccermatics, I have created risk maps which show the location of positive actions in the build-up to all shots taken.  A positive action is one in which possession is maintained, won, or a shot.  These maps are displayed as heat maps (and thanks to Martin from stats4footy blog for the tutorial).

So here’s how Jimmy’s QPR compared to his predecessors during the 15/16 Championship season.


This first point of note is that Jimmy’s QPR have generated shots from higher up the pitch, from more advanced positions. There has also been greater activity in the penalty area resulting in shots, which is encouraging and synonymous with efforts to get the ball into the front man, mostly the BFG.

However, you will see pre-Jimmy that shots were born dominantly from the wings and the right side central / half space. Under Jimmy, there is a clear left side bias. I watched the opening game of this season with interest, versus Leeds, keen to see if this trend would continue.

Here’s the heat map (, playing left to right.


QPR’s main threat, again, came down the left, with Bidwell and Shodipo linking up well with Chery, more centrally. This clustering of activity raises the question; is this too predictable and easy reading for the on looking opposition scouts?

54The only left sided edition Jimmy made to the squad last season was El Khayati, and bringing through Kpekawa, so has he simply got more out of the existing players and identified the creative strength being on the left side?

Looking at the average positions, El Khayati has brought more width on the left under Jimmy. And, judging by the more closely distributed positions of Hoilett and Chery, these players appear to have had more settled roles under his reign, which I’m sure will have contributed to their improvement in the second half of the season.


Drilling down further and considering the location of key passes, defined as those passes leading to a shot, Jimmy’s QPR continue this left side threat, although across the attacking territory, unlike before his arrival when chances were created from discrete zones centrally and deep on the wings. Also, notice how QPR appear more threatening from corners under Jimmy.

67But we know all shots are not equal, some with a higher likelihood of being scored than others, such as those taken closer to goal. In my last post comparing QPR pre and since Jimmy, I looked at expected goals (xG) as a measure of the strength of shots taken, as determined by their likelihood on finding the net. Ben Mayhew kindly shared the results of his xG model which suggested QPR were worthy of an extra 0.02 goals per game under Jimmy, or a goal per season. Not a significant change at all, but considering Ben’s model considers only the type and location of shots, are we missing other signals that QPR may be better in attack than suggested by the xG totals?

A quick way to answer this is to look at shots on target, in which QPR were 0.54 per game better off under Jimmy; 7.5 additional goals over the course of a season based on league average conversion rates. Perhaps Ben’s xG model has undersold Jimmy’s QPR in attack relative to its view of QPR before he arrived.

The following images replicate the risk maps earlier, but only for those shots taken within the danger zone, in this instance 18 x 20 yards directly in front of the goal, the central area of the penalty box from where shots taken are generally more likely to result in a goal.

8 9

Again, the left side threat shines through and the contribution of central midfield and goalkeeper is more profound. Comparing with QPR pre-Jimmy, I would suggest that Jimmy’s QPR move the ball forward more quickly, and it is particularly pleasing to see danger zone shots being created from within the penalty area.




To provide another layer to our interpretation of the quality of chances created in the danger zone, the following describes the type and origin of key passes, and more evidence that QPR are moving the ball forward more quickly as indicated by deeper long passes.

11Jimmy’s QPR have generated a lot of shots from crosses into the box 10and very few from throughballs. Michael Caley has probably the most advanced public study of xG (see here for more) and he has observed that through balls and danger zone passes are the best type of key pass, whereas crosses are, put simply, bad.





For me, the left side threat is too predictable, but hopefully the arrival of Yeni N’Ngbakoto provides options to vary the approach, and it will be interesting to see if this observation continues into 16/17. QPR look worthy of having the thirdworst xGF in the Championship last season and Jimmy’s arrival doesn’t seem to have brought any betterment in this regard. The preference of crosses is not encouraging, however, it’s very good to see shots being created from within the box. In 2015/16, Jimmy had to make the most of what he inherited which, unlike his predecessors, rarely included Charlie Austin. To his credit, Jimmy found a way of maintaining the team’s overall xGF, despite Seb Polter replacing Chaz at the spearhead of the attack. The closing images compare the two strikers and show just how big a hole Charlie left.

12 13

Joel Lynch: scouting by numbers

Transfer activity at Loftus Rd is gathering pace with the signing this week of Joel Lynch, followed swiftly by the arrival of Jake Bidwell, who has made the short trip across Chiswick to join from Brentford.

Joel, the peak age for a centre back at 28, arrives from Huddersfield where he has spent the last 4 seasons. During this spell he played 15 minutes for Wales coming on as a substitute in a friendly v Bosnia.

Now, I’ve got to admit to knowing little about Joel before he signed so I was intrigued to read on the club’s website that JFH has signed Joel for his “ability to bring the ball out from the back and play”. I am currently writing a series of pieces in which I am exploring how QPR have changed / improved since Jimmy’s arrival (see the first here) so I am interested that he is aspiring to play football from the back.

But back to Joel, and a quick visit to for an overview of his 15/16 season:

joel_lynch 1

So, whoscored don’t paint the most exciting picture.  Based on their average rating Joel was ranked as the 86th best centre back in the Championship last season, of the 119 players to play in that position. Particularly, they identify no particular strengths and note discipline and tackling as weaknesses. In terms of his style, the site’s analysis recognises that he often plays the ball off the ground and generally stays on his feet when defending.
However, Jimmy and the team have clearly seen something in Joel which isn’t captured by whoscored’s rating algorithm.

Delving deeper into the numbers, and with inspiration from Mark Thompson’s profiling of centre backs (see Mark’s work here), I’ve compiled a few metrics which I believe are important to good centre back play, and compared Joel and QPR’s centre backs, against all Championship centre backs for the 15/16 season to have played 900 minutes or more (I feel this is a reasonable cut-off to mitigate the small sample biases).

First of all, a look at the playing style, which can be heavily influenced by tactics and strength of team.

joel_lynch 2

Joel was one of the least defensively active centre backs in the Championship last season. This may surprise as he was playing for Huddersfield who finished 19th. However, Huddersfield ranked in 5th place for shot dominance; for every 5 shots conceded, Huddersfield took almost 6 shots. It would, therefore, be reasonable to infer that they spent more time attacking than they did defending.

There are a couple of other interesting observations when looking at defensive activity. Firstly, QPR’s centre backs did far too much defending for my liking. Secondly, the supporters’ player of the year, Grant Hall, was the least active of all QPR’s centre backs. I will look at how successful Grant was with his defensive actions shortly, but this may support there being more to the position than just getting your boot in. Defending a space so as to put the attacking team off entering it is one such example (but I’m not suggesting this is solely down to one player).

Considering defensive line and passing style, Joel’s profile is similar to that of QPR’s first choice centre backs although there is nothing in his numbers from last season to say that his style particularly involves bringing the ball out from the back. That isn’t to say he is not very good at this, but rather he doesn’t seem to have been given particular orders or necessarily taken it upon himself to do so.

As final note on style, it’s probably not hard to see why Clint’s time is up at Loftus Rd. Well-liked by us fans, in part for his willingness to throw himself into the mixer, QPR’s line was somewhat deeper when he played, he liked a long ball, and he doesn’t seem to tick Jimmy’s box of having sufficient ability to carry the ball out of defence.

Now, let’s look at ability across various defensive metrics.

joel_lynch 3

I suppose I had better start my commentary of the defensive metrics considered by stating that I don’t believe in statistical scouting in isolation, equally I don’t believe in the traditional eyeball test on it’s own (unless as a scout you have time to intensively watch a player over half a season).

Based purely on 15/16 Championship data, Joel Lynch looks like somewhat of an underwhelming signing. The only attribute he appears to shine in is aerial ability, in which he was 9th best performer out of 80 centre backs in the Championship last season. Other than that, he was poor to average when tested for other attributes. He was 17th in terms of likelihood of giving away a foul, 66th in long ball success, and only 12 other centre backs gave the ball away more. For other metrics, Joel’s performance could be considered average.

Now, I’m not one to label this a poor signing before Joel has been integrated into the team and tested against Jimmy’s demands. The centre back position is one of the more difficult positions to analyse statistically with publicly available data, particularly because of the importance of good off-the-ball positioning, the data for which isn’t available publicly.  But compared to what QPR have in this position already, Joel looks like a squad filler.  He may be an improvement over those who recently departed, and his importance to the squad may hinge on the club’s recruitment of a right back which will determine the use of the Chief out wide.

Have QPR improved under Jimmy

The close season should provide most fans with renewed optimism ahead of the approaching season as clubs scramble to refresh their squads, build on the previous season’s successes or look to put past horrors behind them.  The last few seasons have been a rollercoaster for QPR, on the back of which the club appear to have settled on a sustainable strategy going forward.  Given the issues that had to be addressed after the 14/15 relegation season, last season’s return to the Championship was always going to be one of consolidation.  A 12th placed finish was achieved but not without a wobble which seen Jimmy Floyd Hasselbaink replace Chris Ramsey at the helm (via a cameo from Neil Warnock).

So what does the future look like for QPR under JFH?  In a series of posts, I intend to delve into the numbers to analyse Jimmy’s impact at Loftus Road and the style he appears to be pursuing at W12.  In future posts, I will consider how QPR’s attacking and defensive play has changed, if at all, but not before first considering whether QPR have improved during his first 27 games in charge (I’m only considering Championship games in 15/16).

The table doesn’t lie

So the saying goes, and so I will entertain the advocates by first taking a look at the league table under JFH and his predecessors.


The increase in number of draws since Jimmy’s arrival is the obvious difference, but by looking at the above as a proportion of the number of games played, a few more differences are obvious.


The increase in draws has had a clear effect on achieving fewer points per game (PPG), driven by lower goal totals, 2.19 goals per game v 2.58 pre-JFH.  It would be reasonable to assume the mechanism for this reduction in goals is the adoption of a more defensive approach, which in turn has impacted on attacking productivity.  I will explore this further when looking at shots and again in future posts when comparing team styles.

In terms of the share of goals (Goal Rate, GR), there has been a slight improvement with Jimmy’s team outscoring opponents, but nothing to get too excited about.

Fine margins

Football is decided by goals; such rare events that using them to differentiate between teams on a game by game basis is limited due to fluctuations in scoring rate and luck. Over a season, these fluctuations smooth out and better teams tend to outscore lesser teams, often driven by a greater share of shots on target (SoT) which in turn is usually driven by the share of all shots (TSR). Shot ratios are a good predictor for goal rates. So, beyond looking at the goal rates alone, is there anything in the shot numbers to suggest Jimmy has improved QPR?

TSR is the share of all shots, SoTR is the share of SoT, Dominance (Dom) is the number of shots taken for every opposition shot (or SoT)

QPR have given up a greater share of both total shots and shots on target since Jimmy took the hot seat. These differences aren’t statistically significant¹ but what is more interesting is that the volume of shots (and SoT), both for and against, have increased. This observation doesn’t support my earlier suggestion that goal totals may have reduced since Jimmy’s arrival due to a more defensive approach. Rather, the increase in draws is explained by an increase in shots against such that shot dominance has tightened, most notably with regards SoT.

1.4Left begins to rank the 15/16 Championship sides for shot dominance. Reading are an outlier as are Burnley who don’t even make the extract with a ShotDom of 0.83, but other than that you will see ShotDom correlates well with success. In fact pre-JFH, QPR were on course to post a ShotDom that wouldn’t have seen them look out of place in the play-off positions.

However under Jimmy, QPR have been more accurate with their shooting having registered a higher proportion of shots to SoT (Acc F), although the benefits of this have not been seen due to converting SoT to goals (score%) at a below average rate. In defence, QPR have allowed a greater percentage of shots into SoT (Acc A), while performing above average in goal.  In fact, the increase in the number of saves made per game under Jimmy is statistically significant (t-test p=0.03) which I will explore further in a future post. Pre-JFH, QPR performed slightly below average in both Score and Save percentages.

Acc Balance = AccF + AccA, PDO = Sc% + Sv%

A few words on PDO which quantifies the balance of Score% and Save%. Pre and since Jimmy, QPR’s PDO is within the typical range expected (90-110%). PDO is often considered a measure of luck, as it tends to regresses over the course of a season, and beyond, to within the typical range described. You could therefore argue that QPR have been more fortune since Jimmy’s arrival.

Not all shots are equal

Expected goals (xG) provide an empirical measure for the worth of shots or the quality of chances created, described by the probability of a given shot being scored. As with the other metrics considered, xG correlates strongly over a season with performance.

Ben Mayhew has kindly provided his assessment of QPR’s xG for the season (check out Ben’s work here which includes some great visualisations to describe the goings-on in the lower leagues).

1.6Keeping with the dominance theme, QPR have fallen short in terms of the sum of the strength of chances created both pre and since JFH, although worse so under JFH. There’s not much between the average strength of shot for and against to put Jimmy’s inferior performance for this metric down to giving up or creating lesser chances. Instead, QPR’s decline in xG is driven by an increase in the number of shots conceded.

The Verdict, so far

On the face of it, QPR have performed slightly worse since Jimmy’s arrival.  Points per game are lower, and they have given up a greater share of shots and SoT.  Goal rate has improved, thanks to a much improved Save%.

However, before we dump on the current regime and concern ourselves that the future is one of Championship mid-table obscurity, the 15/16 season has been one of much transition with a few of the big names having been moved on to be replaced with lesser known potential from the lower leagues.

And, for at least a third of Jimmy’s reign, we may not be comparing hoops with hoops.  It would be fair to question the balance of motivation in QPR’s last 9 games or so, as it was from this point that the playoffs were pretty much out of reach, yet 6 of the last 9 games were against teams with something still at stake.  With this in mind, my next post will focus on how QPR have performed against expectations and attempt to see if they got the goals and points they deserved under Jimmy last season.


Note: ¹ A T-test was used to determine statistical significance

Fancy Stats: The Championship so far

I have been playing about with different ways of presenting team level Championship stats in recent weeks, and settled on the following approach (for now anyway); the Fancy Stat League Table.

These tables show the rank of a team for particular metrics – the higher the rank (or taller the bar), the better a team has performed in that metric compared to their rivals.  An important limitation of these tables is that they provide no scale to the stats presented; Brighton sit one place above Middlesborough in terms of points per game, but by how much?  The tables don’t tell you that.

So, why this approach?

Well, I think (hope) they are easy on the eye.  The majority understand a league table and, personally, I find that the quicker I can understand a graphic, the better the graphic.

I did mess about with graphics displaying the raw numbers but unless you know the typical range of the metrics presented, this approach probably wouldn’t mean too much.  Feel free to recommend alternative approaches, I’m all ears.

Anyway, here’s how all teams have performed so far, presented in the context of this weekend’s fixtures, but first a few words on each metric.

Points/game – this replicates actual league position

Points v Expected – this provides a view of performance in terms of bookies’ expectation, and provides an indication of over/under performance

Actual goals – for, against, and rate (for/total goals) – goal rate correlates very well to league position at the end of a season

Expected goals – as for actual goals but this provides some context in terms of the quality of chances taken and conceded.  This is based on the excellent Ben Mayhew’s model which you can read more about on his blog – go check it out, there’s loads of great analysis across England’s professional game.

Finally Shots, total and on target – again these correlate well to final positioning however treat with caution, score effects, particularly at close game states (where a team is leading or trailing by 1 goal), and strength of schedule can skew these figures.

I’m looking forward to see how teams perform over the season and observe any regression.  I hope also, at some point, to introduce form tables.

Final note, and a generalisation which may be worth exploring further; tables which are left hand heavy (points and goals) may suggest a level of over-performance, right hand heavy (expected goals and shots) might identify teams that deserve a break.

Two teams due a positive bounce. Bristol City performing better than their save % suggests, Forest due some fortune in front of goal
Oh QPR, you pain me – Looking very ordinary and that variation in goals scored v expected is worrying. Birmingham’s results to date look unsustainable
Two teams that, if continue in this vain, are in for a tough season
Burnley very precise with their shooting, but it will be interesting to see how sustainable their performance is given low total shot rate. Bolton look like they are creating good chances but been unfortunate in front of goal.
Derby creating good chances but having difficulty turning shots into SoTs. Opposite true in terms of shots conceded. Wolves may encourage a reverse  in this trend giving up few shots but with a high percentage on target
Score effects likely to be flattering Huddersfield somewhat but need to improve in attack if going to climb the table.                                                                     Ipswich may struggle to reach heights of last season but looking better going forward than in defence.


Brighton, Look, Impressive! They are roughly 6 shots (2 SoT) better of than their opponents per game; I’ll be watching their score rate keenly, which is relatively low. Leeds look mid-table at best.
Score effects likely to be underselling ‘Boro here having only found themselves trailing by a goal in 2 games so far. Only Brighton are posting more SoT/game than Fulham who might find themselves in the league’s gutter if they can’t sustain such shooting precision
Blackburn are posting good numbers and may be a force if their scoring percentage can take off in time. The Dons numbers don’t suggest they will be seeing a resurgence.
I suspect Preston games have seen battle after battle in midfield this season; giving little up in the final third and hardly threatening in attack. Cardiff results possibly flatter so far.
That 6/1 for Reading to be promoted just 3 weeks ago looks massive right now; probably best described as one way traffic. Charlton are enthusiastic in attack, but won’t be anything other than a participant unless improve defensively.
Wednesday look like they are dining out on an unsustainable score rate at present, while Hull are living up to expectations as strong Play-off contenders


Poll: Defining the ‘Best League in the World’

The phrase ‘Best League in the World’ is a term often bandied around, sometimes sarcastically, by many with an interest in the English Premier League.  I’m interested to analyse the factors you consider important in determining the quality of a league.

If you would like to have your say, and help inform a future post on this matter, please complete the short poll, below.

How important are the following factors, on a scale of 1 (least important) to 5 (most important), in identifying the quality of a league.




Boardroom Decisions – If QPR did Moneyball this summer

I had hoped that this post would present the findings of a detailed look at QPR’s better performances this season in order to find the magic formula which will save us from relegation.  Well, I’ve decided that would be a waste of time. The official line of the club might be that QPR can still survive, but it will take three wins, including away at Man City, just to have a chance.  So, rather than focus on the impossible task of staying in the Premier League, I thought I would look ahead to life back in the Championship, and in particular, consider the uncertainty that surrounds the make-up of the squad.

To quantify this uncertainty, here’s a look at the distribution of contracts by expiry (all data, unless otherwise stated, from  Data considers the regular 1st team squad, including those out / on loan):

QPR contract expiry

Almost half of the squad are out of contract at the end of the season.  While this might appear alarming in that it potentially leaves QPR with a massive rebuilding exercise to do, it gives the club options at a time when their financial power is reduced.

However, it is more important to consider the age distribution of the squad, in particular, the ages of those on expiring contracts.

Goalimpact has carried out extensive research into the effect of age on performance, and developed the following ageing curve:

The Football Aging Curve

The curve suggests that ability / performance increases until the age of approximately 26, after which it tails of before really deteriorating from 30 upwards.  It can be argued that peak age is between 24 and 30 years. This is, of course, a generalisation as individual traits, injuries, and position will impact on an individual’s curve.  Note, goalkeepers are excluded from Goalimpact’s curve due to their longevity and tendency to peak later in their careers.

QPR ages

The plot to the left shows QPR’s age distribution for the regular squad.  The age profile of the regular squad, as it stands, does not make pretty viewing.  60% of the squad are between the ages of 24 and 30, so the majority of the squad is at peak age.  However, it is currently skewed towards the elderly rather than youth; worryingly, there is little to no pipeline of players on the rising side of the curve.  (The contributions of some of QPR’s youth / development squad are noted, but I don’t consider their inclusion to have been frequent enough to warrant inclusion as part of the regular squad).

QPR ages 1yr+

To the right, the age profile of the squad excluding those on expiring contracts.  Of the expiring contracts, three are loanees, and eight of the remaining nine will be 32 years old, or older, by the end of the season.  Only Ale Faurlin is at peak age on an expiring contract, but considering his injury record, the view may be that this is not necessarily a bad situation for the club.  Only one player above 30 has a contract beyond the end of the current season, Rob Green.

In search of finding the answer to this issue, I turn to the principles of moneyball and the findings of the excellent statistically informed book, Soccernomics.

Expiring contracts

Serious consideration should be given to allowing these to elapse, and release the players concerned.  This will create a huge whole in the squad, and at a time when the Chairman is looking to tighten the purse strings and not necessarily spend large amounts on transfer fees.  However, Soccernomics suggests that net wage spend is more important than net transfer spend.  The expiring contracts are worth approximately £350k per week (after struggling to find a source for wage data, I settled on using the wages described in the Football Manager 2015 database – I’m hedging the integrity of this post on this wage info being there or thereabouts).  Even if QPR only replace these players with younger players of similar ability, they can expect to do so on reduced wages.

New signings

QPR have a huge shortage in pre-peak players; they need to invest in youth.  Players can be developed more before they reach their peak.  It should be expected that spend on pre-peak players will be an investment as players should hold more value once they reach peak age.  Working within the constraints of financial fair play and the availability of funds, QPR should not be afraid of a negative net transfer spend if a significant investment is made on youth.

Charlie, Leroy, Sandro, Matty and Steven Caulker

These are all players that I would love to see at QPR next season.  They will all be at peak age next season and I believe they would shine in the Championship, providing a decent spine of a team around which to rebuild, but the reality is that they are likely to want to join / be in demand of a number of Premier League clubs.

The club should try to hang on to these players and only sell at the right price.  Soccernomics suggests that you should sell a player if offered more than they are worth, and identifies that centre forwards cost more than they should.  English, Dutch and Brazilian players are often overrated and, therefore, overvalued. values these players at £32.5m.  If the club can recoup £45m+ for these players, or equivalent pro-rata, they will be on the right side of moneyball.

Rob Green

Rob Green has been in good form this season, posting a save percentage which is above the league average.  He also has another season on his contract and will turn 36 during the next season. value him at £880k and he is English and on £50k pw.  Although goalkeepers have a long life at the top of the game, Rob will be on the downward side of the curve by now and QPR have a decent alternative in Alex McCarthy.  If QPR can get near to his value they will boost their transfer kitty and significantly reduce their wage spend with little impact on the quality of the goalkeeping position.

In summary

There are some big decisions to be made in the QPR boardroom over the summer.  The uncertainty surrounding QPR’s squad was always going to be there this close season, regardless of whether they were going to be in the Premier League, or not.  Failure to secure Premier League status will certainly make rebuilding the squad a more difficult task.  However, the contractual situations provide a real opportunity to address the issue of QPR’s age curve. The research suggests the club should be less concerned about the size of any hole they will need to fill in the squad and the associated cost of doing so, but instead strive towards investing in youth and a relative reduced outlay in wages.