This is the final article in a short series, taking advantage of the international break to baseline where the squad are against the performance metrics. So far, I have covered:
As the team prepare to face Hamilton Academical away on the plastic, before embarking on the Champions League campaign, I finish with data around scoring and goal threat. This series, as well as giving you an insight into current performance levels, allows me to highlight the breadth of performance metrics now captured for every player for every match. I will try and get around to providing a Glossary, but in the meantime if you have questions please leave a Comment or engage on Twitter @Alan_Morrison67
A reminder that unless stated, the values are generally “per 90 minutes”.
Goals! Goals! Goals! (or not)
Expected Goals (xG) are now entering the mainstream football lexicon and are on Match of the Day for each English Premier League game. xG is not a uniform value. There are as many xG models as people prepared to build them. Due to the research and maths behind them, people tend to be very proprietary about them and don’t share. So, you either build your own, or find a way to share, or pay Opta. I have hitherto used a very basic xG model that only considers the location of the shot. I have discovered another “free” one that I will be using from the Hamilton Academical game onwards that considers things like the match scenario the shot came from (corner, free kick, fast break), whether it was a header or not, whether the shot is high or low and to the corner or middle of the goals, as well as the location. EXCITING!
For newbies, xG is basically an indication of the probability of the shot being a goal give the historical evidence. An example: a shot from outside the area but central (width of the 18-yard box) has roughly a 3% chance of being a goal. (Which is why long shooting is generally rubbish).
A player can go on a “streak” where everything goes in e.g. both Ntcham’s goals have been deflected in. Over time though their goals record should regress to the mean, or xG. Players can consistently outperform their xG due to being expert finishers or because they play for a team that provides them excellent quality chances – or both.
Over 11 matches I have a small sample (so be warned), but if you are still with me, you want to see some real data!!!
So, an immediate curve ball – xNPG. This simply means Expected Non-Penalty Goals. The reason for removing penalties from the analysis is that the probability of scoring a penalty is around 77% irrespective of who is taking it. Therefore, including penalties artificially inflates the numbers for the penalty takers. Penalty taking is considered as a separate measurable skill.
Despite being behind Sinclair in real, actual goals, no surprise Griffiths is leading the xNPG ratings. As I showed within the CeltsAreHere article, Griffiths exists to take shots and not much else. I am being a bit unkind, but Griffith’s can hit nearly as many shots as completed passes.
Griffiths and Sinclair have dominated the scoring with 11 between them. Sviatchenko and Dembele being 3rd and 4th is a distortion as they have both only played 115 and 111 minutes respectively. The attacking midfielders are bunched together next, although Rogic is on 0.12 reflecting his penchant for long range efforts.
But overall, I like this graph as it shows a lot of players providing scoring attempts. Nine different players (ten if you include Mr o.g.) have scored for Celtic in 11 matches. Sharing the goals around will no doubt please Rodgers.
Looking at the complete comparison of actual versus expected goals, the final column shows the difference between actual and expected. Both Sinclair and Griffiths have outperformed their xG but slightly. Forrest has performed about as expected for his 3 goals as has Armstrong for his 2.
McGregor is the biggest overperformer – scoring 0.28 goals over expected. Ntcham also overperforms as both his goals owed something to large slices of luck, as did his assist for Sinclair’s goal versus Linfield.
Discounting Dembele and Sviatchenko due to playing limited minutes, the biggest under performer, as with assists, is Hayes. This could be due to bad shot selection, great goal keeping, bad luck. But over both the creative and scoring stats, Hayes has not had the “luckiest” start to his Celtic career.
I am recording Big Chances this year, and using the Opta definition: “A situation where a player should reasonably be expected to score usually in a one-on-one scenario or from very close range.”. Think Sinclair versus Hearts when Hughes header fell to him 5 yards out, the ball was low, and he had no one in front of him or challenging him. Or Sinclair’s first goal against Astana, through one-on-one with the ‘keeper under no pressure. Values are again per 90 minutes.
Again, Dembele has only played 111 minutes (his Big Chance came from a corner against Linfield). And no surprise to see Sinclair and Griffiths leading the way – especially Sinclair who has found some great scoring positions so far.
I have added the recording of blocked shots for this season. It is a good indicator of shot quality, as well as telling us something about defensive actions. Here is the breakdown of all shots per player (values per 90m):
Griffiths again leads the way with 2.21 shots on target per 90 minutes. This is a long way ahead of Armstrong in 2nd.
Forrest has not had the most sparkling of starts to the season yet has 3 goals and 54% of his shots have hit the target. It reinforces the view he tends to not do something, whether it is a pass, take on or shot unless he thinks to can complete it.
Hayes is not far behind reinforcing his lack of good breaks from his start with Celtic.
The most profligate appears to be Rogic, who only has 14% of his shots on target (as a team Celtic manage 31% of shots on target). He has the 2nd most shots off target and the 3rd most shots blocked. He attempts 4.51 shots per 90m, but can perhaps select more carefully when to do so.
Most xG models will inform you that it is better to shoot from inside the box than outside, to have a better probability of scoring. Celtic have several players highly capable from distance that can beat those models such as Griffiths, Armstrong, McGregor, Rogic and now Roberts.
I’ve removed those with less than 10 shots.
Last season, Griffiths hit 45% of all shots from outside the box. As well as his movement noticeably improving, so he appears to be making better shooting decisions, with 71% of his 34 shots this season from inside the box.
Sinclair has been rewarded for taking 88% of his shots in the box with 6 goals. Forrest, again careful, conservative, has 81%.
Rogic and Hayes are both perhaps more optimistic than accurate.
Interesting to see Ntcham at the bottom – he does not break into the box very often, playing much deeper than Armstrong who he is often compared too, but plays a very different game.
To get shots in the box, you must receive possession in the box. This season I measure possessions achieved in the opposition box.
Dembele not played many minutes but it does illustrate that of the strikers, he is the penalty box player. Sinclair gets into the box with the ball more than Griffiths.
Encouraging that young Miller got into the box 10 times to receive possession in 2 matches against Kilmarnock.
Hayes, for an attacking wide forward, is well behind others of his peer group, and can perhaps offer more in that regard.
Confirmation that Ntcham has a more conservative game than Armstrong, only getting possession in the box 0.79 times per 90m.
Finally, more evidence that the days of swashbuckling Brown are long gone. He receives possession in the box, on average, once every ten matches.
I have introduced Packing and Impect several times now. The primer article is Concerning Packing. A really like the Receiver getting credit, and therefore data points(!). What the Pack Receive data highlights is:
- Movement / space creation / getting free of markers
- Ability to control passes from various distances and heights / technique
- Work rate
All otherwise quite difficult to quantify with a “stat”.
A reminder that the Impect score is the total number of defenders (3 points), midfielders (2 points) and forwards (1 point) by passed with a pass and, in this case, successfully received.
Dembele has only limited minutes but already you can see what he brings with his strength, power, and positioning abilities. The forwards should have the highest scores as ultimately it is those players the team is trying to get the ball to score.
What is encouraging is that Griffiths is not far behind. I don’t have Packing and Impect scores for previous seasons, but I would posit Griffiths is showing significant improvement is his mobility, movement, and ability to take the ball under pressure.
Rogic is also notable in that he receives an average of 17.79 passes a game that take out opponents, the highest in the squad. Given he is generally tightly marked playing in a congested area between midfield and defence, that is commendable reflecting his touch and power and strength.
Young Miller was found time and again down the left flank against Kilmarnock and has clearly not forgotten his attacking roots. Early days and small data sample.
Tierney is perhaps lower than I would expect. Teams this season have tended to overload the defence against Tierney and Sinclair. This situation has been made worse by the relative lack of threat from the right-hand side from the more defensive Lustig and conservative Forrest. The return of Roberts changes the dynamics of the Celtic attack hugely. Perhaps this will allow more progression for his mate on the left.
Armstrong and Ntcham are low on the list as they are usually the ones sending the passes through.
When you aggregate all the Packing and Impect actions – passes, receives, and dribbles, you can identify the players that contribute most to taking out opponents and thus getting Celtic closer to the goal!
It is only two games, and it was against a poor Kilmarnock side, but Miller looks to have progressed from his appearance last season, and has provided a potent attacking outlet on the left. Small data and let’s assess when, say 900 minutes have been played.
I am slightly surprised Armstrong is not a clear leader. The Pack Pass king, he is also adept at dribbling past players at pace, as well as receiving the ball in advanced positions in the box. But I hope this metric and concept helps people understand the value of Armstrong to the team.
And Rogic. The Wizard of Oz not only creates for the forwards, he is a valuable outlet for defenders to find, and a safe haven for the ball.
I would expect Sinclair and Tierney to up their rates once Roberts is rampaging down the right, providing greater attacking balance.
Forrest has the least Impect of the wingers and McGregor of the central midfielders.
Much more on this over the season.
Expected Points (xPts)
A new metric I am capturing this season is Expected Points (xPts). This gives a value to each goal in terms of the extent it wins the game. Assuming any win is worth 3 points, even a cup game. It is based on the time the goal is scored, the score at that time and whether you are the home or away side. It calculates a value for that goal. For example: scoring the 5th goal of a 5-0 win in the 87th minute is not going to be material to winning the points (awarded 0 xPts). But scoring the only goal of a 1-0 win in the 87th minute will have significant impact on the winning of the “points” (awarded 1.5 xPts).
Sinclair, therefore, contributes 0.26 points to each game he plays with his goals. Most of his goals have been important to Celtic winning in the context of the match at the time. His goals have been “worth” 2.3 points to Celtic.
You may notice that Armstrong is not on the list. This is because his 2 goals made effectively no difference to the match outcomes coming late in already won matches.
Early days on this one and will benefit from a lot more data.
Scoring Contribution / Expected Scoring Contribution (xSC)
Finally (!), a look at overall contribution to the scoring effort. Scoring Contribution amalgamates Goals and Assists to provide an overall productivity measure per 90 minutes. Expected Scoring Contribution amalgamates Expected Assists and Expected Goals to provide a view on what you could have been expected to contribute given the location of the shots taken and created.
Griffiths and Sinclair lead the way with both outperforming their xSC. Last season Griffiths led the team with 1.45 SC per 90m and has improved on that so far with 1.53. That is 1.53 goals and assists per 90 minutes. Quite the productivity.
In 3rd we have our unlucky Irishman. Hayes can have expected to have contributed 0.6 goals and assists per 90 but his one assist gives him an actual SC of 0.26.
McGregor is notable for outperforming his xSC as do Ntcham and Ralston. Armstrong and Forrest are producing as per expected for the chances and shots they have had.
As mentioned many time, early days and small data samples. But I hope I have given some insight to the data now collected and hope that these terms and concepts will become familiar to you as the season unfolds.