Untold Arsenal on Twitter @UntoldArsenal
Untold Arsenal on Facebook here
By Arvind
There is a lot of great work going on at Untold about various kinds of statistics, that affect Arsenal results – both positively and negatively.
Just recently, one article which talked about how Gael Clichy’s statistics were nowhere near as bad as you would think they would be, if you listen to all the bad things they had to say about him. However, while the statistics do tell one part of the story, they seldom, in my opinion tell the complete picture. So for example if a player had 15 shots on goal; it doesn’t necessarily mean he is the best striker in the world – a lot of those shots could be from way out, with very little chance of a goal being scored from any of them.
Similarly you could argue that a winger who seems to cross every 5 minutes is not necessarily good, if there’s no one in the middle who can receive his crosses or if his crosses are too high or too low, giving defenders the chance to come in and clear. There could be numerous similar situations for every player in every part of the park. So effectively, a completely objective analysis for me does not tell the story.
So what could possibly be done to read and make even more sense of all those statistics that we see? I think someone needs to watch the game; end to end, understand how the opposition was playing that day and how we adapted our game to try and win it.
So for example: Let us say we play United and say that Nani is the biggest threat and he is primarily playing on the right wing. Now the LB on that day decides; “Okay I won’t get forward too much today, as Nani is on form. I will try and stop him”. So that day you will see that the LB did not get forward and cross too much. His overall stats hence do not look good at all. The truth however is, that in the context of that game, he’s done a great job.
Now we all look at games just like this, and have our opinions of players after the game. Then when the transfer season comes up we use those opinions we had of certain players and decide in our minds, about whether they should be sold or retained at any cost. I’m saying, these opinions are not necessarily accurate for the simple reason that there are too many factors per game for us to remember at the end of each game..let alone the entire season, to decide whether a player did well or not.
That’s where I thought – “Wouldn’t it be a great idea if someone could make a list of parameters per position and rate every player on each of these against every opposition all throughout the season?” The person would have to actually watch the game, of course and take into consideration all the factors, which decide the player’s performance. If we had all this subjective data, we could use it at the end of every month, maybe to find out how well one has actually played. This could go hand in hand with the statistics that are put out on a regular basis and provide a more complete picture of what is actually happening.
Lets take an example so you’re clear what I’m saying here. The key parameters on which a fullback at Arsenal is judged..considering we play the ‘Arsenal way’ are in order of importance:
— Recovery Pace to get back into position
— Interceptions all over the respective wing; the further up the better
— Width while attacking
— Kick starting a counter attack
— Tackling
— Heading ability
— Defending in the box
There will be other attributes of course, and maybe the order above is not ideal..but the point is.. On any given day the fullback must be judged against all of those parameters and only then can his performance be decided. Some parameters on a certain day might not be relevant at all. For e.g If a team doesn’t launch a lot of diagonal balls or if there aren’t too many corners or if the ball never came into the fullback’s “zone” – Heading ability means nothing and can be a “NA” for that game.
This is subjective of course, and ratings per game would really depend on the viewer, but I thought its an interesting experiment worth conducting for a few games at least, to see if it is beneficial and helps us understand our players better.
What do you think? Is it an experiment worth trying? I would love to hear your thoughts.
Untold Media: Launch of new media section, and we need your help
We need a guy with decent defence skills
Well it sounds like a gud idea 2 me. Then we wuldnt condem players unjustly. Clichy was wrongly judged prior to his exit. Accepted he made a few blunder which in turn cost us goal.dat dosnt make him a bad player.we all av our bad days.
Just watch cashley cole, its not rocket science its football!!!
Arvind,
Spot on: Great perspective you take and points you make. Quantitative measurements alone tell a necessary but NOT sufficient part of the story. The specific position and the match context cannot be discounted. These are qualitative/interpretive components. People act like so-called objective counts provide conclusive results. But you beautifully point out that more is needed to approach anything resembling conclusive. A position-by-position set of parameters would be something that one hopes our scouts and decision makers either have internalized or submit as part and parcel of their internal reports and recommendations on potential signees. I’ve advocated for a second poacher/striker to complement RvP; to clinch the argument I’d turn, in part, to analysis of how “clinical” or “profligate” our current players have been within the 6 yard line, etc. This would require not only counts of goals per shots (goal:shot ratio), but also a situational (qualitative) analysis of each of the players shots in its context. I haven’t given a refined enough argument here; but only mean to agree with you that quantitative (counts) and qualitative (contextual) perspectives must be joined for a far better and usable perspective. Quantity alone appears authoritative, but it too requires interpretation in the end and is only one part (and an important part) of an adequate analysis. I appreciate your insights and a chance to chime in with mine. Kudos.
unfortunatly, you won’t find a left back with all these attributes round the corner and if you do, you ll get the price. anyway great post
great left backs seem to be very rare
Thanks Bob. Glad you think it is useful. Do you think it would be useful if I say made up a complete list and tried to watch as many games[whatever are telecast in India :)] as possible to rate players?
Also yes..your RVP example is precisely what I mean. You got it spot on. The only key is defining context..which is in the end subjective, but it should be okay as long as its consistent.
@All: The point isn’t above left-backs specifically nor are those attributes the only things that need to be looked at. Those are just a sample. Its about judging a player fairly game after game against a defined set of parameters [which are less or more than what I mentioned]. I just wanted to bounce this idea off you guys and see if it is worth putting in effort to devise a system for the entire squad.
The idea is definately good. But it will depend on who does the review. Plus lots of statistics will also be eventually needed, we might even have to take help from Opta Joe. And also many other factors should come into play. Also, it will necessary that the reviewer should not be biased towards any players as well. Overall, A tough job, but great idea.
Arvind,
It’s a bold and super idea. There would be variations within a role of course depending on the overall formation. Why not start with one backline, midfield and advance position. But you could start with a set of parameters for each position you decide to take on and use it as a starting point. (What the sociologist Max Weber called an “ideal type.”) Then the realities of which formation and actual match conditions will lead to variants of the position-model (left back v1, v2, v3). Also, I wouldn’t get hung up about your definition of context being subjective or not. There’s no pure objectivity – that is a myth. There’s an interpretive component to analysis, whether it’s a quantitative survey or a qualitative case study. There’s no pure numbers that speak for themselves. Why? Because their meanings must be interpreted. So I’d say do your thing, man. Make your initial positional model (based on your experience); get some other people’s views on what tweaks are needed to refine it; take on the one’s that make sense to you and modify your positional model, and there you go. Do your statistical counts for any given match and test the model(s) out. If it’s too ambitious and you have no life, then cut back to a smaller number of positions, instead of all 11. Then you could decide to build it back up to what you can handle. It’s a huge project and it may take you a full half season to come up with something that’s more useful for the next half season, etc. I think it’s exciting and while you do need statistical counts, the value of your modeling does not live or die on having a massive amount of statistical counts. A case study that both combines qualitative and quantitative perspectives is honest and might be useful or not. You have to do it to find out. (Anyway, maybe some of this was useful? or not? It’s enjoyable to think about.)
Arvind, I think it is a great idea and it would be very helpful if someone would do such things. I’m sure Wenger does something similar. 😉 Maybe not in person but he will have people on his staff doing this.
Like you say numbers are needed and fine but there is more than left/right back gave 12 crosses in a game. If they result in 6 goals or even chances this is great. If the all go over the goal harmless he has done bad even if the numbers give you the impression he did a great job.
maybe we could brainstorm a bit about what attributes are needed and should be monitored? I think your list is already pretty complete in a way. Cant really think of something to add to that for the moment
I think it should also include the “normal” numbers like the how many times he touched the ball, passes: completed/incomplete, dribbles successful/unsuccessful, winning the ball, losing the ball… you know the more used ones…
And also maybe helping out the centre back or unnecessary rushing to the center leaving his wing open…
Now we just have to find 11 people who each can do one position and we would be in the know… 😉
Untold player watch… is something new getting on his way?
Oh, and as I know a few things about reviewing a game: most of the time it is fun and I can promise you that you will see many things you didn’t see before in each game. Even in the painful games to review….
Arvind, yes, all well and good – from the positive side of defending. But what about, for example, mistakes made that lead to goals – whether mis-positioning, mis-tackling or just not getting back quick enough. After all, those are the ones the fans notice.
And statistics don’t measure the amount of confidence that teammates may feel when playing with certain players, something which might affect their own performances – as a possible example (which may not be a good one) would be performances by defenders in certain matches when Manuel Almunia was in goal.
Or maybe our players are all super-human and have no nerves? 🙂
good idea, untold player watch sounds like an interesting solution. maybe we could start with a new signing, a regular, and a youth that has broken through?
sounds like a massive task to collect all that data as i agree with gooneraside all stats would have to be included, though i wouldnt agree with the confidence claim as being able to control your emotions on the pitch is an attribute sought after on the pitch imo.
would there need to be two people per player? more? sounds huge. different way to watch a game i suppose.
I don’t mean to be a negative Nellie, or a damper on the enthusiasm you, and others have for this. Please don’t let my disagreement turn you from your desire to carry out this analysis.
But I couldn’t just help but think how massive a data set you are talking about, and was reminded of Charles Reep (http://en.wikipedia.org/wiki/Charles_Reep) who spent his life in analysis of where each pass was made from and how goals were scored, but ultimately drew the ‘wrong’ conclusions from all of that data. The kind of mix of objective and subjective data you are talking about is probably the holy grail of football statisticians. I really don’t think it is actually possible to do this independently. Incorporating all the different dimensions of team play, for each player in every position, in all the various events for a football match, and then of opponents and different players they may have, adding the referee’s influence on the game etc etc.. is simply too much data to incorporate, and then present into one meaningful whole. If you could somehow write a computer program to do it, then it is possible to get close, and you would probably make millions off of it if you sell it to the clubs.
As I said, I don’t mean to be negative. It’s just that I’m not convinced it would actually lead to a better analysis of the game. However, if you do want to do it, as a suggestion, you may want to explore the possibility of a tie up with Zonal Marking (who cover the team aspect fairly well in their analysis), and maybe paying for OptaJoe’s stats.
@Shard
Charles Reep was entirely correct in his analysis. His conclusions are the fundamental basis of Wengerball. You’re merely viewing his conclusions through the filter of how they were implemented.
The problem is that Reep deduced that less than three passes was ideal. It would take six or seven short passes to take the ball from defence to the striker, therefore everyone assumed that you had to get the ball upfield as soon as possible, without regards for accuracy. What you need is to be able to consistently execute medium to long passes, as our players are trained to do.
Accurate short and long passing is a prime requirement for any Arsenal player. A twenty yard pass to a midfielder who then executes a thirty yard pass to an attacker remains one of the best ways to score a goal.
Assuming that a long kick by the average player has a 1/4 chance of reaching the intended player and a short pass a 1/2 (where three short passes are required to cover the same distance), the average manager would instruct his players to favour long passes (since four punts will result in one decent long move, compared to eight attempts by short passing). It’s all a matter of percentages.
@Woolwich
The statistical data Reep gathered was correct, and even his conclusion of 3 passes was correct. It was a statistic and as such was correct in a limited context, or rather only in a certain context. That’s why I said ‘wrong’, and not wrong, although it wasn’t just that people assumed something on the basis of the data. Reep himself advocated the long ball.. (I don’t agree with the assumptions of 1/4 and 1/2 though i agree with the point you make by using those figures)
The point still stands that firstly it’s a massive data set to just collect. Which CAN be done even though it is difficult. The problem comes in applying the subjective to it, while retaining the sense of what you are trying to show. All the variables simply cannot be computed in a satisfactory fashion, especially without some internal access to the club. In the case of the LB dropping back if there is an attacking player down his flank. That is just the viewer’s assessment of what is the best thing to be done. But, how do you account for the manager’s instructions in that maybe the idea is to push the attacker back and force him to cover? I just feel that the entire exercise won’t yield any concrete conclusions or even reliable trends, since there will be simply too much unaccounted for.
@All: Thank You very much for all your positive comments.
@gooneraside: Yes, good point about the negative side.
@All: Yes its a massive amount of data. I’m aware of that.
@NegativeNellie (Shard ;)): I think what you’re saying is perfectly logical; but I just want to start small. Its not going to be perfect initially and it may not be until next year. I will just keep it very simple and take the most critical attributes only and not complicate it too much – till it reaches a stable state. If it works well, hey who knows.. lets see..
The subjectivity is there..of course; that is the point of the whole exercise. I’d liken it to our appraisal systems at work; a fully quantitative system wouldn’t be fair and a wholly qualitative system is too lax. The best is somewhere in between. I want to try and see if I can do it .. with your feedback 🙂
As Bob says… the worst case is .. “It Flops”. That’s not too bad.. at least we can try 🙂
@Walter: I will work on making a complete spreadsheet of the most common parameters and share them with you and whoever else is interested (Do let me know). Once this sheet is final in my mind – I’ll try and give it a spin. All your inputs will be most helpful in getting the say 5 top parameters ; per position?
@Bob: Thank You for your encouragement. Hopefully this works. If what I’m trying does work, maybe more of us can get involved.
@Shard
I think I sort of agree with you that the dataset is not readily objectified and will inevitably suffer some subjective bias.
Doesn’t it’s not worth doing.
@Arvind (& Woolwich)
I never meant to say that you should not go ahead with it. I think if you want to and think you can offer us something worthwhile then I will be in line to congratulate you (and will for the effort even if ‘it flops’). I just wanted to give my opinion, which is of scepticism about the effectiveness of any such analysis. I’ll be more than happy to be proven wrong though, and wish you the best of luck in the endeavour.
Just felt like waving this flag again on behalf of Arvind’s journey: There’s no such thing as pure objectivity; quantification does not mean Truth; and there’s always an interpretive element to all analysis (some part of which is called subjectivity – which is hardly a pejorative, as it is sometimes used, as if any taint of subjectivity is enough to disqualify the value of a truth-claim).
@bob
Agreed. And I don’t oppose it at all. Who am I to say it won’t work, or it doesn’t make sense, or anything of the sort.. I just don’t think that the quality of the data such an exercise will produce will be any better or reliable than what we already have.
@bob
Agreed. And I don’t oppose it at all. Who am I to say it won’t work, or it doesn’t make sense, or anything of the sort.. I just don’t think that the quality of the data such an exercise will produce will be any better or reliable than what we already have.. But I could be totally wrong about that, and I even hope I am.
Hi Shard,
yes, for sure, and we’ll all learn a lot about what’s possible from the undertaking… so Arvind, check in and let us know what you’re turning up, or six months of silence will go by all too fast!
It can be quite challenging to have objective metrics that most of the people can agree upon. I think it is advisable to avoid any subjective metrics as it will just lead to argument without base. So perhaps, we can start by identifying some quantifiable measurements that can be implemented. For me, some of the quantifiable measurements are:
— Interceptions with regard to ball position and direction.
— Width while attacking
— Kick starting a counter attack measured as the time it takes for the ball under possession to enter the opponent final-third area.
— Tackling with regards to position and the direction of the ball.
— Heading ability
AnT,
“subjective metrics”? *(I promise most readers this will get a bit tedious, so, my apologies in advance.) Alas, when you put it that way you set up a straw man that’s easy to knock down. Nope, you can’t escape the acts of selection and omission that go in any definition of fact (or that establish a working category as you do), and that always has an element of judgment which means an act of interpretation which has a subjective element. There are no pure facts that speak for themselves. The need for quantification (call it metrics if it floats your boat, or even sabre-metrics if it gives a rush) is real and crucial, lord knows. But quantification is not sufficient to call it Reality. And it’s no guarantee (as you indicate) of something that’s argument-free. And yes, that said, quantify, quantify, quantify. No problem. But please don’t delude yourself into thinking you’ll achieve a sufficient let alone total explanation as a (tentative) result. The point is to be as conscious as possible of your selections and omissions, your method and your model, and then quantify going forward based on these semi-subjective choices. But I’ll tell you what, doing it your way will get you grant money; admitting what I say will invite ridicule and park pigeons for company. Cheers, Arvind, on with the show!
Bob,
perhaps I didn’t put my thought clearly.
The quantifiable measurement is just a starting point. Afterwards is the interpretation phase in which we can try build a theory for analyzing the performance of a player.
As I said earlier, it can be really challenging to get some good quantifiable metrics that we can use as a starting point. As we may have noticed of what Walter has done for his referee analysis, we may need/want to change the way we measure or even the subject to be measured during a season. However, it is best to first define the most applicable metrics before we started so that there is less need to change it in the middle of the season.
AnT,
Thanks, mate. We’re closer to full agreement. And I hope you enjoyed the match.