The ability for us amateur analysts to access Opta data through front end publishers such as StatsZone and Squawka has probably been the single biggest reason for whatever advances have been made in the field of football analytics over the last year or two.

But, I discover that having access to this data now causes me problems.  I find that I now always have a desire to dig ever deeper into the available data; I have a need to go further and further into the Rabbit Hole.  Believe me, it’s a bloody frustrating place to exist in.

So why am I going all philosophical on my readers?

Key Passes metric

This was sparked by the excellent piece published yesterday by Adam Bate on Mesut Özil where he uses metrics such as Key Passes and Key Passes per90.
As Ted Knutson remarked on Twitter, it is great to see the emergence of metrics such as Key Passes being used in the mainstream media and it’s a clear sign that analytics is slowly making its way into football parlance.  The use of this metric helps put an objective measurement around the creativity of footballers.

Before the Key Pass metric came into existence, reporters would have had no option but to gush in subjective terms about the “wand of a right foot that player x possesses” or at best they perhaps could have cited one or two specific examples of great passes.
Thanks to the Opta recorded measure of Key Passes it is accepted that it now possible to move away from relying solely on subjective measurements and be able to construct a synopsis of a player on undisputed facts.  This synopsis can then have the “soft factors” added into it by the writer to complete the picture of a creative player.

So what’s the problem?

I’m not totally happy with the use of Key Passes being used to measure the creativity of a player, because I know we can do better.
Now don’t get me wrong, I’m much happier with using this measure than not having any objective measures available to us at all, but the drawback when using Key Passes is the implicit assumption that all Key Passes are the same.  I say this as the data providers (edit – just to clarify I mean in terms of the standard metrics that are published) do not distinguish between the relative effectiveness of different key passes, it therefore follows that users of this stat just cannot distinguish between different types of Key Passes.

Coutinho v Sissoko

I’ll use the example of the Key Passes created by Liverpool’s Philippe Coutinho and Newcastle’s Moussa Sissoko over the first 3 games of this current Premier League season to demonstrate my point.

Firstly, how did the two players fare in terms of Key Passes so far this season?

Sissokho WS

Coutinho WS

The above images are taken from whoscored.com

We can see that both players have an average of 2 Key Passes per game, or 6 Key Passes in total.

The Analytics Skeptic

Now picture a journalist who was previously an analytics naysayer deciding to finally embrace analytics.  He goes along to the whoscored site and uses the best data (Opta supplied) that is readily available to him and states that, on the objectively measured data, Sissoko and Coutinho were equally creative over the opening 3 games of the season.

Our journalist has taken the stats at face value and penned his piece along those lines.  However, within 10 minutes of his piece being published he gets bombarded with comments by readers asking what planet he lives on if he is suggesting that, even over those 3 games, Sissokho was as creative as Coutinho.

Don’t stop

Let’s go back to the sentiments I expressed at the top of this piece.
I now appreciate that it is wrong to stop at the top level of detail, Key Passes in this example.   It’s just a pity that no one told our journalist friend this as he has now decided that he’s never going to look at an Opta statistic again, and that he was right all along not to trust those pesky stats.
More detailed data is available, so why don’t we use it instead of being content with the top level metric of Key Pass?

The obvious answer for why we tend not to use the whole depth of data that is available to us is that it is so bloody time consuming to digest and interpret.  Thankfully, at this stage of the season with just 3 games played that task is just not quite so onerous.

Let’s now look in detail at the 6 Key Passes that the two players played over their first 3 games, starting with Sissoko.

Sissokho

Apparently Sissoko created 6 chances during Newcastle’s opening 3 games.  I would contend that he actually made 1 very good chance, 1 poor chance and the other 4 are falsely labelled as chances as they all must be 40 yards or more out from goal.

So what about Liverpool’s Brazilian playmaker, Coutinho?

Coutinho

Coutinho’s 6 Key Passes really look like Key Passes.  I would argue that only one of those passes led to what could be classified as a “poor shot”.  The other Key Passes led to very high quality shooting opportunities for Liverpool.

Summary

This piece isn’t intended as a slight on Sissoko, and indeed I don’t think that too many would have argued that the Frenchman was as creative as Coutinho.  However, when looking at the bare Key Pass numbers it appeared that the two players had the same creative impact over their first 3 games.  When delving a little deeper into the data available to us we see that this simply wasn’t the case.

So what’s the point of this?

I guess I have 3 points:

1 –  There just aren’t enough hours in the day to be able to process and effectively use all of the data that is now available to us in one form or another.
2 – On a macro level, even being armed with the data is not enough.  The data needs a lot of TLC to be able to be used efficiently and to release the secrets that it undoubtedly holds.
3 – On a much narrower micro level, it is obvious not all Key Passes or chances created are the same.  Now that the term “Key Passes” is slowly appearing in the mainstream media is it time to go a little further into the Rabbit Hole and attempt to assign some quality measure to the raw, but important statistic of Key Pass?  I feel it is, and the above extreme example illustrates why.

And remember, all Key Passes are equal, but some Key Passes are more equal than others.

  • http://www.optasportspro.com Simon Farrant

    Hi Colin,

    Interesting article, and a good key pass comparison. However, I’d like to clear up a couple of points from an Opta perspective:

    “The data providers do not distinguish between the relative effectiveness of different key passes”

    This isn’t strictly true. Every pass we record has an x,y coordinate associated with it, so it would be very easy to add context to the the Sissoko/Coutinho comparison. ‘Key pass’ in itself is an objective term – a pass that results in a shot, but the pitch location obviously tells you a bit more, as perhaps does the end result of that shot.

    Regarding the journalist example you mention, they could certainly use the data available to them via WhoScored, Squawka, Statszone etc – they all do an excellent job at making the data we collect accessible to the wider public. However, we number many major newspapers and media outlets across the world as clients – the journalists we work with would also have had access to the Opta editorial team. They are skilled at making sure that data used in the media is given an appropriate level of context, ensuring that print and broadcast media avoid the traps you mention above.

    it’s worth remembering that the key pass metric itself is designed for the media – it’s a useful and effective shortcut for our Opta media clients. However, anyone doing more in-depth, OptaPro-level analysis, either within clubs or analytical organisations, would probably only use this as a small part of their wider analyses.

    Please don’t take this as criticism of the article itself – I think discussion around these metrics is very important. I just wanted to, ironically enough, give your article some context…

    Always happy to discuss this type of stuff further though – feel free to tweet or email me.

    Cheers,
    Simon

    • Colin Trainor

      Simon, thanks for your comments.

      I do appreciate that Opta have much more in depth data available to those who choose to use it.

      My piece wasn’t a slight at Opta, if anything it was a slight at the media and the interweb. I know that you have xy data available (ultimately it is from that data via Statszone that I took those screenshots from), but my point is that I haven’t seen any article anywhere where the writer has gone deeper than key passes or chances created.
      Actually, I lie. Paul Riley has written a piece similar to this. But that’s all I can think of right now.

      That was the point of my piece. It almost seems that it’s a huge leap forward when the media use any analytical type metrics, such as Key Passes and that progress is being made. But actually, there is still such a long journey to travel in terms of making real use of the extremeley valuable data that Opta collect.

      We are both saying the same thing Simon.

      Thanks
      Colin

  • http://blogs.columbian.com/portland-timbers/ Chris Gluck

    As an additional note – a key pass is “not” I repeat “not” an direct indicator of a potential assist that goes awry. A key pass is more subjective than I feel and think Ted leads you to believe if the definition provided in ‘fantasy football’ is anything to go by.

    Indeed, as you know I worked extensively on tracking ‘failed assists’ for the Portland Timbers this year until leveraging team passing statistics a bit more.

    1) it was easier and
    2) the correlation is extremely strong when looking at the entire statistical bed across the entire MLS…

    I do believe we see things eye to eye but alas I remain stubbornly steadfast and steadfastly stubborn that the ingredient to individual player success rests with the ‘team’ concept of creating goal scoring opportunities.

    Of note as well is Throw-in’s are not classified as passes in Golazo – but they can be gathered together, along with defensive clearances, tackles won, interceptions and crosses (also not classified by either OPTA or Golazo as a ‘pass’…

    So beware the mines in the fields of statistics – and on this point about ‘key passes’ I must disagree with the learned Simon…

    That’s not a popular view but consider this… in looking at all MLS games this year ‘goals for’ is only a strong indicator for 9 of the nineteen teams relative to points in the league table… 6 teams (to include Real Salt Lake) have a better correlation to points in the league table with ‘goals against’ as opposed to goals for…

    This is a bit winded but it gets back to the earlier point made this summer – a striker is only as good as the players that surround him and that each team must be analyzed differently because each team behaves differently.

    In closing – I have made a statistical and objective decision to eliminate key passes from my future analysis – it is subjective and a decision to call a pass key is made after the fact… in my objective viewpoiint every pass is key because every pass can lead to a shot taken or interception that results in an opponent conducting a counterattack that in turn creates a goal scoring opportunity.

    Perhaps you’ve not read my previous article on this flaw?

    All the best,
    Chris

  • Nick van Gelder

    Firstly, great article!

    I’ve been reading quite a bit of articles on statistical analysis lately and everytime i read a tweet or article using ‘Key passes’, i wondered what the hell that might be. I was rather sceptic, because at first glance it sounds rather subjective which pass is defined as ‘Key’ and which one is not. Now that I know what it is, i am still sceptical and rather surprised that such a random measurement is so widely used in analysis.

    As you point out, there is a rather huge difference in the ‘Key passes’ Sissoko made compared to Coutinho. Without going deeper into those ‘Key passes’, the essence of the measurement ‘Key passes’ is in my opinion almost entirely lost. (And even when looking to the sketches, you don’t have a perfectly good analysis. Coutinho could have made terrible passes which where nevertheless followed by a shot on goal, which made his pass a ‘Key’ pass. I am not saying that was the case for any of the key passes Coutinho made but it does count in weighing a key pass.)

    Now i wonder why this measurement is used so much given the rather uselessness of it without the context. I would expect more scrutiny in the analytics community. In my opinion, the best analyzers are the ones who see and write about the limitatations of their statistical analysis, rather than the ones who are writing an interesting story on a player which simply misses context and therefore is somewhat misleading. That could easily be the case with the measurement key passes, as you point out.

    (Pardon me for my english, I’m Dutch.)

    Anyway, nice article!

    Off the subject. Where can I read more about the measurements being used in footballanalytics? And which football analysis sites are worth visiting?

    • Colin Trainor

      Glad you enjoyed the article Nick, and your English was fine.

      I understand why people use the Key Passes measurement, indeed I have used it myself in some articles. We use it as a proxy for creativity as there no better better stats available to help identify the creativity in players. So for all the weaknesses in the Key Passes metric, you mentioned some of its shortcomings and the entire purpose of the article was to highlight these weaknesses, it’s probably here to stay.

      In terms of helping you understand the measurements used in football I would start with this http://www.optasportspro.com/es/acerca-de/blog-optapro/posts/2012/optapro%27s-event-definitions.aspx definition of Opta terms.
      Opta is the main provider of data for football and the very informative sites such as http://www.whoscored.com, http://www.squawka.com and http://www.fourfourtwo.com/statszone are all driven by Opta data.

      As for football analysis, twitter is a good place to start finding links to interesting blogs, writers and articles. I have created a Public List https://twitter.com/colinttrainor/lists/football-analysts and although I’ll not have included everyone on this list it’ll certainly serve as a good starting point. Just remember to follow me too!!!

      Thanks
      Colin

      • Nick van Gelder

        Haha, I am following you now.

        Thanks for the sites/suggestions. Appreciate it!

  • Anthony

    Hi Colin

    Do you know if in the chain of events; “Pass – Take on – Shot”; is the pass still considered a key pass?
    Take ons seem to be a little more subjective based on what I read on the Opta site. Do you have thoughts on those?

    I guess it has occured to you to weave in your expected goal statistic into key passes. Is that feasible?

    • Colin Trainor

      Anthony, yes I am pretty certain that “Pass – Take on – Shot” is still counted as a Key Pass.

      A successful takeon is simply a dribble past an opposing defender.

      And yes, good spot with the ExpG value being applied to Key Passes. I stopped short of explicity stating this in the article, but it was by using that measure that I was able to provide the examples of Sissoko and Countinho, ie one very good and one poor example.

  • Toshack

    Colin,

    Interesting post and the subsequent “debate” has been equally interesting to read.

    What happens if you apply a “Final third” filter? Will that lather the statistics a lot? (I guess Yes?)
    And if so, will it screw up statistics because you miss out of potential successful long range screamers?

    If you would apply a Final third filter – where would the Exp G value stand? Or it could still be used independently?

    //Peter

    • Colin Trainor

      Toshack,

      I haven’t looked at adding any filters yet. The point of the article was to show that some further data / analyis is required (like @Basstunedtored has been doing the past few says) on key passes for them to be truly meaningful.

      • Toshack

        Ok Colin,
        Sorry if I stepped in an mentioned the “obvious”. Looking forward to the next step in the eternal search for more meaningful statistics.
        Cheers,
        Peter

        • Elliot

          Would it not be more beneficial to look at passes into “key areas” rather than key passes? As can be seen, a “key pass” can be very misleading, the example of Sissoko above being one of many no doubt.
          If, for a very rough example, the 18 yard box was given as a “key area”, the data would now look more relevent; Sissoko goes from 6 to 1, and Coutinho from 6 to 4.
          Just an idea, I’m interested to hear what you all think

  • Elliot

    Sorry,
    My comment above wasn’t meant to be a direct reply to the last message, still getting to grips with things!

    • http://blogs.columbian.com/portland-timbers/ Chris Gluck

      Hi Elliot,

      For me at least I do track and trend passing in the final third and passing across the entire pitch separately – i no longer track key passes – they are subjective as it is the OPTA analyst who makes a decision at the time whether or not the pass was ‘key’ – I’d prefer a completely different stat captured call ‘goal scoring opportunity’ – which would be any ball played from anywhere on the pitch where a teammate can take it – turn/touch it and strike it – this includes any ball that has also been successfully defended against – i.e. cleared or blocked – so in other words it’s an assist ball, a failed assist ball that the striker screws up on, or a failed assist ball because the ball was ‘defended against’ and removed from danger… For me that would clarify who is actually creating chances, blowing chances and defending against chances…

      • Colin Trainor

        Chris, Key Passes per Opta are not subjective. The pass before a shot is a Key Pass. That’s it.

        • http://blogs.columbian.com/portland-timbers/ Chris Gluck

          I think we will agree to disagree on this point but that’s okay – my only question back to you is that the number of key passes in a game (based upon your definition) would mean that every cross in a game is a key pass and when you chart total crosses per game versus total key passes in a game they simply don’t add up – even more interesting in the 600 odd games I analyzed using OPTA data this year there was no single game where any crosses (completed) were logged as ‘successful passes’ yet failed crosses were logged as ‘unsuccessful passes’… however viewed the data sources and data are at best, crude.

          Here is the definition of a key pass = (A key pass is one that leads to a shot at goal) – now go and do a random check of the number of key passes in any MLS game and then compare that to the number of shots taken – they simply don’t match. For example check out this URL:

          http://www.mlssoccer.com/sites/all/modules/custom/mls_matchcenter/mls-gamematrix/index.php?omi=732711&seasonId=2013&competitionId=98&language=en&tag=1&reiter=g&efltblosd=JnBsYXllcnM9MTUzMTAsNTAxMzYsNDI4OTMsMzcyNTgsODMzNzgsNDE2MDYsNDE0OTEsNDc0MDcsMTU1MTcsODMwNzMsOTUyNzgsMTEwOTc3LDE1MTk0Niw0MTY2NiwxNzc2MiwzOTIyOCw5MjQzMCw0MTUyOCw0MTUzMiw5MjQyOSw1NTgxNyw5NjYyMiwzOTI3NiwzNzI1OSw5NTI2MSwxMTAwMDAsMTA0NDUsNjAxNTMmZmlsdGVycz0xMA==

          of the most recent game between RSL and PTFC – 18 shots taken/11 key passes vs 7 shots taken/5 key passes but the one key pass that led to Portland’s second goal – the header by Piquionne is not shown as a key pass (up high – right side of the pitch in added time – note no key pass from a Timbers player in that chalkboard diagram…

          • Elliot

            I tend to agree here, would a long, hopeful clearance from defence that finds a striker, who then scores, be classified as a key pass? According to the definition then yes, however I’d guess that most people would agree that this is technically not a key pass.

            I’m currently competing a dissertation at University along similar lines, focusing on contributions of central midfielders to a match, and looking to see if there is a (if any) correlation between how a CM plays, and the amount of success achieved by a team. Theoretically, as the CM is the hub of the team, there should be some sort of positive relationship.

            From my data collection and analysis, one thing has struck me; the incompetence of commentators and studio analysts. I analysed a match from Euro 2012, looking at the midfield of Russia. One of their CM’s went through a 10 minute spell of not touching the ball at all, quite rare for a CM of any team. At the end of this 10 minute period, he scored (albeit a lovely finish). The commentator and analyst proceeded to talk about how his influence was growing on the game, how he was taking charge of the midfield area…all after not touching the ball for 10 minutes.

            What annoys me that real analysts can do all the fantastic research they can, when amateur commentators and studio analysts can make up and presume things. Sadly, the latter is what the majority of the public will hear, and believe

          • Colin Trainor

            Chris,

            By no means am I an Opta spokesman, but the total shots will pretty much never equal the number of key passes in any game. A shot that has ocurred after a defensive mistake, ie an interception or a successful tackle willl not have a Key Pass as a pass wasn’t made directly before the shot was struck. This is why numbers of shots will be greater than number of key passes.

            Also, how can every cross be a key pass?

            A cross may be unsuccessful – this is not a key pass. And even if a cross is successful (ie it reaches the attacker in the box) but the player receiving the ball doesn’t shoot (perhaps he lays it off) then that cross will not be a key pass.

            And in relation to the specific example that you give in RSL v PT. Have a look at Squawka (http://mls.squawka.com/real-salt-lake-vs-portland-timbers/11-11-2013/us-major-league-soccer/matches) click on the Crosses made by Jack Jewsbury and you can see that does have a Key Pass for that cross in the last minute. This means that Opta did record that pass as being a Key Pass. Perhaps it is because Opta seem not to display crosses on their chalkboard that has caused this confusion for you.

            Thanks
            Colin

          • Colin Trainor

            Chris,

            I asked Simon Farrant of Opta why Jewsbury had no Key Passes and it is because the MLS chalkboard separates assists from key passes. So you need to also add in assists for your analysis.

  • http://blogs.columbian.com/portland-timbers/ Chris Gluck

    Elliot, anytime you want to discuss your research let me know – I have logged all successful and unsuccessful passes for each team in MLS this year for all 34 games by all 19 teams (both inside and outside the attacking third); I also, for a short period of time logged crosses (good or bad) plus fouls and other details. My own research is focused on ‘team’ indicators as opposed to individual indicators but in your approach I would offer one thing to consider in how effective a CM is – track their 1st touch (success or failure) and passing accuracy ‘within’ the attacking third and ‘into the attacking third’ – two very different statistics. The idea on tracking their first touch is not my own – this suggestion came up in a discussion I had with a Dr. of Sports Conflict with the University of Oregon – an extremely sharp man named Dr. Pendleton. For MLS you can do this in the OPTA Chalkboard by selecting one player and then plotting the two different ‘distribution’ ratios by using the pointer to filter between the attacking third (internal) and those passes that ‘enter’ the attacking third from the ‘external’. pop me an email or comment here if you have questions on how to do that and good luck with your research. PS: As odd as it may sound a clearance is a ‘known’ action by a player to clear the ball out of danger – clearances are practiced and trained to – so in the off chance that a clearance does get all the way forward (very unlikely but some defenders and even goal keepers do get assists every now and then for their own end) where a striker can turn and get a shot off then while called a clearance it is also called an assist – therefore it was a key pass in the game whether someone wants to define it as a pass or not doesn’t matter. Another consideration – sorry for going on so long – is that if a defender muffs a pass and it is taken by the now attacking team and struck home for a goal then it is in fact a key pass based upon the definition… again many will disagree but for me a key pass is “any” pass that results in those three actions I provided above. In statistics it is critical, for me, to understand the context and meaning and intent of the statistic – so in validating that definition the numbers need to add up in the OPTA CHalkboard – they simply don’t – therefore the statistic is invalid and not representative of the intent – hence I do not collect it as my own analysis with then use faulty logic.

  • http://blogs.columbian.com/portland-timbers/ Chris Gluck

    Colin, Thanks for the link to squawka – this is the first time I have viewed this site – I will work my way through the diagrams and mechanics the next few days to get a better feel for this and how it can help my research. And no I’m not goofy enough to think key passes will add up exactly to the number of shots taken – it is the way the data is displayed that does not make reasonable sense to me – if a key pass is a key pass then when you click on it it should show as a key pass on the OPTA chalkboard regardless if it was a cross, flick-on, throw-in, corner, through-ball, whatever – the idea that I have to click on two separate boxes in the OPTA Chalkboard to get one piece of data does not make sense. And this is also the case with squawka – I have to click on crosses to get key passes from crosses and then click separately on passes to get key passes from passes??? If this makes sense to you then great – for me it doesn’t make sense based upon my coaching experience – a cross “is” a pass and therefore should be counted as a pass – now separating them out also has it’s benefits in digging deeper into the subset of passes but a cross “is” a subset of a pass. A throw-in is a pass as much as a corner is a pass as much is a free-kick a pass – a pass is defined as an exchange of the ball between one player and another – an unsuccessful pass is defined as a failed exchange of the ball between one player and another. An assist is where that successful pass, in any shape or form directly leads to a shot taken where a goal is scored… an assist is a subset of a pass.

    anyhow – interesting discussion – best in your calcs and thanx again for the link to squawka….

    • Colin Trainor

      Chris,

      Don’t shoot the messenger!!

      I never said it made sense to me, just that I was able to work out what was going on.

      You’re now having a gripe at how Squawka and the MLS chalkboard display the data. That is different to your initial claim that Key Passes are subjectively decided by and Opta analystst. So at least we’ve made some progress.

      • http://blogs.columbian.com/portland-timbers/ Chris Gluck

        Colin,
        I don’t actually think I have shot the messenger at all – if you read through what I have offered (albeit long winded) I would submit that my additional analysis/feedback/information actually supports your three primary takeaways in the article; the most important of which is #3 in my opinion.

  • Uttam

    shouldnt a key pass be one where there is a considerable difference in expG value between a player taking a shot(instead of passing) and the expG value of his teammate taking the shot after he has passed it to the teammate.

  • Pingback: Euro – Tactique – Une expression pour briller en parlant foot : les occasions créées | Le Bon Article – Hespress News()