98.4% of all referee decisions are correct, claim PGMO. And amazingly the Independent believes them.

By Tony Attwood

Just two weeks after Untold published its 10 point plan for how an open, honest and honourable referees’ organisation would look, and PGMO allowed a journalist into their hallowed offices for a chat.

Coincidence?  Of course it was.  Untold is just a blog run by a few enthusiasts who feel that PGMO is one of the key reasons that there is something fundamentally wrong with football in England.  The notion that we can have any influence is fanciful.

A number of Untold readers wrote in and noted the interview in the Independent, which took place in the offices of PGMO.  I’ve held back on any commentary until now, just in case there was more to follow.  But all seems to have gone quiet again, so it is time to take a closer look at the report and what it implies.

The article was an interview with  Howard Webb technical director of Professional Game Match Officials, and looks to be like an opening salvo in re-imaging PGMO into your nice, warm, friendly, cuddly bunch of everyday football lovers.

There’s lots of talk about Webb refereeing the Christmas truce game in Ypres as part of the commemorations of the First World War, and refereeing an under 15 schoolboy game that his son was playing in when the appointed ref did not show.

It is all homely stuff, and of course such everyday story of refereeing folk can be of interest, but it is hard not to think that this is little more than a smoke screen, following on from the sudden first move by PGMO, when they followed Walter’s commentary on the lack of action on having video refs at games with a press release that the Telegraph published.

What happens is the none of the big issues of fanatical secrecy levels and the lack of enough refs to avoid anyone handling any team more than twice a season, are addressed.  Instead we get Jack’s Under-15s game at Oakwood High School in Rotherham.

And, hey, all those allegations on Untold about the utter and complete incompetence of their referee analyses which suggest near perfect referees striding the turf each day of the year, well, no need to worry about them because after the interview with the Independent the father of Jack (of under 15s fame) is “due to see a software developer about a project to help analyse the select group referees in even greater detail.”

But wow, they don’t need to – it is just that they’ll always go that extra mile…

“Not that they are not already under the microscope. Webb can tell you from the statistics on his phone that match evaluators in the Premier League have assessed every game in minute detail and say that up to the round of 15 matches so far this season there have been, collectively, 38,718 decision moments for referees. Of those, they say, 98.4 per cent have been made correctly.”

Yipee.  Everything is perfect.   All we have to do is take their word for it, because they sure as hell aren’t going to reveal their analyses in detail so that we can see exactly what happens.  Conclusions without evidence.  I can think of quite a few dictatorships that have worked the same way.

And that is the difference between the Untold analyses of referee performances and PGMO’s.  We publish the analysis line by line by line.  Event by event by event.

And what do they do?  They tell us they have 98.4% correct.  And utterly amazingly, the Independent, a newspaper whose very title suggests it is, sort of, Independent, just believes it.

But, let’s not get into this too deep, because in a trice, the interview moves on to the homely stuff, moving jobs and how he misses the games.

“You’d be crazy to say you don’t miss those big games or that great feeling of satisfaction when a game has gone well and you can relax a bit,” he says.   “I could have carried on for 10 more years. I could have coasted but that was never the way.”

You can almost feel the honey dripping off the walls.

The one thing the “interview” does it let us know that PGMO are most certainly not going to give in to the suggestions made by those of us deeply concerned about the standard of refereeing and the way that PGMO is run as a secret society that makes MI5 look like open house.

Of Webb the interviewer says, “He is passionate about the high standards of the 17 select group referees…” so we know we are in safe hands.

The whole PR exercise is just that – diverting attention from the fanatical secrecy of PGMO and its refusal to engage in the real debate.   Instead we hear that there will be a DVD to clubs sent out about holding in the penalty area.  Oh yes, and subsequent to Walter’s piece about video assistants Webb has been to the Netherlands to take a peek at video assistants.  Well now, what a coincidence.

But in essence this is a piece about diverting attention.  Telling us that they are already on the case we suggested they were not on (video refs) and reiterating their statistical claims without a single, solitary shred of evidence.

Instead we get a review on Clattenburg leaving the ground on his own.  That’s what the Independent, or maybe PGMO, think is a BIG STORY.

I guess, having had this sort of thing in the Telegraph and now the Indy, we are going to see more drip feeding of PGMO stories across the press.

The only issue is, will each newspaper be taken in by the chance to have a free article, rather than do any serious investigative reporting.

Untold Index

25 Replies to “98.4% of all referee decisions are correct, claim PGMO. And amazingly the Independent believes them.”

  1. Thanks for writing about this Tony.

    I was glad to see the article in the Independent, and sad that it was exclusive. I was hoping that other news organizations had also been there, as then we might have seen more specific numbers. In any event, PGMO is convinced that they have found essentially all of the incidents in that first 140 games of the 2014/15 season.

    I went looking for some data to help in reviewing those numbers, and that data appears to be available to subscribers only. As I can’t afford to subscribe, no data. If you have the data, or know where I can find it, please tell.

    Having read commentaries for the most part in following Arsenal, I know there are situations where the commentary doesn’t mention trivial details. I need details. How many passes in each game by each team (don’t have to break it down to per player). Throwins. Goal kicks. Corner kicks. Dribbles. Handballs. Fouls. Yellow cards. Red cards. Simulations.

    As I mentioned in another thread, PGMO is saying that they only made 619 or 620 mistakes, and that there were only 9 instances of simulation in those 140 games. These 2 numbers (619.5 and 9) have nothing to do with the referee calling it or not, they have evaluated all of those games, and those are how many incidents of those 2 specific types that are present. Called or not.

  2. We know better than to believe that lot at PGMO……..off topic Tony or still on the Refs,should’nt the Ref have ordered Skrtel off the picth to get treatment in the game yesterday.you and walter know better than me when it comes to officiating but i was always of the impression that only the goal keeper is allowed to be treated on the pitch or someone who is in a life or death situation.please correct me if i’m wrong.

  3. It must be that the 1.6% of wrong decisions happen against us then !

    I can’t blame the ref for yesterday though. We were poor and we nearly did to Liverpool what man u did to us. Thing is Lvg was a genius for getting a result against us. We however are hopeless according to the media

    Physio Room indicates that all our players bar Jack wiil be back by the 1st of January. Some for Boxing Day, others for 28th and Mesut on New Years Day

    I hope they will be match fit and ready to play. We play West Ham away and Southampton away in the next ten days. I don’t think we’ll win if we have the same midfield and defence as yesterday. I’m confident that when we are near full strength we can go on a good run but losing those two games will give us a lot to do in chasing them both.

    My hope and ray of light is looking forward to a rested and keen Mesut playing with all our pacy forwards 🙂 🙂

  4. i was expecting a proper article on our performance yesterday and not just a ‘bad day at the office’ sort of stuff as we seem to be having quite a lot of them and move on to the favourite ref bashing.

    this is the same pool team who played mid week and were not expecting the energy levels they showed. was also shocked by AW’s statement, ” we deserved a point….really?ng.

    @Tunnygriff: yes, Pool did what we did to manure, but the difference, was that we conceded in both, the manure and pool game. so its quite apparent that we know where the problem is, but don’t see to rectify it.

  5. Gouresh

    Agree totally. We desperately need to get some of our midfielders back . We were so poor at keeping the ball yesterday. Flamini is so so poor, I’d rather play Coquelin. We need Arteta and one of Ramsey or Rosicky back for the West Ham game. I’m worried what Austin will do to us. How far is Kos away ?

  6. Tony,
    You forget that Christmas is a time for fairy stories, make-believe Santa Claus and general good will, comfort and joy. And particularly pantomime.
    The Independent and the PGMO have jointly set an example.

  7. I like your ref analyses but I think you are a bit over-imaginative if you think they are doing anything because of your blog.

  8. By accident I turn my tv on the Stoke-chelsea match. I wonder if the pundits will praise Stoke for their kicking at all that moves this time? What has this got to do with football?
    I don’t know how the comments will be in the UK during this match and compared that to the Arsenal matches at Stoke

  9. I guess a handful of UA readers may be comfortable doing all the statistical rigamorole (sp?). I am still trying to convince myself on something, so maybe explaining a little about why I think Howard Webb did us a favour willhelp me get the aha moment I need.

    Pretty much the default probability distribution to use with processes that involve counting, is the Poisson distribution. There are a number of requirements that should be met when using a Poisson distribution. But a lot of the time, if the average of some counted random deviate is about the same as the variance of that same random deviate, people will try Poisson distribution. That distribution has as a defining characteristic, that its mean equals its variance (and is a 1 parameter distribution). One of the counting exercises surrounding the development of this distribution, was counting how often soldiers were kicked by horses in the French army (during Napolean?).

    Where a lot of us get a hands on use for statistics and counting, is flipping a coin and counting how often we see heads (or tails). If the coin never lands on its edge, this is a binomial process (only 2 outcomes). The binomial distribution is generalised in terms of the probability of the one event (heads, pass, success, …) called p, and the other event, which occurs with a probability of 1-p. The mean of a binomial process is the number of events (n) times the probability (p). The variance of this distribution is np(1-p). Or, rephrased the number of successes is np. In general, the variance is always smaller than the mean, but becomes like the Poisson when the mean is small.

    There is another distribution, called the negative binomial which is also often used in counting problems. The binomial always has a variance less than the mean, the negative binomial always has a variance greater than the mean.

    The Poisson is the default for counting problems. The binomial is considered when the data is less dispersed than a Poisson. The negative binomial is considered when the data is more dispersed.

    I don’t as yet know much of anything about the PGMO data. I do know that they claim to have found ALL the simulation events (called or not) in the first 140 games of the 2014/15 season. I will be acting conservatively, if I consider that the variance of this estimate of 9 is larger than a Poisson distribution would allow for.

  10. I think it was Gord who first drew my attention to Webb’s erroneous assertion that 98.4% of ref decisions were correct.

    I don’t think Webb or the Independent produced evidence to back this up – in contrast to what Walter has done in publishing his ref reviews for the last few years.

    Deception is (now) being used by the PGMO to cover up for and ensure the continuance of incompetence.

  11. Oh, now this thread is getting visitors. 🙂 Just when I put up a bit longer note. Oh, thanks Andrew.

    The Poisson is about counting in general, whereas the binomial and the negative binomial are about counting successes. And if success is possible, you can be sure that there are some failures around as well.

    The binomial distribution is all about counting successes. If we have a probability of detecting success of ‘p’, and we go through ‘n’ trials, we expect the number of successes to be n times p (or np).

    The negative binomial is a little bit different. In doing ‘N-1’ trials, we have seen ‘n-1’ successes and ‘x’ failures (having done N+x-1 trials in total). If the ‘N+x’th trial is success, we have seen a negative binomial process. It is important to realize the last trial done, is a successful one.

    The standard deviation is the square root of the variance. Most of us know of correction factor version involving how much data is involved. A field of physics called statistical mechanics, says that in the absence of better information, we can usually assume that the standard deviation will scale by the square root of the number. If N increases by a factor of 4, we expect the standard deviation to double. Along that line of thought, if we only have a single N, we can guess that the standard deviation will be the square root of N, times some unknown number. And we will hope that this unknown number is small.

    Well, we have this number from PGMO, that there were 9 acts of simulation in the first 140 games of the 2014/15 season. This number (supposedly) comes from an undetermined number of people looking at ALL the video feeds of the games (not just the one video feed which was broadcast on TV). The square root of 9 is 3.

    The data we are looking at is one sided, you can only observe zero or positive integer numbers of occurences.

    In trying to show that something is very unlikely in a statistical sense, we need to show that a value observed is so far out in the tails, that it is more likely to be an outlier than a true data point. In going out into one (or both) tails, we talk about how many times the standard deviation we are going (out of habit for the most part).

    If our unknown multiplier was 3, a single standard deviation in the negative direction gets us to 0. Which is not a situation which “feels” right. There is no real reason why our “characteristic measure” of disperson has to be the same for positive and negative movements away from the mean.

    In my write up of a couple of days ago, I picked a multipler of 2, which would give a standard deviation of 6 (2 times 3) and a variance of 36 (6 squared).

    As 36 is more than 9, using a negative binomial distribution could be considered.

    The mean of a negative binomial distribution is pr/(1-p) and the variance is pr/(1-p)^2. Or, the variance is equal to the mean, divided by (1-p). And p is the probability of success in a single trial.

    36 = 9 / (1-p)

    (1-p) = 9 / 36 = 0.25

    p = 1 – 0.25 = 0.75

  12. If an independent (and highly skilled) analysis of those same 140 games determined that there were in fact 0 instances of simulation, it would still be very difficult to show that the determination of 9 by PGMO was unlikely. Hence, for us to call their analysis into question, we want to find more instances of simulation than 9.

    How much more?

    Well, we are going to go far out to the right (towards positive infinity). Let’s say we had a list of simulation possibilities. One (or more) experts would look at this, and give their opinion as to how many instances of simulation is present. We believe there are (at least) 9, we think there are more.

  13. > Oh, thanks Andrew.

    I guessing I am telling damn lies. I should have put a smiley up.

    Hmm, deleted something too early.

    We are using a negative binomial distribution, and we have found the probability of detecting a single instance of simulation to be 0.75 (or 75%) if our variance is 36 (for a mean of 9). In terms of Bernoulli trials, we had 12 trials to find 9 successes. Because it is a negative binomial process, the 12th trial was a success.

    Wikipedia explains things through a candy selling example. So, let’s sell candy.

    Pat is required to sell candy bars to raise money for the 6th grade field trip. There are thirty houses in the neighborhood, and Pat is not supposed to return home until five candy bars have been sold. So the child goes door to door, selling candy bars. At each house, there is a 0.4 probability of selling one candy bar and a 0.6 probability of selling nothing.

    What’s the probability of selling the last candy bar at the nth house?

    The number of houses in the neighbourhood is not used for most of this example. To properly use the negative binomial distribution, the last house visited (which ever trial number that is) is a sale. As worded, either 0 or 1 candy bars are sold at each house, there is no possibility of selling more than 1 candy bar.

    If we switch back to the binomial situation, and look at this probability of selling equalling 0.4, we would expect that visiting 12.5 houses would be required in order to sell 5 candy bars. A side effect, of the probability of success being low, is that when a person looks at the some of the individual probabilities which go into a sum, we see a fairly long string of increasing probabilities.

    Wikipedia calculates the probability of selling the 5 candys at the 10th house, and the chance of finishing on or before the 8th house.

    f(10) = 0.1003290624.

    f(5) = 0.01024
    f(6) = 0.03072
    f(7) = 0.055296
    f(8) = 0.0774144
    \sum_{j=5}^8 f(j) = 0.17367

    You can see above, that f(6) is greater than f(5), and it appears that this tendency to increase continues for quite a way. What happens if we increase the probability of success to 0.75?


    And that sum of f(5) to f(8) is 0.8861846923828125

    The explanation, requires you to remember that the sale of the 5th candy bar is at the last house visited.
    When the probabiliy of success is high, we see the peak (mode) in the distribution earlier, for a success probability of 75% we are seeing the peak at 6 houses. For Pat to have 5 failures by the end of the visit to house 9, before making the last sale at house 10 has a fairly low probability. We also see that it is quite likely that Pat will finish somewhere between house 5 and 8 (inclusive).

  14. Hopefully all those numbers hasn’t bored you silly. The following is all numbers, and it deals with the PGMO simulation data.

    We start at 9, because you can’t have 9 successes with less than 9 trials. The peak (mode) in probability is at 11. There are 2 failures in detecting simulation 9 times. This assumes the variance is 36 (for the mean of 9).

    You have to get past 14, before you start to see probabilities less than the probability of getting 9 success in 9 tries.

    There are a number of places to “draw a line”: 0.95, 0.99, 0.999, 0.999 9, 0.999 99, 0.999 999. The 0.95 mark is vageuly related to 2 standard deviations and 0.99 to 3 standard deviations. After that we have 4N (4 nines), 5N, 6N and so on.

    For the PGMO finding 9 simulation incidents, we want to work with n=9 instead of 5 (for candys), and we want to find out how far we need to go to get a sum of 0.95 (or higher). As we need to find 9 simulation events, we start at f(9).

    f(9) = 0.075084686279296875 _____ Sum 9..9 = 0.075084686279296875
    f(10) = 0.16894054412841796875 ____ Sum 9..10 = 0.24402523040771484375
    f(11) = 0.21117568016052246093 ____ Sum 9..11 = 0.45520091056823730468
    f(12) = 0.19357770681381225585 ____ Sum 9..12 = 0.64877861738204956053
    f(13) = 0.14518328011035919189 ____ Sum 9..13 = 0.79396189749240875242
    f(14) = 0.09436913207173347473 ____ Sum 9..14 = 0.88833102956414222715
    f(15) = 0.05504866037517786026 ____ Sum 9..15 = 0.94337968993932008741
    f(16) = 0.02949035377241671085 ____ Sum 9..16 = 0.97287004371173679826
    f(17) = 0.01474517688620835542 ____ Sum 9..17 = 0.98761522059794515368
    f(18) = 0.00696300019626505672 ____ Sum 9..18 = 0.99457822079421021040 2N
    f(20) = 0.00135303753813785479 ____ Sum 9..20 = 0.99906460842066734071 3N
    Sum 9..23 = 0.99994549223455901363 4N
    Sum 9..25 = 0.99999263645867515585 5N
    Sum 9..27 = 0.99999907046876110124 6N
    Sum 9..30 = 0.99999996242314353185 7N

  15. This being the same Stoke that manhandled the lads, right Walter? Just checking. It’s impressive that the numbers bear out a certain truth, just not yours. When is a majority not a majority, one might ask? Ask Walter, he sees things where they ain’t!

  16. Even if this note gets lost to most people in a series of gobbledegook notes, I should keep it here, maybe reference elsewhere.

    Mourinho has a problem with “independent” reviewers, reviewing diving incidences. It is possible he has multiple arguments, one of his arguments is inconsistency. I don’t think the fact that the reviewers has anything to do with inconsistency. Look at Walter (and others) reviews on UA. Some of the officials are all over the ballpark in terms of what they might decide in a given circumstance.

    I think the problem Mourinho has with independent reviewers, is that they are independent. They are not subject to his influence, that are not open to Mike Riley 😈 influence, and they are not open to FA influence.

    I think independent reviewers are necessary. If nothing else, it opens the door to other independent analysis of PGMO/EPL.

    There are 2 kinds of simulation to be aware of. There is the “on the ball” simulation, of which diving in the penalty box in the hope of “winning” a penalty simulation. There is also the “off the ball” simulation, where a player goes down pretending there was an “off the ball” incident.

    The off the ball incidents are harder for many reasons. All the officials may be concentrating on issues in a different part of the field, and so it takes a while before they become aware that there is something to deal with, but they probably have no idea what. It will take the officials longer than usual to decide to bring play to a stop, and then they have to figure out what happened. Replaying recordings of the TV feed may not be useful, as the TV feed may not show the incident either (nominally the same reason as for the officials). It is the video feeds not currently being part of the TV feed that may show the incident. Yes, the TV people can access those feeds, but they may decide they aren’t going to. Hence, it may only be PGMO which sees this, if indeed they do analyze every video feed of every game.

    I would suggest that the people operating TV cameras at EPL games, be given a button. If they see a player go down and not move anywhere on (or near) the field, they press the button. Most of the incidences I’ve seen writeups about, are heart problems. I suspect concusions could also fall into this category. But in general, if a player goes down and doesn’t move, it should be considered serious. And it doesn’t matter who gets the attention of the 4th official (who can inform the referee).

  17. What’s up Ernest Reed?

    Have to get in your 2 or 3 stupid postings to get paid for doing your job of interfering with the smooth functioning of AFC?

    How much do you get paid for writing this crap in a blog where you are perceived as being opposite of the published intent of the blog?

  18. FIAF; UEFA; FA; PGMO – the titles are famous; the behaviour is the same!!

    All milking the finances Football brings; all corrupt in their reporting.

  19. I think one thing to bear in mind about the media is that in the wake of the phone hacking scandal most papers would prefer to retain self-regulation rather than allowing an independent external body undertake it. This makes it hard for them to criticize organizations such as the PGMO or FIFA also prefer o police themselves.

  20. If PGMO detected only 9 cases of simulation during the course of 140 games,then Hazard is responsible for 4 of them….which i find to be utterly false cos we still have Rooney,Ashley Young Drogba still playing in the epl.PGMO are a bunch of liars….lol

  21. He’s bald, He’s red,
    He sleeps in Fergie’s bed

    Just had to repeat that song. What is Mr Webb’s job at the PGMO anyway? Lieas and propaganda, of course, but officially, what does another crooked PGMO ref do after his retirement?

  22. Howie always was and still is a smarmy git.

    From my analysis in front of the TV I can assure you that at least one decision by the officials is a game/result changer in EVERY game in the PL.

    It makes a complete mockery of any kind of competition.

  23. I just find the attitude of Webb, Riley et al completely patronising. Just how gullible do they think we are……………..

Leave a Reply

Your email address will not be published. Required fields are marked *