PDO analysis of the elimination of Poland and Russia

In a short term competition such as the main tournament of any international championship, luck may be more important to have on board than skill. Arguably, one can’t arrange for luck to be on one’s side, while for skill certain arrangements can be made. Think of player selection, formations, tactics, motivation and a series of other factors involved.

Given the ‘surprise’ results of yesterday, where Russia and hosts Poland were eliminated from group A, it is interesting to see how these two teams fared with respect to skill, and more importantly, luck.

This is where the concept of PDO comes into the equation. PDO originated in the world of ice hockey analytics, but is slowly gaining ground in football analysis. Without repeating the recent introductory post on the PDO concept in full, PDO is a stacked measure of shot conversion and saves percentage.

 

PDO = 1000 * SUM ( (goals scored / shots for) + (goals conceded / shots against) )

 

The key concept to get your head around is that shot conversion and saves percentage are not essentially different between strong and weak teams. There is compelling evidence that the difference between strong and weak teams is mainly found in the net amount of shots created, rather than the net conversion of those shots. Conversion rates in itself seem, at least partly, related to match situations like the score line in relation to pre-match expectations.

Therefore, in the long term, when enough matches (and thus shots) keep being added, PDO regresses to a mean level which should be somewhere around 1000.

 

Russia

Let’s look at a few examples in Euro 2012. We all remember Russia’s matches, right? Judging by the reports after their first group game against Czech Republic, a solid 4-1 win, Russia suddenly were outside favorites to even win the tournament. And now, after a 0-1 loss to Greece they’re out, empty handed. The fact that their 4-1 win over Czech Republic was based on a near 25% shot conversion, which wasn’t bettered by any team so far, already indicated that their success wasn’t based on repeatable factors indicating skill, but rather on unrepeatable factors indicating luck.

Russia’s PDO after the first game was 1173, dropping slightly to 1114 after the draw against Poland, but returning to average levels after the disappointing loss to Greece, to finish at 1008. The rather large swings in PDO already tell that this parameter needs time to settle, and that a stretch of three matches is a small dataset to draw significant conclusions, but in general PDO’s tend to move towards the average of 1000, so a drop in performance was always around the corner to Russia.

 

Poland

If Russia are a fine example of results going exactly the path that the numbers tell us, Poland forms an example where it didn’t. Obviously, this has to do with the small number of matches played, or in other words, the best teams don’t always win.

Poland created more shots than their opponents in every single game, yet are eliminated without winning a single game. Their PDO was always substandard, regressing toward average from 934 after the draw with Greece to 957 after the draw with Russia, but dropping  to 946 after the 0-1 loss against Czech Republic.

Were this a 30-ish league competition, Poland and Russia, assuming they would have continued to create about 60% of the shots in their games would have stood the best chances of finishing first and second, but the short-term nature of the tournament allows a fair chance for upsets to happen.

 

In the end

No statistical model is perfect though, and PDO is merely a model to look at the balance in skills and luck among different teams. But PDO is a simple method that allows an excellent discrimination between results that have a great chance to hold over a longer term, and results that don’t. There is no general agreement on whether using total shots or shots on target, but in the long run this doesn’t seem to make a difference. In order to increase the numbers I prefer using total shots, particularly in these short competitions.

Obviously, the main area where PDO can be improved lies in the fact that shots are very heterogeneous. In the present model, every shot counts as one, be it a speculative forty yard effort, or a penalty kick. Hopefully, in the near future, we will be able to assign ‘evidence based’ values to these shots and look at goal scoring chances, or even better, expected goals scored.

 

The table below presents the three-match PDO and Total Shot Rate of all four teams in Group A.

PDO

TSR

Czech Republic

977

0.49

Greece

1084

0.28

Russia

1008

0.60

Poland

946

0.60

 

For more on PDO, check the excellent blog by @JamesWGrayson.

4 thoughts on “PDO analysis of the elimination of Poland and Russia

  1. Tim Modise

    Here’s my concern with PDO: Greece’s TSR is 0.28 which according to jamesgraysonmetrics means that they are skill challenged (and lucky). However, I dare say that Greece deliberately played that way(supa defensive means less shots at goal) and were successful AND they did it that way in 2004. Now if you play that way deliberately, and are successful, are you not displaying skill,
    perhaps not the same sort of skill but skill nevertheless?

    James Grayson also knocks Chelsea’s 2 Trophy games – they were lucky. But the same holds, Chelsea played that way deliberately and were successful (as did Inter in 2010). If they had opened their game and had more shots at goal they would (here’s the kicker) have lost!.

    So in certain circumstances, deliberate low TSR plus success seems to indicate skill and not luck.

    Reply
    1. 11tegen11 Post author

      Hi Tim!

      Thanks for touching on an important area of concern with regard to the PDO.
      I partly agree with your comment, and as mentioned in the article, the weakness of PDO is that it treats all shots equal.

      The part of your comment I don’t agree with is to aim for deliberate low TSR. We should aim to measure the quality of chances, and do so without any form of human judgement in order to prevent bias and interobserver variability.
      Ideally, we would rate shots accoring to match situation, location, header vs shot, etc.
      Then, we would be able to assess which team created the cumulative most scoring opportunities, or expected goals scored.

      Regarding Greece, please look at the shooting map from their match against Russia. Greece had three attempts from within the six yard box, Russia two. Greece had three attempts from elsewhere in the 18-yard box, Russia more than ten. Finally, Greece had one attempt from outside the 18-yard box and Russia more than fifteen.

      Now would you seriously think that Greece is happy with that and aim for this shot distribution to happen again when playing Germany in the quarter finals?

      Greece were lucky to avoid a goal here and reasonably lucky to score one themselves. Creating the highest cumulative expected amount of goals scored should be the aim, not a “deliberate low TSR”.

      Reply
  2. aolsh

    forgive me if this is a stupid comment, but on your equation for PDO shouldn’t it be (goals scored/ shots) – (goals conceded/ shots conceded)instead of (goals scored/ shots) + (goals conceded/ shots conceded)? Otherwise everyone would presumably have a PDO of over 1000.

    Reply
  3. aolsh

    I actually ran this analysis on MLS season. Interestingly the PDO results almost perfectly correlated with the standings. I ran it again, except the second time with SOG instead of just Shots. There was a little more variance, but basically it showed that the best teams were the luckiest and the worst teams were the unluckiest. It is a little tough to swallow this, but in MLS there is so much parity that there may be some truth to that statement.

    Reply

Leave a Reply