Tuesday, July 26, 2011

Which Goals Matter? The High Value of Transition Play in the Premier League

While some goals are certainly more beautiful than others, do some actually matter more than others? I've been trying get a bead on this question in part by comparing the relative value of goals created in different ways. My earlier analysis of Guardian Chalkboard/Opta data for the Premier League from the 2010/11 season suggested that open play goals had a stronger statistical connection with points or wins than set piece goals did (The Guardian defined set piece goals as resulting from penalties, free kicks, and corners). The correlation between open play goals and wins/points was about twice as strong as the correlation between set piece goals and match outcomes. So both contributed positively to points and wins, but the data also indicated that open play goals had a significantly stronger connection with match outcomes.

The natural next question is why. Part of the answer seems to be statistical and lies in the distribution of the data. Curiously, the results suggested that the odds of winning the match are better with 0 set piece goals than with 0 open play goals - but how can you have positive odds of winning with 0 goals of any kind, you might ask. The reason for this simply has to do with the fact that there are many more open play goals than goals from set pieces, so in every match with zero set piece goals, the chance of there being at least one open play goal are still very good. In fact, considering all matches played by teams last year (in the EPL, that's always 760 matches, since we have 20 teams playing 38 matches per season), there were 560 matches with 0 goals from set pieces; and of the 560 matches with 0 set piece goals, there were 358 that had at least 1 goal from open play).

So let's try to account for this. For starters, we want to isolate the true, independent, value of a goal of a certain type (open play v. set piece). One way to do that is by calculating how goals change the odds of winning if only one type of goal is being scored in a match. So for the fun of it, I took out my statistical magic wand, ran some regressions, and set each of the goal categories - open play and set piece goals - to zero while allowing the team to score with the other goal category. This allows me to simulate different scenarios, given the underlying patterns in the data. I then calculated the odds of winning the match when we score only from one type of goal (statistically speaking, that is). This tells us what the value of an open play goal is, assuming that there are no set piece goals in a match, and vice versa.

When we graph these results, we get the following picture of how scoring each of these types of goals changes the odds of winning, assuming there are no goals of another type in a match.

When no goals are scored of either type, the odds of winning are, well, zero, as the graph shows. What's more interesting, though, is that the line for open play goals is a bit steeper than for set pieces. There is only a small difference in how much each type of goal matters to the match outcome when it's the only goal a team scores in a match. But once you get to two or three, the odds of winning increase more rapidly with open play goals.

So far so good; but why would that be, I wondered? One possibility is that not all open play or set piece goals are created equal. In particular, as you may know, I have been wondering about the uselessness of corners and free kicks and the usefulness of fast breaks in football.

To be able to tell the differences between goals created in different ways, I examined the effect for each goal type separately. So I ran a set of linear regressions as I did in the original analysis and plotted the predicted values of points and wins for each type of goal created. But this time I split the data up by open play goals, fast breaks, corners, free kicks, and penalties. Here is what it looks like when we use different kinds of goals for predicting points for the average team (the plot for wins is similar).

The results are pretty obvious, if you ask me. By far, the most valuable goals in the Premiership are goals from fast breaks. So the strong effects for open play goals as a general category owe a lot to the big effects of fast breaks on match outcomes: it's particular kinds of open play goals that matter greatly, as we can see from the high intercept and steep slope. As I have pointed out previously, fast breaks are rare but also much more likely to lead to accurate shots and, eventually, to goals. Keeping in mind that, on average, a single goal is worth about 1.14 points in a professional match in the big leagues, it is clear that a fast break goal is easily worth twice as much. Not a bad return on some investment in being able to play a coordinated fast counter attack.

As Don K, a reader with more football wisdom than I can muster most days, pointed out to me, there is a potential problem with 
showing that counterattacks are more productive. And this has to do with what we make of this kind of finding. Specifically, he rightly noted that some teams might view this as tacit support for (unattractive) Route 1 football. This is not exactly how I would suggest the results should be implemented on the coaching end of things. As Don noted to me "A quality counterattack is entirely under control by passing into the path of a teammate and not the football equivalent of 'dumping the puck' seen in ice hockey."

I couldn't agree more. Beautif
ul examples of such play abound, and they look nothing like kick and rush. Think of the U.S.'s Abby Wambach
scoring against Brazil just a few days ago in the women's World Cup, or the Netherlands' 2nd goal against Italy in Euro 2008. My favorite so far is 
Landon Donovan's goal that put the U.S. up 2-0 against Brazil in the 2009 Confederations Cup final. Take a look.



These are all terrific examples of controlled end-to-end counterattacks. And the numbers show that they have incredibly high value. It's an amazing and all too rare thing when goals that are beautiful also turn out to be the goals that matter.


PS: With thanks to Don.