Thursday, July 7, 2011

Do Open Play Goals Matter More? Data From the 2010/11 EPL Season

As you may know by now from some of the other analyses of goal creation I've done these last few weeks, most goals in the Premier League come from open play, while some types of set pieces (especially free kicks) only rarely yield goals. And, as I've noted, some teams create goals in very different ways from others - perhaps the most telling examples were Arsenal and Wigan last year.

But what these analyses don't tell us is which goals matter. Comparing Arsenal's and Wigan's fortunes last year, you may think that open play goals are associated with more wins. But why would, say, an open play goal be more valuable than a goal from a set piece for producing points and wins? A goal's a goal's a goal, right?! Actually, I have some ideas (more of a hunch, really) about why this might be, but before I even get to that, one question is whether there actually is a difference in the probability of wins or points earned, depending on the type of goal created.

Using the Guardian Chalkboard/Opta data I have collected for this last EPL season, for starters, I simply added up all goals from set pieces as defined by the Chalkboards (corners, penalties, and free kicks) and all goals from open play (open play, fast breaks).* As a first step I wanted to know what the correlation was between each of these types of goals created and points as well as wins. Recall that a correlation tells us whether there is a (linear) pattern in the data, where more of one thing (say, goals) goes hand in hand with more of another (positive correlation), or more of one goes with less of another (negative correlation).

So here are the correlation coefficients (Spearman's rho if you care) between number of goals of a certain variety on one hand and points and wins on the other for the 2010/11 season.

             |   Points    Win
         Win |   0.8797   
   Open Play |   0.5775   0.5506
  Set Pieces |   0.2518   0.2333  

The correlation between open play goals and points or wins is well over .5 around .55-.57, while the correlation between set piece goals and points is roughly (less than) half that at .23-.25. So both contribute positively to points and wins, but open play goals seem to have a significantly stronger connection with match outcomes.

Why would that be? Part of the answer seems to simply lie in the distribution of the data. There are many more open play goals than goals from set pieces. Teams scored between 0 and 2 open play goals in 90% of all matches, while they scored 1 or fewer set piece goals in 93% of all matches (they score zero set piece goals in about 74% of matches). This means that the chance of goals being associated with wins are better for open play than set piece goals. So while both have a positive relationship with wins and points, this relationship is stronger for open play goals.

Another way to look at the data is to run a regression and plot the predicted values of points and wins for each type of goal created. Here is what it looks like for wins (the plot for points is similar).

Here you can se that the open play (orange) line is steeper than the set pieces (blue) line. This reflects the stronger correlation I talked about. But interestingly, the odds of winning are higher with zero set piece goals than with zero open play goals. This seems to suggest that set piece goals are more valuable across the board. Why? Because the odds of a team winning the match with 0 set piece goals are 29%, whereas the odds of a team winning with 0 open play goals are only 7.6%. So this has little to do with 0 set pieces winning matches, I would suggest, but simply the odds of creating goals from open play v. set pieces (these are obviously not conditional probabilities).

Another thing to keep in mind is that these are imaginary (i.e., predicted from the regression model) odds. As I mentioned above, the real range of goals scored is between 0 and 2 goals (and really, 0 and 1 for set pieces). In the next couple of weeks, I hope to explore the value of different kinds of goals under different match conditions and for different kinds of teams. Till then, enjoy the speculation and intrigue of the transfer market.

* One thing to note here is that the Chalkboards use a slightly more restrictive definition of set pieces compared to what Opta does for some of the other data they publish, depending on the number of touches allowed after the set piece was taken.