Finishing in MLS Part 2: Is Finishing Real? Heading Towards a Conclusion
/By Sean Steffen (@seansteffen)
The topic of “finishing” is always a fun one in the analytics world, and, last April, it’s one I studied using data going all the way back to the beginning of the league to see if I could find evidence for a statistically significant gradient of repeatable finishing skill in MLS. Click the link to read the piece in full, but the short of it was, while there were many instances where a forward outperformed their xG by a wide margin or converted an unusual number of their shots on goal, these seasons were rarely repeated within a player’s career as you would expect if such numbers were tied to a skill.
After such a long and arduous study, you can imagine my consternation any time I read a piece praising or criticizing a player’s finishing skill within the league. In fact, when Jordan Morris told the New York Times, “my finishing is still raw,” I nearly had an aneurysm. Doesn't anyone read long winded statistical articles anymore?
Of course, I accept that long held beliefs don’t change overnight, and, although I am generally a cynic, when I look at what has happened with baseball over the last few years and how analytic concepts have worked their way into the broadcast booth and the columns of newspapers and blogs, I also recognize that this is not a hopeless cause.
All we analysts can do is put our noses to the grindstone, publish works, and champion the works of others. With this in mind, I recently set about exploring an aspect of finishing which I have long wondered about: heading.
Let me explain. While Michael Caley was able to find evidence of finishing among a small handful of elite, high volume shooters in Europe, I found that such a tier of player does not seem to exist in MLS. One thing which we both found was a clear delineation between the goals above expected goals of defenders as compared to the goals above expected goals of forwards. In both the EPL and MLS, forwards are more broadly better at putting the ball in the back of the net than defenders.
In the broadest of sense, finishing skill does exist when comparing attackers and defenders, and the best hypothesis as to why has to do with the amount of finishing practice that forwards take in training. Now, for those who feel I may have contradicted myself, it should be clarified that most of my claims about finishing are not made at such a macro level.
Defenders and Forwards may display finishing skill differences, but once you start comparing forwards to each other you’ll find, outside of a very small handful of players in Europe, little to no evidence there is any gradation of the skill among forwards. Presumably the repetition of practice largely standardizes the skill.
But the vast majority of our data supporting this comes from shots taken with the feet. An interesting question to raise is, what about headers?
Going into this study, I hypothesized that the skill of accurately heading the ball could be less standardized than striking it with the foot, which could mean that players who specialize in heading the ball (Kei Kamara and Alan Gordon spring to mind) might be able to consistently outperform the expected goals model on their headers in a way that we do not see from forwards in this league when we look at all shots.
So, let’s dive into the numbers.
From 2011 to 2015, MLS saw 6,562 headed shots. From those shots, 2,145 were on target and 723 of them were goals. That gives us an average league conversion rate of 11.02% headed shots and 33.71% for headed shots on target.
Headed Goals vs xG in the aggregate
The first test is a simple one. We need only take the xG from all 6,562 of these shots, which, in this case comes out to 715.67 expected goals, and compare that to how many goals were actually scored—723. That’s only 7.33 goals over xG over 5 full seasons of soccer.
But how does that compare to shots taken with the foot from the run of play?
| Action | xG | G | Percentage off |
|---|---|---|---|
| 715.67 | 723 | 1.01% | |
| 1486.57 | 1515 | 1.88% |
As you can see, there isn't much difference between the accuracy of the model when it comes to headers versus shots taken with the right foot in the run of play, and, when we look at the distribution of individual season performances in regards to goals minus expected goals (G-xG), we get a fairly nice little bell curve with a tendency for the extreme outlier to fall more so on the positive side of the equation.
As you can see, heading appears to be largely standardized, however, the fact that all the extreme outliers fall on the positive side of things could suggest we are seeing a separation caused by skill.
One problem with such hypothesis is that we see this same tendency for outliers to fall on the positive side of the bell curve when we look at all shots as a whole.
As we know from our previous study, however, players do not tend to repeat their performances when breaking the bell curve when looking at all shots. But is this true for headers?
To examine this further, let’s look to individual performers and see if the outliers are the same players season to season.
Accuracy
In our data set we see that 32.67% of headers are on target. Why is this shot metric relevant? Normally, it’s not, but in the case of headers where the variance in distance within the data is a lot smaller than with normal shots (nobody is taking 30 yard headed shots) it can actually prove to be quite useful in terms of measuring accuracy. So, the question becomes, are certain players year in and year out beating the league average for headed shots on target?
The below is a list of every forward who has taken 15 headed shots or more in a season and has outperformed the league average for headed shots on target by at least 10%.
| Player | Year | Min | Headers | On target | % on target |
|---|---|---|---|---|---|
| 2012 | 1394 | 15 | 11 | 73.33% | |
| 2015 | 1083 | 16 | 9 | 56.25% | |
| 2012 | 1894 | 16 | 9 | 56.25% | |
| 2013 | 1555 | 17 | 9 | 52.94% | |
| 2011 | 1228 | 19 | 10 | 52.63% | |
| 2011 | 3053 | 21 | 11 | 52.38% | |
| 2014 | 2619 | 20 | 10 | 50.00% | |
| 2011 | 2168 | 18 | 9 | 50.00% | |
| 2012 | 2535 | 19 | 9 | 47.37% | |
| 2011 | 2817 | 19 | 9 | 47.37% | |
| 2014 | 2600 | 19 | 9 | 47.37% | |
| 2014 | 1063 | 19 | 9 | 47.37% | |
| 2013 | 2369 | 30 | 14 | 46.67% | |
| 2013 | 2564 | 15 | 7 | 46.67% | |
| 2012 | 2541 | 24 | 11 | 45.83% | |
| 2013 | 1841 | 20 | 9 | 45.00% | |
| 2012 | 2356 | 20 | 9 | 45.00% | |
| 2012 | 1816 | 29 | 13 | 44.83% | |
| 2014 | 2311 | 18 | 8 | 44.44% | |
| 2015 | 2634 | 25 | 11 | 44.00% | |
| 2011 | 1929 | 16 | 7 | 43.75% | |
| 2014 | 1392 | 16 | 7 | 43.75% | |
| 2014 | 2977 | 28 | 12 | 42.86% | |
| 2013 | 2786 | 33 | 14 | 42.42% |
It has happened 24 times by 17 players. As you can see below, the 3 repeaters are Alan Gordon, Chris Wondolowski and Steven Lenhart.
| Player | Year | Min | headers | on target | % on target |
|---|---|---|---|---|---|
| 2012 | 1394 | 15 | 11 | 73.33% | |
| 2015 | 1083 | 16 | 9 | 56.25% | |
| 2013 | 1555 | 17 | 9 | 52.94% | |
| 2014 | 1392 | 16 | 7 | 43.75% | |
| 2011 | 2817 | 19 | 9 | 47.37% | |
| 2014 | 2600 | 19 | 9 | 47.37% | |
| 2013 | 2786 | 33 | 14 | 42.42% | |
| 2011 | 1228 | 19 | 10 | 52.63% | |
| 2014 | 1063 | 19 | 9 | 47.37% | |
| 2012 | 1816 | 29 | 13 | 44.83% |
For those knowledgeable in league history, you’ll know these three actually played on the same team in 2012 and were known as the “bash brothers”. A team which turned the usual inefficient recipe of mass crossing (they had a cross to through ball ratio of 14:1) and turned it into the highest scoring offense in league history with the three forwards scoring 50 goals between them. Perhaps this offers some insight into how that was accomplished.
But putting them on target in one thing. What about converting? The league average for conversion of headed shots on target is 33.71%. Below is a list of players who had 15 headed shots or more in a season and converted 10% or over the league average for shots on target conversion.
| Player | Year | Min | headers | on.target | goals | Conversion |
|---|---|---|---|---|---|---|
| 2012 | 3010 | 25 | 9 | 8 | 88.89% | |
| 2012 | 2243 | 33 | 11 | 9 | 81.82% | |
| 2014 | 2637 | 18 | 4 | 3 | 75.00% | |
| 2012 | 2535 | 19 | 9 | 6 | 66.67% | |
| 2013 | 2641 | 19 | 6 | 4 | 66.67% | |
| 2012 | 2618 | 15 | 6 | 4 | 66.67% | |
| 2013 | 2369 | 30 | 14 | 8 | 57.14% | |
| 2011 | 2817 | 19 | 9 | 5 | 55.56% | |
| 2012 | 1394 | 15 | 11 | 6 | 54.55% | |
| 2015 | 2890 | 44 | 17 | 9 | 52.94% | |
| 2013 | 3081 | 20 | 6 | 3 | 50.00% | |
| 2012 | 2605 | 18 | 6 | 3 | 50.00% | |
| 2011 | 2760 | 16 | 4 | 2 | 50.00% | |
| 2013 | 2289 | 16 | 4 | 2 | 50.00% | |
| 2014 | 2942 | 16 | 2 | 1 | 50.00% | |
| 2013 | 1430 | 15 | 4 | 2 | 50.00% | |
| 2013 | 1841 | 20 | 9 | 4 | 44.44% | |
| 2012 | 2356 | 20 | 9 | 4 | 44.44% | |
| 2015 | 1083 | 16 | 9 | 4 | 44.44% |
Of those 19 players, five were able to repeat. Gordon, Wondolowski, Alvaro Saborio, Eddie Johnson, and Gyasi Zardes.
| Player | Year | Min | headers | on.target | goals |
|---|---|---|---|---|---|
| 2012 | 1394 | 15 | 11 | 6 | |
| 2015 | 1083 | 16 | 9 | 4 | |
| 2012 | 2535 | 19 | 9 | 6 | |
| 2013 | 1430 | 15 | 4 | 2 | |
| 2012 | 3010 | 25 | 9 | 8 | |
| 2011 | 2817 | 19 | 9 | 5 | |
| 2012 | 2243 | 33 | 11 | 9 | |
| 2013 | 1841 | 20 | 9 | 4 | |
| 2014 | 2637 | 18 | 4 | 3 | |
| 2013 | 2289 | 16 | 4 | 2 |
While Gordon and Wondolowski are on both lists, Lenhart disappears from this list and three new names are added which we do not see in the first chart. But why?
The most obvious thing to point to is sample size. Both of Zardes’ seasons jump out for the low amount of shots he put on target, and you can say the same thing for one of Saborio’s seasons. Their data is not consistent enough to be noteworthy. Eddie Johnson’s 2013 season could also be a problem. Had a keeper made one more save and he would have fallen below our mark.
Given the sample sizes for shots on target conversion, this is clearly not the best way to approach this problem, however, our data size for Wondo and Gordon seems significant enough to note, especially since it corroborates what we saw with the percentage of headed shots on target.
To look at this problem closer, we can turn to xG.
Expected Goals
The separation in heading skill becomes fairly pronounced once we start looking at xG over-performance per shot as demonstrated by this bell curve.
Things start to really separate around .08 so that’s where I have decided to draw the cut off line for over-performance deemed statistically.
As in our last paper, it is necessary to look at Goals Minus Expected Goals (G - xG) per shot, since each player takes a different amount of shots.
As you can see below, it is a mark which has been surpassed 11 times from 2011 to 2015 amongst forwards with 15 headed shots or more.
| Player | Year | Min | headers | On target | goals | G-xG | Goals -xG pshot |
|---|---|---|---|---|---|---|---|
| 2012 | 1394 | 15 | 11 | 6 | 4.12 | 0.27 | |
| 2012 | 3010 | 25 | 9 | 8 | 4.7 | 0.19 | |
| 2012 | 2535 | 19 | 9 | 6 | 3.49 | 0.18 | |
| 2012 | 2243 | 33 | 11 | 9 | 5.12 | 0.16 | |
| 2015 | 1083 | 16 | 9 | 4 | 2.23 | 0.14 | |
| 2011 | 2817 | 19 | 9 | 5 | 2.59 | 0.14 | |
| 2012 | 2618 | 15 | 6 | 4 | 2.02 | 0.13 | |
| 2013 | 2369 | 30 | 14 | 8 | 3.99 | 0.13 | |
| 2013 | 2641 | 19 | 6 | 4 | 2.11 | 0.11 | |
| 2015 | 2890 | 44 | 17 | 9 | 3.77 | 0.09 | |
| 2013 | 1841 | 20 | 9 | 4 | 1.66 | 0.08 |
And, unlike when we do this with normal shots, this list has a high percentage of repeaters. Of the eight players to do it, three did it in multiple seasons, and, as you’ll see below, it’s the same cast of characters: Gordon, Wondolowski and Johnson.
| Player | Year | Min | headers | On target | goals | G-xG | Goals -xG pshot |
|---|---|---|---|---|---|---|---|
| 2015 | 1083 | 16 | 9 | 4 | 2.23 | 0.14 | |
| 2012 | 1394 | 15 | 11 | 6 | 4.12 | 0.27 | |
| 2012 | 3010 | 25 | 9 | 8 | 4.7 | 0.19 | |
| 2011 | 2817 | 19 | 9 | 5 | 2.59 | 0.14 | |
| 2012 | 2243 | 33 | 11 | 9 | 5.12 | 0.16 | |
| 2013 | 1841 | 20 | 9 | 4 | 1.66 | 0.08 |
The one issue I see here is that each of the three have either a 4 or 5 goal season, which means a lot of G-xG per shot is hanging on a single goal. Gordon’s 2015 performance, for instance, could be dragged slightly below the 0.08 mark (0.0768) had he scored one fewer goal. Eddie Johnson’s 2013 season would fall well below our mark if he had one fewer goal, while Wondo would have stayed 0.0038 above the mark with a goal less in 2011.
It’s also worth noting that if we drop our threshold to 10 shots instead of 15, we can also add another season for Alan Gordon, which may lend further credence to skill being the underlying factor for what we are seeing from him.
Conclusions
While evidence of repeated model breaking finishing in MLS for all shots is near non-existent, this does not seem to be the case with headers.
Taking a few sample size issues into account, it’s hard to characterize the evidence as “strong,” but we do seem to see that, while, on the whole, heading appears to be largely standardized as a skill, as we see from the aggregate look at G-xG, we also see evidence of varying degrees for repeatability in each of our tests, and, most notably, by the same players.
While the evidence isn't lights out, and there is still a great deal of noise in our data in terms of players doing exceptionally well one year but falling off the map soon after, we also have a small handful guys like, Chris Wondolowski, Alan Gordon, and, to a less convincing extent, Eddie Johnson, showing repeatability within our parameters in multiple tests.
If finishing skill is indeed the culprit, it could open the door to questions about other types of non-standardized shots. Giovanni dos Santos’ G-xG performance, springs to mind, when you consider the amount of chipped shots he has taken, which, similar to headers, one would suspect are not standardized.
As more seasons go on, hopefully we can build on this work, come to better answers, but the one thing which we can confidently conclude is that, once again, fan perception of finishing seems to be completely off the mark from what we see in the data. I was, for instance, shocked to see that Kei Kamara did not show up in our tests as an overly impressive finisher with his head and I’m sure I am not alone in this shock.
Ultimately, it seems we must divorce ourselves from our perceptions of finishing or at least refine them heavily. Even with heading, the amount of separation we are seeing from random noise seems to by minimal and only isolated to two or three players specifically.
As much as soccer fans may not want to equate each shot to a roll of the dice, that’s where our evidence continues to point. Of course, this may ultimately be a harmful belief for players, which is why I am ultimately happy to have Jordan Morris think that his “finishing is still raw,” as that may ultimately be what drives players and coaches running drills to practice these skills so much, causing the standardization which makes the skill irrelevant in the first place.
Despite this chicken-egg conundrum, I still feel that, for fans and those who write about the game, an evolution is needed in terms of we think about shots in MLS.