Finishing. This is not the fist time I’ve approached this topic on this here blog. But as we drive deeper into the season, it’s an increasingly difficult topic to ignore for Atlanta. While its easy to lament missed chances and their impact on points dropped in the table (think of DC home match), the truth is that Atlanta United are a statistical outlier, finishing an exceptionally high percentage of their chances relative to the rest of the league, and outperforming their expected goals by the widest margins we’ve seen in MLS. It begs the question…
When it comes to converting shots into goals, are Atlanta luckier than most or better than most? Stick around. I have charts!
Evidence has shown that in soccer, consistently creating good chances and limiting your opponent’s chances is a key success factor. The rate at which your chances are converted into goals may come and go, but whatever it is that you do that creates more and better chances and limits the quantity and quality of your opponent’s chances is the secret recipe, the valuable IP that makes a team good. In fact when trying to predict a team’s future performance, the number and quality of its past goal scoring chances compared to those of its opponents is more often a better predictor than its past goal differential.
And it’s because of this that when you see a team with an exceptionally high shot conversion rate, you are supposed to be worried that the team will perform worse in the future than it has in the past — that its past results are due in large part to something that won’t be repeated in the future (chance). This fear could be comforted if the team were taking higher probability shots than the rest of the field (the sort of thing you might measure by looking at the team’s expected goals per shot – EDIT: STOP WHAT YOU’RE DOING AND WATCH THIS VIDEO), but generally speaking high conversion percentages are red flags.
Atlanta United’s historically high shot conversion and exp. goals over-performance
Atlanta United has an exceptionally high shot conversion rate. And it has achieved this all season long. Atlanta are on top of almost every flavor of shot conversion chart (shots, shots on target, open play shots, etc). As an example, immediately below is the shot on target conversion metric through the 21 games or so in 2017:
And here are the shots on target conversion rates for all teams from 2012 – 2017 YTD:
So it’s not just the highest SOT conversion rate in MLS at the moment, but it is historically high (only 2014 Dallas and 2013 Red Bulls top Atlanta’s 2017 figure), although curiously a few teams this year are also very high . Does Atlanta have a high shot quality per shot (xG/shot) number to prop up this high conversion percentage? No, they don’t. It’s middling at best. See a helpful graphic of MLS shot quality and volumes from AmericanSoccerAnalysis here. Another way of looking at this is to look at Atlanta’s goals scored compared to their expected goals. Atlanta are averaging 1.9 goals per game against an expected goals per game figure of 1.3. They’ve scored 15 more goals (41 vs 26) than MLS teams historically have based on the volume and location and other attributes of the shots they have taken (based on publicly available data). And again to prove the point, this will look familiar, but here’s how Atlanta looks historically in terms of their goals scored above expected goals per game (it is the highest recorded):
In Search of an Answer
So what do we say about this? As we’ve discussed before on the blog, the possibilities seem to be as follows, and a combination of these is likely:
- Atlanta are luckier than most.
- Atlanta are more skilled at shooting than most.
- Atlanta are taking better shots than most, which is propping up the high conversion number; however, the models we have for measuring shot quality (expected goals) are struggling to properly value the particular shots Atlanta are taking.
Shot Accuracy (also historically high)
Quickly, on #2: Skill, I’d like to point out that not only are Atlanta putting a higher percentage of their on-target shots past the goalkeeper, they are also putting a higher percentage of their shots … on target — testing the keeper at a higher rate than anyone else. And while, like shot conversion percentages, this isn’t a perfectly repeatable stat (repeatability suggests skill is involved), it is more repeatable than conversion. First, here’s the shot accuracy of 2017 MLS teams (shots on target / shots). Atlanta kills it.Some pretty good teams at the top, but some also dispersed throughout if I’m honest. And below are the shot accuracy rates by team from 2012 to 2017 YTD:
Once again, that’s historically high. Only the 2015 Sounders put a higher percentage of their shots on target. And again curiously, another 2017 team is right there with Atlanta, this time the Fire.
I’m tempted to suggest this very high accuracy number is evidence of either #2 (shooting skill) or #3 (Atlanta’s chances being better than the models suggest). I can’t really prove that first idea but it seems plausible. The second one — I don’t know — I think I remember something about shots on target including some bit of embedded information about the quality of the chances (since its easier to put a high quality chance on target than a low quality one). To my eye, watching the team, it doesn’t feel like they are lucking into a high percentage of shots on target. The team really seems to prefer to not shoot until they’ve set up an open look. Many good opportunities end before a shot is generated as the team works to find a better shot by passing or dribbling — this sounds a little bit like *gaming xG* but I can’t be too sure.
Shot Openness: Dissecting all the goals
I haven’t looked at every single shot taken this year by the Five Stripes, but if we look at all the goals (yep all of them, many of them here: thanks Whitecaps), a very high percentage of these are 1v1 against the keeper (and several are empty net opportunities), or there are very few opposition outfield players between the shooter and the keeper. This type of “openness” of a shot has a big impact on its chance conversion, but most expected goals models can only “guess” at the the openness of a shot based on other event data context. They don’t “see” the openness, but they might see that the shot was assisted by a through ball or that the buildup was defined as a “fast break.” I tried to record a bunch of admittedly judgmental shot quality attributes for all these goals. See below for a summary of what I compiled watching every goal, looking out for certain “openness” attributes and then I’ll compare it to another team:
I haven’t done this for all MLS teams because of time, but what jumps out to me are the 26 1v1 chances, and the 35 chances with 1 defender or fewer between the shooter and the keeper/goal. My gut is that these figures are high compared to the league average (which I do not know). For comparisons sake, I picked another team, New York Red Bulls and gave them the similar treatment of watching every goal scored and capturing “openness data.” I chose them for two reasons, first they are currently running neck and neck with Atlanta for the 4th seed in the playoffs (and a first round home match), and secondly (and more importantly), they are an example of a very good team that’s not crushing its expected goal numbers like Atlanta is. In fact they are slightly under-performing. So I was interested to see if there’s a significant difference in the composition of the goals scored. There is.
Across the board NYRB’s goals feature slightly fewer “openness” attributes, except they do a great job of finding BWP in the 6 yard box. If we compare them side by side it’s somewhat easier to see.
Is this the difference between a team overperforming its expected goals by historic margins (+0.6/game) and a team slightly underperforming its expected goals (-0.11/game)? Here are the distributions of defenders in between the shooter and the keeper for Atlanta vs Red Bulls.
Again, we can feel confident it’s either 1) chance/luck and Atlanta really aren’t as good in attack as the results show, 2) Shooting skill with Atlanta being better at finishing than most other teams, or 3) the quality of Atlanta’s chances (primarily related to their “openness”) not being picked up in the event data based expected goals models.
I would suggest it is a mixture of all three with the above “goal quality attributes” hinting gently towards #3. And perhaps Atlanta’s three designated players being signed into the attacking front 4 hinting gently towards #2.
Let me know your thoughts. What am I missing?
I should note that watching all of a team’s goals is a fairly weak attempt at trying to identify other shooting attributes that might be contributing to outliers in shooting efficiency and that looking at every single shot would be a much better exercise (and even still a flawed one). But I don’t know how to get the footage and honestly I don’t know if I could do it. This is at least something. I think it’s better than throwing one’s hands up and saying “welp, Atlanta are lucky” or “welp, Atlanta are master finishers.”
Expected goals data comes from AmericanSoccerAnalysis. If you read this regularly you probably know that by now.
Post credits Scene hinting at more shared universe titles to come
While I was digging through finishing metrics (primarily goals minus expected goals figures), I noticed something interesting that I’m not sure how to interpret. There’s reason to believe its further proof that there’s significant “openness” of shots not being appropriately valued by expected goals models. Basically, if you look at G-xG distributions between home and away matches you see notable differences. The home distributions stretch out more positively with large totals of significant overperformance games showing up whereas the away matches look more like a normal distribution but tilt slightly in the direction of under-performance.
If you take all these distributions as a percentage of the total and chart the differences in percentage share by bin between home and away you get the below image, which more clearly shows the lean towards an over performance in finishing for the home team relative to the away team.
This is 2017 data for all MLS teams over 22 or so weeks of the MLS season, so it’s a good number of games and it runs the full spectrum of good and bad teams. It would suggest that either 1) teams shoot better at home, or 2) when playing at home teams get better chances than the models are able to capture in xG. The first should be dubious to anyone that thinks finishing generally averages out across all shots. The second one makes more sense with what we know about home field advantage in MLS, that it definitely exists. It doesn’t seem too wild to think that teams create better chances at home, and perhaps even chances that are better than the expected goals models give them credit for. But that’s just 2017 data. What does the 2011-2017 data show? Well, it’s both a) much less pronounced (smaller differences between bins between home and away), and also b) home teams seem to both significantly over-perform and significantly under-perform more often than away teams.
So that leads us back to something strange going on in 2017, which could very well be a skew driven by Atlanta United, or a shift in the landscape of finishing for MLS in 2017, or an increase in the types of good chances that are undervalued by models? Or… variance working its way out over the course of a full season.
Appendix: Assorted resources on finishing skill in football
There are many others out there, plus I’ve discussed it occasionally in ATL match previews and recaps on this blog.