Not-Soccer is Hard
Four games into the 2019 season, we’re still searching for glimpses of contrary evidence to suggest this 2019 Atlanta United team will grow to be anything more impressive than the below average side we’ve seen so far this year. I want to examine what we’re seeing at face value and build a framework for trying to understand its place in the proper context of a sport that is difficult to predict. The natural place to start would have been the trip to Mapfre Stadium this past weekend — a long anticipated return to action and the first fixture this season not book-ended by matches in other competitions. Unfortunately, as the manager eloquently explained shortly after full-time was blown, starting any sort of analysis about the Columbus match is problematic:
“I think we can only talk about the first half hour - those circumstances were also not great but you could play a little bit. Of course, we started very bad to concede the goal, but I think afterwards we play really well - we created our chances - 4 chances - 3 very good chances. They had 2 chances and scored 2 times. That is also football. But I think we can only talk about the first 30 minutes because that looks like football and the rest you know was yea - and also with luck - and also dangerous for our players and for the other team... I wouldn’t call it football anymore.”
Before you accurately point out in the comments that the weather impacted both teams (I’m begging you), accept for a moment that while your reaction is grounded in something true, it also does not give us much information at all about how good Atlanta United is or more importantly how good it will be over the remaining matches — which is what I want to build towards here. Since I’m half-filling in for Payson Schwin’s weekly “De’s Nuggets” post, I will first take Frank de Boer up on his offer and attempt to comment on the first 30 minutes, before I disappear into the crystal ball of chaos and math.
The Half Hour of Soccer at Mapfre Stadium on March 30, 2019
In the opening few minutes, Atlanta United were challenged mightily by a deliberate and unrelenting Columbus Crew team who found a nice ball in behind between the left back Mikey Ambrose and the left center back Leandro Gonzalez Pirez. As Joe Patrick, points out on the latest Five Stripe Final show, it’s the concession of that attacking move in and of itself that is the mistake. Atlanta defended the remainder of the sequence mostly adequately, but the danger itself had already been awoken, and the ball fell to the Crew in the rare spot where you score more often than you don’t, and they did.
Over the next 30 minutes — these are the minutes de Boer is proud of — Atlanta, mostly through Julian Gressel, created 2 shots inside the box, a nice transition move ending with a ball played through by Josef Martinez for an off target strong-footed shot from the German, and a well-worked cross set up by Josef from Mikey Ambrose to Gressel for a header saved low by Zach Steffen. By no means were these free throws, at least according to American Soccer Analysis who had the two shots together at a combined 25% likely to score based on historical shots from similar locations in similar situations (weather effects excluding), but shots in the box are good, and you want those, and they’ve been elusive for the Five Stripes so far this season.
Atlanta created a couple lesser value shots from Ezequiel Barco and Hector Villalba. And perhaps just as important, during the half hour after Columbus’ goal, the home side didn’t create a shot. Josh Bagriansky and John Fuller do a much better job breaking down the tactics and performances of the match here and here, respectively, and they’re definitely worth your reads.
If we accept what de Boer says here - and I do - that this part of the game was football, and that “the rest was falling down in water,” one thing we could do is extrapolate out this period of time to imagine a full 90 played under poor but playable conditions. From an expected goals perspective, Columbus’ single very good chance (which they scored) roughly doubled the value of the 4 shots Atlanta created over this period, and this is a problem when you try to imagine a full “normal” 90 minutes played on the road against a good team — you end up with something like Columbus putting up 1.3 goals worth of shots and Atlanta putting up 1.0 and Columbus winning 45% of the time and drawing 30% of the time based on these distributions of chances. And yea, this is both abstract and rough around the edges — it is definitely a straw man, but it’s not a terrible one. I am completely OK with throwing out the last hour of the game, and it’s up to the coaching staff to honestly assess how that game goes for the remaining hour in normal conditions, but I’m not convinced the result is any better.
Did the team look better in the opening half hour against Columbus than other times this year? Yea I guess. If you close your eyes and ignore the opening 2 minutes, does it look better? Yes, definitely. Does it look as dangerous in attack or as defensively sound as last year’s team? No. And as we look out towards the rest of the season (I’m getting there), that’s the form the team aspires to and the fans expect.
Speaking of the 2019 Atlanta United season to date, de Boer was asked whether the team deserved more than 2 points from the opening 4 games based on its performances:
“Sometimes we did and sometimes not, but I think today of course when you concede an early goal, that’s your fault of course, but afterwards you know, I think we weren’t the worst team on the pitch, maybe sometimes we were equal and sometimes we were better and but we have to — how do you say in English — sometimes it is also not luck, you have to get that luck to give it your way, and we can only do it with hard work together.”
This is a good quote from the manager. I think the English idiom he’s searching for is “you make your own luck,” and I see this as an encouraging response to a question that most managers of struggling teams latch onto as an opportunity to make excuses. For, while it is true that the team has scored fewer goals than the number and quality of the chances it has generated would suggest (2 goals on 4.2 expected goals worth of shots), and while it is true that over time these two metrics tend to converge such that expected goals is a better predictor of future goals than past goals (I’m getting there), it is also true that the chances the team has created have been middling and much lower than those of the 2018 Atlanta United team, and frankly it is also true that these chances have been swallowed up by the number and quality chances it has conceded to its opponents so far. The fact remains that even if Atlanta had scored a couple more goals so far (e.g. friendlier keepers or better luck), the underlying output from the team both in attack and defense has not been near good enough.
So while Frank might well have said “yes, we’ve been unlucky to only have 2 points through the first four games based on how the team has played,” and been technically right, what he actually said has more truth in it, and it should resonate better with a fan base that is currently intolerant of excuses. In short, to “make your own luck” in soccer is to prepare and execute on a plan that BOTH creates shots from good areas AND prevents your opponent from doing so. Atlanta has some work to do here. Now let’s take a shot at forecasting it.
Chaos and Correlation
It’s hard to predict things using small sample sizes, but in soccer certain metrics correlate with future results better than others, and some need less weeks of data than others to be do the job. Here are a couple snapshots for context. The first is a time-based graph of the historical correlations in MLS between various predictive metrics after X number of games played and the actual points earned per game over the remaining games played from that week onward.
As an example, a way to read this would be that after 4 weeks of MLS games, the correlation between a team’s points per game and their points per game over the remaining 30 weeks is a very poor 0.06, while the correlation between their goal difference over those 4 games and their points per game over the remaining 30 weeks is also a very poor 0.08. But the correlation between a team’s expected goal difference over the first 4 games and their points per game over the remaining 30 weeks is noticeably better - up to 0.27 (!). As you can see, none of these correlations are particularly high in the early stages of the season because soccer is hard to predict and 4 soccer games is not enough games to predict the future with any level of accuracy; however, the chart makes very clear that the predictive power of expected goals revs up significantly faster than that of the other metrics. After about 10 weeks of match data, xG hits its stride nicely, and it takes another 12 weeks of data for the signal in a team’s average goal difference or its points per game to shine through the noise. This is why people talk about expected goals in soccer. It works better than the other stuff.
From an Atlanta United perspective, it should be encouraging that regardless of how you slice it, the chart above tells you that 4 games of data is still an under-powered sample to be representative of the rest of the season. If Atlanta is still underwater at the 10-game mark, let’s regroup and have a sterner discussion, but it should also just be encouraging that the overall correlations here are so weak. Lots can happen in an MLS season and their are countless stories of teams making lung-busting runs up the table in the second half of the season.
This is now, now
Here’s another chart. Since we’re standing here after week 4 with 2 points in the bag and averaging an expected goal difference of -0.5/game, this is the dot plot of all the teams from 2011-2018 with their expected goal difference (per game) through 4 games on the X axis, and the points per game over the remaining 30 games on the Y axis.
Because it’s so early in the season, the range of future points outcomes for a given sliver of xGD performance to date is very wide. If we focus in on teams at the four-week mark with negative expected goal differences, we can see plenty of notable examples of teams powering on to good seasons. DC United and Portland Timbers from last year stand out, but as you can see from the slope of the line, for every one of those, there are multiple Minnesota, Colorado and Orlando teams that start off cold and go on to struggle mightily. It’s still early but I should point out that the top left on this chart (for teams that falter out of the gate but ascend to great heights) is quite a lonely place. There’s very little precedent for teams doing what Atlanta has done so far and going on to rack up 2 points per game afterwards. To the extent that this is the only future Atlanta fans will tolerate, they best prepare themselves for disappointment.
Visualize success, but don’t believe your eyes
So, when internet historians a year from now pull the same graph as the above but add in the 2019 data points, we know roughly upon which X coordinate, 2019’s Atlanta United will sit (-0.4 xG/game through 4 games), but how high on the Y axis will ithave climbed (or fallen) after 30 matches under the current regime?
To form a coherent answer to this question you kind of have to answer a separate question of “fundamentally, what is it that will be different from the past 4 weeks in the remaining 30 weeks to come?”
For most of these historical use cases, the teams that started off poorly in the opening few games did so in part because they started the season off on the road. In 2018, DC United and Portland were undergoing stadium construction and renovation, respectively, neither team playing matches in their true home stadium. Even Toronto FC in 2017 started its first 3 games on the road before the team opened at BMO. This year’s Atlanta team has played an even schedule in this regard, half at home, so that doesn’t quite fit.
Or there’s Seattle in 2018, just above even in xGD after 2 home games and 2 away games, having been dealt the Jordan Morris long term injury blow, but who skyrocketed over the remainder of the season, adding Raul Ruidiaz in the second half of the year and generally finding a level of health and fitness the team lacked early on. This doesn’t quite look like Atlanta either unless you were to count Pity Martinez finally finding his form in the latter half of the season as a new DP signing coming on in the summer. Perhaps, I could be convinced the lack of fullback depth in the opening weeks is something that will not recur, bur frankly, (no pun intended - Oh God, you, can close this tab now and I won’t even blame you), this is a franchise that’s never had seasons at a time of fullback depth.
You might also imagine a team running a veritable gauntlet of elite MLS powerhouses in their first 4 games, and improving against a more balanced schedule down the stretch. But that isn’t this Atlanta team either, a team that dropped the ball against an expansion Cincinnati side at home.
No, it really seems that in order for the next 30 matches to look different from the first four, it’s going to really need to be the case that the team’s underlying performances to date were just not indicative of the team’s potential - that the team has faced headwinds that will not persist, that it was ill but will now recover, lost but will find its way. And this begs the question: well, what the hell are we talking about then?
It’s a tough question. My best guess is we’re talking about the unique confluence of the following factors, most of which could be temporary:
- New head coach and coaching staff seeking new team identity and play style: It is reasonable to expect this headwind to fade as the season progresses.
- The loss of the best player and creative hub from last year’s team: While it’s possible the team won’t fully recover from the loss of Miguel Almiron, it is reasonable to expect his replacement to find his footing more and more as the season progresses.
- Fixture congestion in a period where the team should’ve been gelling around a new identity and coaching instructions: it is true that the additional CCL matches will not persist; however, MLS compressed the schedule in 2019 such that 15 of the remaining 30 matches will be played with less than 5 days rest. While fixture congestion is not necessarily going away, at the very least when two teams meet from here on out, they’re more likely to be on similar amounts of rest.
FiveThirtyEight projects Atlanta to bag just under 1.6 points per game going forward, a pace that for context bests last year’s Columbus Crew and is just shy of the 2018 MLS Cup finalist Timbers, while a pace of just over 1.4 points per game (think last year’s Galaxy) over the remainder of the season would likely land Atlanta in the 7th and final playoff spot in the East. But to imagine a world where the club powers on at a rate of 2 points per game from here on out (nearly matching 2018’s Championship pace) you have to imagine a team with a settled identity that consistently creates danger for and guards against danger from bunkering opponents and opponents who come to play alike.
And the settled identity thing is tough since the captain of the team was just benched after a full week of rest in favor of the younger more mobile center back. And it’s a problem since the team has played probably 5 formations in 4 games, and it’s a problem because we don’t know who the left back is or the backup right back. And the history of Atlanta United’s success usually involves Jeff Larentowicz, who was also out of the team this past week. And sure, I get it, that it’s hard to fit Tito Villalba and Julian Gressel and Ezequiel Barco and Pity Martinez and Josef Martinez into one attack, but it’s also hard to imagine a team with a settled identity scoring goals without playing all of those guys, and yet Tata Martino accomplished it last year.
Inventing Soccer Again
All in, if I were to summarize the complexities of Atlanta’s problems at the moment, I would say that the first four weeks have looked like the first four weeks of an MLS expansion team, albeit a well-funded one. Maybe as a comparable, what we should be looking at is the 2018 LAFC team. They opened 2018 with a new coach and a new star player (OK everyone was new). It’s not perfect because these guys played their first 4 games on the road also, but ultimately LAFC put up weak expected goal differences early and as the team came together over the course of the season and its identity formed, it earned 1.7 points per game from the remaining 30 fixtures, making the playoffs and just narrowly missing out on a first round bye — a clearly talented team that ran into an RSL buzz saw in the playoffs.
Expansion teams have an opportunity to band together and create an elite supportive and winning team culture from scratch. If it is in fact true that the issues here are not the permanent, structural kind but the transient, cyclical ones, then rebuilding this culture over the next few months will be critical and keeping the players’ motivations and general comradery at a high level will be the only way to transition the group from the over-worked but under-drilled squad of talented footballers we suspect of them to the focused, well-conditioned, defending champions we expect of them. Standing at the beginning of April, there are plenty of matches left to play, and there’s plenty of time for things to start going right, and there’s still plenty of risk that the things that need to go right won’t.
Turn the lights out, say goodnight. No thinking for a little while.
Let’s not try to figure out everything at once.