What Expected Goals tells us about the 2018 Quakes so far

The Expected Goals (xG) metric is the all the rage in soccer analytics right now. To some, xG is the Swiss army knife of metrics: it can be used for everything from predicting game outcomes, to setting fantasy lineups, to being the cure for any ailment a team has. While that is likely overblowing it a bit, xG is certainly useful in the appropriate context and can be used as an indicator if a team may be over-performing or under-performing in a game, month of games or entire season.

In a previous article, we mentioned xG is a derived metric, not a statistic. Various analytics websites and companies have applied their own xG models on top of shot data to create xG values, and these models are being constantly improved. Of these sites, AmericanSoccerAnalysis.com (@AnalysisEvolved on Twitter) is the one who applies their xG models to Opta shot data for MLS and shares the resulting xG values, along with much of their model methodology, with the public. All xG data in this article comes directly from or is calculated from ASA public data. Other xG models may provide different values for the same shots and goals.

Expected Team Goals

The 2017 Quakes were an interesting expected goals enigma. Both offensively (39 GF, 46.6 xGF) and defensively (58 GA, 47.1 xGA) they under-performed against xG. In fact, the Quakes -18.5 goal difference when subtracting expected GD (xGD) from actual GD was the biggest under-performance in MLS. The only other team lower than -12 GD-xGD was DC United at -18.1. The significant xG over-performing teams in 2017 were Atlanta United at 27.6 and Toronto at 25.7. No other team was even close to them. Contrast this with the 2012 Quakes who performed at 18 goals over xGD.

San Jose Earthquakes Goals For/Against and Expected Goals For/Against since 2012

GF=goals for; xGF=expected goals for; GA=goals against; xGA=expected goals against; GD-xGD=goal differential minus expected goal differential

The 2018 Quakes started off the season in a somewhat different direction, actually over-performing against their expected goals for five of the first seven games before significantly under-performing against the Portland Timbers last Saturday. Many thought this early over-achievement was a warning sign, and perhaps it was. Currently the Quakes 12 goals through eight games matches fairly well with their expected team goals for (xGtF) of 10.64. They sit at 11^th overall in goals per game and 12^th overall in xG per game per ASA. According to a tweet on May 11 from Matt Doyle (@MattDoyle76) at mlssoccer.com, Opta, who uses a different and undocumented xG model, has San Jose at 9.46 xG.

Just brutal from Philly pic.twitter.com/HfyWG4vi3m

— Matthew Doyle (@MattDoyle76) May 11, 2018

The potential reasons for having a higher goals than expected goals through the first few games include converting low-probability shots, poor opponent goalkeeper performance, a better-than-average shot on target ratio, a bit of shot luck (like Hyka’s deflected goal against Houston) or a combination of these factors. The combination of factors is the most likely reason here. The Quakes currently are 5^th in the league in percentage of shots on target with 38.6%. Putting shots on frame is a good start – one can easily argue they just aren’t shooting enough of them – but in actuality, while the Quakes sit 15^th in shots per game, they are 11^th in shots on target per game, which is not too bad. Shot quality is not a big issue – the Quakes sit in the middle of the pack in xG per shot and xG per shot on target. There’s nothing in this data to get excited about or to predict a pending turnaround without changes in the summer transfer window. But at 1.5 goals per game, the attack’s ability to score goals currently outpaces any Quakes season since 2012, even when accounting for an overall league goals inflation of 11% on a per game basis since then. While not scoring a goal against Portland at home is concerning, the recent Orlando and Portland games respectively had the second- and third-best xG of the season, which an optimist could see as a sign of better things to come.

Expected Team Goals Against

Of course, the biggest concern of the season is the defense — and here we will examine what expected team goals against (xGtA) tells us. In 2017, the Quakes infamously tied the 1998 team record for goals against with 60 (1.76 goals per game). The 2018 Quakes are on pace to break that record with 2.0 goals against per game, currently 21^st out of 23 teams this season. Real Salt Lake is also at 2.0 goals against per game, while Impact Montreal currently have the league’s worst defense at 2.44 goals against per game. But if you are looking for the silver lining, it would be the expected goals against. The Quakes 1.34 xGtA is 7^th in the league. In other words, it is possible the Quakes have been a bit unlucky. Andrew Tarbell started off the season near the bottom in keeper xG with 7 goals against but only a 5.23 xGA, meaning he had given up 1 or 2 more goals than he probably should have given the quality of the shots. Starting with the Philadelphia Union away game Tarbell has again shown much of that same promise he displayed in last season’s US Open Cup, giving up 9 goals but with a 10.05 xGA in the 5-game span, plus he has had a lower GA than xGA in 4 of the last 5 games to go along with 20 saves in those games – an encouraging reversal.

Head-to-Head xG

If games were determined by xG, the Earthquakes would currently have 4 wins and 4 losses.

What’s telling though is the average margin of victory of those hypothetical 4 wins is 0.62 xGD (expected goal differential), but the average margin in the 4 losses is -1.44 xGD. So in xG terms, the losses feel twice as bad as the wins feel good, and that sums up this team fairly well to this point. Minnesota United aside, even when they play well, the Quakes seem to barely outplay the opposition (much of the NYCFC and Portland games), but when they play poorly the opposition looks significantly better (much of the Houston, Orlando City and Columbus games). xG bears out this general feeling.

Expected Goal Probabilities

Expected goal probabilities tell us, based on individual shot xG values, the likelihood a team should have won or lost a game based on the quality of their chances. Eliot McKinley explains this concept in a recent article for ASA. For example, the chart below tells us the probability the Quakes should have gotten the 3 points against Minnesota in game 1 is 61% based on expected goals for both teams.

This chart for game 8 against Portland indicates the Quakes had a 55% chance of getting a win and were quite unfortunate not to at least get a point.

Goal probability charts credit to Eliot McKinley (@etmckinley on Twitter) using ASA source data

However, based on social media reactions during the match, Earthquakes fans didn’t feel good overall about the quality in the game. Note the Quakes didn’t get a clear xG advantage until the 55 minute mark. Add on the Diego Valeri free kick golazo (indicated by the dot) which was a low xG chance, and it just leaves a bad taste about the quality of the play, even though the end xG is reasonably good. Note by the quick upticks in the xG trend line both teams had several better chances throughout the match than the Valeri goal itself was. In these probability charts, McKinley uses xGp on the y-axis to denote he is using the aggregate of player expected goals instead of team expected goals, which are often the same but sometimes are not.

Player Expected Goals

xG at a player level is limited to essentially gauging the effectiveness at finishing chances provided to attacking players. While this provides some usefulness, when you consider the Quakes have four attackers up top who are expected to provide both goals and assists, a more useful way of looking at them would be their expected goals plus their expected assists (xA). For example, while Danny Hoesen and Valeri “Vako” Qazaishvili were grabbing the early headlines with actual goals and assists against Minnesota United in the first game of the season, the highest combined expected goals and assists belonged to Magnus Eriksson with a 0.134 xG and 0.84 xA. This was an early indicator Eriksson was already fitting into the team and would be an asset. Indeed, Eriksson now has the most combined points on the team with 5 (2 goals and 3 assists) and also leads the team in key passes (passes which end in a shot, shot on goal or goal) with 19.

Minnesota United FC 3/4/18 Game – Goals, Assists, Key Passes, xG and xA

Conclusion

Whether you view xG as the ultimate soccer metric or just one indicator to support your viewpoint, there is no doubt it tells us something about the 2018 version of the Earthquakes. For those who wanted a change from 1-0 predictable Dom-ball to something more exciting, xG tells us these Quakes may be in the highest scoring games this side of 2012. For those seeking hope for this season, xG tells us the defense and goalkeeping are improving and maybe have been a bit unlucky so far. For those seeking reality, xG also says maybe the Quakes are right where they are supposed to be in the table. With three-quarters of the season remaining, expect much more analysis from Quakes Epicenter using xG and other metrics and data in the coming weeks.