Photo credit: ISI Photos. JT Marcinkowski directs his defenders during one of the San Jose Earthquakes’ few clean sheets of 2022, a 2-0 win over the Vancouver Whitecaps on September 4th.
Special acknowledgment to Quakes Epicenter patron Trevor Wojcik, a data scientist in the economics field, for constructing a season and player simulation technique for me that will be used in this series and beyond. Trevor’s insightful work and feedback made this preseason series concept for 2023 possible.
Introduction
In the first article in this series, we revisited the 2022 season — 10,000 times. Using a season simulator, we demonstrated that the 35 points won by the Earthquakes in 2022 were bang on to their expected number, but that teams often over-achieve or underachieve the value they should have gotten. We demonstrated Austin FC is an example of a team that overachieved their true performance by a few points last season.
Those 35 points were 12 under the final Western Conference playoff spot held by Real Salt Lake. In 2022, the Earthquakes continued to concede far more goals than they scored, a habit that has been in place since 2015. In order to bring order from the chaos of MLS, targeted improvements have to be made to switch those two numbers.
To switch those two numbers in 2023, San Jose can do one of two basic things:
1) Score more goals than they give up. In 2022, they scored 52 but gave up 69. For the Earthquakes that would mean scoring 17 more goals just to break even. An even goal differential might put them on the bubble of the playoffs but more likely out of them.
OR
2) Reduce 69 goals against to a number less than they are able to score. Third-place FC Dallas scored only 48, but they also only gave up 37. Fifth-place Nashville scored 52, but they only gave up 41.
Based on their moves in the 2022 Secondary Transfer Window and in the 2023 Primary Transfer Window so far, the Quakes have chosen to address #2.
Starting from the Front
Although they have chosen #2, it is important to maintain at least around 50 goals this season. The lowest goal count a playoff team had in 2022 was 43 by seventh-place RSL. To maintain 50 or more goals, new San Jose head coach Luchi Gonzalez is going to need to creatively squeeze a little bit more out of players like Jeremy Ebobisse, Jamiro Monteiro, and likely either Benji Kikanovic or Cade Cowell (or both, depending on who becomes the regular starter on the left).
If there is a concern about Jamiro Monteiro, it’s not for a lack of effort. He covers the pitch like an in-his-prime Nico Lodeiro. The problem is he produces offensively like a past-his-prime Nico Lodeiro. That’s a joke — Lodeiro produced a lower 0.46 xG+xA (Expected Goals + Expected Assists) per game last year (16th among attacking midfielders) but still produced much more than Monteiro’s 0.29 xG+xA per game (22nd out of 26 attacking midfielders). As I explained last preseason, this is a key metric for measuring attackers that are Designated Players.
Monteiro’s actual goals and assists numbers were slightly higher but nowhere close to enough to move him up the list. The Quakes did not get enough threatening attacks from the middle of the pitch in 2022.
That’s not a concern if the Quakes have another player in the top 15 or so in the league, but their first player sorted by xG+xA per game is Jeremy Ebobisse — 29th in the league. It’s a challenge to articulate how this team scored 52 goals like that. A better set piece year? Can you count on Nathan to score three again?
Here are the top 2022 scorers:
- Ebobisse – 17 (15 non-penalty)
- Espinoza – 7 (2 non-penalty)
- Kikanovic – 6
- Monteiro – 4
- Yueill – 4
- Cowell – 3
- Nathan – 3
- Several tied for 2 or 1
The 2023 strategy to get 50 goals feels like it’s going to have to be:
- Hope Ebobisse stays healthy.
- Hope Kikanovic isn’t sent to Greece (or is replaced by a DP…whoops…high TAM player after Gruezo signs).
- Hope Monteiro or Yueill suddenly flourishes under Luchi as Zone 14 merchants.
- Hope Cade just finally has that year despite still not being the legal drinking age for another two years.
I’m skeptical. Asking Ebobisse to score 20 (at least 18 non-penalty) goals feels like squeezing blood from a stone. Espinoza is who he is, one of the top assist players in MLS for the last four years but almost certainly hit his peak in 2022. I would say the Quakes can’t rely on another seven penalty goals, either — that’s a bit on the lucky side. Kikanovic is more of a scorer than a player who will find Ebobisse as often as Espinoza does. If Luchi doubles Benji’s minutes (which he should if he stays), maybe he scores 10 or more. I’ll take it.
But someone whose name isn’t Cristian Espinoza needs to also get the ball to Ebobisse, Kikanovic/Cowell, and backup-striker-to-be-named-later more often in dangerous positions. That has to be Monteiro and Yueill or Monteiro and 17-year-old Niko Tsakiris. There’s a Yueill-ephant in the room, but we’ll have to address that another time.
So, my colleague Trevor Wojcik and I got to work. We wanted to find out if Ebobisse should have scored more in 2022, and what would it take for him to score more in 2023. Trevor modified the season simulator we created for Part 1, and adjusted it so we could look at individual players.
Rats. It looks like Ebobisse pretty much hit his peak in 2022. Asking more without changing anything is very difficult to do or requires a lot of luck. The best way to improve his probabilities is to give him more shots (i.e., more service) and/or better shots (i.e., better service).
Let’s see how this works. Ebobisse shot 101 non-penalty shots in 2022, good for third among strikers behind Muhktar (119) and Arango (106). That means there isn’t much upside to more shots, although maybe 0.3 to 0.5 a game is possible. The focus needs to be on better shots.
That’s where Monteiro, Yueill, and, hopefully, Tsakiris have to come in. Currently, they aren’t a threat around the box. There are only going to be so many perfect crosses a team can do in a season. The improvements have to come from attacking patterns in the final third to give Ebobisse a chance to finish more.
13 of Jeremy’s 15 non-penalty goals came from open play, and 10 from the type of progressive passes I talk about a lot — but only one came on a through ball and none came from a cutback. What that means is the team was having to score a lot without “getting behind” defenders, and that drives down the probability (what we geeks call “xG”) of each shot.
Fortunately, Luchi is well ahead of us. Watch the following video from 5:25 to 6:45.
Sound familiar? USMNT fans will immediately recognize this language. It sounds like Luchi-ball from FC Dallas will be morphed into more Berhalter-ball. That’s great for the defense (more on that in Part 3), and probably good overall for the offense.
While Berhalter was very successful at improving the USMNT defense, he was a good bit less successful on the offensive side. But playing in the World Cup is like playing the first four games of a new season and not playing the next 30 to see how it turns out, particularly since the lineup was healthy enough to play together for the first time ever at the World Cup. We don’t know what we don’t know about that young team and their ability to execute Berhalter’s attacking philosophy against some of the better defenders on the planet.
We’ll talk more about the tactics in Part 3, but the attacking philosophy here is to win the ball higher with the press and counter-press, attract with the ball, play quick, play vertical when possible, and get in behind the defense — all while having a rest defensive structure when the attack fails to produce a shot. In theory, it should result in some better shots, while protecting from wide-open counterattacks.
Ebobisse had a pretty good 0.141 xG per shot (14.1% expected conversion rate) in 2022. The league average for strikers was 0.135. There were a few strikers, all with fewer shots, with 0.15 xG / shot, 0.16 xG / shot, or even higher. Let’s say we can bump Jeremy 0.01 xG higher on average. This may come from just a few shots with much better xG. For example, an 0.15 xG shot from a progressive ball with defenders between him and the goal might turn into a 0.45 xG one behind the defense or a 0.35 xG cutback from a run outside the box that freezes the defense. This won’t happen all the time, but even once every 2-3 games can make a big difference.
As you can see, we’ve bumped it up a bit. Whereas 14 goals were the highest peak before, now 16 goals are. It’s no guarantee he’s going to get 18 to 20 goals, so it might take a 0.02 xG per shot improvement to increase the probabilities that the team will need them to keep their goals around 50. That sounds like so little, but it’s really tough to do. This will be a metric to watch as we get a few games into the season.
There are other “simple” ways to improve the attack, and we’ll also get into these in Part 3.
Big Changes in the Back
It’s a bit frustrating to hear the constant “where are the signings?” question. We are now in a situation where five of the six basic defensive starters — defensive midfield, back four, and goalkeeper — are expected to be different from 2021, pending the Carlos Gruezo signing as of publication and a determination if new signing Daniel (De Sousa Brito) will start opening day. Only Nathan may be the constant from 2022.
That’s a lot of signings and a lot of change. Even in the SuperDraft, the Quakes went defensive with Daniel Munie in the first round. Luchi and Co have their work cut out for them to get these new starters working as a cohesive unit in his new defensive game model by the start of the season, and I expect some rough patches as they iron it out and evaluate these players.
Now, there will be a lot of fans very concerned about changing their favorite goalkeeper, JT Marcinkowski, for Daniel, if that happens. I understand the concern. JT is a leader. We enjoy his candid answers in press conferences and the way he takes accountability. But there is no denying that something was amiss in 2022 with JT’s performances as the season went on if we are objective about it.
According to American Soccer Analysis’ public data, JT gave up 10 goals more than expected in 2022, while saving almost 5 goals more than expected in 2021 (shown in the right red box below). The number of shots he faced, the post-shot xG aka “xG Keeper” (acronyms: PSxG or xGK — I’ll be using xGK in this article) of the shots, and his minutes are very close to each other (see the gold boxes). However, he gave up 15 more goals in 2022 than in 2021 (left red box).
Now xG models are all different, and so it’s only fair to look at other publicly-available models. FBref.com now uses StatsPerform (formerly Opta) data for their models. They are slightly less critical of JT in 2022, but less positive about his 2021 as well. StatsBomb aligns with StatsPerform as well, with roughly +3.5 in 2021 and -8 in 2022.
So what gives?
First, JT faced more shots in a 2-year period than any keeper in MLS since 2013 when this data started being tracked. Facing 186 attempts in 2021 and 182 attempts in 2022 is #2 and #3 respectively since then.
JT also faced 12 one-v-ones in 2022 according to StatsPerform, most in a league where the typical starting keeper faces three or four (William Yarborough faced 10, conceding four). This resulted in seven goals.
In my conversations with former goalkeepers, they are concerned this barrage has been taking a toll on him mentally and physically. He’s giving up at least twice as many goals than any keeper on a team facing half of those shots and thereby it has the ability, even subconsciously, to affect his performance.
Secondly, JT lost his legendary goalkeeper coach, Carlos Roa, when Matias Almeyda was dismissed. Goalkeepers I’ve talked to tell me that they value continuity in their goalkeeper coach. There are many examples of a top keeper losing their coach to another club, and seeing their performance dip dramatically. In moving from Roa to an interim goalkeeper coach, we can almost see it in the numbers when we follow JT’s xGK-minus-goals conceded trendline for the last two seasons.
There’s a huge difference after Roa left, and it was almost immediate. While there are peaks and valleys during the rest of the season — not every game was an under-performance — the trend is steady and the decline was steep.
The metrics across xG models from ASA, StatsPerform, and StatsBomb say that JT‘s conceded goals–given the volume and quality of both pre-shot locations and shot types plus shot placements and ideal keeper positioning–should have been somewhere between 56 and 59, instead of the 66 that it was.
Finally, questions about Emi Ochoa’s future remain and are now magnified by the hiring of Gaga Slolina’s goalkeeper coach, Adin Brown, from Chicago Fire FC.
Can Brown help Emi Ochoa, who feels likely to at least be in the US squad in the U-20 World Cup May 20-June 11 in Indonesia, make a big move to Europe as well? I think Chris Leitch is banking on it.
That means either the Quakes can hope that JT will regain his form under Brown for 2023, while facing many fewer shots, and also bring in a run-of-the-mill MLS backup. Or something had to be done and done immediately.
And to be clear, even a JT who is “in-form” is a 72% shot-stopper, which is just barely a top-third season in MLS. JT is an MLS-quality goalkeeper across all parts of the game, but it’s very difficult to argue he’s above average as a shot-stopper. His career save percentage is ~69% both for the Quakes and at Reno 1868 FC. Jon Busch and David Bingham turned in 74% save seasons between 2014 and 2016.
Although there have been worse recent seasons such as Andrew Tarbell in 2018 and Daniel Vega in 2020 (both of those keepers lost their jobs during the season), I favor doing something immediately and not taking a chance. Apparently so did Chris Leitch and his scouting/video/data analytics staff.
Enter Daniel De Sousa Brito, known as just Daniel in the Brazilian way like Judson, Nathan, and Rodrigues. The Quakes had a chance to get Daniel, one of the two best shot-stoppers in the Brasileiro Serie A in 2021 and 2022 since Internacional’s backup goalkeeper went nuclear while Daniel was nursing an injury in the last fourth of the 2022 season. With Internacional looking to cut expenses and bring in new attackers, they’ve gone younger and cheaper at the keeper position, and Daniel became available to San Jose.
Sure, it’s not typical to use an international slot and some TAM on a goalkeeper in MLS, but I strongly believe that’s a market inefficiency. The most valuable position on the pitch in MLS is the goalkeeper.
Let me demonstrate this in a couple of different ways. Here are the top position field players in the American Soccer Analysis’ Goals Added (g+) metric by total Goals Added above the average player in their position.
While g+ isn’t perfect by any stretch, it does a pretty decent job in the aggregate explaining the goal difference in a team and attempting to allocate it to each player across six categories.
g+ says that Hany Mukhtar was the most valuable field player in 2022 at the striker position with Carles Gil at the attacking midfielder position. It’s tough to argue with that and most of this list, honestly — except, well, for Michael Bradley who is weirdly valued highly due to a very big contribution in one game (that’s what intercepting a pass, dribbling past defenders, and scoring all in one situation will do for your value). You can also see that Mukhtar and Gil get their points in different ways. Mukhtar gets value mostly from Receiving and Shooting, while Gil gets value from Dribbling and Passing.
Okay, now let’s look at the goalkeepers.
What you’ll see from this (along with the absence of former chart-topper Matt Turner due to his move to Arsenal) is that the top four keepers are more valuable than any field player in terms of their impact on their team’s goal difference above the average MLS player.
The other thing you’ll notice is that almost all the value comes from the Shotstopping category. Six of the top seven keepers even have negative passing values. Guess what? It doesn’t matter. The value of passing, so long as a keeper doesn’t make more than a few big mistakes a season, is low, as is every other category. Ball-playing keepers may feel important to Twitter and some coaches’ systems, but the models don’t really see it. That’s not a Goals Added flaw: the other possession models see things the same way.
Now, while the Brasileiro Serie A is not available on the public ASA site, I have the data (although not for Sudamericana or Libertadores, sorry). Among the goalkeepers, Daniel was #2 in Shotstopping (5.83) and #3 (5.40) in overall Goals Added in 2,021 minutes which is two-thirds of the MLS keeper minutes. If he maintained that pace in 3,000 minutes he would be third on the MLS list, assuming a directly equal translation of performance between the leagues.
If you believe Brazil has the better league, as some do, well then the Quakes may be getting a top-tier MLS keeper. According to ASA’s xG models, the shots are a little tougher on average in the Brasileiro Serie A and the keepers are considered a little bit better overall, making Daniel’s over-performance even more impressive.
Over 10,000 simulated seasons, Daniel significantly overperformed expectations in both seasons as the starting goalkeeper with Internacional.
Remember that earlier chart with JT’s xGK minus Goals? Here’s Daniel’s.
Now the thing you will notice quickly is that he’s faced fewer than half of the shots JT has faced. Yes, he’s had a marginally competent defense in front of him as well, finishing second in the league, but he’s faced tougher shots than JT according to the average xG and xGK (PSxG) of the shots.
You may say, “that’s all well and good, Jamon, but JT was above the line for 2021/2022 as well until around 200 shots”. That’s a completely fair argument. Daniel has to prove it for another 200 shots or so, but even in Daniel’s shots, his trend is very consistently up. Every time he starts to dip, it reverses course.
This is due to a very consistent save percentage. While JT has moved between 67% and 72% for his career at both USL Championship and Major League Soccer, Daniel finished 2021 with 79.6% and 2022 with 78.5%. He’s been very consistent overall. While we don’t have the data for these games, in over 12 games of minutes between Sudamerica and Libertadores, he has an incredible 88.2% save percentage according to FBref.com, which gives us even more confidence in this analysis. The league average is 74% as the black horizontal line shows below.
There have been some rumors that Daniel De Sousa Brito made a couple of Daniel Vega types of mistakes for Internacional that caused fans there to turn on him. While that’s possible, I haven’t seen the video proof yet. But, frankly, that’s just noise and not signal. Even if a keeper was directly responsible for three mistakes leading to a goal in a season, but saved 10 goals over expected, they would still be +7. Fans will focus on the three mistakes, instead of the +7 overall.
Daniel Vega not only made mistakes, but he was average in 2019 in both xGK-Goals and save percentage and well below average in 2020 in both. The mistakes made him a net negative, even in 2019 when his xGK-Goals were barely positive.
Now, you may still not believe me, so Trevor and I have done something I don’t even know has been done before. We have built an xG keeper model just for Daniel and upscaled him to ~5,000 shots to match the total for the other keepers in Brasileiro Serie A in 2021 and 2022. Then we took 3,333 shots from Daniel and 1,667 shots from the other keepers to account for a potential 50% regression to the mean. And then we ran all of the shots JT faced in 2021 and 2022 against the “Daniel Model”, to make Daniel face JT’s shots for 10,000 seasons. Yeah, it’s a bit crazy.
Even making Daniel take 33% of his model from other keepers in his league, he was still likely to improve on the Quakes’ 2021 by a couple of goals and on 2022 by probably 16-20 goals. This is why Bruno Costa scouted him, and then Chris Leitch went and got Daniel.
For fun, I created a Matt Turner-only model. Turner is 2x better than any other MLS keeper since 2013 on an xGK saved per shot basis. He’s 2x better than second-place Nick Rimando who is ahead of the other 150+ keepers (Houston has the #3 in Steve Clark). The Matt Turner Model says if Matt is between the pipes for the Earthquakes he most likely concedes 46 goals in 2021 (vs. JT’s 50) and 47 in 2022 (vs. JT’s 66).
Matt Turner would maybe have put the 2022 Quakes into the playoffs with a +2 goal differential. That’s why the goalkeeper position is the most important position in MLS.
No other position can do that unless you are getting prime Messi. Prime Wondo, Vela, or Josef doesn’t even do that.
And remember we made Daniel’s model worse on purpose. I cannot wait to see how this turns out in real life.
Conclusion
Look, if you are upset the Quakes went out and tried to get better at the goalkeeper position because you like JT, I don’t know what to say to you. We all like JT — love JT. JT is the man, and he’ll always be the man.
But if you want to see these Quakes make the playoffs in 2023, I hope this has bolstered those hopes. In Part 3, we will take a look at the game model that we are seeing in the preseason and are likely to see during the season and make a call if we think these Quakes will make the playoffs under Luchi Gonzalez in year one.
Enjoy the preseason games.
If you want to know more about our simulation methodology, Daniel Model, and Matt Turner Model, we will be discussing them on our analytics channel on the Quakes Epicenter patron Slack. Join our Patreon for $5 or $15 per month to get access to our Slack throughout the 2023 season and join the crazy game day chat and other chats we have throughout the week.