Morgan Rogers does not have that kind of profile. He is an attacking midfielder who can play either wide or central. He does a fair amount of ball progression with passes and with carries, but falls well short of the league leaders at these positions. Rogers made news with his goal-scoring, including some gorgeous finishes, but he ended up with 10 goals last season in well over 3,000 minutes, and his xG was somewhat behind that. His shot creation numbers fall into roughly the same zone: average to above average depending on how you look at it, but short of elite. Now, it could be the case that despite not having one standout metric, the aggregate of his production might place him among a rarefied set of players. That is exactly what the comparables score I developed in the Anderson piece can help identify. How many players have put up numbers similar to those that Rogers most recently produced in the 2025–26 season?

How the Comparables Work: Ball Progression by Carry

With ball progression, I am using the same numbers that I used in the Messi and Ronaldo articles.

• “Progressive” actions defined as those that move the ball at least 25% of the distance toward goal

• “Expected threat” or “xT” added, comparing the likelihood of goal-scoring from the point where an action began and the point where it ended

• Passes and carries into the penalty area

For each of these, I’m looking at both the player’s raw numbers, and their share of team production. “Share of team production” refers to the player’s individual share of the team’s total in that statistic while the player was on the pitch. So if a player completed 10 progressive passes, and in those minutes the whole team completed 100, the player would be responsible for a 10 percent share of team production. For comparable players, I am looking for players with at least 1800 minutes played at central and wide attacking positions, excluding strikers. (For further elaboration of how these numbers were produced, see the Appendix on Method below.)

The story across all of these metrics is that while Rogers’ raw numbers do not distinguish him very much from the median qualifying attacker, he did clearly carry an outsized load of the work for Aston Villa.

These are the charts for progressive carries.

One clear thing that stands out in the numbers is that Rogers, despite being known for his ball-carrying, is not a huge volume ball progressor. Playing in a more central role, he has fewer opportunities to get on the ball and drive into space than a true winger like Jérémy Doku, Savinho or Alejandro Garnacho. Even within Aston Villa’s system, he has not been asked to do an unusual amount of carrying given his position.

Now, this could be an issue with the progressive action definition, which tends to favor true wingers who get more space to drive the ball forward. Expected threat typically gives far more credit for actions taken in and around the penalty area, where there are much larger differences, from one spot to another, in the likelihood of goals being scored. Here the numbers tell a slightly different story. Rogers produces no more than an average amount of penalty area danger from his carries, but he is crucial to Aston Villa’s tactical approach as their core ball-carrier near the penalty area.

The numbers for carries into the penalty area are similar, with Rogers becoming increasingly important to Villa’s attacking plans.

As in the Anderson piece, I use a blend of raw and team share to produce an overall score. Here I am also mixing together the three statistics, with higher weight on progressive actions and xT added, to produce a “carry progression” score and comparables.

Broadly, this is one of the stories I have found with Rogers. His ball-carrying produces the occasional highlight-reel moment, but in the aggregate it is hard to distinguish his statistical profile from most Premier League attackers.

But he does begin to separate himself, at least somewhat, when the full breadth of his attacking production is accounted for.

Passing, Shot Creation, and Those Goals

Read more

This profile has three fundamental pillars. Anderson gets on the ball at an exceptionally high rate. He is effective moving the ball forward through the defense, primarily via passing. And he does a good amount of defensive work. It sounds simple, but very few young players can manage such production.

In this newsletter, I will go through Anderson’s development and produce a set of methods for identifying comparable players over the last 17 years. It will take some time to explain, but by my preferred method, these are the six most comparable seasons to Elliot Anderson’s 2025–26 in my database:

• Jorginho, Napoli 2016–17

• Granit Xhaka, Gladbach 2015–16

• Hakan Çalhanoğlu, Inter 2023–24

• Toni Kroos, Real Madrid 2022–23

• Marco Estrada, Montpellier 2012–13

• Marco Verratti, PSG 2022–23

It is an impressive set of names, and the full comps list has another Arsenal connection, fantastic Serie A playmakers to remember, and a few more world-class pivot midfielders. Let’s go through the numbers and see how we get there.

Elite Involvement in Play

A simple way to think about touch volume is involvement in play. How often is this player showing for the ball and receiving a pass when the team is in possession? Over the course of the last three seasons, Anderson went from, at Newcastle, a player who was relatively uninvolved in possession to, last season with Forest, the Premier League’s leader in involvement among qualifying central midfielders.

What is notable here is that the other leaders in this statistic all played for Manchester City. Not just Rodri, but also Bernardo Silva and Mateo Kovacic. It is much easier for a player to get a lot of touches in a team that dominates possession and circulates the ball in midfield. In the next chart I look at a player’s involvement in context: what percentage of the team’s possession touches does that player account for while they are on the pitch? Strikingly, Anderson’s 2024–25 season now appears as among the elite, an early indicator of his breakout season to come.

The only Premier League midfielder over these three seasons who was as integral to their team’s possession play as Elliot Anderson in 2025–26 was Rodri. Reporting on Manchester City’s interest in Anderson has consistently mentioned the importance of his ability to constantly look to receive passes, and the data confirms this skill.

Of course, Rodri is not known just for his ability to find space and show for the ball but what he does when he gets it. And on this measure, Anderson’s development has been equally impressive.

Elite Ball Progression, Too

When Anderson was first coming up through the Newcastle system, he played as much more of a free eight looking to drive the ball upfield. He used his athleticism and technique to move the ball forward at his feet. The classic problem for a player like this is two-fold: they move themselves into a new zone where they may not be as ideally positioned for recirculation of possession, and of course they are no longer in their defensive position should the ball be turned over. Midfielders who progress the ball by carry at high rates tend to be free eight types, often involved in play near the penalty area. As Anderson developed at Nottingham Forest, he cut back on ball-carrying.

Sometimes Anderson’s ball-carrying gets mentioned among his elite skills because he still does flash the ability to move the ball forward at his feet. But this is no longer a large portion of his game. Anderson has shifted his ball progression responsibilities from carrying to passing.

Read more

PADDLIN’ won the contest.

Notably, PADDLIN’ and Nate Silver’s PELE model managed to defend the honor of the modelers as the “BLUESHIRT” predictions were made by a person plugging in their own opinions.3 Along with a few “human” picks, the contest also included a number of baseline models, made using LLMs or various simple aggregates.

The various baselines, whether a pure agnostic model or a Claude or ChatGPT prompt, underperformed the models and the humans. Individual people making subjective picks had a great deal of variance, ending up with three of the top ten overall, and also with two scores lower than any football-based model.4 The Polymarket futures made the top five, but fell slightly behind the Fantomen model as well as the top three.

It is fun and validating to win a contest, and I could just end the newsletter here, but I think there are some lessons to consider about modeling and about what “winning” a contest means.

For one thing, I wrote an entire World Cup semifinals preview in which I talked about why my subjective analysis disagreed with my model. So I cannot take too much credit here. In this newsletter, I’m going to look at how PADDLIN’ won, and consider to what degree I can learn from my disagreement with my model.

I wrote up my model and methodology in the Introducing PADDLIN’ newsletter here, and added continued discussion of its workings across several of the World Cup blogs. Admirably, Nate Silver has continued to publish full explainers of his models, and you can read his exhaustive write-up on PELE here.

The Winning Models Took Similar Risks

The most notable characteristic of the winning models in this contest is that they made precise and confident calls. PADDLIN’ stands out in particular in the group stage, with unusually confident predictions of a group stage exit for some of the worst teams at this World Cup: Curaçao, Qatar, Jordan, Iraq, Haiti, New Zealand, and South Africa. The PELE model also stands out for its aggressive calls on the top four teams at the World Cup: Spain, Argentina, France and England. And both particularly took clear positions on the gaps between better and worse teams likely to meet in the round of 32.

Now, not all of these risky calls paid off. In particular, PADDLIN’ and PELE missed on some of their more off-consensus calls in the group stage and had to catch up.

This is probably the most interesting thing about the prediction contest results, to me. Top contenders may succeed in different ways, and indeed BLUESHIRT made the most of group stage qualifying calls and was pegged back during the knockouts. But PADDLIN’ and PELE just had very similar outputs.

At base, it is not terribly surprising that PELE and PADDLIN’ had similar team ratings. Both models were based on a mix of a weighted Elo for team performance and adjusted Transfermarkt value for player quality. The processes of weighting Elo and of adjusting and integrating Transfermarkt value were different, but the shared tendencies are easily identified.

First during the group stage:

All of the teams where PELE and PADDLIN’ showed up off consensus in the groups were teams whose Transfermarkt values made a significant difference in their projections. Norway had one of the five largest boosts from Transfermarkt value in the field, and all of the others were significantly downgraded by the player value adjustment.

Of course, the misses here on South Africa and Australia cost both systems points.

And then in the round of 32, both models had a number of strong bets that paid off, with a decidedly different pattern:

Here it seems there are two clear patterns. First, the two models trusted CONMEBOL results and gave notably high odds of advancing to some teams that had not impressed outside observers: Brazil, Colombia, and Paraguay. In the cases of Colombia and Paraguay, these teams were dragged down by their Transfermarkt squad value, but their records in qualifying and in Copa America were nonetheless enough to justify a higher rating than the consensus.

But the most obvious factor driving these calls is home field advantage. PELE and PADDLIN’ were far above consensus on Canada, the United States, and Mexico. This must be the host effect. Both systems gave a large effective Elo boost to World Cup hosts (about plus-60 in PADDLIN’ and about plus-90 in PELE) and both gave about the same additional boost for home teams playing at elevation (an additional plus-60 or so). I spent a long time working on home field advantage, and I talked on the Double Pivot podcast about what a nightmare it is to estimate it, and it seems this work paid off.

In the end, the results in the final three matches broke my way. PELE was much higher on England and somewhat higher on Argentina and lower on Spain than PADDLIN’. There are many versions of these semifinals where PADDLIN’ does not pass PELE, and ultimately the contest has to be extremely close for it to come down to the final.

What I find more interesting, and what made most of the difference here, are the similarities between these two models’ output.

Both made aggressive calls throughout. One of the hardest things in international football modeling is simply trying to get the model to predict that bad teams will lose to good teams at very high rates. The expanded field in World Cup 2026 meant that there were more teams with very low chances of success, and PADDLIN’ and PELE were designed to identify those teams.

Both models took home field advantage seriously. Even though the calculations came out somewhat differently—I decided to regress my home field advantage finding somewhat more—ultimately having an objectively derived home field advantage factor made a big difference. And finally, the two models bet on the teams which had demonstrated levels of success in South American competition. This had mixed results over the tournament, but was ultimately a winner given Argentina’s run to the final.

These choices and findings were far from exciting, but they were no less crucial for that.

Good Advice Is Often Boring Advice

When I look back on my doubts about PADDLIN’s picks for the semi-final, the consistent argument that I was making was that the model was updating too slowly. Argentina’s rating was based to a significant degree on the side’s performances from 2022 to 2024. France’s rating had not updated quickly enough based on their complete dominance at this World Cup. Perhaps England should have been given more of a reset to their rating after hiring a new coach.

The results—and in both matches the team that won had the better of the chances as well as the goals—suggest the longer-duration bet was the right one. It is easy to get out over your skis during a tournament and believe that just a few matches tell you more about what a team can do than its previous years of performance.

These are, I think, classic examples of the insights that you can get from model-building. Sometimes you discover something strange and exciting. But usually what you find is far more prosaic. Home field advantage matters a lot in international football and it’s a good idea to put a lot of effort into calculating it precisely. You should not be too sure that a few matches at a tournament give you a better idea of how good a team is than its demonstrated level of play over previous years. International teams with consistent results in high level competition, even if they don’t have squads stacked with the most famous players in the world, should be good bets to succeed at the World Cup.

PADDLIN’ and PELE came in at the top of this contest based on a shared set of bets that no one could ever call counterintuitive or groundbreaking. But by building models and testing them against past performance, we had good reason to bet than these particular boring theories about international football were the right ones to depend upon.

What’s Next for Expecting Goals?

The World Cup was a whirlwind for me. I built the model in a month, published it the morning of the opener, and blogged my way through five weeks of matches. I came out exhausted, but also confident that I wanted to keep doing this. I think the World Cup showed a way to balance the larger studies this newsletter was originally created to publish, with blogs that put an analytics spin on the latest news in the soccer world.

The first big project I plan to tackle is a club version of PADDLIN’. Much of the model’s design came from studies of league data, and rather than saving everything for one giant write-up, I’ll be looking to publish the findings in progress as they are relevant, just as Uruguay’s draw with Cabo Verde gave me an opportunity to write up adjusted xG. The new PADDLIN’ won’t be ready when the Premier League kicks off August 21, but it will be underway.

I also came out of the Messi–Ronaldo series with a deeper historical player database, and I’ll be putting it to work on player analysis. This will start with an Elliot Anderson piece on Monday. If there’s a player you want covered, let me know. Thank you for subscribing. I’m looking forward to this new era of the newsletter.

As we head into the World Cup final and what could be Messi’s last match on the World Cup stage, it is a good time to continue the work of quantifying and then marveling at his greatness. I knew more or less what I would find when I began this project, and it is still incredible to see laid out specifically.

But this newsletter has a secondary purpose as well. By looking at the top-15 peak and career lists across a variety of statistics, it is possible to begin pulling out some of the other greats of the modern era and comparing or categorizing them. I see no particularly reasonable arguments for anyone but Messi at the top, it is just as clear that Ronaldo stands in second. Below that the discussion gets much more complex, but I see in the data particular cases for Luis Suárez, Mohamed Salah, and Eden Hazard, as well as the more obvious ones for Robert Lewandowski and Kylian Mbappé. And as we go through the different statistics, players with more specific cases as top passers, carriers and receivers also stand out.

I want to pull one chart out here, along with a question asked in the Double Pivot podcast Discord.

Mbappé is fifth here in non-penalty goals and assists in his five best seasons, across top-flight league and Champions League play. His peak already stands among the greatest of the modern era. And Messi scored or assisted about 23 more goals per season in his best seasons.

Who are some players, it was asked, who are about as far behind Mbappé in their peak seasons as Mbappé is behind Messi? These would be players who put up about 100 non-penalty goals and assists in their peak seasons.

Some notable examples: Iago Aspas (104), Lorenzo Insigne (104), Jamie Vardy (103), Philippe Coutinho (98), Papu Gómez (97), Mario Götze (95). All excellent players, of course, but obviously none of them comparable to Mbappé in their attacking production. And that’s the size of the gap between Mbappé and Messi.

The ball progression statistics only serve to further highlight this point. Messi is not the single greatest player of the last seventeen seasons by every one of these metrics individually—although he is for many of them—but in the aggregate no one else shows up as high, as often for either peak or career production, and it is not close. And these are all statistics at least one step removed from scoring or assisting goals.

These are the core statistics that will be covered, across passing, ball-carrying, combined progression, and ball-receiving.

• Penalty area entries by pass or carry should be self-explanatory. The one thing worth noting here is that the standard definition of a “carry” requires the player to move the ball at least five yards, and so a player who simply happens to receive the ball just outside the penalty area and move it one step in will be excluded, but that is very rare.

• For progressive actions, either pass or carry, I will use John Muller’s definition of actions that move the ball at least one-quarter of the distance to goal from when they began. For progressive passes, I am excluding all passes played into a contested aerial, and in all cases crosses are excluded. I will have more to say about these definitions in the passing section.

• Expected threat added by pass or carry uses the definition established by Karun Singh, measuring the likelihood of a goal being scored from any position on the pitch and using the chaining of events to calculate a threat value from every other location on the pitch. Players receive value for moving the ball from a location with a lower xT value to a location with a higher xT value. I have excluded negative xT events, as a player retreating (or passing backwards) should not be penalized for that action more than a player who fails to move the ball at all.

Finally, before we get to the charts, I should repeat a point from the last newsletter about how to think about “peak” greatness:

For peak studies, I am using a similar process to the study of peak in baseball. Jay Jaffe’s JAWS system provides a first measure of greatness by combining a player’s career totals with their totals over their seven best seasons. There is of course nothing magical about the number seven, and Bill James often looked at three, five, and seven year peaks. I found in going over these numbers that all three selections told similar stories, and I preferred the five-year peak because soccer careers tend to be shorter and it gave players who may have a few seasons outside the data coverage window more room to make the lists.

I am looking at numbers by season here because the season is the proper focus for analysis at the team or player level. Trophies are handed out based on team performance over a season, and players consistently see their jobs as contributing toward these team achievements. A player’s peak should be the seasons in which they did the most to improve their team and move them closer to winning these trophies.

Peak and Career Ball-Carrying

I have used four statistics for ball-carrying, three of which are found above, and the fourth is a straightforward single number.

• Successful take-ons are the one statistic with no comparison among passes. These are actions where a player matches up one-on-one with a defender and gets past them toward goal.

• Messi is just absurd here.

• Eden Hazard and Neymar clearly appear well beyond all other competitors as #2 and #3 in peak and career take-ons. It is possible a fuller database that captured Franck Ribéry’s early seasons at Bayern and Marseille would add him to that group.

• A few pure specialists appear on this chart who will not show up anywhere else: Allan Saint-Maximin and Adama Traoré among the Premier League specialists, as well as Raffael from the Bundesliga.

• Roberto Firmino, clearly underrated as a one-on-one specialist.

• It should be noted that the definition of the take-on appears to have been tightened in recent seasons, as total successful take-ons across all leagues have dropped notably. This is an issue that fortunately will not arise with any of the other stats here but is worth noting. Jeremy Doku, under the older definition, might well be near the top of the leaderboard.

• There may never have been a penalty area carry specialist like Vinicius Junior. He is still only 26 years old and he is already at the top of the career chart. His peak dominance is something that only Messi achieves on any of the other charts.

• This appears to be an overlapping skill with take-ons, which makes sense. Both in that it will often require beating a defender to get into the penalty area, and because defenders may be more likely to sit off and allow entry to the penalty area to a player they are scared might blow by them if they square up. Wilfried Zaha notably shows up here, despite playing mostly for much less ball-dominant teams, although he will not again. Gervinho too.

• As we step back a little bit further from the penalty area, Messi and Ronaldo shoot up the charts. Both did a lot of long carrying for their teams to move the ball from midfield into the attacking area, which appears here under progressive carries.

• Vini Jr seems likely to surpass them over his career if he can avoid injury.

• Arjen Robben joins Franck Ribéry here from the great Bayern sides of the 2010s. I have always been slightly annoyed by the “all Robben does is cut inside and shoot” meme. Robben in fact was a highly varied and effective ball-carrier, who drove to the endline more or less as often as he cut inside, and his appearance here helps demonstrate that fact.

• Jesús Navas and Papu Gómez, two otherwise extremely dissimilar players, appear here. Navas was an elite long-carrier from more defensive wide positions, while Gómez had a free, inside-forward style role for Atalanta where he could receive the ball all over the pitch to drive forward or play around the penalty area to create shots.

• It is interesting to think about the extent of Messi’s advantage here and what it tells us about expected threat in comparison to the other statistics. What xT does most notably is focus on actions very close to the penalty area and give them outsized rewards. There is a “goals funnel” on the pitch according to xT, and being able to move the ball into the central zone near the goal will provide more xT than half a dozen penalty area carries. These specific actions, where Messi managed to weave through penalty area traffic for a clear shot on goal or a pass to a teammate for a tap-in, make up an outsized number of Messi highlight-reel clips. No one comes remotely close to Messi in these actions, and that comes through in expected threat.

• I am confident that with full career coverage, Ronaldo would establish himself as #2 here in peak as well as and career.

• Raheem Sterling has probably become somewhat underrated over time as his early decline has masked the memory of his peak. At his best he appears here as comparable to Robben and Ribéry, Mbappé and Salah, in ball carrying by xT.

Creative Passing: Different Systems and Different Results, and Always Messi

Obviously one way to measure creative passing is simply in terms of assisting shots, as laid out in the previous newsletter. But I can add to that passing into the penalty area, progressive passing, and xT added by pass. These highlight notably different sets of players including a large array of names that haven’t appeared previously on these charts, perhaps they highlight some inadequacies of each stat as a single metric, and also they highlight Lionel Messi.

Read more

1. Can I reasonably argue that Messi is the Greatest Of All Time with a database that only tracks goals and assists back to 2002 and only tracks the other underlying statistics back to 2009?

2. In comparing Messi to other modern players in underlying statistics such as progressive passes and expected threat (xT), was I only making a career value argument? Was Messi simply the best in these numbers because he was so durable, or was he the greatest season to season as well?

This newsletter will be organized around answering those questions.

First, it is true that statistical analysis of football faces terrible data availability problems. The advanced on-ball statistics are not available comprehensively before the mid-2000s. Assists were not consistently tracked until shockingly recently. The only reason we have assist totals for the Premier League before 1999 is because some heroes at Opta went back to watch tape of every single Premier League goal and assigned the assists after the fact. It is not even trivial to assemble databases of goals scored in top league competitions before 2000. What I was able to do here was to expand my databases by one good step beyond the previous newsletter. I now have assists back through the 1999–2000 season, and goals through the 1987–88 season. This data obviously cannot answer a number of questions: it cannot treat the game before the late 1980s, and it still only allows me to dig so deep before 2009. But the results were telling nonetheless.

In the data now extending back 15 years before Ronaldo’s top league debut, I find no one with remotely comparable goal-scoring to Messi or Ronaldo. When that is combined with the findings on Messi’s dominance in most on-ball creation statistics, it makes it very hard to believe anyone from either era came close.

As I said in the last newsletter, going further back into the history of the game faces not only severe data availability problems but also runs into fundamental problems of differences in tactics, and more than that, differences in quality of competition. The concentration of top talent into the biggest leagues, and thus the ramping up of opponent skill levels for the greatest players, took off with the globalization of football as a business in the 1980s into the 1990s, and was not fully achieved until the 2000s. Even if somehow a database of statistics appeared for the earlier era of Pelé and Beckenbauer, contextualizing those numbers properly for comparison would verge on impossible. Any statistically oriented discussion of the GOAT in football must carry a disclaimer that era comparison beyond the relatively recent past has these two fundamental and probably unsolvable confounders.2

There are questions about “all time” that may be unanswerable. But Messi’s (and indeed Ronaldo’s) greatness appears all the clearer when you move the window of analysis back into the 1980s.

Updated Goals and Assists Charts: Messi and Ronaldo Still Far Ahead

The most striking thing I found is that not only did no goal-scorers in their own era compare to Messi and Ronaldo, the same was true for the generation of stars that preceded them.

• As noted in the previous newsletter, Ronaldo’s lead here is a function of penalty goals, and when penalties are removed (or credited relative to expected finishing), Messi also passes him in goals.

• Among the new additions here are Samuel Eto’o, Raúl, Thierry Henry and Alan Shearer. All of them put up goal-scoring numbers in their era competitive with the greats of the 2010s, but none came close to Messi and Ronaldo.

• This chart highlights what has made Lewandowski different. Through his early 30s, he was something like first among equals. But he remained an elite goal-scorer far longer than any of his rivals other than Messi and Ronaldo and locked up the second place on these charts. It will be interested to see if Harry Kane can have a similar late career run.

• It is easy to see here how most of these lines begin levelling off at age 30, besides Messi, Ronaldo and Lewandowski.

• The early career trajectories of Kylian Mbappé and Erling Haaland are clearly unprecedented in the modern game besides Messi and Ronaldo. It is unlikely for obvious reasons that either could maintain this level of production well into their 30s, but they it is impressive to be at this level in their 20s regardless. (And as I will note in the peak chart, there are still gaps to make up.)

• Adding in assists, with particular added coverage in the Premier League, gets Ryan Giggs’ exceptionally long career onto the list, along with David Beckham and Francesco Totti with another couple seasons of Serie A assists.

• Unlike with goals, a number of players did keep pace with Messi on assists through their 20s. But at different points along the path, Mesut Özil and Kevin De Bruyne began to play fewer minutes, while Cesc Fàbregas picked up a new role playing further from goal.

And of course when these two data sources are combined, that puts Lionel Messi into a clear lead.

• The greats of the 1990s all fit comfortably next to the greats of the 2000s and 2010s, but none of them come close to Messi and Ronaldo. Expanding the scope of coverage throws into further relief what outliers both players were.

• The new additions to this chart are Wayne Rooney and Antoine Griezmann, both great players whose combined goals and assists push them onto the leaderboard where they would have otherwise fallen a bit short.

• Thierry Henry and Samuel Eto’o likewise appear competitive with the greats of the 2010s.

• It’s also clear that Kylian Mbappé and Erling Haaland right now are on paths unprecedented besides Messi and Ronaldo.

These contemporary stars also raise the question of peak value. The way that Messi and Ronaldo both kept putting up numbers through their 30s will be incredibly hard to replicate. But this is not the only measure of greatness. Have any of these players, in their best seasons, reached a similar level?

In this newsletter, I will present the peak goals and assists leaderboards, and then begin to consider some of the more advanced stats on shot creation. Tomorrow’s Part III newsletter will then include the full leaderboards for the various ball progression statistics.

Peak Goals and Assists: It’s the Same Story

For peak studies, I am using a similar process to the study of peak in baseball. Jay Jaffe’s JAWS system provides a first measure of greatness by combining a player’s career totals with their totals over their seven best seasons. There is of course nothing magical about the number seven, and Bill James often looked at three, five, and seven year peaks. I found in going over these numbers that all three selections told similar stories, and I preferred the five-year peak because soccer careers tend to be shorter and it gave players who may have a few seasons outside the data coverage window more room to make the lists.

I am looking at numbers by season here because the season is the proper focus for analysis at the team or player level. Trophies are handed out based on team performance over a season, and players consistently see their jobs as contributing toward these team achievements. A player’s peak should be the seasons in which they did the most to improve their team and move them closer to winning these trophies.

With these charts, I was also able to separate penalty and non-penalty goals for almost every season under consideration, and so I can present the peak and career lists with penalties both included and not included.

As I noted in the first newsletter, one of the hardest problems in soccer analytics is accounting for context. How did a player’s role and teammates affect their ability to produce? Strikingly, context becomes much less of a problem in comparing Ronaldo and Messi, who played for the best teams in the world, surrounded by the best teammates, in systems designed to funnel the ball into the situations which best suited them. But as we move down these charts, and especially as we look more deeply at the on-ball creativity statistics, the question of context must arise again.

In this newsletter I will not be accounting for this problem specifically. It is one of the great questions in soccer analytics, with limited treatment in the public sphere. I hope to be able to return to this data in the next year with a context-adjustment system, but that will be a whole series of newsletters to produce.

Note the marks on these charts for data availability.

• * for players whose careers were still active at the end of the 2025–26 season

• † for players who have 1–4 seasons at age 22 or older for which data was unavailable.

• ‡ for players who have 5 or more seasons at age 22 or older for which data was unavailable.

Read more

In the knockouts, both teams have underperformed PADDLIN’ model projections overall, and in just about every match. A red-card adjustment is necessary here as otherwise England’s performance against Mexico looks worse than it was given Mexico’s man advantage, and the reverse is true for Argentina’s win over Switzerland.

On the other side of the bracket, the story is different. France have been clearly the best team at this World Cup match to match, with Spain comfortably second-best.

It is not surprising, then, that PADDLIN’ rates the winner of France-Spain as significantly more likely to win the tournament than the winner of the other semifinal, by a margin of 56 percent to 44 percent.

But beyond that finding, PADDLIN’ produces different results from some of the other notable projections. The gambling markets have gone all in on France as the most likely winner. The Silver Bulletin PELE model is closer to the gambling markets than PADDLIN’ is, but clearly in between the two on most issues. (PELE likes Argentina even somewhat more than PADDLIN’ however.)

This is probably the question I have heard more than any other on social media. Why Spain? Why not France? Why Argentina?

And as I have watched the games, my own opinion has somewhat diverged from the model, too. In the two-part World Cup semifinals preview on the Double Pivot podcast, we agreed that there were good reasons to favor France, at least at the margins.

And I was even more bullish on England’s tactical edge against Argentina in the preview podcast for that match.

So when I find my own model’s results relatively suspect, obviously it is time to write about it.

The purpose of this newsletter is not to justify the PADDLIN’ results or to reject them. It is certainly not to advocate for any particular gambling strategy.1 I want to explain how the model works, and I put together these little studies in great part to understand it better myself. I know all of the various individual modeling decisions that went into the model—and you can read about them in the Introducing PADDLIN’ article—but in practice how the pieces fit together and what they combine to build is always a little bit new even to the modeler.

I want to know why my own eyes—as well as my own reading of the data in less structured ways—are leading me away from the model and what goes into its determinations and why they are different.

What I found is that to grasp the differences, I have to look deeper than the more obvious modeling decisions that made PADDLIN’ stand out. Rather it was a series of broadly defensible modeling choices adding up, over time, to results I do not entirely agree with. And it makes me think about new ways to continue to refine these models to produce slightly more intuitive results.2

The Obvious Explanations: xG and Player Value

The biggest decisions that set PADDLIN’ apart from most models are the incorporation of advanced statistics, including xG, where available to evaluate international teams, and the inclusion of Transfermarkt data to supplement those evaluations with player value estimates. It would have made sense if those distinct components were the primary drivers of these findings, but they are not. In fact, the addition of xG and Transfermarkt value significantly boosts the European teams in relation to Argentina. The gaps between Spain, France and England appear to be primarily driven by the results lens in the model.

Both Spain and France gained a bit more from the addition of xG and Transfermarkt data than England did (plus-86 and plus-85 for Spain and France, compared to plus-58 and plus-24 for England and Argentina. The larger effect here is that Spain’s and France’s performances at the tournament itself have pushed them further ahead while France have gained somewhat on the model’s favorites.

It is not a case where expected goals provide a massively different signal, or where the Transfermarkt values shifted anything too notably. So to find the gaps we need to look more granularly at exactly which results mattered, and how. With Spain and France, those differences are somewhat harder to explain and remain more equivocal. With Argentina and England, I can see what the model is doing and I can state my disagreements more clearly.

Spain and France: Understanding Elo, Player Value Choices

Read more

Obviously winning trophies is a terrible metric for comparing football players, as the sport is fundamentally a team competition given the enormous size of the pitch. One player can only be involved in so many plays, and no one can have the outsized impact on winning of a basketball superstar, or even probably an NFL quarterback. But there is no escaping the question of the GOAT.

If you work in analytics, eventually you will be asked about Lionel Messi and Cristiano Ronaldo. Who do you think is better? Who is the greatest of all time? And people will often be surprised that analysts don’t hem and haw, don’t talk about all the complex, dynamic, still-unanswered questions about how the game works, but instead respond quickly, “oh, it’s Messi.”

In this newsletter I plan to explain why there is an effective consensus within the analytics community on Messi–Ronaldo, but in some ways I think the question of why a consensus exists is the more important one.1

In the very first Expecting Goals newsletter, I offered my own definition of “sports analytics” as distinct from simply talking about statistics. It has three parts: valuation, the attempt to quantify a player’s contribution to winning; context, the attempt to identify the various ways a player’s teammates, coach, league, home field, and so many other factors impact measurements of any one player’s contributions; and randomness, the consideration of how the inherent variance of people and of complex, contested physical actions may show up in the statistical record. In soccer analytics, the latter two confounders are so richly varied, and so baked into the reality of the game, that analysts often struggle to make determinative statements on anything. But in an oddly precise way, all of these problems are minimized in the Messi-Ronaldo debate.

This is Part I of a three-part series. Parts II and III serve to respond to some questions about this newsletter as well as to extend the analysis across a larger set of players.

Messi-Ronaldo Part II: A Larger Database, Thoughts on “All-Time” and Peak Value

Messi-Ronaldo Part III: Ball Progression Peak and Career, and Remembering Some Guys

Messi and Ronaldo: The Reverse Image of Everything Difficult about Soccer Analytics

The first and most obvious point here is randomness. Messi and Ronaldo played long careers and maintained near-perfect fitness, so that the samples of their production are just about unmatched in the data era. Many people would have said in 2022 that it was a major achievement that both were still starters for their national teams at 35 and 37, respectively, at the World Cup in Qatar. (The age curve in football is peculiarly vicious: in a sport demanding extraordinary fitness, outfield players are typically in decline at age 30.) Four years later, Ronaldo has started every match for Portugal while Messi missed the starting XI only in Argentina’s dead rubber against Jordan.

And through their careers, neither faced serious injury issues until they were well into their 30s. Further, the tracking of in-depth event data in soccer began in earnest in the 2000s, just as both were kicking off their careers. For these reasons, the statistical sample on both is unusually robust. The two players in my database with the most shots attempted in domestic league and European competition are, by far, Cristiano Ronaldo and Lionel Messi. Ronaldo has about 3,300 shot attempts since 2009–10 in my database, and Messi about 2,850, with Robert Lewandowski the only player over 2,000 behind them. The sample allows analysts to make confident claims about these two players which responsible commentators would not on the basis of just a few seasons of data.

Their longevity also highlights another problem that bedevils greatness debates across sports analytics but does not apply here: how to weight peak against career value. This is a concept developed in baseball analytics. As one example, two of the greats of mid-century baseball were Mickey Mantle and Hank Aaron. Mantle was clearly, to an analytical lens, better than Aaron in his best seasons. But Aaron was impossibly durable, eventually breaking Babe Ruth’s career home run record while Mantle’s career was first slowed and then derailed by injury. Who was better? It is not a question with any simple answer. And fortunately for us, it does not apply here. At their best, Messi and Ronaldo were clearly the greatest goal-scorers of their era, and both maintained that level of performance their entire careers.

On the question of valuation, once again the Messi and Ronaldo question sidesteps the largest problem in the sport. While the tracking of on-ball statistics has allowed analytics to move back from goals and assists to better quantify how players contribute to chance creation and ball progression up the pitch, as well as to ball retention within possession systems, the on-ball data offers much more limited insight into defensive contributions. Defensive “actions” such as tackles and interceptions are tracked, but they are contextually variable and, more importantly, only a subset of the actual defensive contributions that individual players can make. With the advent of statistics based on tracking data, more progress is being made toward understanding defense, but that work is still in process.2 Fortunately, once again, this is of only minimal relevance to Messi and Ronaldo. Both players had few defensive responsibilities and were the on-ball focus of their team’s attacks, and both made their obvious and most-famous contributions in goals, assists, and other actions around the penalty area which are best tracked by the on-ball data.

These issues of defensive responsibility and tactical role then raises the third point, that of context. Messi and Ronaldo both played as the focal point attacker in systems built, obviously reasonably, around their unmatched abilities. But more to the point, they played almost all of their peak seasons in the same league, Spain’s La Liga. And their clubs, Barcelona and Real Madrid, were probably the top two clubs in the world over that period. At their peaks, both were surrounded by teammates rated among the best in the world.

To quantify this, I collected the data from Transfermarkt on team and league estimated player value, and unsurprisingly it is Barça and Real at the top for virtually the entire period the two shared.3

Further, each of these teams worked to funnel their scoring chances to their star players, a clear shared tactical context. Despite being surrounded by the best teammates that money could buy at the time, Messi and Ronaldo each accounted for about one-third of their team’s shots, expected goals, and non-penalty goals.

This chart also demonstrates a side point—that Messi was one of the greatest finishers of the era, relative to expected goals, while Ronaldo converted his chances at an above average but not elite rate. My data here accords with Marek Kwiatkowski’s more rigorous study of finishing.

In my writing for this website, I have primarily focused on these hard problems of analytics, seeking a deeper understanding of team and tactical contexts, of substitution effects and the age curve. But with this topic, most of those difficult issues can be set aside. We can get a clear sense of what Lionel Messi and Cristiano Ronaldo brought to their clubs through their on-ball statistical profiles, and we do not need to present the findings with excessive modesty because so many of the confounding issues do not apply to this particular problem.

It’s fun. We can just start with goals.

Goals: Before you stand two GOATs

Between domestic European and Champions League play since 2000, no one has scored more goals than Cristiano Ronaldo and Lionel Messi. Ronaldo has scored 639 and Messi 625.4

Robert Lewandowski, who this summer has decided on a move to Major League Soccer’s Chicago Fire, is the only player within even 100 goals of the two. And Lewandowski was a pure striker his entire career, unlike Ronaldo who played wing and wide forward for many years, and Messi who played a variety of more on-ball roles from wing to 10 to a false 9 when he played nominal center forward.

Even now, as a number of new greats climb the list, what is most striking is how far they still have to go. It is noteworthy that Kylian Mbappé and Erling Haaland are keeping pace with Messi and Ronaldo at ages 27 and 25 respectively. But the chart shows that Mbappé and Haaland benefited from earlier starts as elite goal-scorers, and in their peak seasons now neither has been scoring goals at a fully competitive rate. To keep pace with Messi and Ronaldo, they would need to begin scoring more goals now and keep it up well into their 30s. It is unlikely.

Now, raw goals can be misleading. In particular, these numbers have included penalty kicks, a peculiar set of goals which are scored primarily according to the team’s choice of penalty taker rather than due to any actions taken with the ball in play. Ronaldo scored 34 more penalty goals than Messi, a total larger than the gap between them. Messi actually scored more non-penalty goals than Ronaldo over his domestic and Champions League career in Europe.

As this newsletter is going to be highlighting any number of measures where Messi shows up as the greatest of all time, it is worthwhile to start with one where he is decidedly not. Messi has been a merely average penalty kick taker over his career, scoring 79 percent while Ronaldo scored 85 percent, compared to a 77 percent global average. One way to account for a more successful penalty kick taker is to credit players with penalties scored above average, which would net Ronaldo about 11 goals. Crediting penalties in this way would place Messi ahead of Ronaldo by a handful of goals.

And this is the fundamental case for Messi over Ronaldo. The goals are about even. Everything else, including chance creation and ball progression, puts Messi far ahead.

Everything Else: Messi leads in chance creation and ball progression

What are the various ways to measure player production apart from goals? The most obvious is assists, passes that assist goals. But even over relatively long periods of time, assists can be misleading as they depend on other players finishing chances. For this reason we can also look at expected goals assisted, the xG of shots assisted, as well as total shot assists, to confirm the results.

To measure ball progression beyond just shots, there are a number of statistics that capture some of this. We can look for passes and carries into the penalty area, recognizing that defenses seek to prevent players from reaching the 18-yard box. Take-ons measure when a player gets past their defender one-on-one. Progressive passes and carries capture actions where a player moves the ball, by a pass or carry, at least 25 percent of the way to the goal. And there is expected threat, a measure of a player’s successful ball movements by pass and carry that take possession into areas where goals are more likely to follow.

And across nearly every one of these stats, Messi produced far more ball progression and chance creation for his teams than Ronaldo.

Note that for these numbers, we cannot go back as far into the past as with goals, but with a starting point in 2009–10, still the large majority of their shared careers are covered.

I included progressive passes received here to help break down the two players’ responsibilities. Ronaldo played as more of a ball-carrying outlet winger than Messi, and so more often when a pass was played forward out of defense or midfield to the attack, it went to Ronaldo.

If we look at the overall statistics that both players put up, once again Messi’s dominance across categories of chance creation and ball progression is clear.

Of course, Messi’s role involved more passing, while Ronaldo’s was much more concentrated in ball carrying. For chance creation, which necessarily involves passing under these definitions, that simply shows Messi’s greater contributions. For ball progression, the distinct shapes of Messi’s and Ronaldo’s production become clear when these statistics are split into carry and pass numbers. Ronaldo was an elite ball-carrier, one of the greatest of all time. He did provide penalty area entries by carry than Messi did, although he falls a little behind on progressive carries and xT added by carry. Most of Messi’s advantage comes from his massively superior passing metrics. And Messi also shows a large advantage in take-ons: while Ronaldo may have moved the ball further forward at his feet, Messi much more often unsettled a defense by getting past his defender in the midst of a successful carry.

This is why analysts find it so easy to argue for Messi over Ronaldo. They played similar focal-point attacking roles for their teams, and they scored similar numbers of goals. But Messi assisted about 60 percent more goals for his teammates, moved the ball into the penalty area twice as often, made progressive actions 60 percent more often, and created 80 percent more expected threat. While there is no agreed-upon method for adding together these statistics into one unified value metric, it is hard to imagine any method that would not credit Messi with far more value to his teams.

It would be one thing if Messi held this advantage in chance creation and ball progression over Ronaldo, his closest competitor in goals scored. But when we expand the scope to compare Messi to all other players of the same era rather than merely to Ronaldo, these statistics suggest Messi was also probably the greatest chance creator and ball progressor of his era. He does not merely tie Ronaldo in goals and surpass him everywhere else. Messi roughly ties Ronaldo in goals while beating everyone else, and also surpasses everyone else in shot creation and ball progression metrics.

Everthing Else: It gets a little bit absurd at this point

The one statistic that, like goals, can be isolated further back to the 2000s is assists. Messi’s lead in assists against all competitors, over this period, is similar to his and Ronaldo’s advantage in goal-scoring.

This also serves as another simple way of establishing Ronaldo as the clear #2 to Messi’s #1. While Ronaldo stands far behind Messi in goals assisted over their careers, he and Luis Suárez are the only other players to grace both charts.

When you zoom in to the other, more detailed metrics, incredibly, Messi is the career leader in almost every single one.

I particularly enjoy Toni Kroos showing up as the player with the most progressive passes in this period. While Kroos did more than Messi on a per-90 level, it captures something of Messi’s greatness that the player whose core job for most of his career was to cover ball progression responsibilities on Ronaldo’s teams, ended up only slightly ahead of Messi on progressive passes.

This chart also helps narrow the focus on why Messi was the greatest ball progressor of the era. It is not necessarily that he was the best passer. Cases can be made for players like Kroos or Kevin De Bruyne. It is not necessarily that he was the best ball-carrier, as either Vinícius Júnior and Ronaldo could stake a claim. But as a player who was equally dangerous moving the ball forward by pass or carry, he clearly had no equal. Eden Hazard, at his peak, was perhaps the only player who compares, but Messi was able to maintain his top level for far longer.

The Greatest

In many ways, what this analysis shows is that the GOAT debate can serve to obscure the greatness of Cristiano Ronaldo. It is easy to imagine how this article would read, in a world without Messi. How incredible is it, we would write, that the best goal-scorer of this era, a player so durable and with such a long peak, was also among the best chance creators and ball carriers of the era as well. No one else comes close to the top of all three component charts.

But as is so common in these sorts of discussions, even trying to narrow the lens on Cristiano only serves to throw Messi’s greatness into starker contrast.

Lionel Messi was not the best player of his era. He was the three best players of his era.

He was roughly tied with Cristiano Ronaldo as the greatest goal-scorer. He stands alone as the greatest chance creator, with Ángel Di María, Neymar and De Bruyne his only competitors at his peak level. And he was the greatest ball progressor, with only perhaps Eden Hazard in the same class in his best seasons.

While reaching further back into time to compare Messi to people who played in earlier eras is necessarily speculative, it is difficult to imagine that any other players dominated the game this thoroughly for this long. Certainly none of them did it against a sustained and consistently elite level of competition, as such a concentration of talent only occurred in the modern era.

Lionel Messi is the greatest of all time and, despite how uncertain so much of our statistical knowledge of soccer remains, the data really does demonstrate that.

This is Part I of a three-part series. Parts II and III serve to respond to some questions about this newsletter as well as to extend the analysis across a larger set of players.

Messi-Ronaldo Part II: A Larger Database, Thoughts on “All-Time” and Peak Value

Messi-Ronaldo Part III: Ball Progression Peak and Career, and Remembering Some Guys

One question is whether the furor around the White House’s intervention in Folarin Balogun’s red card suspension created a more difficult atmosphere for the players. If you want to hear my opinions on and my breakdown of the Balogun situation, I was interviewed on KCRW, a Los Angeles public radio station, by Madeleine Brand before the match. But the question of whether the USSF, FIFA, or the White House were justified in their actions, or followed any kind of fair process, is a separate one from how the larger context affected the game on the field.

And that is an impossible question for analytics to answer. Notably, manager Mauricio Pochettino denied at his press conference that the political situation made any real difference, and he also emphasized that he and his players were responsible for their performance regardless. When you get into questions of human psychology and group dynamics, in a situation none of the players have or likely will experience again, it becomes impossible to test any particular theory. And that is before getting into a discussion of Belgium, who came out with a clear tactical plan and executed at a high level. The number of unknowable potential causal factors should inspire humility among commentators. We will probably never know the answer.

But there are certain baselines that analytics can offer to the discussion. Did the United States play unusually poorly against Belgium, or was this simply what normally happens when a team at the USMNT’s level faces its first top level European test in a tournament? That is an objective question that the PADDLIN’ model can help answer. If the team’s performance was not below par, then there is little need to ask further questions about context. But if it was significantly below par, then at least there is a football outcome that needs explanation.

Likewise, perhaps the question is not one of the team rating based on their performances but the talent level overall. Is a team with talent like the United States’ expected to fall short against Belgium? Is this loss just another in a long line of USMNT losses built on the same problem, a shallow talent pool? Here the Transfermarkt ratings used in the PADDLIN’ system can be useful. These ratings are not foolproof, but they have been shown to predict team quality and results, and given that the ratings are crowd-sourced from a mostly-European set of evaluators, there is little reason to think Americans would be overrated.

In these studies, I found that the United States’ talent has improved significantly over the last decade, and that the performance against Belgium was surprisingly poor. This does not mean that any particular explanation is justified, but it does show that the match and the performance require explanation.

Belgium 4-1 United States: A Paddlin’ Indeed

The United States came into the round of 16 having won three of four contests with only a defeat to Türkiye in a dead rubber match with a mostly-rotated starting lineup as the exception. In normal competitive situations at this World Cup—excluding matches with nothing to play for and periods of time played at a man advantage or disadvantage, the United States was dominant. And then against Belgium, they were dominated.

A United States team that had allowed opponents few opportunities within close sight of goal conceded three huge chances to Belgium in the first half (the two Charles De Ketelaere goals and Youri Tielemans’ first-minute miss), all within 12 yards of goal. An attack that had created good chances at high rates fell silent, despite chasing a result for most of the game.

Of course, Belgium are the best team the United States have faced. Even after some disappointing performances, the Red Devils were still on the edge of the tournament top 10 in the PADDLIN’ ratings. No one should have expected the United States to run Belgium off the pitch like they did to Paraguay. But this is a question that a projection system is perfectly set up to answer. What were reasonable expectations for the United States coming into these matches, and did the US exceed or fall short of those expectations?

In both cases, the PADDLIN’ projections reinforce the initial observation. The USMNT played at an unusually high level in their opening matches, and then collapsed well below expectations against Belgium.

The Red Devils were clearly the best team the United States faced. But with home field advantage and a competitive team rating, the difference between the two teams should have been only fractions of a goal, not the 4-1 result and 2.5-0.8 adjusted xG estimate of the match itself. Likewise, while the United States was rightly favored in their group matches and against Bosnia, the easy victories in live matches all reflected performances well above expectation.

Winning against Paraguay, Australia, and Bosnia, and then losing to Belgium, would not in itself have required any explanation. But the results can mislead here. It is not merely that the United States won the matches they should have won, but they did so comprehensively and rarely faced any risk of defeat. And likewise, Belgium did not merely proceed to the quarterfinals as expected, but dismantled the United States to such a degree that even a fluke deflected goal could not make the match competitive.

The results may have been to expectation but the underlying performances were not.1

When you account for the extent of Belgium’s superiority, it was a surprising result that opens up further questions for analysis.

If we come at the question from a different angle, looking at estimated talent level rather than the team’s rating and projection, we find similar answers. The USMNT has added significant talent over the last decade and clearly progressed up the charts of international teams, and a purely talent-based model would be just as surprised by this result. This data also puts a different spin on the 2014 team that took Belgium to extra time and the 2018 team which failed to qualify for the following World Cup.

The Improving USMNT Talent Pool

Read more

In the adjusted xG, the result reads like a fluke. Argentina ultimately attempted 22 shots, the large majority of them inside the penalty area and 6 within 12 yards of goal, rated at over 2 xG. None of Cabo Verde’s 16 shots came from such close range, and they ended up with about 0.5 xG.

But this is not going to be a newsletter about how, actually, the dramatic match that everyone loved was a fluke of finishing. To a meaningful degree, of course, it was. Cabo Verde’s extra time equalizer was one of the greatest shots in World Cup history, and everyone watching knew they had seen a truly special finish, one that commentator Stu Holden noted was probably the greatest Sidny Lopes Cabral has ever managed to pull off in his life.

But in analytics terms, it is not exactly news that Cabo Verde needed a bit of magic to push Argentina to the brink. What is far more impressive is that the Tubarões Azuis had made it to that point, and that they were by no means played off the pitch by Argentina. As noted in an earlier Expecting Goals World Cup blog, a deeper analysis of the event data and the adjusted xG methodology show that Cabo Verde played Uruguay close to even and did not require much in the way of good fortune to pull off that draw. And in the final group match against Saudi Arabia, needing a draw to reach the World Cup knockouts for the first time in their nation’s history, Cabo Verde did not sit back and play for the point but took the game to the Saudis. By xG, adjusted xG and field tilt, the match was Cabo Verde’s to win, and for once they could not make one of their many good scoring chances count.

One scoreline can always be a fluke. The PADDLIN’ model is built on that suspicion, and it actually docked Cabo Verde rating points for the famous draw with Spain even as the rest of the world celebrated it. But I found that if I take all the underlying data seriously in the aggregate, even the numbers that call into question most of the praise for Cabo Verde’s opening draw, what appears is a story much larger than a single result. When a team keeps passing every test that international competition provides—at the level of the players, the program, and the tournament—the flukes stop being the story.

This is true first within the arc of the 2026 World Cup, and as I will argue, also over the past several years. In this World Cup, Cabo Verde played an unusually difficult schedule and managed to avoid being completely dominated by two of the best teams in the world and went at least toe-to-toe with their other opponents. By PADDLIN’ expected points relative to schedule difficulty, Cabo Verde’s numbers at this tournament are competitive with Croatia, Belgium and Egypt.

This chart throws into relief the larger reason why Cabo Verde’s run has captured the world’s attention. This is a tiny archipelago nation of about 530,000 people, with a diaspora population, primarily in the United States and Portugal, only slightly larger than that. The Saudi Arabia team they were unfortunate not to dispatch draws its players from a population nearly 70 times larger—still roughly 30 times larger even if you credit Cabo Verde its entire diaspora.

To understand the incredible nature of this run requires consideration of several different points: how much football talent Cabo Verde has produced for a nation of its size and wealth, how how far the team punched above that talent for years, and how this World Cup exceeded even what that record promised. This is a story of player development, as well as recruitment from the diaspora, a story of tactics and player relationships leading to improved play and a testament to the work of team manager Bubista, and finally a story of how it all came together on the biggest stage.

Three Levels of the Cabo Verde Miracle

Read more

It was a controversial call, but FIFA’s head of refereeing Pierluigi Collina explained that this decision was consistent with new guidelines at the World Cup for corner kicks.

“Although keeping a position is not a foul per se, when an attacking player is not interested in the ball and deliberately moves, even marginally, with the clear intention of obstructing opponents’ movement and prevents him from defending, then referees, and VAR when needed, should carefully analyse the incident and intervene,” Collina said.

“This is especially the case when the tactic aims to prevent the opposing goalkeeper from being able to defend the goal.

“Coaches and players were informed, so it should come as no surprise that referees will punish these fouls.”

Such a standard offers a major change from the norms of refereeing in the Premier League, where the standards of acceptable contact with goalkeepers on corner kicks were exploited in radical new ways in the 2025–26 season.

The Meat Wall, Explained

The logic behind the tactic that this blog christened “the meat wall” is easy to follow.1 Scoring headers is difficult, and the further from goal you are the harder it is to get the power or find the angle on your shot to beat a goalkeeper. On corner kicks, headers attempted from inside the six-yard box are scored about 15 percent of the time, while headers from outside the six-yard box are scored about 5 percent of the time. If you could simply play all your corner kicks to the area right in front of goal, you should be able to score more goals. But there is an obvious problem: the goalkeeper. Keepers are expected to “control their area” and rise to claim crosses hit close to goal, and so a strategy of more six-yard box crosses will likely lead to more claims not more goals. The solution, developed most influentially by Nicolas Jover and Mikel Arteta at Arsenal, was to station attacking players near the keeper, who would in turn be marked by defenders, to create a wall of bodies that the keeper could not fight through to claim the cross.

The meat wall worked. During the 2025–26 season, the crosses hit to the six-yard box became a dominant strategy across the league, but the rate of claims by goalkeepers nonetheless decreased slightly. I created a model of keeper claims based on cross location, and this suggested claims were down more than 50 percent compared to expectations.

This analysis has also been confirmed with player tracking data, as David Reed showed on Sky Sports that the number of attackers in the six-yard box on corners has increased by about 70 percent over two seasons.

And of course the goals came too. There were 183 corner kick goals in the 25–26 season, up from an average of about 150 per season in the previous four seasons. And it was not just corners—there was also a massive increase in goals from throw-in set pieces. But while set piece goals exploded, open play goal-scoring actually decreased. Instead of set piece goals adding greater excitement to the game, it seems they crowded out other goals. Teams chose to play a more conservative style, taking fewer risks to create chances because they could count on set pieces.

All of these points are drawn from larger studies from Expecting Goals over the previous season:

• Long Throws and Meat Walls: The Set Piece Revolution in the Premier League

• Who Made the First Meat Wall? The Origins of the Set Piece Revolution

So one of the big questions heading into the World Cup was whether this strategy would transfer from the Premier League to the international game.

Corner Kicks at the World Cup

So far in 2026, through the matches on July 1, the rate of goal-scoring from corner kicks was a mostly unremarkable 0.38 per match, compared to about 0.35 per match over the last three World Cups. Indeed, set piece goal-scoring appears unremarkable overall. At a more granular level, looking at xG because of the smaller samples of matches involved, it appears that corner kicks are providing somewhat more danger than in the past. It is just not an enormous change.

A deeper analysis of corner kick statistics suggests that teams have adapted to the changes to the rulebook implementation, and the insights behind the strategies that changed the game in the 25–26 Premier League may still be seen behind the numbers.

There has been an increase in corner kicks taken to the six-yard box, from 33 percent in the 2022 World Cup to 39 percent here, but that is a far cry from the absurd 59 percent of corner kicks crosses taken to the six-yard box in the Premier League last season.

It makes sense that the new refereeing standards would have such an effect. Attacking players purposely taking up positions near the keeper are essential to the meat wall strategy. If they are at risk of being called for a foul for “even marginally” moving toward the keeper, and if referees have been instructed to pay particular attention “when the tactic aims to prevent the opposing goalkeeper from being able to defend the goal", teams will reasonably decide to back off from the approach. Even Germany’s very mild attempt to obstruct the keeper with one man in his space was identified and punished.

And when you look more closely at corner kick location, the story moves into tighter focus. Teams have learned from modern set piece tactics and are trying to play more corners to more dangerous locations, but the ultra close-range targets the meat wall tactic made possible are no longer available.

In past World Cups, a larger number of crosses were played to very poor areas, well outside of the six-yard box. The blue areas in the 2026 graphic show the corner kick deliveries that have become rarer this cycle. Fewer teams have played crosses to the penalty spot or the wide areas outside the six-yard box in 2026. But teams have not been able to concentrate corner kicks to the most dangerous areas just a few yards from goal, with more corners also being taken to the locations about seven to nine yards out as well as toward the edge of the six-yard box. Whereas in the Premier League in 2025–26, those corners also nearly disappeared in favor of a concentration of crosses hit to just a couple yards out.

I find these results even more compelling than if corner kick delivery had not changed. If the data were just the same as it was eight years ago, it might suggest that international coaches had simply failed to incorporate new insights about set pieces into their preparation. Instead, these charts display a tendency toward optimizing corner kicks, reflecting those new insights, but with the rules functioning effectively as a dampener and preventing the development of the radical strategies dominant in the Premier League.

The new directives on corner kicks seem to be working.

But what about the other new rules? FIFA introduced a set of new rules aimed at speeding up the pace of play and preventing long delays in the action. Several of these are somewhat harder to capture analytically, but the rules aimed of speeding up restarts of play should be easy to study.

Here’s how Dale Johnson at the BBC summarized the rule change:

Countdowns for goal-kicks and throw-ins (five seconds): If a player deliberately delays the restart of play, a goal-kick could become a corner or the throw given to the opposition.

The count will not begin when the ball goes out of play. A referee will choose to activate it if a player is taking too long.

The rule relies on the referee’s discretion. Referees at the World Cup have called for turnovers of possession in several matches, but the real question with the effectiveness of these rules is not whether punishments are handed down. Have there been fewer delays before restarts of play?

And this also raises a question about set pieces. Throw-in set pieces have caused an explosion in delays in the Premier League. Nearly half of all throw-ins taken from attacking areas in the 25–26 season required a delay over 20 seconds. That rate of delay was more than double the rate from previous Premier League seasons. If the “shot clock” for throw-ins is enforced, and if these frankly ridiculous new delays are nipped in the bud, will that also limit the effectiveness of the long throw strategy?

The New Rules Have Crushed Time-Wasting, But Have They Curbed Long Throws?

Read more

There are still so many matches to play. The way powers of two work means that the expansion of the the knockouts by one round has doubled the number of win-or-go-home fixtures from any previous World Cup.

And I have a new round of PADDLIN’ projections for the knockouts. There is a big new knockouts chart that will be updated after every match. I have new tables walking through how each contender can win the World Cup based on their paths through the bracket. And finally there are a new set of “how well have they played?” tables looking at every team’s adjusted xG performance in all their matches to date.

But before I get into the new charts and new tables rollout, the first slowdown day at this World Cup is a great time to catch up on all the soccer analytics content from Expecting Goals.

This newsletter was originally built to be a place for larger studies in soccer analytics. In the World Cup blogs, the concept has been to merge this focus on the study of the game through data with more of-the-moment daily coverage. So each of these blogs covers not just some of the events of the group stage but also larger questions in analytics that the games raised.

• June 27: What Happens When Teams “Should” Play For a Draw?

• June 26: On “Wanting It More” — What Happens When Teams Have Nothing to Play For?

• June 24: Explaining the World Cup Third-Place Matrix

• June 22: Adjusted xG and Big Chance Regression

• June 21: Adjusted xG and Game State Factors

• June 19: (Ok This One Was More Just Fun With Charts In Retrospect)

• June 18: Why Should You “Adjust” xG and How Does PADDLIN’ xG Work?

And of course there is always the Introducing PADDLIN’ explainer for the full World Cup model. On my podcast, the Double Pivot, we also put together a three-part series on what it takes to build a World Cup model and thinking through the whole process of model-building from the broadest question (“what is a ‘model’ anyhow?”) through to the specific PADDLIN’ projections. You may also be interested in our “How to Watch Soccer Like a Nerd” episode going over how we watch and think about the game.

We’ve also covered England being good, the USMNT being good, Ronaldo and Yan Diomande, England not being that good (and Colombia and Brazil), and for subscribers we profiled Elliot Anderson, kept track of the USMNT, and talked about the economics and the politics of the World Cup.

Really the point of today’s newsletter is that I have been exceptionally busy for a few weeks.

The New Tables

The most important new table isn’t a table at all, but the new World Cup bracket. It shows the current active knockout round and the odds in all of its matches, plus the favorites to make it to the next several rounds and the current champion odds.

This graphic will be updated on the primary PADDLIN’ projections page.

Along with this, I have put together “How They Win The World Cup” charts for the 14 favorites (all teams with a projected chance of winning it all of 1 percent or higher). These show for each team their projected chances of winning not just their upcoming match but also their most likely future knockout matches, and how their odds of winning it all would increase with each step and each different opponent.

This is what the model sees as Argentina’s current path:

And this is what it looks like for the United States:

These charts will be updated on two pages, with the teams in alphabetical order, so you would find Brazil on the A-G page and Mexico on the H-Z page:

• How They Win the World Cup: A-G

• How They Win the World Cup: H-Z1

Finally there are the Team Performance tables, which use PADDLIN’ adjusted xG to evaluate team performance. In these tables the teams’ performances are compared to the projection of an average round of 32 team against the same opposition, with the same home field advantage or disadvantage. Dead rubbers are excluded. One table shows adjusted xG difference, and the other shows expected points for every match compared to that round of 32 baseline.

Both tables will be updated on the World Cup Team Performance Index page.

The Detailed Knockout Paths page has been updated as well with all the upcoming rounds.

End of newsletter.

Share Expecting Goals

• Main World Cup Projections Page

• Expecting Goals World Cup Daily(ish) Blogs

• World Cup Fixtures and Results (including Projections and xG)

• How They Win the World Cup: Teams A–G

• How They Win the World Cup: Teams H–Z

• Detailed Knockout Paths, Path Difficulty, and Likely Matches by Venue

• World Cup Risers and Fallers: Who Has Improved Their Odds the Most?

• PADDLIN’ Team Ratings

• Introducing PADDLIN’: Newsletter on Modeling and Methodology from Expecting Goals

• How to Build a World Cup Projection Model: Double Pivot Podcast (YouTube), Three-Part Series on PADDLIN’

PADDLIN’ Team Performance Index

A rating of how well teams have played in this World Cup according to Adjusted xG. Dead rubbers excluded. Numbers for each match are compared to how an average round of 32 team would be projected to do in that match, given the home field advantage situation and opposition. (That’s why the average is negative, the teams are being compared to an above average team for the tournament, one that qualified for the knockouts.)

First, expected points compared to an average round of 32 team’s projected points in that match.

And adjusted xG difference compared to an average round of 32 team’s projected xG difference in that match.

Navigation

• Main World Cup Projections Page

• Expecting Goals World Cup Daily(ish) Blogs

• World Cup Fixtures and Results (including Projections and xG)

• How They Win the World Cup: Teams A–G

• How They Win the World Cup: Teams H–Z

• Detailed Knockout Paths, Path Difficulty, and Likely Matches by Venue

• World Cup Risers and Fallers: Who Has Improved Their Odds the Most?

• PADDLIN’ Team Ratings

• Introducing PADDLIN’: Newsletter on Modeling and Methodology from Expecting Goals

• How to Build a World Cup Projection Model: Double Pivot Podcast (YouTube), Three-Part Series on PADDLIN’

• Main World Cup Projections Page

• How They Win the World Cup: Teams A–G

• Expecting Goals World Cup Daily(ish) Blogs

• World Cup Fixtures and Results (including Projections and xG)

• World Cup Team Performance Index: Who’s Played the Best?

• Detailed Knockout Paths, Path Difficulty, and Likely Matches by Venue

• World Cup Risers and Fallers: Who Has Improved Their Odds the Most?

• PADDLIN’ Team Ratings

• Introducing PADDLIN’: Newsletter on Modeling and Methodology from Expecting Goals

• How to Build a World Cup Projection Model: Double Pivot Podcast (YouTube), Three-Part Series on PADDLIN’

How They Win the World Cup

For each team, their projected odds of reaching each round, which opponents they would be likely to face, their odds of beating those opponents, and how their chances of winning the World Cup would change depending on which opponents they draw.

Mexico

Morocco

Netherlands

Portugal

Senegal

Spain

Switzerland

United States

Navigation

• Main World Cup Projections Page

• How They Win the World Cup: Teams A–G

• Expecting Goals World Cup Daily(ish) Blogs

• World Cup Fixtures and Results (including Projections and xG)

• World Cup Team Performance Index: Who’s Played the Best?

• Detailed Knockout Paths, Path Difficulty, and Likely Matches by Venue

• World Cup Risers and Fallers: Who Has Improved Their Odds the Most?

• PADDLIN’ Team Ratings

• Introducing PADDLIN’: Newsletter on Modeling and Methodology from Expecting Goals

• How to Build a World Cup Projection Model: Double Pivot Podcast (YouTube), Three-Part Series on PADDLIN’

• Main World Cup Projections Page

• How They Win the World Cup: Teams H–Z

• Expecting Goals World Cup Daily(ish) Blogs

• World Cup Fixtures and Results (including Projections and xG)

• World Cup Team Performance Index: Who’s Played the Best?

• Detailed Knockout Paths, Path Difficulty, and Likely Matches by Venue

• World Cup Risers and Fallers: Who Has Improved Their Odds the Most?

• PADDLIN’ Team Ratings

• Introducing PADDLIN’: Newsletter on Modeling and Methodology from Expecting Goals

• How to Build a World Cup Projection Model: Double Pivot Podcast (YouTube), Three-Part Series on PADDLIN’

How They Win the World Cup

For each team, their projected odds of reaching each round, which opponents they would be likely to face, their odds of beating those opponents, and how their chances of winning the World Cup would change depending on which opponents they draw.

Argentina

Belgium

Brazil

Colombia

England

France

Germany

Navigation

• Main World Cup Projections Page

• How They Win the World Cup: Teams H–Z

• Expecting Goals World Cup Daily(ish) Blogs

• World Cup Fixtures and Results (including Projections and xG)

• World Cup Team Performance Index: Who’s Played the Best?

• Detailed Knockout Paths, Path Difficulty, and Likely Matches by Venue

• World Cup Risers and Fallers: Who Has Improved Their Odds the Most?

• PADDLIN’ Team Ratings

• Introducing PADDLIN’: Newsletter on Modeling and Methodology from Expecting Goals

• How to Build a World Cup Projection Model: Double Pivot Podcast (YouTube), Three-Part Series on PADDLIN’

One question this raises,1 is whether the model was wrong to begin with. As the third round of matches continues, and teams gain a clearer sense of what results will give them a better chance of qualification, are these teams motivated to secure specific results which they know will gain them a berth in the knockouts?

Sometimes, such a competitive situation can lead to greater excitement. Senegal, for example, did not stop attacking against 10-man Iraq because they knew every additional goal increased their chances of qualifying for the round of 32. As discussed in yesterday’s newsletter on “dead rubbers” and match projections, Ecuador’s dramatic upset of Germany reflected a situation where Ecuador knew they needed a win while Germany had nothing to play for.

However, in many cases the incentives cut against football excitement and even border on collusion. This is the story of the famous “Disgrace of Gijón”, and the scene is set for such a result in Kansas City. (Obviously should it play out, we should call it “the new Missouri Compromise.”) Algeria and Austria are both on 3 points, and 4 points would secure second place for Austria and for Algeria, a guaranteed place in the round of 32 and one of the best third-place teams. Because Austria has even goal difference and Algeria has minus-2 GD, both teams would be at serious risk of elimination with even a one-goal loss.

A similar but not identical situation arises with one of the earlier matches, Croatia–Ghana, where Ghana have secured a knockout berth on 4 points, while Croatia need a draw to guarantee survival. While Ghana could go all out for a chance to top the group and surpass England, it is also possible the Black Stars will seek to avoid injury and prepare for the next round.

I have wondered, as I’ve watched these matches play out, whether my model has it wrong. Although PADDLIN’ does incorporate handling for matches like Ecuador–Germany where only one team has something to play for, was it a major oversight not to include handling for potential biscotti and quasi-biscotti like Algeria–Austria and Ghana–Croatia?

For this newsletter, I looked back over results from the last 15 years of international tournaments, and my conclusion is: probably.

Seeking Biscotti: The Math of Potentially Compromised Results

Read more

Coming into the match, Germany had clinched first place in Group E and had nothing to play for. Because of the new head-to-head tiebreaker system, it did not matter what the goal difference was in the final group matches: Germany’s victory over Ivory Coast sealed the place for them. Ecuador, by contrast, could only qualify out of the group stage with a victory. The pattern of the game clearly reflected these different priorities.

The second half began 1–1, and typically one would expect the match favorite to exert control and push for a winner. But instead, from the kickoff of the second half through to the Plata winner, it was Ecuador who had the advantage in possession 54 percent to 46 percent and the advantage in field tilt (passes in the final third) 67 to 33 percent. Germany used their periods of possession mostly to slow down the game and Ecuador used theirs to push for a goal, exactly in reverse of what would be expected in a match where both teams had something to play for.

The two other matches in which a team had already clinched their group, Mexico–Czechia and United States–Türkiye, did not tell quite as clear a tactical story. The United States conceded late to lose 3–2, but that Türkiye goal followed a long stretch where the USMNT were playing on the front foot chasing a goal. And Mexico won 3–0 in a match where Czechia created little danger.

In the aggregate, however, these three matches do reflect a more general trend. The United States had by far their worst defensive match of the tournament, conceding three goals on just about 3.0 xG.2 Germany likewise were leaky against an Ecuador side that was projected as a primarily defensive team. Even with Mexico’s solid play, the three teams combined conceded 2.3 more goals and 1.6 more xG than they would be projected to under normal circumstances.

And of course the three teams went a combined 1–0–2, hardly meeting expectations for two tournament hosts and one European powerhouse.

These results were already predicted by the PADDLIN’ model.

Dead Rubbers and “Nothing to Play For” Adjustments

The model obviously did not predict Ecuador to win, but it did drive down Germany’s odds of winning significantly due to the “Nothing to Play For” adjustments. In all of the final round matches with at least one team with nothing to play for, the model shifts the odds of winning slightly to the team with real stakes in the match.

In the development of the PADDLIN’ model, I ran a study to break down tournament and qualifying matches into three separate types:

• normal matches, in which both teams have competitive stakes, whether advancement from a group or qualification to a tournament

• “nothing to play for” mismatch contests, in which one team has competitive stakes and the other does not

• “dead rubbers” in which both teams lack competitive stakes

What I found was, first, that when there is a mismatch in stakes there is typically also a mismatch in results and level of play. Exactly as anyone would expect, and as anyone who watched Ecuador–Germany saw demonstrated on the pitch, it matters if a team has something to play for within a tournament structure.

Further, in dead rubbers, I found a significantly greater tendency toward randomness, as favorites were more likely to draw or lose to underdogs.

These results accorded more or less perfectly with an earlier Expecting Goals study of “Nothing to Play For” matches in league competition. In that study, I found that in matches at the end of the league season, teams that have clinched a title or are guaranteed one of the European places but cannot win the title, and teams that are guaranteed to be relegated, all play worse than they did earlier in the season.3

Notably, this pattern of diminished quality of play also has more specific causes, and these causes are just as intuitive as the broader effects. Teams play worse when they have nothing to play for because they defend more poorly. There are only limited, and harder to identify effects on attacking play. Defending is hard, thankless work, and it does appear to be the case that when soccer players have the option, they will often choose to let their defensive intensity lapse. But attacking is fun, and players mostly do keep attacking.

So when I ran the study on the international data, I knew what to expect and I had a clear hypothesis.4 We should expect to see worse performances, and in particular we should expect to see problems primarily in defense.

It might seem right now that the Germany and United States results were exactly in line with my expectations, but the story does get a little more complicated than that. It turns out that tournaments may have different dynamics from both leagues and qualifying competitions, and this history then raises questions about why matches in the 2026 World Cup have not quite followed that pattern.

Read more

But its exact dynamics became important following the results on June 23. Ghana’s draw against England puts the Black Stars on four points, which is more or less guaranteed to be enough to qualify for the knockouts.

In the late game on June 23, DR Congo lost to Colombia, putting their qualification from the groups at somewhat more risk.

The combination of these two results, within the bizarre matrix math of FIFA’s new third-place system, meaningfully increased the chances that England will draw Senegal in the round of 32.

It will take a while to get back around to this point, but we will get there. To begin with, here are the very broad strokes of the new knockout qualification system.

Eight third-place teams qualify for the knockouts, from twelve groups. But no one knows in advance which groups will see their third-place side qualify. Arithmetically, there are “12 choose 8” different possible ways that the eight qualifying third-place sides can be distributed among the twelve groups, for a total of 495 different combinations.1 For every single one of these 495 different combinations, FIFA created a distinct way those eight teams would be lined up to face eight different group winners.2 That is, four group winners are pre-assigned to face a specific second-place team from another group, just as the knockouts used to work in the 32-team system, and the other eight face the ludicrous math of the third-place matrix.

So if we are trying to figure out the third-place system, four group winners can be set aside immediately: Groups C, F, H, and J. These are the groups of Brazil, the Netherlands, Spain and Argentina. This structure explains why there was so much discussion of the possibility of a round of 32 tie between Spain and Argentina. If one of Spain or Argentina failed to win their group, the other would be guaranteed to face them, while the winners of other groups get a selection from the third-place pool.

Full PADDLIN’ groups projections are here, for reference.

So, finally, how does FIFA’s new process distribute teams into the knockouts in practice? I fed the entire third-place matrix into my projection system and it gave a few clear answers to start. As applied, there are certain groups whose third-place team is more or less guaranteed to face a specific group winner, should they qualify, and others where that third-place team’s opponent will remain unknown until late in the third round.

There are four groups whose third-place team is more or less guaranteed to face a specific group winner: Group B (likely Bosnia), Group L (Ghana or Croatia), Group K (likely DR Congo), and Group A (likely Czechia). The above chart shows that these four teams would line up against England, Colombia, the United States, and Egypt, should the favored teams win those groups. But there are also four groups whose third-place team, should they qualify, still has massive uncertainty about their most likely opposition: Group E (likely Ecuador), Group H (Uruguay, Cabo Verde or Saudi Arabia), Group J (likely Algeria) and Group I (Senegal).

Senegal, who are now more likely to face England than any other group winner, offer the most important story here. The Lions of Teranga had perhaps the toughest opening two matches of any team at this World Cup, losing to France and Norway in succession. But Senegal could still qualify for the group stage with a win over Iraq and a little bit of good fortune in the third-place results elsewhere. According to the PADDLIN’ model, Senegal rates as both a roughly 70 percent favorite to make the knockouts, and as the strongest of the likely third-place sides. Their odds of facing England have increased, but it is still highly uncertain where Senegal end up.

And that is what this newsletter was originally intended to explain.

This Will Be a Different Kind of Expecting Goals Newsletter

When I began this project, I thought I would be able to summarize the basics quickly and give fans a simple overview of their teams’ likely paths. But as I began working through the matrix logic, and in particular drafting paragraph after paragraph following the almost impossibly arcane combinations of results that may deliver Senegal to a round of 32 berth, I realized this was a different kind of newsletter. You will follow me into madness, and we may not make it out alive.

Should you choose to continue reading, however, we will gain a better understanding of what factors will be determinative in drawing the round of 32.3

But there are two issues to clarify first. For one, not all of these teams will qualify for the round of 32, and in some groups there may be another team that qualifies in third. In Group B, for instance, Qatar would finish third with a win, and would be more or less guaranteed a place in the knockouts with that win. And if that happened, Qatar would be guaranteed to face the United States, the winners of Group D. PADDLIN’ makes Bosnia big favorites in that match and so they projected as the most likely third-place qualifier, but that has not been fixed yet.

In this newsletter, for ease of reading, I will often refer to the current favorite for a group’s first or third-place spot as the team which will take that place in the knockouts. There is uncertainty on all these projections, but the points apply whether Canada or Switzerland win the group or whether Bosnia or Qatar finish third.

Further, if Bosnia and Qatar draw, then it is very unlikely either team would qualify for the knockouts. Should that happen, the United States would face a different third-place team in the round of 32. Because no team has yet locked up a third-place finish, that means that every one of the eight group winners in the chart faces a reasonable amount of uncertainty about their knockout opposition.

The winner of Group K (Colombia or Portugal) is now more or less guaranteed a match against the third-place team from Group L (Ghana or Croatia).

Along with Colombia, the winners of Group I are highly likely to face the third-place team from Group F. In Group F, both Japan and Sweden have blowout victories over Tunisia on their records, and so whoever finishes third is massively favored to make the knockouts. This gives France (or possibly Norway) an unusual amount of certainty in their likely opposition. It will be Japan or Sweden, unless something highly unexpected occurs.

That more or less does it for group winners who have a straightforward round of 32 opponent. Everyone else faces the question of whether one or two results in other groups might give them a different round of 32 tie, and could even deliver Senegal to one unlucky side.

For the United States, Canada, Mexico,4 Germany, England, and Belgium or Egypt, there are many different combinations of results that could throw off their round of 32 plans. Likewise, Scotland, Senegal, Algeria and a few other likely third-place finishers, will await the results of other matches to see which group winner they are likely to draw.

For each of these teams, the permutations are complicated but fixed. And so I have collected the data on which results will lead to which opponents for each of these teams, and the likelihood of each outcome according to PADDLIN’.

Third-Place Combinatorics, or, the Esoteric Punishments for England 0–0 Ghana

Coming into the day, England stood a 25 percent chance of drawing Senegal as their round of 32 opponent, and this increased to 36 percent with the conclusion of Colombia–DR Congo.

To explain how this occurred requires answering what seem like basic questions, such as “who will face England in the round of 32?” and “where would Senegal be placed in the knockout bracket?” However, the answers to these questions become complex in ways that stretch the capacity of language to properly and coherently draw out.

Take my hand to begin the descent, if you dare.

United States and England: Linked Fates, Long Tails of Insanity

Read more

But in this case, the headline statistics may actually be hiding something. While Cabo Verde’s draw against Spain looks very much like a classic story of variance, there are real indicators of a competitive team performance in the match against Uruguay.

Reconsidering Uruguay 2–2 Cabo Verde

The headline numbers here initially suggest it was randomness again. Opta’s xG gives Uruguay nearly as much of an advantage in this match (2.3–0.9) as it did for Spain against Cabo Verde (2.1–0.2) in the first round of group matches.

But when you look beyond the topline xG, clear differences emerge. Spain completely strangled the match, dominating not merely possession but also access to the penalty area and final third, and taking many more shots on goal. Uruguay certainly had the advantage in field control, but they were far from equally dominant.

These underlying Uruguay–Cabo Verde statistics seem to reflect a match in which one team had most of the ball and most of the territorial advantage, but not in a way that left the play on the pitch uncompetitive with only a question of whether one of that team’s shots would hit the back of the net.

Especially Cabo Verde within striking distance on shot count (17–12), these numbers begin to look like a match in which an underdog side defended stoutly and counterattacked efficiently. My subjective observation in watching the match accorded more with that story, as I saw relatively few moments when Uruguay appeared one pass away from unlocking the defense and a reasonable number of Cabo Verde counterattacks that offered some early danger before being corralled by a defender.

So what is going on here? If the match felt competitive, why does the xG look so lopsided?

I would not argue the xG is incorrect here. It is always possible for even a well-tuned statistical model to mislead about the quality of a few individual chances, as I have argued was the case in USA–Paraguay, but here that seems unlikely. The three best chances for Uruguay by xG were Maxi Araújo’s goal-line, open net, headed equalizer (rated at a reasonable 0.89), Agustín Canobbio’s free volley over a scrambling keeper from six yards (0.56), and Fede Valverde’s fast break miss from the left channel early in the match (0.25). These chances make up the bulk of Uruguay’s summed xG (1.7 out of 2.3) and they seem properly tuned.

But if you have been following my blogs at this World Cup, this description might sound familiar. The PADDLIN’ model’s Adjusted xG has two components. The first is a game-state adjustment which downweights the value of chances created by a favorite who is chasing the match, as Uruguay were before Araújo equalized. And the second component is a big chance regression.

Read more

Oops.

But even beyond how unexpected it was, the most notable thing about Türkiye’s shock exit is exactly how it happened. Twice Ay-Yıldızlılar1 were favored, twice they conceded an early goal, twice they piled on enormous numbers of mostly low-xG shots and failed to score a single one. The combined barrage was one of the most impressive and also one of the most ineffectual in recent World Cup history.

Between the 0-2 loss to Australia in the opener and the 0-1 defeat to Paraguay in the late night match on Juneteenth, Türkiye attempted 62 shots. The combined barrage was one of the largest in recent World Cup history, and by any account it suggests Türkiye were unfortunate to be eliminated so abruptly. But this barrage also showed exactly the kind of shot profile that football analytics has long viewed skeptically: huge volumes of low-quality attempts from poor shooting positions.

No other team at this World Cup has attempted more than 45. In fact, no other team at any World Cup since 2014 had attempted more than 45 shots in their first two group matches before Türkiye blew past that standard. And it was not just the number of shots that stood out. Again and again Turkish players settled for low-quality chances from outside the penalty area and through traffic. The shot total is inflated by massive numbers of attempts from outside the penalty area and even from over 25 yards.

One thing you can also see in this graphic is the evolution of football tactics over the last two decades. The wisdom of such barrages has been challenged both by tactical experts who sought to pry apart opposition defenses with new attacking plans, and by statistical analysts whose expected goals systems demonstrated the typical futility of long range attempts. Other than Türkiye, the graphic shows almost exclusively teams from the 2014 World Cup at the top of the long-range leaderboard.2

Because of how few good chances Türkiye created, other than a couple late in the match against 10-man Paraguay, the expected goals numbers massively downgrade the rating of these performances. My system estimates the xG total between the two matches at 3.9, for a unimpressive 0.06 xG per shot. The following shot map displays all the chances, with the size of the chance relative to the area of the marker, and goals in pink.3

Yet even 3.9 xG may overstate how impressive these performances were. Readers following the PADDLIN’ charts will have noticed that Türkiye’s posted numbers are much lower. My “Adjusted xG” credits Türkiye with only 2.4 total xG, not 3.9.

Adjusted xG, Game State, and World Cup Comebacks

In my World Cup blog from Thursday, June 18, I talked about the new Adjusted xG system that runs behind PADDLIN’, and how it incorporates big chance and game-state factors in its calculations. Türkiye offers the best example yet of how this system works, and in particular how much difference the game-state adjustment can make.

Read more

• Main World Cup Projections Page

• Expecting Goals World Cup Daily(ish) Blogs

• World Cup Fixtures and Results (including Projections and xG)

• How They Win the World Cup: Teams A–G

• How They Win the World Cup: Teams H–Z

• World Cup Team Performance Index: Who’s Played the Best?

• World Cup Risers and Fallers: Who Has Improved Their Odds the Most?

• PADDLIN’ Team Ratings

• Introducing PADDLIN’: Newsletter on Modeling and Methodology from Expecting Goals

• How to Build a World Cup Projection Model: Double Pivot Podcast (YouTube), Three-Part Series on PADDLIN’

Detailed Paths to the Quarterfinals

Projections for which knockout opponents are most likely for every team with a projected chance of reaching the quarterfinals over 25 percent. Top four most likely individual opponents for each match.

Note that the percentages here are the percentage chance assuming the team makes that round. Canada has only a 29 percent chance of making the quarterfinals, but they are 40 percent to face Argentina in the quarterfinals. That means that 40 percent of the time, if Canada makes the quarters, they will face Argentina.

Path difficulty to the quarterfinals for each team.

Likely Matches by Fixture and Venue

The most likely matches to be played in each scheduled knockout fixture, for the round of 16, quarterfinals and semi-finals.

Quarterfinals matchups still remain heavily a function of the randomness in early knockout matches. The most likely single quarterfinals matchup, Spain vs USA in Inglewood, is still only a 24 percent probability.

Share Expecting Goals

Navigation

• Main World Cup Projections Page

• Expecting Goals World Cup Daily(ish) Blogs

• World Cup Fixtures and Results (including Projections and xG)

• How They Win the World Cup: Teams A–G

• How They Win the World Cup: Teams H–Z

• World Cup Team Performance Index: Who’s Played the Best?

• World Cup Risers and Fallers: Who Has Improved Their Odds the Most?

• PADDLIN’ Team Ratings

• Introducing PADDLIN’: Newsletter on Modeling and Methodology from Expecting Goals

• How to Build a World Cup Projection Model: Double Pivot Podcast (YouTube), Three-Part Series on PADDLIN’