Why Moneyball Will Not Work in Soccer

In the last decade, the philosophy known as ‘Moneyball’ has revolutionised the way baseball teams evaluate players. Based on the field of ‘sabermetrics’, the approach is designed to reveal players’ ‘true’ value by judging them solely on their performance-related statistics. Given its remarkable success in baseball, how long will it be before Moneyball becomes common practice in the Premier League?

The use of sabermetrics was pioneered by baseball teams looking for a way to overcome the vast financial inequalities within their sport. The system seeks to identify the most cost-effective players, and so teams who are under the greatest pressure to use their limited resources efficiently have found the greatest use for it. In theory, this makes Moneyball an ideal antidote to the problem of the financial divide between the haves and have-nots of the Premier League. Yes, a salary cap would even things out once and for all, but how likely is that? As far as we can see, the league’s ‘poorer’ teams are going to have to come up with a solution of their own and this could be it.

Due to the subjective nature of player evaluation, there are huge inefficiencies in the football transfer market which are just waiting to be exploited. At the moment, players are valued in a non-scientific way; without using data, clubs are forced to predict their future output through hunches and guesswork. There are thousands of players out there who have been underestimated by the conventional wisdom, and so their true value is greater than their market value. Theoretically, if a system able to accurately predict players’ productivity was discovered, even a club of meagre means could assemble a squad strong enough to challenge for titles. This is the great attraction of Moneyball.

One Premier League club has already tried to implement a sabermetric approach into their transfer market dealings. John W. Henry, owner of Liverpool since late 2010, has already used the Moneyball philosophy to great effect with the Boston Red Sox. It was unsurprising then, that he quickly announced similar plans for the Anfield club upon his arrival. Liverpool have since installed an American-style hierarchy whereby transfers are controlled, in part, by a Director of Football. Damien Comolli, previously of Tottenham Hotspur and Arsenal, was handed the role and proceeded to sign a draft of new players to rebuild the failing club. He claimed to have used a system inspired by sabermetrics in the acquisition of these players, most notably Stewart Downing.

You don’t have to be the most avid follower of English football to know that Liverpool haven’t enjoyed much success with this strategy. After a dismal season, in which an absurd number of their high-profile summer signings flopped, both Comolli and Kenny Dalglish promptly lost their jobs. Andy Carroll, Jordan Henderson, Charlie Adam and Downing have quite frankly been awful, especially when you consider the Reds splashed out a combined £78 million on them. Based on the example Liverpool have set, it would be surprising if other clubs were lining up to try their hand at this Moneyball game.

Defenders of sabermetrics may say the problem was that Comolli and Dalglish went about using the system in the wrong way, and they would certainly have a point there. The fundamental ethos of Moneyball is to purchase productive players for less than they are really worth. Comolli and Dalglish did the exact opposite of this; they bought overrated players at enormously inflated prices. The fact that they bought primarily native talent, all of whom came with a ‘British Tax’, is indication enough that they didn’t fully appreciate the aim of the game they were playing.

Regardless, Moneyball, however exciting it may be, will not work in football. It works in baseball because statistics play a much bigger role. Anyone who has ever watched a baseball game will testify to having been bombarded with figures for batting averages, on-base percentages, slugging percentages, earned run averages, strikeouts and walks to name just a few. British viewers tend to find the prevalence of statistics in American sports downright bizarre.

Not only this, but the two sports are, by nature, worlds apart. Football is much more fluid; each player’s performance is dependent on the play of others. A striker can’t score unless he is provided with service from supporting players. A goalkeeper can’t keep a clean sheet without the help of his defence in front of him. Baseball, on the other hand, is a more structured game. Each play follows the same basic format and results in players being either credited or debited. This means that statistics can measure individual performances more accurately and each player’s worth can be judged more precisely.

The statistical evidence which supposedly justified Stewart Downing’s £20 million signature is reason enough to disregard the potential role for sabermetrics in football. In his final season at Aston Villa, Downing completed 24% of his crosses, an impressive number. He also made nine assists that season which, compared to teammate Ashley Young’s 11, suggested he would have a reasonably productive first season for Liverpool. However, statistics in football are ultimately misleading and largely irrelevant. While Young had a respectable debut season at Manchester United, Stewart Downing’s stat line read: 36 appearances, zero goals, zero assists. A fitting quote here is one from the Danish former player and manager, Ebbe Skovdahl: “Statistics are just like miniskirts – they give you good ideas but hide the most important things”.

Football, unlike baseball, isn’t about the raw numbers; it’s about how players gel and complement each other. In a baseball team you can, in most cases, take out a player and replace him with a superior one without issue, regardless of the team’s playing style and tactics. The same cannot be said of football.

That’s why we won’t see owners of Premier League clubs turning to statisticians and number-crunchers anytime soon, as is the norm in Major League Baseball. Although most fans would love to see something (anything!) to help football’s underdogs level the playing field, unfortunately it won’t be Moneyball.

19 thoughts on “Why Moneyball Will Not Work in Soccer”

  1. I don’t think the conclusions follow from the reasons you’ve offered. It is no doubt true that baseball lends itself to statistical analysis in a way that football does not, precisely because (a) baseball games are composed of discrete events and (b) these events focus *mostly* on two players. This makes it easier to distinguish the contributions of individual players.

    But it does not follow that sports where (a) and (b) are not true are not amenable to statistical analysis. Plenty of other sports that involve complex interaction between players have been studied fruitfully in this way. Take American Football as an example (I’ll use NFL as a stand-in to keep things straight). Like the NFL, football involves a number of players interacting to yield results, and those interactions are dependent upon the individual performance of each player as well as the schema adopted by the team. A running back’s performance in the NFL depends upon the blockers ahead of him, a lineman’s performance depends on whether he is consistently double teamed, etc.

    Yet, in the NFL, there is a lot of interesting statistical analysis – both by teams themselves and by groups like Football Outsiders. It’s more difficult, but it is entirely possible to tease out the contributions of individual players. Or take fielding statistics in MLB. For a long time these statistics were next to useless. The most commonly used one was “Errors,” which reflects the number of mistakes that a fielder makes which result in bases gained by the opposition. This is a poor statistic because players who have the range to make a play on more balls tend to make more errors than players who never make an attempt on it at all. Modern statistics, however, divide the field up into zones and look at plays in those zones, and even use advanced tools to track the trajectory of the batted ball, the movement of the player, etc. These sorts of advanced techniques could be applied to football to track the specific types of balls that a player is able to play (e.g., what kind of crosses do they put in? under what conditions? from what zones of the field?) which can be related to the players you already have.

    I’m not sure why we should simply throw up our hands in football – the task is difficult, and there are always going to be confounds, but that does not mean it could be a fruitful endeavor.

    Finally, two side notes.

    (1) A single case where statistical analysis fails does not show that statistical analysis will not be successful. As you noted at the outset of your piece (maybe this is your considered view and your remark about Downing was hyperbole), it might simply be that Liverpool did a *poor analysis* of Downing. After all, if we are going to learn anything from statistics, it’s the importance of a significant sample size.

    (2) Statistical analysts in MLB and other sports have *never* recommended analyzing players solely by reference to their statistics. In MLB for example, one still needs to have scouts to get a read on things like physical tools, technique, etc. It’s not a matter of replacement, but rather a matter of supplementation.

  2. I am disappointed this post didn’t make any effort to bring in the NBA, specifically the Dallas Mavericks, in discussing the moneyball concept. The NBA and the flow of its game is more similar to football. The Dallas Mavericks field a very competitive team annually and won the NBA Championship last year using moneyball type tactics.

    1. NBA is salary cap, and each team spends the max or thereabouts. Mav’s even spend more often because of luxury tax funded by Cuban.

      In an even playing field, the principles are less applicable.

      1. The NBA uses a lot of statistical analysis to rate players and how they would fit into your team. Check out the NBA section on ESPN.

  3. Moneyball means spending money in ways other teams are not spending money. Looking for undervalued gems on the market that other people overlooked, because you cannot afford to spend for what other people are looking for.

    It works to a certain degree in soccer- see Udinese and Newcastle.

    Swansea & Norwhich have executed Moneyball principles to great success last year but finding cheap players, even Championship level players, that fit the system and other teams overlooked.

    It just has to be executed well, because it won’t be long before your gold will be the gold that the clubs with money will come calling for.

    Udinese lost Sanchez, Zapata, and Inler, and followed it up with a better season. This year they will lose Isla and Asamoah. Every year must restock the tank.

  4. I am astounded this article can properly describe “moneyball” admit that liverpool didnt follow the principles or spirit at all and then discount the application of the ideas underpinning moneyball based on something unrelated to moneyball principles. Sabermetrics and advanced statistics may not be ready for exploitation but the idea of exploiting inefficiency in the market isnt dependent on statistics. As johnsager pointed out Newcastle managed to unearth demba ba and tiote and other cheap players who are good. They found bargains in African players and maybe thats a segment of the market overlooked. American players may also be a segment of the player pool undervalued for whatever reason. As the article points out this is what moneyball is, unearthing undervalued assets. So i dont understand why that idea wont work in the epl.

  5. Thanks, John Sager for beating me to the point I was going to make about Swansea and Norwich! And Newcastle for that matter. (I kind of think of Napoli in Italy as a club like this as well.)

    The soccer way to “Moneyball” would seem to be less about emphasizing underappreciated statistical measures (i.e. On-Base Average) and more along the lines of what Norwich seems to do – find relatively unknown guys like Holt and Pilkington and the like, plug them into the system, get results, and eventually make money selling them off. Then discover another Pilkington playing for another Atherton Collieries and repeat the process.

    Liverpool seems to have missed the point entirely – try to throw money around like Chelsea or the Macunian clubs (but not nearly enough to really compete) but at the same time get fixated on offbeat statistical measures that in and of themselves aren’t highly indicative of undervalued talent. Worst of both worlds, when you think on it. International football does not have a magic statistic (like on-base average before “Moneyball”) that is significantly underappreciated by professional clubs.

  6. I’ve done work in baseball sabermetrics in the past for a popular baseball sim, so I have a bit of experience here. It is much easier to do this work in baseball because the events that lead to actionable decisions (large sample sizes) are enforced. There is a reason that fielding is the most difficult part of baseball to quantify… because it is the one component that is not enforced. Fielders are not involved on every play. However, as the first comment mentioned, big strides are being made to remedy this situation using a zonal tracking system.

    My guess is that soccer will be solved this way as well. Even though the players are all dependent on each other, there are some reasons to be hopeful. One, there are a large number of possession changes per game, each leading (in theory) in an attempt to score. That gives a pretty good number of events to work from. Two, with minimal substitutions, we also have a decent number of events to evaluate player interactions over multiple offensive or defensive events. Tag each event (pass, tackle, shot, defensive header, etc) spatially and relative to other players on the pitch. Get enough of this information and you can crunch the numbers to find discrete elements that have a probability of leading to a good or bad result.

    The biggest difficulty is also one of the main reasons many of us enjoy soccer. Typically low scoring, each game has a fair amount of variance. Each season has a fair amount of variance for individual players. A few lucky bounces and some players look superb for half a season. A few shots off the post and a player can look pretty poor. Focusing on goals as a successful result is probably a bad idea for this reason.

    If I were working in this field I would come up with a PlayResult concept. Each possession would eventually get a score based on the quality of the termination of the possession. A turnover at midfield would be slightly negative. A turnover in the opposing penalty area would be negative as well, but less so. A shot from the right side, 35 yards out, would be a small positive. A shot from six yards out would be a large positive. All players on the pitch would contribute to the result based on proximity, touches, etc. Then, at the end of the match, you would just sum up the players weighted totals and you would have a Success rating for each player.

    Of course, the labor involved here is tremendous. Without automation there is still some subjective assignment that takes place, even if the various weightings follow a rigorous step by step breakdown that has been tested against thousands of matches. But I imagine automation is coming. Tracking technology has vastly improved and automating the data break out would follow from that.

    Unfortunately, my guess is that the early adopters will not have this level of sophistication and will get fooled by short term aberrations. Perhaps that is what happened with Downing. And, once this technology starts to come online, my further guess is that the richer clubs will be the ones who can afford to implement it, leaving poorer clubs that much further behind. By the time the technology becomes widespread the Moneyball benefit is gone, since the whole point is to look for the previously unknown undervalued characteristic. When everyone knows the same things, that advantage is gone and the advantages that remain have ever decreasing returns.

    And, just to show how much variance our subjective opinions have, I though Spearing, Henderson, and Downing were all awful for Liverpool this season. But I thought Adam did fairly well. Most people strongly disagree with me on Adam. I would love for sabermetrics to improve for soccer, not for the teams, but so that all of us fans can have another thing to argue about! Cheers.

    1. One additional note about early adoption: The A’s found success in part because they were the first ones doing it (or at least, an “early adopter”). They only had to exploit one single, under-valued component of a baseball player – his ability to get on base – in order to have an advantage over their competition. Once other teams followed suit, and now that every team in baseball uses an advanced statistical approach for part of their player analysis (even my Mets, though it took hiring Sandy to do it), the success is far harder to achieve.

      Only teams that do it best (Tampa Bay, I’m looking at you) can benefit because the playing field has leveled quite a bit since the 1990s. As a result, the harder-to-get statistical measurements, like fielding, are now in play because every little advantage matters.

      Why does this matter to soccer (err, football)? Because in football, no one is doing it yet. ANY single inefficiency in the market that can be found will provide a great advantage to whichever club can find it. For example, if a club figures out how to measure the success of crosses from 35 yards or closer, and finds that this is an undervalued trait in the market, it will be exploited to much success*. Clubs don’t need to figure it all out just yet; they don’t need to reach the level of complexity baseball has reached. They simply need to get the ball rolling (no pun intended).

      Maybe Liverpool has. Their results aren’t good as of yet, but if ever there was a small sample size that should be ignored, this one season for Liverpool is certainly it.

      *side note: the inefficiency doesn’t need to be realized by a statistic. Maybe you find that large-legged Koreans are under-valued in the market. I dunno. That’s the other important point here. Moneyball the story was about finding an inefficiency, not about statistics. It turns out sabermetrics found the inefficiency, but they are not one and the same (I realize the poster I’m responding to knows this, but it’s not called out by the original article).

  7. As baseball and American football leagues in North America have very little competition for players except for the Japanese baseball leagues and the CFL applying statistical analysis to players has more validitity as you have a virtual closed system
    . Football has many top flight leagues across the world, and factoring in different styles of play, officiating, climate and football culture would make statistiical comparision more difficult. As others have pointed out football is a more fluid game and is more difficult to quantify. Competition for player signings in an open competitive market between clubs and leagues drives up player costs as well.
    That is not to say that good scouting and signing players from lower leagues or overseas has no statistical basis but one merely one factor in that process.

  8. If you want to point to a team that “buys low/sells high” and has been able to triumph over wealther rivals you could name Porto. They regularly win domestic silverware and have a champions league and two uefa/europa league titles in the last ten years. Look at the some of the players they have bought for relatively low sums and sold for rockstar money. Deco, Lisandro Lopez, Carvalho, Lucho Gonzalez, Radamel Falcao, Hulk, etc.

    I don’t think you can find a moneyball-esque reliance on stats in their approach. At the core of their model has been a policy of finding cheap talent (often abroad) getting the most out of those players and selling at the right time, even if it feels a year too early. They will never have the financial power of United or Real Madrid but this approach has allowed them to stay afloat financially and have a shot at winning trophies.

  9. Money ball will not work unless it takes every statistics into consideration.

    Statistics will work in every sport but you have to take other aspects of a player’s game into account.

    Pep Guardiola(Barcelona) used stats. His logic is if you press, keep possession and create lot of chances then you should win. Unlike moneyball they didn’t use stats to buy players, rather they found players to achieve the statistics required to win.

  10. I think if you are not using some type of statistical analysis to help you rate players you will fall behind the other teams. Their is more than moneyball.

  11. It won’t work for young players because scouting people rely on eye popping talent and not the players juvenile track record.
    There seems to be a belief that raw talent can, hopefully be massaged to the clubs advantage.
    It’s a hope and a prayer that a grain of sand can turn into the pearl that completes the team and gives the clubs coffers some bartering leverage.
    There is little to lose and everything to gain!

    The evaluation of established talent, however, is another matter, and here a smart evaluation process based on proven performance stats is key to a big financial investment.
    The fact that a given player can score goals in one league doesn’t mean that he can in another, that a finesse type midfielder can dance through team after team – but not every team, that a dominant central defender comes unglued against a specific type of center forward – it all points to the fact that a combination of factors must be computed to make the process more fail-safe.

    It’s all about risk management and it’s something that most clubs are very poor at right now!!

    Billy Ball is not with football yet, but with FFP looming and the emergence of wiley-er management and ownership, this kind of thinking will replace the conventional status quo.

  12. The moneyball ethos of getting more from less can work in some ways.

    Look at Norwich they took underperforming players from the lower leagues and inspired them and played them in the right way and went from League 1 to 12th in the Premier League without breaking the bank.

    Sam Allardyce did the same at Bolton by using statistical analysis more in his play to give them 3 or 4 seasons punching well above their weight.

    Moneyball is harder to apply to football, but there are bargains to be had if you mix together old school gut instincts and feel with statistics and an eye for a bargain.

  13. Let me help everyone understand how bad this article really is, by linking to someone else’s infinitely-better criticism of this article:


    There you go. Damien Comolli doesn’t play Moneyball. His (and John Henry’s) only “novel” idea was to make sure that their net transfer spend wasn’t too high, which they did because they made a boatload of money off of the Torres sale, among other things.

  14. In baseball, you’re always depending upon the rest of your team to win. If you’re a pitcher, and you only let up one run (which is pretty good), and your offense doesn’t score, you lose. Moot point.

    Arsene Wenger is obsessed with statistics. Just because it’s not publicized and stamped with a “Moneyball” seal, doesn’t mean that he doesn’t use sabremetrics. Wenger has been purchasing under-valued players for years.

    The idea of finding under-valued players (the heart of “Moneyball”) will work if properly applied in ANY sport. It just so happens that LFC, as you pointed out, paid over-the-odds for their players, and made it look like a shite strategy.

Leave a Reply

Your email address will not be published. Required fields are marked *