Moneyball: Applying Sabermetrics to Football Manager

For those who don't know Moneyball: The Art of Winning an Unfair Game is a book about American baseball coach Billy Beane. He is coach of Oakland Athletics and they struggled to compete with the likes of big spending New York Yankees. He developed a statistical model, known as sabermetrics, which allowed him to identify undervalued players and compete with the best teams in the league.

Fenway Sports Group bought the Boston Red Sox and applied a similar transfer strategy which resulted in winning the world series. Since then they have purchased Liverpool FC and are trying to emulate the successes of the Red Sox buy buying players based on rules outlined in Moneyball and using statistical analysis to identify players.

Sabermetrics is based around the premise that the traditional way of determining a players worth is flawed. It also aims to find an individual players contribution to a winning performance.

Baseball consists of many discrete events (hundreds per player per season), each measurable as a contribution to run scoring. In football, the relevant events are goals, but they are rare. And, for more common events, like "passes completed," the exact contribution to goal scoring or goal prevention is often not obvious.

Historically the following stats have been used to evaluate how well a player has performed.

Goals

Assists

Clean sheets

Pass completion %

Shots on target %

Tackles completion %

All of these are dependent on the skill level of other team mates. Imagine if you put your favourite premier league midfielder into a non league side, do you think his assists would be as high as if he was playing for the league champions? Somebody has to finish those chances for him.

What stats are available in Football Manager that allow us to more accurately evaluate a players contribution.

Tackles per game

Passes completed per 90 mins

Shots on target 90 mins

Headers won 90 mins

Conceeded per 90 mins

Distance covered per 90

Dribbles per game

Using a combination of the above stats with the desired position we should be able to identify some undervalued players.

Rules of Moneyball

In addition to identifying players based on statistics there are rules on transfer policy.

1. Draw opinion from several sources.

So you have identified a player using sabermetrics, now you need to scout him and then scout him again with a different member of staff.

2. Sell a player before buyers see a deterioration in his game.

Arsene Wenger is the example to follow here, can you think of a player Arsenal have sold who has gone on to be as good as he was for Arsenal?

3. Sell any player for the right price.

Every player you buy is an investment, and its about maximising your return before that players value will start to decrease.

4. Buy players in their early twenties.

Teenagers are underdeveloped and may never reach their potential and as such are high risk investments. Older players have less chance of an increase sell on value. Players in their early twenties are sufficiently developed to gauge whether they are likely to reach their potential, and will have increased sale value when they reach their peak.

I plan to play a career game using these principles and perhaps ill update this thread to let you know how it works, i would be interested if other people used these rules and to see how much success they had.

I started a game and holidayed a year to compile a first eleven using the stats to identify players and filtering out anyone over 25. So if money was no object this would be my first 11.

(I used an updated transfers db)

GK - Neto - Fiorentina

DR - Rafael - Man utd or Dejan Lovren - Lyon (stats almost identical)

DL - Gael Clichy - Man City

DC- Gerard Pique - Barcelona

DC- Mamadou Sakho - Paris Saint Germain

MCd - Yann M'Villa - Stade Rennais

MCa - Javier Pastore - Palermo

MR - Lionel Messi - Barcelona

ML - Stevan Jotevic - Fiorentina

ST - Edinsaon Cavani - Napoli

ST - Diego Costa - At Madrid

I hope to apply all of this to the lower leagues and see how far i can go.

Good thread and I look forward to seeing how this pans out.

Nate Silver tried to apply his PECOTA version of sabermetrics to the World Cup after he used another modified version to accurately predict the state-by-state results of the US presidential election.

Here were the results of his initial pre-cup tests:

Not bad obviously, but as I recall, the major problem to applying PECOTA to football was the relatively ambiguous nature of "form" compared to the fairly clear cut statistics of baseball. In baseball, the game is primarily a one-on-one affair with the success of each match-up being cut and dry (strike out, base hit, fly out, homerun, etc.). Football is a more complex, team game where the extent to which each individual act contributed to a team's overall success (or lack thereof) is much more difficult to determine. While you're still generally able to track the quality of teams, tracking the quality of individual players is far more difficult.

Still, I'm very interested in seeing how your experiment pans out.

Reminds me of a book which I read, I lifted some of their principles in order to get to my transfer market principles. Obviously some aren't very applicable in FM, but it's still a good list to work from:

1. Don't waste money on transfers - Typically new managers want to make their mark on a side, so they clear out the previous manager's purchases at a discount. They then bring in their own men. A new manager is generally allowed to buy big on the pretence that he is shaping the club for the future while in reality he leaves very quickly. He doesn't get a bonus for making a profit, so why should he care?

2. Don't buy after a major tournament - What a player did last does not equal what he will do next. Everyone has seen how good he is but he is probably happy with his success.

3. Don't be a sucker for fashionable nationalities - Brazilians and Dutchmen are incredibly overvalued. Buy unfashionable nationalities for less, even though they are probably just as good. It's easier to sell a **** Brazilian than a decent Mexican.

4. Identify and abandon sight based prejudices - Scouts are obviously more likely to report back on players who stand out. Blondes for example. Gentleman generally prefer blondes, sometimes without realising it. Scouts are also suspicious of short players, or stocky players, or players who are too tall. If you want to get the most out of your talent scouts you need to abandon these prejudices.

5. Be as eager to sell as you are to buy - If you stop upgrading you are ****ed. No room for sentiment. Guess when a player is peaking, like when the stock markets peak, and you will reap the rewards.

6. Older players are overrated - Conventional wisdom states that a player peaks in his early 30's. Right? Well that's wrong. Studies by the University of Loughborough shows that a player peaks at 27 years old. Sell at or after that age, before other clubs notice the subtle declines in the player's performances.

7. Buy players with problems and help them deal with them - Players with gambling problems, or players who drink, are sometimes remarkably underpriced by their clubs. If you can help them, you will reap the benefits.

8. Help players relocate - A player won't reach his potential unless he has adjusted. If he is being taught the language, his kids have a nice school, he has a nice house, nice car, etc. Hire a relocation consultant. It may cost £15000 for an expensive consultant, but they will be a friend to the player, and will get them what they want. Hardly any teams do this, however those that do tend to be more successful. For example: AC Milan, Lyon, Ajax. Don't trust hand the player all the responsibility. Bad examples of relocation include Drogba to Chelsea and Anelka to Real Madrid. Don't fall into that trap.

9. Use the wisdom of crowds - Discuss everything as a group. Bounce ideas off of eachother. Self explanatory really.

10. Buy players in their 20's, not their teens - Very few players are at the top by 20. Brilliant schoolboys tend to disappear. You can only be confident of a player's potential when they are mature. Bigger clubs demand stars, just buy young players because they are good.

11. Centre forwards are overpriced - And even though they have long careers, keepers are underpriced.

12. Sell if you are offered more than a player is worth - Players are untransferable, that is until you get an offer which is higher than you expect. No point in clinging on to players when you could buy two replacements with the money you could get from their transfer fee.

I tend to find that I can live by those principles and win my fair share of games and tournaments. And having looked at some of your points they look pretty similiar so you should be reasonably good too.

And as for your stats you perhaps don't just want to be looking at shots on target, since the more shots you take the more that will end up on target due to simple probability. You'd really want to be looking at on target as a percentage of total shots. In the same way you'd perhaps be better off looking at goals conceded as a percentage of shots faced for keepers. Percentages are better than just stats because they don't lie as often.

Anyhow, good luck. I'll certainly be following; you may want to set up a career update thread though.

I decided to start my game after the first season holidayed so i have some stats to start with. Southampton have been relegated to League 1 so I thought, they would be perfect to test the system out and have the infrastructure in place if i move up the leagues quickly.

I am using a basic framework to identify players:

Only players who have 1800 mins played will be considered for analysis

GK

G/90 < 1.25

DC

H/90 > 5.0

T/pg > 4.0

DRL

T/pg > 3.5

DC/90 > 5.0

Pc/90 > 20.0

DM

T/pg > 4.0

DC/90 > 5.0

Pc/90 > 25.0

AM

Pc/90 > 25.0

Dr/pg > 2.0

ST/90 > 1.0

ST

Pc/90 > 15.0

Dr/pg > 2.0

ST/90 > 1.5

Sh% > 50%

Anyone who meets the criteria is then scouted by each of my scouts

Have you thought of a formation to use these players in? Also, I have to say that I think you should be using percentages for your shooting and dribbling stats.

And as an aside, Liverpool aren't really using sabermetrics, Henry might be but I'd bet you any money that Dalgleish is building a team using "the secret" not any statistical analysis...

I decided to start my game after the first season holidayed so i have some stats to start with. Southampton have been relegated to League 1 so I thought, they would be perfect to test the system out and have the infrastructure in place if i move up the leagues quickly.

I am using a basic framework to identify players:

Only players who have 1800 mins played will be considered for analysis

GK

G/90 < 1.25

DC

H/90 > 5.0

T/pg > 4.0

DRL

T/pg > 3.5

DC/90 > 5.0

Pc/90 > 20.0

DM

T/pg > 4.0

DC/90 > 5.0

Pc/90 > 25.0

AM

Pc/90 > 25.0

Dr/pg > 2.0

ST/90 > 1.0

ST

Pc/90 > 15.0

Dr/pg > 2.0

ST/90 > 1.5

Anyone who meets the criteria is then scouted by each of my scouts

im interested in using this,,,but could you be a bit clearer as to what all this means? The stats are all in percentages on the player screens not in the type you've put here,,,(or i dont understand anything you are on about lol)

I tend to line up with a 4-1-2-2-1

There is no dribbling success stat in the game only made and made per 90.

Regarding using shots on target percentage, i have looked into this a bit more and there is a strong correlation between shots on target % and shots on target per 90 mins, as you might expect. It would appear to me that its actually quite hard to find a player that has more than 1.5 shots per game on target and below 50% accuracy. I have updated the framework i use to include shots on target %.

Regarding Liverpool, we dont really know 100% what they do but im pretty sure it goes something like:

Dalglish identifies what the team needs

He tells Comolli

Comolli creates a shortlist of players (well known for using stats to identify players at former clubs)

Dalglish chooses player off shortlist

Have you thought of a formation to use these players in? Also, I have to say that I think you should be using percentages for your shooting and dribbling stats.

And as an aside, Liverpool aren't really using sabermetrics, Henry might be but I'd bet you any money that Dalgleish is building a team using "the secret" not any statistical analysis...

Charlie Adam is a very good example of a statistically "overvalued" player. He has a poor pass completion rate, and I believe got more cautions than goals last season. Definitely not a "moneyball" player, compared to David Vaughan, for example.

Reminds me of a book which I read, I lifted some of their principles in order to get to my transfer market principles. Obviously some aren't very applicable in FM, but it's still a good list to work from:

<snip>

If anybody is interested in reading further, the book is called Why England Lose, and is by Simon Kuper (of Football Against The Enemy) and Stefan Syzmanski.

Coolestrock

If you go to player search you can create a custom view with stats such as Headers won per 90 mins, Distance covered etc etc.

Coolestrock

If you go to player search you can create a custom view with stats such as Headers won per 90 mins, Distance covered etc etc.

ok cheers,,,,can i just make sure ive got it right,,,,,,,,H= headers won,,,,,,,,,,T= tackles won,,,,,,,,,,,Dc= distance covered,,,,,,,,,,,,,,Pc= passes completed,,,,,,,,,,,Dr= dribbles made per game,,,,,,,,,,ST= shots on target,,,,,,,,,,,,,,,,,,,,,,,,,for the gk does the G/90 stand for goals conceded??

ok cheers,,,,can i just make sure ive got it right,,,,,,,,H= headers won,,,,,,,,,,T= tackles won,,,,,,,,,,,Dc= distance covered,,,,,,,,,,,,,,Pc= passes completed,,,,,,,,,,,Dr= dribbles made per game,,,,,,,,,,ST= shots on target,,,,,,,,,,,,,,,,,,,,,,,,,for the gk does the G/90 stand for goals conceded??

yes mate

H/90 = headers won per 90mins

T/pg = tackles per game

Dc/90 = distance covered per 90mins

Ps/90 = passes completed per 90 mins

Dr/90 = dribbles per game

ST/90 = shots on target per 90 mins

I have always been sceptical about the usage of stats and sabermetricism in sports other than baseball. While statistical analysis is a hobby of mine and I love to do it in my spare time I have never professed to know how it works in building a winning team. Saying that because a certain thing (sabermetrics) works in baseball, it will work in football too is like saying that acupuncture can cure cancer because it works on arthritis. Two completely different conditions and if it does work then it is either a fluke or a placebo...

Baseball is completely different to football in terms of its sample size, for any given season the team which has the team which fits best statistically to the model is probably going to win a significant amount of games. However, as Billy found out to his cost once the sample size fell dramatically (in the playoffs) they couldn't win a world series. Boston needed the financial clout of a big franchise working in roughly the same way to accomplish the WS win. Straight away that shows me a problem with your strategy, since it works on probability. If you assemble a squad which works well together like the A's did then you'll win about 100 games a season. At best. Out of 162 games that is 5/8. If we transfer that to the premiership that means that you'll win 23-24 games a season when your squad is at its peak. That gives you an average of about 70 points on average. Assuming a 50-50 ratio between draws and defeats you get about 77 points a season. Which is enough to win you the Premier League twice since its inception. Assuming that this squad will only stick around for a few years at its best you'll need a lot of luck/a lot of undervalued talent to make a success of it.

And that of course implies that your figures are correct. I have already raised concerns about the stats which you have chosen to pick in order to represent the players better. And it just seems that your allocation of favoured stats is a tad random. As if you've sort of missed the point a little. Just because those stats aren't used as much to calculate a player's worth doesn't mean that they should be used to do so. In order to find out the true measures of value Beane's number's runners had to run so many multiple regression models that they probably bled from every possible orifice for a few weeks after it was all done. And while that was happening they went through some bad patches. 75 win seasons and all that. They didn't just randomly pick stats and choose to use them, which in some cases flies in the face of all logic. The reason why nobody uses the distance a player runs in a game to calculate their worth is because doing so is pretty silly. I'd rather have Matt le Tissier than Usain Bolt on the wing...

Baseball is the most structured of any game ever invented, with the possible exception of cricket. Every play can be seen as a series of steps to a run, a base hit or an out. In football that doesn't happen, most passages of play are pretty meaningless apart from those which lead to goals. It's difficult because of the fact that there are so many non-events and to get a true picture you'd have to watch every minute of every game. This isn't possible, expecially in football manager.

Instead of comparing football to baseball compare it to basketball, nobody has ever figured out how to do it in that sport either. Statistics don't make a good football team, nor does probability make a good basketball team. Although baseball is a "team" sport there is only ever one batter hitting the ball at any one time and thus influencing the play. Therefore probability can be used to some extent to create a favourable series of actions and reactions for your team. However "real" team sports aren't about stats, they're about people. That's why I'd have Bill Russell over Wilt Chamberlain any day of the week, and that is why I would rather have Messi than Pelé in my side. The last time I checked, there was no stat which can measure warrior-like intensity, single-handedly dragging a team from the bottom to the top and making the fan at home think: "I would want to be in that guy's foxhole in a war." Team sports aren't won by the number's guy in the back room with his excel document, they are won by the team on the field. And if you can make a success of this team remember that it is down to your team gelling well and having compatibility as a unit, not due to your statistical theories and possible connections and correlations and all of that crap.

Even though I have more arguments against this system, it's late so I'll sign off for the time being and I'll check back later tomorrow. I admire what you're trying to do but I can't help but feel that you're paddling up the wrong creek with no paddles on a burning boat carrying a megaton of TNT.

And as a final response to your case for Liverpool using this system, Dalglish could be compared to a toilet roll user and Comolli to the manufacturer. The manufacturer may use statistics and science to create the product but the consumer has the final choice. And does that process mean that the toilet roll usage is sabermetric? No.

If anybody is interested in reading further, the book is called Why England Lose, and is by Simon Kuper (of Football Against The Enemy) and Stefan Syzmanski.

funny, in america its called soccernomics, its a great book

I tend to line up with a 4-1-2-2-1

There is no dribbling success stat in the game only made and made per 90.

Regarding using shots on target percentage, i have looked into this a bit more and there is a strong correlation between shots on target % and shots on target per 90 mins, as you might expect. It would appear to me that its actually quite hard to find a player that has more than 1.5 shots per game on target and below 50% accuracy. I have updated the framework i use to include shots on target %.

Regarding Liverpool, we dont really know 100% what they do but im pretty sure it goes something like:

Dalglish identifies what the team needs

He tells Comolli

Comolli creates a shortlist of players (well known for using stats to identify players at former clubs)

Dalglish chooses player off shortlist

pinched straight off of RAWK

Reminds me of a book which I read, I lifted some of their principles in order to get to my transfer market principles. Obviously some aren't very applicable in FM, but it's still a good list to work from:

Soccernomics, eh? That book was an interesting read.

EDIT: Didn't see ThaChamp already beat me to it.