World Player Rankings

'Goalimpact' have published their list of the world's top 50 footballers, based on their algorithm.

1) Müller
2) Özil
3) Benzema
4) Messi
5) Lewandowski
6) Fabregas
7) Walcott
8) Neuer

Selected others:
21) Ronaldo
25) Rooney
30) Hazard
33) Robben
46) Alexis

http://www.goalimpact.com/2015/04/top50-football-players-april-2015.html

Naturally, what stands out is Ronaldo at #21 (down from #1 the year before). Walcott at 7th seems extremely high and, World Cup-winners or not, few people would consider Thomas Müller & Mesut Özil the two best players in the world.

Thoughts?

This must be a joke? Or someone must be extremely drunk.

So they work out using an algorithm, can't find it on the site (haven't looked overly hard mind) if the maths is sound then fair enough.

These things are always a load of bu**sh**. Read one the other week, Gael Clichy was the best full back in Europe. Just a load of nonsense.

hampo171 wrote:So they work out using an algorithm, can't find it on the site (haven't looked overly hard mind) if the maths is sound then fair enough.

Goalimpact measures the extent that a player contributes to the goal difference per minute of a team. If a player has a high GI value, then his team has a good goal difference when he plays compared to matches where he doesn’t play. An average player in my database has a value of 100. The best player in the world approximately 190.

The system uses an entirely different approach compared to most other player rating systems, as it does not record single player actions and try to value them. Instead, it just rates the player by the outcome and is unaffected by how the result was achieved. For the more mathematical oriented readers; Goalimpact is a kind of Shapley Value for team sports.

The reasoning behind this can be explained by using car race prediction as an example. Imagine you try to predict the result of a race and you are free to choose your prediction method. One way could be to collect data on all the cars. For example, the engine power, weight and body shape. This data may provide a picture of which car should be fastest. Unfortunately, even a race is too complicated for this to work. You are probably going to miss one or more crucial parameters. Did you consider the weather, the tactics of the opponent or the quality of the box crew? What about the driver’s skill? Predicting a car race is incredibly hard. Predicting a football games is orders of magnitude harder.

Given that, I don't think it is be feasible to come up with model that combines aggregated bottom-up football data, such as successful tackles and passing percentage etc., and expect it to provide a meaningful player rating any time soon. Instead, I go the other way round and look from the top-down, from the perspective of the result. Back to the racing example, you could measure the average speed of each car in previous races. The average speed is already an aggregate of many factors that eventually lead to this average. Just sorting the cars by the historical average speed will be a good estimate of the outcome of the next race.

The footballing analogue to the average speed in racing (distance per unit time), is the goal difference per minute playing time. Taking goal difference instead of goals scored ensures that both the offensive and defensive capabilities of the player are reflected. The team success depends on scoring more goals than conceding, in short, having a positive goal difference. It is of secondary importance whether that is 1 - 0 or 4 – 3.

The Goalimpact score averages the goal difference per minute for all games the player plays for any team, season and league. This is his ‘previous race speed'. However, taking this simple average can be easily misleading. A player with many games in lower leagues can achieve a high positive goal difference far easier than if he played in a top league. Therefore, the real world meaning of the Goalimpact score improves dramatically if corrected for the Goalimpact score of team mates and opponents, and consideration is given to home advantage and many other important factors. Unfortunately, this makes the Goalimpact of a player dependent on the Goalimpact of the rest of his team, creating a mathematical problem that is very computational intensive to solve. However, this can be done and you see the results on this page.

When interpreting Goalimpact value, keep the example of average speed in mind. It doesn't tell you the 'maximum speed' ('talent') of the player and it doesn't tell you the current speed ('form'). Instead it tells you the average performance the player actually had. The score may be low for many reasons, the player is just bad is only one of many possible explanations. Maybe he was just very unlucky, or the Manager played him in a position that didn't fit him (imagine Messi as goalkeeper as an extreme example) or maybe he played through a slight injury.

However, high Goalimpact scores provide easier conclusions. If a player has a high score, than we can conclude, with confidence, that he has/had a high contribution to his team. Meaning, under similar playing conditions, he is likely to have a high impact. To conclude that a player with a high value must be good is easier than to conclude that a low value makes a bad player. That said, even a high score can be the result of a very lucky player. It is just statistics, after all!

The database with all games and players is updated monthly and therefore the Goalimpact values of all players change in monthly intervals. For young players with few games considered, the Goalimpact will change faster than for older players that have played many games in the database.

Surely this is an MVP type gig? Bit like the PER rating in basketball? Ronaldo will be dismayed to learn that Luiz Gustavo is a better footballer than him

Seems to me to be predominantly how a team has done then reverse engineered scores/ratings for the individuals involved.

But all it needs is a high profile manager to take it on board, then another, then another and before you know it Walcott is captain of Bayern Munich, at centre back, with gola boots and a rucksack of bricks on. About 6 months before someone adopts and films the football equivalent of "The Money Game" with Wesley Snipes playing Lionel Messi.

As you can probably guess, it gets a "pfft" from me.

I knew Walcott was world class!!!

Indeed Stella. The problem was we were looking on the field, and not on the PC for his quality.

alexis sanchez at 46 ? Unbelieveable

Messi number 4, anyone who does a poll and puts Messi at anything other than #1 maybe, maybe #2 is an idiot.

My top 5

1. Messi
2. Ronaldo
3. Suarez
4. James
5. Neuer(ok he is a goal keeper)