Wednesday, October 29, 2008

Mr. Blatter needs a new ranking system

Sepp Blatter was asked during his visit in Israel about Israel's unrealistic FIFA ranking.

Here's his answer (from The Jerusalem Post):

"We used to calculate the ranking with a university formula and help from experts, but then we decided to implement a similar system to the one the ATP uses," Blatter explained.

"I don't know why you're complaining? You're 16th in the world.

"If anybody's got a better system to rank the teams, please tell me about it."

Perhaps Mr. Blatter should take a look at these ranking systems:

World Football Elo Ratings

Based on the Elo rating system, developed by Dr. Arpad Elo. Here's the system behind the rankings.

AQB International Soccer Ratings

Developed by Michael Cameron of the University of Waikato (New Zealand). More details on the system.

Voros McCracken's International Team Ratings

Developed by Voros McCracken. He's famous in the world of baseball for his Defense independent pitching statistics work. The starting point was KRACH (Ken's Ratings for American College Hockey), "an implementation of a sophisticated mathematical system known as the Bradley-Terry rating system". It is the single iterative system of the three I've listed. It is also my favourite, but unfortunately Voros stopped updating it after May 2007 due to other commitments and right now I don't have the time to implement it.

Update 30 October 2008:

As I already mentioned, KRACH is just the starting point. Here's what Voros said about the system:

"However a bunch of modifications were made to how to do the iterations, home field advantage, partial wins awarded based on scores and also a separate system that evaluates teams based on goals instead of wins and losses. These systems also account for how long ago the match took place and also the type of match it was (IE, friendly, World Cup Qualifier, World Cup Final, etc.)

The idea is at the end of all of the iterations, the predicted win percentage for the teams in their games based on their ratings is the same as their actual. So that if a team wins 85% of their games, the ratings are such that if you add up all of the expected results that's the win percentage the team will have.

Ultimately I believe iterative methods are the best and most reliable way to encompass all of the data. If you don't do an iterative method, it's difficult to go back and interpret past results based on new information (eg, if you lose to Ghana one week when they're rated low and then next week they beat Brazil, that information should be used to re-evaluate what losing to Ghana means. A system like ELO doesn't do that).

Testing of the system against actual results tends to have it do slightly better than ELO and significantly better than FIFA."

About me:

Christian, husband, father x 3, programmer, Romanian. Started the blog in March 2007. Quit in April 2018. You can find me on LinkedIn.

23 comments:

  1. Very interesting, Edgar..
    I didn't know before the mckraken's ranking but it seems a very good system.
    It's a pity that ranking is no longer updated.

    ReplyDelete
  2. I don't see how is that similar to the ATP.
    The ATP don't use average at all.

    ReplyDelete
  3. Can you explain how "Voros McCracken's International Team Ratings" works?

    ReplyDelete
  4. OK I found it - sounds like a really interesting system, but it is very complicated to understand and calculate.
    I agree that it might be a good system, but it ignores home advantage and more important, it ignores importance of the match.

    So I still think the ELO system is the best.

    ReplyDelete
  5. I've updated the description for Voros' ranking.

    ReplyDelete
  6. Thanks Edgar.
    Where can I see what modifications were exactly made?
    And the presentage that comes off is for a match on neutral ground I guess. How can you can you calculate the presentage with a home advantage?

    ReplyDelete
  7. When I talked with him about the rankings, he was negotiating with a website owner to post them there so he couldn't give me the details.

    ReplyDelete
  8. I still have a bunch of personal stuff to resolve before I can get back into the swing of things, but I've been talking to Edgar a bit on this issue.

    Anyway, in terms of HFA, the factor is applied to the predicted outcomes of matches. And since those predictions drive the next iteration of ratings guesses (with the goal being have those predictions match actual results), they affect the overall ratings. The factor is exponential since that way handles win% of 1 and 0 well. In other words, you raise the neutral site win% by an exponent to adjust for HFA.

    I had been thinking recently about a few new mods (a system designed to deal with outlier results is one) to the system. I guess I have to put in a phone call to FIFA now. :)

    ReplyDelete
  9. Thanks, Voros!
    Please keep updating us!

    ReplyDelete
  10. What do you think about taking the best 18 results in the last 4 years (with each match points being exactly like it is now in FIFA's system), like the ATP ranking system (maybe the WC and continent main tournaments matches should always count, just like GS and Masters).
    Edgar, Can you please make a
    rankings that shows what these rankings will look like? I don't think it should be too hard to implement.
    If these rankings work, I think FIFA might like this idea.

    ReplyDelete
  11. Or maybe just taking the best 5 results of each year could be a solution (that would work better, because the reason Israel is ranked so high right now is that in the last year it made a lot of points because of an easy schedule. By taking the best 5 results each year that easy schedule advantage will be gone.
    Please implement that too if you can Edgar... :-)

    ReplyDelete
  12. I meant better than today, not better than taking the best 18 results...

    ReplyDelete
  13. Amir,

    IMO, the ATP-like ranking won't work. In tennis - it's based on the stage you reached - Win, final, semifinal etc. In FIFA is match based. The GS and Masters are mandatory - not so with the WC and continental finals. Each confederation has its own tournament: CONCACAF and CAF every two years, the rest every four years.
    The 5-best results would be a step back. Besides, many nations don't have 5 matches in a year. Wasn't the previous system based on the best 7 results, averaged with the rest of them? Or something like that.

    ReplyDelete
  14. In my opinion, considering 5 or 7 matches (in general a low number of matches) the ranking is too mutch influenced by occasionals particolar results (i.e. Luxembourg win against Switzerland could put the Luxembourg at a relative high ranking, considering its real strength).
    I think the 18 results is a better idea (each result can be considered with actuals multipying factor 0.2, 0.3, 0.5 for old results). For small nations that have not played 18 matches, the total can be simply divided for 18.

    ReplyDelete
  15. Maybe edgar, if you have the time, you can test the idea of Amir. Points calculation for single match is the same, the application should return best 18 results in 4 years.
    I don't like the actual FIFA method(average of results) because it is too mutch influenced by friendlies.
    In a ranking we espected that after a win the rating of a nation improve or at least remains the same. But if a team play friendlies against lower ranked teams it can decrease its ranking despite wins(i.e. DPR Korea after AFC Challenge Cup).

    ReplyDelete
  16. Another consideration: If Far Oer Islands lost only 1-2 with Italy in Euro2008 qualifing (Italy was in the 1 position of FIFA ranking when the match was played)it takes 0 points; if Grenada beat US Virgin Island 10-0 in Concacaf qual. for 2010 WC it takes 318.75 points. It's logical?

    ReplyDelete
  17. OK, here you go guys: 18 best matches ranking. Enjoy!

    @jos235

    No, it's not logical. You need to take into account strength of schedule - FIFA doesn't.

    ReplyDelete
  18. Edgar, The GS and Masters are only mandatory to whoever qualifies for them, just like I suggested the WC and continental finals would be.
    CONCACAF and CAF finals are held every two years, that's true - and making them mandatory won't make it a huge advantage.
    The 5-best results won't be a step back. It is just that one good year with an easy schedule (like happened to Israel this year) won't make them go so high in the rankings (the system here is without mandatory matches).
    I hope you can make both rankings sometime (in the best 18 results system the time multiplier should be canceled).

    ReplyDelete
  19. Thank you very much. The 18 results version seems pretty good to me, maybe a bit outdated.
    Actually, I don't see why each year is a different time frame in FIFA's rankings, because each campaign lasts 2 years.

    ReplyDelete
  20. You know what would be interesting? To iterate the FIFA's rankings with the new rankings of the teams until you get a stable rankings. I wonder how different will it be than the current one.
    I must stop coming up with these ideas :)

    ReplyDelete
  21. Amir, I've updated the 18 results version.
    Interesting idea, but I really don't have the time now. Besides, the formula behind the FIFA Ranking is already bad so...

    ReplyDelete
  22. Thanks a lot. I completely understand that you currently have more important things in your life.

    ReplyDelete