The Bradley-Terry Rating System applied to international women's volleyball 2013-2014

  • I'm sure the FIVB has put a lot more thought into their world ranking system than I have so theirs is bound to be better, but maybe just for fun you can compare the results anyway. 8)


    The theory: If Team 1 with rating r1 plays Team 2 with rating r2, I expect Team 1 to score r1/r2 times as many points as Team 2. I iteratively calculate the ratings until for every team the points they are expected to score equals the points they actually scored.


    Included competitions:
    - African, Asian, European, South American and NORCECA Championships 2013
    - World Grand Prix 2013 and 2014
    - World Championship 2014
    - Asian and European Qualification to World Championship 2014
    - NORCECA Qualification to World Championship 2014 (Round 3 and 4 only)
    - Qualification to European Championship 2015
    - World Grand Champions Cup 2013
    - Montreux Volley Masters 2013, 2014


    If you think I should include more competitions, please suggest them. Keep in mind that I only want competitions where countries will generally send their A team, and I'm not keen on adding ones which will just include a bunch of tiny countries that don't appear elsewhere (e.g. NORCECA Qualification).


    The results:


    Update Feb 12: Added NORCECA WCh Qualification Round 3&4

  • fivb is bound to be better...mmmn hhmn we all know governments like to inflate statistics in order to mask their ineffectiveness so fivb is not immune on it either


    OK, maybe I was too subtle... I actually think the FIVB ranking is arbitrary and near-useless and mine is much better. :teach:

  • I think that the FIVB's system is actually worse. The matches going into the World Championship for qualification do not get counted until after the entire tournament is over.


    I think that your system works, but you need an entire 4 year cycle to confirm it.


    In my system, I break out the last four years by a percentage 25,50,75,100. Plus with there being few matches played in the early part of the year, I recalculate my numbers in 1/2 years. My system is based on set wins first (similar to FIFA with wins) an then a little bit on points ratio. Then within region vs. international match, and lastly rank of opponent. Right now, going back to matches from 2002, on a match by match basis for the women's game this system is right 78.7% of the time.


    International match value is taken from the entire region performance in international matches and then compared to the regions. Region is given a percentage of the top region score (1 region score/top region score) and then added to .75 (best score is .75). Then divide the 2 regions score for your final value (ex. (1.75 + 1.50)/2 = 1.625).


    Even in this system Brazil is still #1. They've been #1 since July 2006. They never lose in South America and they play very well outside of their region. Getting to many finals and winning. US got close in 2012, but lost the Olympics and were a little down in 2013 so a gap has opened up to Brazil.


    Here is my top 10 right now.


    Brazil
    USA
    China
    Japan (benefit of 2012 Olympics and will probably drop a little in 2015)
    Russia
    Turkey
    Italy
    Serbia
    Germany
    Dominican Republic

  • Hi Brahmin,


    I'm not sure using 4 years of data would be any more accurate than using two years. The performance of the Takeshita-led team of 2012 surely will not be very predictive for Japan's results in 2015. ;)


    I didn't quite understand what you said about "international match value" ... are you giving a higher weight to matches played by teams from different regions? If so, I would claim that it is only necessary because you are basing your system on less data (sets rather than points). But I would love to compare the lower levels of our rating lists to see how well they agree.


    About sets and points: I hope anyone would agree that if Team A wins against Team B by a margin of 25-22 18-25 26-24 17-25 15-13, Team B was actually the better team and would be more likely to win a rematch. The argument against using points is likely to be more along these lines: if a team is leading 20-5, do they play more sloppy volleyball than when they are leading 20-19? I believe this is very likely the case at lower levels of play, but almost certainly not the case at the professional level. In American Football, for instance, studies have repeatedly shown that blowout wins against weak teams are a much better predictor of future wins than wins/losses in close games (games decided by a touchdown or loss) - the latter is essentially random and subject to regression to the mean.


    One improvement I would love to add is rating teams seperately on their points scored on serve and sideouts. Unfortunately, I was not able to deduce this information from the standard match reports (P2 and P3). If anyone can tell me where to find this information for a significant number of matches please let me know.


  • The reason for the 4 years was due to World Championship and Olympic cycles. As they overlap, some countries will play in Africa will play in the Olympic qualifiers, but not the World Championships and then sometimes not even in the African Championships. This would give a fuller breadth of teams playing.


    The reason for the international match value was because the basis for my points system was taken from the FIFA Soccer World Ranking, but modified for to FIVB. Smaller countries usually never get to play countries from a different continent in a competition. It may during a friendly, but volleyball does not really keep track of friendly matches (can do some with the internet, but not all report). If Brazil did not win its Continental Championships or Olympic Qualifiers, then they would be at a lower level of competition. When you play in the World Grand Prix, World Cup, Olympics or World Championship, it is a higher level of competition, hence the international match multiplier.


    Right now I've got Europe as the strongest continent, followed by South America (Brazil is the main reason), Asia, North America (very close to Asia) and then Africa. Africa's score is only 20% compared to Europe and that is over a 4 year average. This year will be the first time it has been that high.


    As for the sets to points choice? Volleyball is unique because you can score fewer points than your opponents and win. Going back to 2002, it is 3.5% of matches (151 out of 4373) that this occurs. So points are important, but if you win sets, more often than not you'll outscore your opponents. I do take the final score (points ratio) into account as well. I reward a team that wins by 25 points with an additional set and little bonuses for their points ratio (24 points and under).


    The serve vs. sideouts is a wash. The team serving more would win due to you having to win a set by 2 points. You'll have to had scored 1 more time on your serve somewhere in the set. Only in a match where you scored less, or a five set tie score would it be different I think.


    Believe me the P2 & P3 numbers don't give us enough information. I've been working on my own advanced stats using the P2 & P3 numbers, but the FIVB does a horrible job of organizing their data for consumption or even outside work. Who is the best blocker? Just look at the block per set, right? No, how many times did the blocker have a chance to block a ball (spiked at him and part of the block) and actually blocked it! The P2 and P3 numbers give you the number of times the player touched it, but a player can jump and not touch a ball right? Let alone, what position is each player on the court. What position did the subs come in, front row, back row. How many points did they play? How many points was the libero on the court? Only the national teams get this information and that is from their own data team. How many possessions did each team have (in addition to spiking, blocks, block rebounds, digs (not kept on your side of the net), etc.


    This some stuff that I've done in the past


    2014 Men's World Championships (go down further on page)