Isn't it a time for "official" BigSoccer World Cup projections? Here are some background facts before we jump to the numbers. To create projections I used online simulator tool here: http://pirozhok.somee.com/sim.aspx http://aspspider.info/pirozhok/sim.aspx Online tool is based on Excel simulator model built by dlokteff http://forums.bigsoccer.com/members/dlokteff.19598/ We are using ELO ratings, formulas for ELO% come from their site. The games are simulated via a random number combined with those formulas. The Winning % formula was derived by Dr. Lokteff using results of group games from the past 4 World Cups. Brazil gets extra points for being a host, their opponents have 100 points subtracted. FIFA and SPI rankings converted into an "ELO" based on fitting it to a curve based off the ELO ratings and ranking. There are 2 models: without scores and with scores. Main diferences between models are: 1st model (no scores): All ties (group stage and knockout games) are broken randomly (we don't simulate scores of games, so no GD). 2nd model (with scores): we convert Win Expectancy to expected Goals scored, and then assign points based on the score. Group stage ties are broken according to FIFA rules: a) greatest number of points obtained in all group matches; b) goal difference in all group matches; c) greatest number of goals scored in all group matches. If two or more teams are equal on the basis of the above three criteria, their rankings shall be determined as follows: d) greatest number of points obtained in the group matches between the teams concerned; e) goal difference resulting from the group matches between the teams concerned; f) greater number of goals scored in all group matches between the teams concerned; g) drawing of lots by the FIFA Organising Committee. Knockout stages tie-breakers (AET or PKs wins) are using random numbers combined with historical data To create projections I ran each model 100,000+ times to eliminate ambiguity and provide an accurate projection. Here are USA projections: 1st model: Advance from the group: 26.94 % Group winner: 4.71 % Advance to 1/4: 17.78 % Advance to 1/2: 10.00 % Advance to final: 4.52 % 4th place: 2.61 % 3rd place: 2.87 % 2nd place: 2.64 % Champion: 1.88 % 2nd model: Advance from the group: 35.43 % Group winner: 8.78 % Advance to 1/4: 19.73 % Advance to 1/2: 8.30 % Advance to final: 2.56 % 4th place: 3.75 % 3rd place: 1.99 % 2nd place: 1.95 % Champion: 0.61 % Full projections are here (you can sort tables by any column): http://pirozhok.somee.com/wcp.aspx http://aspspider.info/pirozhok/wcp.aspx Enjoy!
Interesting but I wonder what's the point of doing a projection without tie-breakers, and one with tie-breakers - and they look so different, for certain teams! Which one would you consider statistically more likely to pay off? Of course, the downside of this exercise is that it takes into account past performances in WC and while tradition counts, how a team did 4 and especially 8 or more years ago has little to no bearing on a team's current form. Also, FIFA rankings are a joke, so, I wonder if even including them at all doesn't badly distort the data. As much as I'd love for it to happen, I'd be pretty surprised if the actual odds that the United States will survive the group stage are 35.43%. Again, this is probably artificially bumped up by the United States' artificially high FIFA ranking - a case-in-point to show how FIFA rankings are distorted, by favoring official games when CONCACAF is so weak, which bolsters the US ranking by beating up on lowly Central American teams. So, if we were to restore the United States ranking to its true numbers, I bet we wouldn't get to 35.43% of likelihood of surviving the group stage.
Answering your questions. 1. Original simulator model was w/o scores, random tie-breakers, then dlokteff created model which uses scores simulator, so we decided to keep them both Which one is better - i can't tell, probably model author can answer this question. 1st model simulates results w/o scores and breaks all ties (group ties and knockout games) randomly. 2nd model generates winning% using ELO first, then uses Poisson distribution to predict the most likely scores (from 0 to 9 for each team in the pair), points are assigned based on score. 2. We are using ratings and past WC performances, ratings alone or past stats alone won't be accurate enough. 3. For World Cup projections we used ELO ratings only, for simulator and match predictor you have a choice to use any of 3 ranking systems 4. Agree with you on USA rankings, i too think they are usually artificially high prior to WC due to our success against weak opposition in Gold Cup and WCQ and warm-up friendlies. 5. If you had a chance to check Match Predictor option, you could see that you can override ratings and past performances by using weighting factors: http://pirozhok.somee.com/gamepredictor.aspx
Really? Interesting. Well, even though I spoke against it, obviously I like to know that I was wrong because I root for the United States.