I was looking at the chances of being seeded for various European teams and I realized my software was using the October fixtures twice, because in one place I was still using the October 7th release date. This is fixed now. My apologies.
I also changed the way I generate the probable results using the Elo ratings. Instead of using the result that would have teams exchange the least amount of points (more info), I now use a method suggested by Virtus, with a slight change. If the win expectancy is between 0.33 and 0.66, that game will be a draw.
So, here you go:
FIFA Ranking: October 2009 preview (V)
FIFA Ranking: October 2009 probable ranking (IV)
This probable results method is much better (0.25-0.75 for a draw was way too much).
ReplyDeleteHi Edgar, I realize this is a little late but I did finally finish comparing results for all September matches. The most shocking thing was that using either method, 34 of 98 matches had the same prediction AND were wrong.
ReplyDeleteIn all, 19 of 98 matches had different predicted outcomes based on the different way of using the ELO ratings. Of those 19, using a 37.5-62.5% win expectancy for a winner predicted 9 matches correctly. The least points exchanged method (25-75% as Amir says? admittedly I did not bother calculating but used your reported predictions) correctly predicted 7 of the 19 matches. The remaining 3 matches were predicted incorrectly using both methods, although the least points exchanged method came closer to the actual results, predicting draws when the "underdog" teams won.
Ultimately I think this shows that any method used to predict matches will miss a fair amount of times.
Thanks for the input, Virtus!
ReplyDelete