Out4Blood & El_Cap's Rise of Nations Strategy

Saturday, November 15, 2003

Big Huge Ratings explained
If you want to know how the BHG ratings work check here.

Me, I'm happy.

BHG is using Arpad Elo's formulas (with some slight modifications). Elo's formulas were rigorously tested with decades of data, and they've been in use for decades (US CHess and FIDE both use them). There are very good reasons for all of the components of the rating system.

1. You want to approximate the real rating as quickly as possible. Thus provisional ratings are calculated differently and will have huge swings until they are established. One of the worst thing you can have is to have ratings correlate with games played over time. You should need to play a few games to get your rating established. But then it should "stick" unless your skill dramatically increases for some reason.

2. Ratings should never reset. Ratings are meant to assess skill. Skill doesn't reset. Ratings systems that correlate strongly with games played are not designed to measure skill as much as they are designed to reward constant play. Therefore, they occasionally reset to encourage players to play agains for hte "money title." For example, Battlenet rewards players who rack up tons of wins more so than the prodigy who comes in and wins his first 30 games against top players. Even if the latter player is twice as good. A better example is the money title in golf vs. the Vardon Trophy. Tiger won the Vardon trophy (for lowest adjusted stroke average) while Singh won the money title. Tiger's the better play because he shoots better when he plays. Singh just played in a LOT more events than Tiger.

3. The "k-factor" declines and then remains stable at 16. Even after the provisional period, skill is likely to increase fairly rapidly as the player gets more experience. In that case, you'd want K to start out higher (about 31 for 25 games played vs. 16 for 50 games and beyond.)

4. Once both players are established, if you are evenly matched, then the winner would get 8 points and the loser would drop by 8 points. If you beat someone 400 points below you, then you'd probably get ~1 point (not sure how they do fractions). Established ratings are exceedingly accurate at predicting winning chances. At a 400 difference, I have a 90.9% of beating you. At an 800 difference, that chance is 99.0%. If by some miracle you beat me, you're rewarded with all 16 points and I lose all 16.

UPDATE: For geeks (like me) interested in comparing the BHG system to the USCF Chess system can review US Chess Federation: Rating Calculations.

UPDATE 2: Errr ... that's the old version. Unbeknownst to me, the USCF has modified their rating system. You can read the detailed algorithm here. Turns out the BHG system is EXACTLY like the new USCF system. As far as I can tell. They just have not adopted the "floor" concept.

Comments: Post a Comment