From britdisc-owner@csv.warwick.ac.uk Mon Jun 8 22:13:02 1998 Received: (from daemon@localhost) by pansy.csv.warwick.ac.uk (8.8.7/8.8.8) id WAA28201 for britdisc-outgoing; Mon, 8 Jun 1998 22:00:23 +0100 (BST) Received: from wol.ra.phy.cam.ac.uk (bee.ra.phy.cam.ac.uk [131.111.48.99]) by pansy.csv.warwick.ac.uk (8.8.7/8.8.8) with SMTP id WAA28196 for <britdisc@csv.warwick.ac.uk>; Mon, 8 Jun 1998 22:00:22 +0100 (BST) Received: by wol.ra.phy.cam.ac.uk (UK-Smail 3.1.25.1/15) id <m0yj91h-00001dC@wol.ra.phy.cam.ac.uk>; Mon, 8 Jun 98 22:00 BST Message-Id: <m0yj91h-00001dC@wol.ra.phy.cam.ac.uk> Date: Mon, 8 Jun 98 22:00 BST From: mackay@mrao.cam.ac.uk (David J.C. MacKay) To: britdisc@csv.warwick.ac.uk Subject: Tournament ranking systems Cc: mackay@mrao.cam.ac.uk Sender: owner-britdisc@warwick.ac.uk Precedence: bulk Hi all, I welcome Reservoir Fish's message about tournament ranking systems. First off, I would like to apologise again to Fish for (a) not making sufficiently clear at CAM98 that all game scores were taken into account in inferring the final ratings; and (b) for cocking up one of the entries in the computer so that the announced rankings were wrong; and (c) for not ensuring that the *closeness* of the ratings was announced: we should really have called it a tie for 3rd place between SBVarsity, Fish and Mud Culture. Here are the correct ratings again: # Team Infera rating 1 nf 11.1 +/- 0.2 2 wh 10.9 +/- 0.2 3 sv 10.3 +/- 0.2 (10.29 +/- 0.17) SBVarsity 4 rf 10.3 +/- 0.2 (10.27 +/- 0.18) Reservoir Fish 5 mc 10.2 +/- 0.2 Mud Culture 6 st 9.8 +/- 0.2 7 al 9.6 +/- 0.2 8 sl 9.4 +/- 0.2 9 sd 9.3 +/- 0.2 10 dg 9.2 +/- 0.2 So, to the discussion: The Fish have hit the nail on the head -- one of the key issues is upsets: `lower' teams ending up above `higher' teams because either (a) they pull off one excellent unexpected win, or (b) because the initial seedings were wrong and put too many good teams in one pool. The question is, what should then happen? In traditional tournament formats, the upset team proceeds to do well in its games in the plate division, say, and maybe wins the plate, and is declared to have come 9th; and everyone suspects that they are actually better overall than the teams in positions 7 and 8. The lucky `lower' team ends up with a nice final rank, and the unlucky higher team, no matter how well they play on sunday, are stuck downstairs. If you like this system, please stick with it and forget about infera. I wrote infera because I thought that this system sometimes was unfair and that there should be an alternative. The idea of infera is to offer an alternative in which a team's ability is evaluated based on *all* its games. So when infera is used, you can survive one upset game as long as you show excellent form in the rest of the tournament. A bit more about infera ----------------------- Infera has the advantage that it can be used with any tournament format. I believe it will give sensible answers in any tournament structure. It does NOT depend on any assumption that pools are equal; it can give ratings to teams that play only a few games (their ratings will have larger error bars on them); it appropriately weights the outcomes of long games more heavily than short games (a 20-10 victory contains a lot more information than a 2-1 victory); and it allows you to invent new tournament formats, in which, for example, not all the games in a round robin need to be played. What I really like about infera is that it rewards the effort of low-ranking teams when they get a few points against teams that they don't beat. If for example, team Fish gets 5 points off Shotgun in a game to 15, and team Blue gets only 1 point in their game with Shotgun, infera infers that Fish is probably a bit better than Blue. In most tournaments, all that effort to get those 5 points just disappears without trace. I think infera is fair. But I agree with the Fish that it removes the epic concept of that crucial single sudden death point that determines whether you go up or down. Yours, David For more about infera , see here: http://wol.ra.phy.cam.ac.uk/ultimate/infera/