An Alternative Tournament Structure: The Rolling Rumble

[ollie]

Originally Posted by RobertDD

By the way, Ollie, I wasn't suggesting win/loss percentage as a tie breaker, but just as the first and only way of determining the outcome of the tournament.

That's how I was interpreting it: badly phrased on my part, sorry. My internal representation has everyone starting tied and so I think of "primary tie-breaker" and "first way of determining the outcome of the tournament" interchangeably.

[dok]

Originally Posted by RobertDD

If you count points left on the board (1 players' army, because you played to the end) and assign that amount as a positive number to the winner, and as a negative number to the loser, you could use that as the 2nd element in your scoring system, or even drop win/loss altogether and just use this metric. The d20 might still be nescessary, but it becomes much less likely. You might discard outcomes of unfinished games at the end of the day, or slightly complicate scoring: points left on the board for you - points left on the board for your opponent.

I like all of this. I actually prefer counting the differential in points killed, as this means that people playing armies that are 5 or 10 points short are not automatically penalized 5 or 10 points every time they win. Similarly, if you eliminate an opponent, you automatically get credit for killing the maximum points allowed at the event (so the opponents of smaller armies aren't penalized either).

This approach does complicate things when it comes to soul devour/mindshackle/stab in the back, as you not only count them as a "kill" for the figure-stealing player, but you have to subtract surviving turncoat figures from the kill total for the betrayed player (basically, you have to double-count the turncoat). But overall, this is slightly more fair.

Is there a link anywhere to the "standard" partial scoring method?

[dok]

Originally Posted by ollie

Being able to avoid counting points is one of the big plus-points for me with this system. I really don't think points remaining at the end of a victory are a very meaningful measure of much.

I guess that's a philosophical difference as much as anything. I tend to think it is pretty meaningful, at least in a statistical sense. I see point differential as being strongly correlated with a posteriori win probability. People who are winning their games in blowouts are people who weren't likely to lose those games, even if they had had some lousy breaks with the dice.

You will of course have exceptions - RobertDD's Kato vs. Q9 is a perfect example. But that's really a function of the time limits. If you play that out, you'll most likely end up with a wounded Q9, and 90 points for a 2-wound Q9 standing at the end sounds about right to me.

I would be happy using some function of point differential as oppose to pure point diff. The idea would be to try to make the function nearly linearly related to the probability (over 50%) of the same player winning if that match happened again. Throwing point diff on a log scale seems like an easy approximation. To pick something out of a hat, say winner gets log(1+diff/50): so winning by 50 would be half as good as winning by 150, which would be half as good as winning by 350.

Originally Posted by ollie

Interesting stuff. It could work very well in this system, though I see a couple of issues:

a) It won't be intuitively clear to everyone how the outcome of a game changes the rankings, and so you lose the clear "win is this good, loss is this bad" feeling of an objectively less good system.

b) It requires that the TD does not play. If I'm going to the trouble of organising a tournament, I want to play some Heroscape!

I think that (a) sort of goes to the same "false sense of a clear result" issue that we both agree on.

As far as (b) goes... I think that if I put some effort into it, I could code a user interface that would be pretty easy to use. Enter player 1 ID number, enter player 2 ID number, enter point diff (P1 - P2), verify that everything is right ("You are entering that Ollie defeated dok with a point differential of 75. Is this correct?") and out spits a new updated ranking list. So, I suppose it could be set up to run on its own.

EDIT: I got that first-order update wrong. I was trying to make it simple, but it actually breaks on edge cases. So I need to do a real weighted average, which is actually a bit more complicated:

New ranking = old ranking*(games played-1/games played) + 
(old ranking - opponent's ranking - game point diff)/(games played)

Seems complex enough that you'd want it done on a central laptop, rather than having people do it by hand.

[ollie]

Originally Posted by dok

<fascinating nerdy goodness>

While I love this type of stuff as a player and observer of Heroscape (see, for example, the NERV), as a TD I really don't like it. Ease and clarity easily trump everything else, as long as winning is at least roughly correlated with finshing higher up the standings. I want people to be able to weigh the importance of a game before/while it is happening if at all possible (at the "If X wins this then she goes into the lead; if Y wins he'll pull into equal first place" level of analysis).

Analysing events in this way after the fact could be interesting though.

EDIT:

Originally Posted by dok

I guess that's a philosophical difference as much as anything. I tend to think it is pretty meaningful, at least in a statistical sense. I see point differential as being strongly correlated with a posteriori win probability. People who are winning their games in blowouts are people who weren't likely to lose those games, even if they had had some lousy breaks with the dice.

I agree. Where we differ is, I suspect, in what we think the tournament structure should be measuring. In my view, you should progress in the tournament for actually winning; the probability of you doing so (based on your skill, army, match-ups etc) is interesting, and someone for whom these numbers are higher should do better, but on the day itself you win and lose for whatever reason and the structure should reward the winners. The unlucky dice are just as good a reason to fall down the rankings as bad play.

[dok]

Originally Posted by ollie

I agree. Where we differ is, I suspect, in what we think the tournament structure should be measuring. In my view, you should progress in the tournament for actually winning; the probability of you doing so (based on your skill, army, match-ups etc) is interesting, and someone for whom these numbers are higher should do better, but on the day itself you win and lose for whatever reason and the structure should reward the winners. The unlucky dice are just as good a reason to fall down the rankings as bad play.

You've just nicely summarized the difference between "descriptive" metrics and "predictive" metrics. Both have value, of course.

[dok]

Originally Posted by dok

I like all of this. I actually prefer counting the differential in points killed, as this means that people playing armies that are 5 or 10 points short are not automatically penalized 5 or 10 points every time they win. Similarly, if you eliminate an opponent, you automatically get credit for killing the maximum points allowed at the event (so the opponents of smaller armies aren't penalized either).

This approach does complicate things when it comes to soul devour/mindshackle/stab in the back, as you not only count them as a "kill" for the figure-stealing player, but you have to subtract surviving turncoat figures from the kill total for the betrayed player (basically, you have to double-count the turncoat). But overall, this is slightly more fair.

It occured to me that the above "points killed differential" is actually equivalent to "points surviving differential", if any player with figures left is given a bonus for however many points lower than the maximum allowed they were. So if you started with 495 points in a 500 point event, and you have figures left at the end of the game, you add 5 to your partial score.

I suppose (if you want them to be equivalent) you also need to also add the proviso that undropped AE count for 110 points if the player still has other figures on the board, and same for Rechets if the player has Iskra on the board.

So, this way of looking at it has its own quirks, but it avoids the extra book-keeping for mindshacklers.

[ollie]

Originally Posted by dok

It occured to me that the above "points killed differential" is actually equivalent to "points surviving differential", if any player with figures left is given a bonus for however many points lower than the maximum allowed they were. So if you started with 495 points in a 500 point event, and you have figures left at the end of the game, you add 5 to your partial score.

I suppose (if you want them to be equivalent) you also need to also add the proviso that undropped AE count for 110 points if the player still has other figures on the board, and same for Rechets if the player has Iskra on the board.

So, this way of looking at it has its own quirks, but it avoids the extra book-keeping for mindshacklers.

I don't think you should add that 5, particularly when deciding the winner of an unfinished game (I care less about what points get carried forward into the tournament structure). The principle in play when deciding an unfinished game is that we are trying to determine who would go on to win. Any method, as you say, should be equivalent to points-surviving-differential.

While no-one is claiming that having the most points remaining is an especially reliable indicator, I haven't heard a better one. (Apart from this one .)

[dok]

Originally Posted by ollie

I don't think you should add that 5, particularly when deciding the winner of an unfinished game (I care less about what points get carried forward into the tournament structure). The principle in play when deciding an unfinished game is that we are trying to determine who would go on to win. Any method, as you say, should be equivalent to points-surviving-differential.

I think this comes down to whether you think the goal of a game is to have your army survive, or to kill your opponent's figures. These are more or less equivalent in any game that is completed, but not in games that are called on time.

If you think the goal is to kill opposing figures, then "kill differential" would be your incomplete game standard. If you try to back that out to an isomorphic "surviving point differential" measure, it requires adding in 5 points for a 495 point armies. If, on the other hand, you think the goal is to survive, then you wouldn't support the add-on points, except perhaps when calculating point differential on a completed game (i.e. when one army is entirely wiped out).

Originally Posted by ollie

While no-one is claiming that having the most points remaining is an especially reliable indicator, I haven't heard a better one. (Apart from this one .)

That's another way of looking at the "what's the goal of a game of heroscape" question. If you believe that having more points when a game is called on time is meaningless, then draws make sense. Of course that only applies to a descriptive measure; a predictive measure would use point differential even if it was considered meaningless in and of itself.

[ollie]

Originally Posted by dok

I think this comes down to whether you think the goal of a game is to have your army survive, or to kill your opponent's figures. These are more or less equivalent in any game that is completed, but not in games that are called on time.

If you think the goal is to kill opposing figures, then "kill differential" would be your incomplete game standard. If you try to back that out to an isomorphic "surviving point differential" measure, it requires adding in 5 points for a 495 point armies. If, on the other hand, you think the goal is to survive, then you wouldn't support the add-on points, except perhaps when calculating point differential on a completed game (i.e. when one army is entirely wiped out).

I'm still disagreeing, I think.

Beyond my fundamental point that it's who has the most points on the board when time is called that is more likely to go on to win, whether you think killing or surviving is the point of the game, I see those missing 5pts as already killed. If your opponent doesn't bring them, you can't kill them. In the same way that someone who leaves a rat off the board gives her opponent a 10pt head start for "killing" that rat, if someone comes with just 495pts he gives his opponent those 5pts in any reasonable system as far as I can see.

[dok]

Originally Posted by ollie

Beyond my fundamental point that it's who has the most points on the board when time is called that is more likely to go on to win,

You sort of toss this out in passing, but it's the most convincing argument to me. If a goal of the scoring metric in an incomplete game is to correlate well with who would have won that game had time not been limited, then points left on the board is obviously more meaningful than points left on the board plus difference between maximum points and starting points.

Originally Posted by ollie

whether you think killing or surviving is the point of the game, I see those missing 5pts as already killed. If your opponent doesn't bring them, you can't kill them. In the same way that someone who leaves a rat off the board gives her opponent a 10pt head start for "killing" that rat, if someone comes with just 495pts he gives his opponent those 5pts in any reasonable system as far as I can see.

I certainly understand this argument, and it's definitely reasonable, but I don't see it as the only logical way to think about those 5 (or 10) points.

[ollie]

Originally Posted by dok

You sort of toss this out in passing, but it's the most convincing argument to me. If a goal of the scoring metric in an incomplete game is to correlate well with who would have won that game had time not been limited, then points left on the board is obviously more meaningful than points left on the board plus difference between maximum points and starting points.

Ahhh, then that's where the problem was. That is the bedrock of my argument---looking back I see that I didn't express myself clearly, never making the explicit connection between determining who would go onto win and the points on the board. Sorry about that.

And, to attempt to steer the thread back towards the topic of the proposed system which essentially does away with the need for this (or, at any rate, relies on it a lot less than all of the standard structures (except one---I won't provide the link this time )) I'm planning to reformulate a few things based on all of this excellent discussion and have a more polished set of guidelines soon. My intention is to use this format at a tournament in the Fall.

[Messenger]

Originally Posted by ollie

Rather than win percentage, how about number of wins minus number of losses? (Win difference? Does this have a name?) This means that someone who goes 6-1 beats someone that goes 4-0, which I think is how it should be. Number of wins (or, equivalently, number of games) would still be the second splitter (So 7-2 beats 6-1). D20 to split equal records at the prize table.

I like this. It solves the 'stalling' tactic which could be utilized if win percentage were the winning criteria.
**If I win two games slowly (100%)I beat the guy who went 6-1 (86%)**
That way of thinking is lame. Your system defeats that.




Consider having another splitter being kill points per game (Kpg). What's that?

Kpg = Total points killed in all games / # of games played.

Come to think of it, the Kpg method sounds better than the traditional point differential method. It levels out the 'blow out games' and makes the 'close' games not as detrimental to overall scoring...hmm.