Re: Re: Owls and Voles --- Rankings

I should clarify - the ranking will be entirely based on final score, and
have nothing to do with a player NMRing or how communicative they are etc.

For scoring, I'm planning on using the same system Thorin used for the Owls
games which was:

* 35 points for an 18+ centre VICTORY
* 1/4 point each year of survival in ELIMINATION or LOSS (maximum of 2
points)
* 1/2 point per centre for survivors in a LOSS (OWLS variation)
* 1 point per centre in a DRAW
* 3 points for surviving in a DRAW
* 1 point for being the outright leader in a DRAW
* Score x1.0 for Open Games; Score x1.25 for Invitational Games (OWLS
variation)

Though that might be subject to change if I feel that the numbers aren't
quite right.

From the final scores, players will be ranked in score order for that game,
and it's that ranking of highest to lowest score that is actually used as
the input to the ranking algorithms I'll be using.

For those not familiar with Elo or TrueSkill, it works on the basis of two
numbers for each player. Firstly, you have the player's rating which is a
number in a preselected range (I'd be tempted to use 1-1000). Then you have
a number which represents how certain you are that that rating is accurate.
With a new player, that number will be large to represent the fact that his
skill level is as yet unknown, the more games you play, the more that value
will change to represent the increasing chance that the rating is correct.

The two numbers don't represent a whole lot by themselves, but what they do
do is let you compute the probability that Player A will beat Player B in a
game. If the rating and certainly factor are the same, then each player has
a even chance of winning, and as the numbers diverge, it represents one
player being more likely to win than the other.

That last part is important for how the whole system holds together. After
each game, the score order rankings for that game are used to adjust each
player's rating and certainty factor, but the amount by which they are
adjusted depends to a large extent on the difference in rating between each
player such that the ratings go up or down based on whether or not you
outperform your expected result in a game. In simple terms, what is means
is that if a low rated player were to lose against a high rated player, the
neither player's rating would change by very much, since the higher rated
player would have been expected to win anyway. Conversely, where the highly
rated player to lose against the low rated one, you'd expect a large rating
adjustment to both players, since the result runs contrary to the expected
outcome.

Since the magnitude of a change in rating is entirely depended on the
relative skill levels of the people you're playing against, a lot of the
incentives to game the system are removed - and if they're not, it's easy
enough to add disincentives to people trying to abuse the system - such as
repeated NMRing to the extend that the player is resigned from the game and
replaced causing an identical rating change to if the player had got the
lowest score in the game.

Sorry for the long post, and I hope that it all made sense!

Ed



2012/1/4 Matthias Görgens <[email protected]>

> **
>
>
> I welcome a ranking system for groupings players into interesting
> games. I agree that we should be careful about altering incentives.
> Thus I recommend making punctuality in handing in orders and
> communications a big part of the score: Because we actually want
> people to alter their behaviour so as not to NMR and to talk to their
> fellow players. This will also blunt the impact of whatever choice
> you make of how to value solos vs draws vs survivals vs eliminations.
>
> Just limit the points you get for talking to one message to each
> surviving player per season. Players should of course write more of
> that in general, but they shouldn't flood their fellow players just
> for the ranking.
>
> Also, when announcing the score, make explicit that it is about
> grouping players into enjoyable games, and not a judgement about
> skill. To drive the point home, perhaps keep a second ranking, that
> only includes game results and not punctuality and press.
>
> For the good-person-to-play-with ranking, you might even ask players
> who they'd like to play with again at the end of a game. Just don't
> weight that too highly in the final score to avoid a popularity
> contest.
>
> What do you think?
>  
>