Classicists League Table

[InterMPC]

If we use the Elo system I recommend using weightings of 1 for a solo, 0 for a loss and 1/7 for a 7way draw. Everything else should be something in between.

For example if we want to weight it heavily towards a win, you could make all draws worth 1/7. Meaning that you only get significant points if you win a solo, you lose points if you lose, and if you draw you either gain or lose a small amount depending on how you rate compared to the other players.

Elo was designed for average scores around 1500, peeking at around 2500.

[InterMPC]

So,

I'd like to implement something into our table regarding the Elo style system. I have updated the testing spreadsheet to incorporate the k-factor and result weightings that I think are most appropriate.

https://spreadsheets.google.com/ccc?key ... y=COvq2tcK

Increasing the k-factor from 25 to 30 effectively increases the amount you lose or gain no matter what the result.

The results weighting changes I've made increase the benefit of going for a solo. A solo is worth more than just twice what a 2 way draw is worth.

Raphtown, if you are okay with these I'll implement it into our table. Of course this is not set in concrete and if we decide in reality that it doesn't work then we can change it.

The other thing that needs deciding is what people's starting values are to be. I proposed starting everyone off at 1500, others have suggested 500 and others have suggested using stats from the site to calculate starting values(such as win%). How would you like this decision to be made? As debate has died down, perhaps I should just implement one option (1500) and if people have issues they can raise them in this forum to get it changed.

[Mapmaker]

I like the idea of having solos weighted by more than just twice a 2-way draw. And I think that if you're increasing the k-factor, surely average points should increase as well - so everyone would start off at a higher score. Or is that unnecessary?

[raphtown]

If ELO is made for 1500, then start everyone *who plays* at 1500. You can leave the others as blank values or 0 for now.

And go ahead with those weighting changes.

[Ceebs]

Will there be a prominent explanatory post for those not familiar with how this ranking system works?

I for one can't be bothered to flip through all this discussion on different editions of these rankings. Since you people seem to be in the know about these sort of things, i trust that you will come up with something good. When that happens, a layman's terms description in a different thread would be most useful.

[raphtown]

Will there be a prominent explanatory post for those not familiar with how this ranking system works?

I for one can't be bothered to flip through all this discussion on different editions of these rankings. Since you people seem to be in the know about these sort of things, i trust that you will come up with something good. When that happens, a layman's terms description in a different thread would be most useful.

To be honest, I have been trusting Mapmaker and InterMPC with the coding behind the system, and have merely been trying to provide feedback. I am sure one of them will be able to provide a clear and concise explanation of how the system works.

[fellowes]

I like the idea of having solos weighted by more than just twice a 2-way draw. And I think that if you're increasing the k-factor, surely average points should increase as well - so everyone would start off at a higher score. Or is that unnecessary?

Wait. This could have very serious consequences: think about it - most people consider a loss to a solo or a loss by exclusion from a draw (presumably having already been eliminated first), equal losses. This schema seems to suggest that losing to a solo is WORSE than being eliminated and a draw being eventually reached. I think it's hard to make the case that these are not both complete losses, and as such should be scored the same. Moreover, any ranking system will have to be 0-sum, since otherwise we'd be favoring those who play more (otherwise we could just give "bonus" points for solos). So solos won't be able to be worth significantly more than draws, if we want to preserve the idea that a loss is a loss (though, it is still possible to "scale" wins, somewhat).

Two options seem to make the most sense, then, instead. One gives a *slightly* escalating value to solos/draws, and the other treats them all the same. Personally, I prefer option 1 strongly, but I suppose a case could be made for option 2, or some similar system. I go over the details and pros/cons for each below.

Option 1: 7 players enter the game. Anyone who loses (is not part of an N-way draw/solo) is charged 1/N losses (the number of players who won) to each of the winners. This ensures they always lose once and only once per game. The value of a win (or loss) against a certain player will be determined by the ELO formula (this holds for option 2 as well). This allows for scaling of the value of solos/draws in the fashion of the table below

# of winners 1 2 3 4 5 6 7
prop of solo points (per winner) 1 5/12 2/9 1/8 1/15 1/36 0
total "value" of game 1 5/6 4/6 3/6 2/6 1/6 0

So we can see that a draw (that you are involved in) will NEVER lower your rating, no matter how much "better" you are than your opponents. Whether this is good by discouraging ganging up on high-raters or bad by encouraging draws is not clear. It has the scaling of the "value" of the game many people hoped for, discouraging draws, especially many-player draws.

Option 2: 7 players enter the game. EVERYONE "pays" a loss to play, and loses equally amongst. Thus, take the (hopefully very hypothetical) situation of a 2-way draw. The two winners are each credited a half win gainst each loser. Then they each give each other half a win (and half a loss, of course). Thus the lower-ranked player gains significantly more. One possible drawback is that a highly rated player who draws with lower rated ones could very well see his rating drop, despite a decent (say, 3-way draw) finish - conversely, low ranked players have a lot to gain from draw-mongering. Another is that there is no "scaling" for making solos more than 3 times more desirable than a 3-way draw. While I'm fine with this, I'm not sure how many others are. On the plus side, this system is probably more intuitive than the system above.

But it is basically the system above, except that everyone loses to the winners, including the winners amongst themselves, whereas in the first system only the losers lose to the winners.

# of winners 1 2 3 4 5 6 7
prop of solo points (per winner)* 1 1/2 1/3 1/4 1/5 1/6 1/7
total "value" of game 1 1 1 1 1 1 1

* excludes the fact that they also lose some, depending on the size of the draw, as well.

PS As the only guy with a solo on the board, it may seem odd that I'm coming down so hard against rewarding them excessively, but I just don't think it's a teneble system, because it adds at least as many perverse incentives as the site's official system, and is one of the problems I have with another player-run ratings system, in the Diplomacy League here on the forum.

EDIT: ARGH the tables look miserable when posted but great before. It condensed my spaces

[InterMPC]

Raphtown,

I can put together a detailed explanation of the system used once it's locked in concrete. Also I would suggest giving people access to a rating calculator so they can calculate their potential shift before and during the game. This could help them weigh up risk v return for attempting a solo etc.

Fellowes,

You raise some interesting points which I need to look at in more detail later in the day. One thing that may need clarifying about the currently proposed system is that when you lose, you rating changes based on all the players who started the game regardless of their results in that game. That means an elimination is equal to being alive when other people draw or solo no matter how many people draw or their ratings. Wins are dependant on the ratings of all people who started the game and the number of winners. Not the rating of the losers. I wouldn't want to encourage ganging up on the highest rated player.

I look more in depth at you proposal and give my thoughts later today. - I appreciate the input!

[InterMPC]

I think Fellowe’s suggestions are all well thought out and easy to implement.

I think the right thing to do before deciding on what the best method for our League Table should be to gain agreement on what we want it to achieve.

These are the issues that I see as being pertinent.

1. In general there is not enough aggression towards going for the win, or reducing the amount of people in a draw. People would rather do someone a favour who they have grown an affinity with during the game, than remove them to get higher points themselves.
2. People should gain a higher reward for getting a result when they aren’t expected to, or get a higher penalty for not getting a result when they are.
3. We should encourage people to play more games. (Not too many). If everyone was willing to play 2 classicist games at a time it would make it easier for people to find classicists games to play. Therefore improving the value that this group offers.
a. Losing should be worse than not playing at all.
b. It should be possible to overtake someone who worked their way to a high position but then has become stagnant.
4. I see no benefit in a 0 sum system. The Elo system which is widely popular has no need for it, and it is opposed to point 1 above.
5. The solution should be one which is easy to update and maintain, with as little amount of human input as possible.
6. It should be as simple to understand as possible while achieving the above points.

Fellowe’s suggestions are contrary to some of these points (point 3 in particular). So it appears we probably won’t come to a consensus on what the optimum outcome is without understanding the pros and cons of the above points.

For my own personal understanding, can I ask what the benefits are for not favouring people who play more games?

Cheers.

[raphtown]

The trouble with favoring people who play many games over those who only play a few is that the ranking system will no longer be solely based on skill. But then the real question is do we want a ranking system purely derived from player skill level? I would tend to say yes, but it seems a few people have differing opinions.