Mafia Elo System - Thoughts so far? | Mafia

#1

My personal view is that I think we'll be able to assess the results better after 20-30 games, but the results seem to make more "sense" than I would have expected at this point, I guess. I considered having player votes on performance, but while I think this would improve short-term results, it introduces other sources of bias that I think are more likely to be avoided long-term in a system that only rewards wins/penalizes losses. It's also difficult to noob snipe, because in contrast to debating, there's a lot more people in each game, and you have to play a lot of games to improve your score significantly.

What are your general thoughts, and what do you expect the rankings to be after some large number of games—say, after 100 games or so?

I didn't know who was going to be on top, but Pie makes sense to me. I expect him to stay pretty high up (even if an 83% winrate likely isn't sustainable). The gap between him and the second-place spot on the leaderboard seems telling even this early on. I think Mikal and Lunatic are underrated at the moment (just got unlucky with the small sample size) and will end up at least above average. As a general rule, I think experienced players being consistently at the top will be a sign that the system is becoming more accurate. As for how they'll rank against each other, I think that's harder to predict.

Edit comment

Mentions Text*

My personal view is that I think we'll be able to assess the results better after 20-30 games, but the results seem to make more "sense" than I would have expected at this point, I guess. I considered having player votes on performance, but while I think this would improve short-term results, it introduces other sources of bias that I think are more likely to be avoided long-term in a system that only rewards wins/penalizes losses. It's also difficult to noob snipe, because in contrast to debating, there's a lot more people in each game, and you have to play a lot of games to improve your score significantly. What are your general thoughts, and what do you expect the rankings to be after some large number of games—say, after 100 games or so? I didn't know who was going to be on top, but Pie makes sense to me. I expect him to stay pretty high up (even if an 83% winrate likely isn't sustainable). The gap between him and the second-place spot on the leaderboard seems telling even this early on. I think Mikal and Lunatic are underrated at the moment (just got unlucky with the small sample size) and will end up at least above average. As a general rule, I think experienced players being consistently at the top will be a sign that the system is becoming more accurate. As for how they'll rank against each other, I think that's harder to predict.

#2

--> @Savant

I think the only think I noticed is

town wins don’t count a lot

mafia wins count a lot

town loses also don’t count a lot

but mafia loses count a lot

im assuming it’s the number of players in the game

but if you had 2 mafia wins compared to someone who had two town wins my assumption is that the mafia wins would weigh significantly more

Edit post

Mentions Text*

#3

--> @Savant

If I understand correctly, a player's Elo rating is equivalent to their expected winrate against a player whose Elo rating is a certain amount relative to theirs, e.g., 150 points lower or 300 points higher. Since Mafia is a team game with a large random component, I doubt that any player will ever have a win-rate above a certain percentage that's nowhere close to 100%. So isn't the possible Elo range effectively restricted to a few hundred points on either side of 1500?

Edit post

Mentions Text*

Forum signatures are bullshit. Nobody wants to waste their time and bandwidth reading something they have read 29,998,559,671,349 times before.

#4

I think it’s fine. I do think we should get a policy in place for replacements though.

Edit post

Mentions Text*

#5

--> @Mikal, @Thoth

if you had 2 mafia wins compared to someone who had two town wins my assumption is that the mafia wins would weigh significantly more

That's correct. Keeps total elo balanced, and also I think it's more telling how you perform when there are fewer other team members to dilute your ability. This kind of expects that you perform similarly as scum and as town, which feels necessary if we're going to have a single elo score. It may be an oversimplification, but if I need to change the system later, I can retroactively rescore everyone so it's consistent.

a player's Elo rating is equivalent to their expected winrate against a player whose Elo rating is a certain amount relative to theirs, e.g., 150 points lower or 300 points higher. Since Mafia is a team game with a large random component, I doubt that any player will ever have a win-rate above a certain percentage that's nowhere close to 100%.

Elo here is actually an estimation of the winrate of a team with an average elo of some value. So if I'm 1600 and you are 1500, it estimates that a team made of me and other people at my skill level would beat a team of you and others at your skill level a certain amount.

So isn't the possible Elo range effectively restricted to a few hundred points on either side of 1500?

I'd expect so if opponents are just random. However, once the spread gets large enough, we may start having people play rated against others with similar elo. That could allow for a bigger range of scores. There's also a decent amount to be gained by carrying a team of others with low elo (which allows your elo to rise faster). But your team is basically random, so you should expect to be paired up with people of average elo within the group.

Edit post

Mentions Text*

<blockquote>if you had 2 mafia wins compared to someone who had two town wins my assumption is that the mafia wins would weigh significantly more</blockquote>That's correct. Keeps total elo balanced, and also I think it's more telling how you perform when there are fewer other team members to dilute your ability. This kind of expects that you perform similarly as scum and as town, which feels necessary if we're going to have a single elo score. It may be an oversimplification, but if I need to change the system later, I can retroactively rescore everyone so it's consistent. <blockquote>a player's Elo rating is equivalent to their expected winrate against a player whose Elo rating is a certain amount relative to theirs, e.g., 150 points lower or 300 points higher. Since Mafia is a team game with a large random component, I doubt that any player will ever have a win-rate above a certain percentage that's nowhere close to 100%.</blockquote>Elo here is actually an estimation of the winrate of a team with an average elo of some value. So if I'm 1600 and you are 1500, it estimates that a team made of me and other people at my skill level would beat a team of you and others at your skill level a certain amount. <blockquote>So isn't the possible Elo range effectively restricted to a few hundred points on either side of 1500?</blockquote>I'd expect so if opponents are just random. However, once the spread gets large enough, we may start having people play rated against others with similar elo. That could allow for a bigger range of scores. There's also a decent amount to be gained by carrying a team of others with low elo (which allows your elo to rise faster). But your team is basically random, so you should expect to be paired up with people of average elo within the group.

#6

--> @ILikePie5

I do think we should get a policy in place for replacements though.

My policy so far has been hoping it's rare enough not to affect the scores too much. But there's probably a better way, like treating them together as a single player and splitting the gain/loss. Could say that leaving early is just an auto loss for that player, but that seems a bit harsh and kind of penalizes their team if there's no replacement.

Edit post

Mentions Text*

#7

--> @ILikePie5

I think it’s fine. I do think we should get a policy in place for replacements though.

Agreed. I lost Elo in one game because I played in two slots. One slot was town and the other mafia. My elo got dinged because town, my original team, lost. Honestly if that’s the case then when I slide in from town to mafia then what’s the point in playing to the new mafia win condition? Elo would just encourage me to out the entire team and sabotage the mafia team.

Edit post

Mentions Text*

"Kohai, your entrance to the thread was legendary, you had a solve and a winning PoE right when you showed up, super impressive stuff, even more so considering it wasn't even your objective!!" - Zrein

#8

Don't care about elo too much.
For replacements, have the new faction affect elo.

Edit post

Mentions Text*

I still don’t know what to type!

#9

Seems fine to me, if it included Dart elo would be awesome but I understand that would be a lot of work

Edit post

Mentions Text*

In loving Memory of Lannan and Ajab, fallen debaters.

#10

--> @Lunatic

It's possible to include DART, but I'm not sure that would be fair since ELO wasn't being recorded on DART.

Edit post

Mentions Text*

#11

It's sad to me that real life finally genuinely is busier than ever before, I really am becoming an actual man with responsibilities and now would have been the time I could have gone down in history as the GOAT Mafia player stats-wise. What a shame.

Edit post

Mentions Text*

The US is a definitively secular nation, founded on complete separation of the State and religion. The Left Wing of US realise this but don't realise that pure freedom requires as low tax and free access to guns as possible. The Right Wing of US don't realise that secular pure freedom means the LGBTQ+ and abortion agendas both end up in line with US's liberty-first ethos.

It is left-wing to be pro-life. You are backing poor babies being born at the inconvenience of the already-born. Go figure.

#12

--> @Lunatic, @Savant

When the archives are done I can run some stats

Edit post

Mentions Text*

I still don’t know what to type!

Conversation

Edit comment

Edit post

Edit post

Edit post

Edit post

Edit post

Edit post

Edit post

Edit post

Edit post

Edit post

Edit post