Stuff

[Shanba]

Too tired to say anything else. Unpick it all yourselves.

I did not include games where it was unclear to me whether there were multiple scumgroups. I did not include games where it was unclear which faction won or which ended in a draw. Games in new york or theme park with multiple scumgroups were not recorded with the rest of the data, but I did not check for multiple scumgroups in the mini games. About a quarter of the way through theme park, the final section I did, I suddenly realised I should not be counting SK wins as mafia wins - from then on, I ignored sk wins entirely, but before that sk wins are counted as mafia wins.

Mafia wins are 1, town wins are 0. Samples were chosen by taking finished games off the first page of each forum.

LITTLE ITALY

1101 3/4
0111 3/4 6/8
0111 3/4 9/12
1001 2/4 11/16
1000 1/4 12/20
0100 1/4 13/24
1011 3/4 16/28
1111 4/4 20/32
0010 1/4 21/36
1011 3/4 24/40
1011 3/4 27/44
1 1/1 28/49

57% scum win

19.6% cumulative binomial (p(scumwin = or < 28/49))

NEW YORK
11001010 (2 scumgroups)

1011 3/4
0111 3/4 6/8
1111 4/4 10/12
1001 2/4 12/16
1011 3/4 15/20
0010 1/4 16/24

16/24=2/3 probability of this happening by chance 7.579%

Cumulated so far: 44/73 60% 5.03% cumulative binomial (p(townwin = or < 16/24))

CONEY ISLAND

1110 3/4
0000 0/4 3/8
0100 1/4 4/12
1110 3/4 7/16
0100 1/4 8/20
1111 4/4 12/24
1011 3/4 15/28
0001 1/4 16/32
1000 1/4 17/36
1010 1/4 18/40
01 1/2 19/42

45.23% 32.2% cumulative binomial (p(scumwin = or < 19/42))

63/115 54.78% scum win 17.55% cumulative prob (P(townwin = or < 52/115))

THEME PARK
1100000 (multiple scumgroups)

1101 3/4
0101 2/4 5/8
1111 4/4 9/12
0110 2/4 11/16
1111 4/4 15/20
1111 4/4 19/24

79.167% scum win

OVERALL:
82/139 58.99% scum win
LARGE GAMES:
35/48 72.9167% scum win
THEME GAMES:
38/66 57.576% scum win

These numbers seem high to me.

[Xylthixlm]

I've found that the IRC bot, which automatically balances itself to keep a 50% town win rate, gives setups that I would naively expect are town-favored. This is evidence that might not be just a quirk of irc play.

[Yosarian2]

Meh. 58% scum win really seems pretty close to balanced to me.

It's interesting town does so poorly in large theme games, but then again, I wonder how much of that is related to how there have been a lot of bad idea games and different varients, which town tends to do horribly in, probably more for psycological reasons then for actual "balance" reasons.

[Kmd4390]

My setups always look town-heavy at first glance. Maybe I should leave them how they are instead of trying to do something to favor scum.

[Elmo]

lol @ town winning 20% of Theme Park games. That I can believe. I did not expect town to win more mini themes than mini normal, though.

I should totally get around to finished that thing I was going to write that could automate this. I say this every year

[Shanba]

lol @ town winning 20% of Theme Park games. That I can believe. I did not expect town to win more mini themes than mini normal, though.

Town tends to have a lot more power in mini themes than mini normals, but I don't think the same thing is true with large normals and large themes.

I also excluded a bunch of large theme games - those with multiple scumgroups, suicide bomber mafia, ehobanohar, since it's already pretty clear that bad idea type games massively favour scum.

Serial killers won more large themes than any other setup when I was trawling. Probably just a statistical anomaly, but it might also be that serial killers in large themes get juicier bonuses than in other games.

I should totally get around to finished that thing I was going to write that could automate this. I say this every year

I don't know how such a script would work, tbh. When I was trawling the games, the thing that really annoyed me was the utter inconsistency of the announcement of who won. Sometimes it was in the thread title, sometimes it was on page 1, sometimes there was a link to the final reveal on page 1 and sometimes I just had to trawl through the last couple pages and find the actual ending.

The mini themes are the only ones where town are winning more than scum, though this is a limited data set. I don't know enough statistics to calculate the relevance of these results, unfortunately.

58% doesn't seem like much but it translates to 89 scum wins vs 57 town wins. If Mini themes, the only ones that town won more of, are excluded, we get instead 63/97 wins for scum, or almost two thirds. I think mini themes, bizarrely, are the best balanced of the lot.

If we exclude, instead large themes and normals, we get 47/91, or about 51% - it may just be that it is the large games which are always unbalanced. It surprised me that scum won so many new york games.

Possible theory explanation - a scumgroup only needs one player to survive in order to win the game. A scumgroup is therefore as good as its best player. A town, on the other hand, is much more of a team effort, and although one player can break a game, catch all the scum, if the rest of the town aren't listening to him it's kinda irrelevant. Half the surviving townies have to catch any given scum to get a lynch. A town, therefore, is as good as the mean of its players skill. So the more players you have, the more scum there are, and the better the chances that they have one or more good players who won't be caught.

[ortolan]

wow, yer, looking back over large theme games I thought exactly the same thing; and just observed that in another thread.

By my count I've played in 6 large theme games as town, one was abandoned, we won one (Silent Mafia) and lost the other four. So that is exactly 80% actually. The two large theme games I've played in as scum we've won both.

oh yer, and there was Martyr Mafia, which I wouldn't count.

Did you include SKs as another faction?

[Elmo]

Town tends to have a lot more power in mini themes than mini normals, but I don't think the same thing is true with large normals and large themes.

This is true. I wonder if the "no doubles" meta hurts large games, since you run out of strong unique roles pretty quickly.

I remain really interested by the fact that most of what the modding community generally knows about balancing large (esp. theme) games is de facto incorrect. </provocative>

[Yosarian2]

Well, Little Italy games and Coney Island games both look to be within the statistical margin of error of 50% to me.

Question: How far back does this go?

[Shanba]

Yeah, you're right about the small games, yos.

Games were collected from the most recent finished game I could find to the bottom of page 1 in little italy and coney island, and for new york and theme park it went from the first finished game down until 24 1 scumgroup games had been reached (basically, I got bored at that point.)

[Yosarian2]

I also excluded a bunch of large theme games - those with multiple scumgroups, suicide bomber mafia, ehobanohar, since it's already pretty clear that bad idea type games massively favour scum.

So, we're not properly balancing large theme games with only 1 scum group then, hmmm? Interesting. Then again, I imagine large theme games with just 1 scumgroup often have strange mechanics.

I tend to think kingmaker style games also favor scum, IMHO.

[Shanba]

I also excluded a bunch of large theme games - those with multiple scumgroups, suicide bomber mafia, ehobanohar, since it's already pretty clear that bad idea type games massively favour scum.

So, we're not properly balancing large theme games with only 1 scum group then, hmmm? Interesting. Then again, I imagine large theme games with just 1 scumgroup often have strange mechanics.

I tend to think kingmaker style games also favor scum, IMHO.

Last kingmaker style game was BM's President Maker which is before my data set.

It's possible what you're saying is true, but the trend is also there (if less pronounced) in the new york forum. I should probably have excluded Traitor Mafia too, on second thoughts.

[mith]

This doesn't surprise me all that much... partly because of mod requirements and partly because we just play more small games, most of our mods have a good idea of what's balanced for 12 players, and I expect there is a tendency to try to use the same ratios with larger games - which doesn't work, since scum power isn't linear with group size.

(The 80% Large Theme result is a bit surprising though. I'd be interested in going through them in detail sometime... I'd expect some have a decent balance and some are virtually unwinnable for the town. Might be helpful to have some examples of what doesn't work and why. I also expect a few are just examples of bad/careless play because of weird mechanics.)

[Thok]

I suspect that large games (especially large theme games) also suffer a lot from personality problems. Stuff like frustration with DGB and zwet (nothing personal and I'm probably missing several dozen other similar feuds, but these do tend to come up a lot in games they play) seems like it would generally hurt town more than scum (since those players are more likely to be town than scum.) Large games also have a higher tendency to lurking/stalling, just because they are much longer.

[MonkeyMan576]

The games I play offsite have more town power/minor roles. I've noticed in the game design, game approvers seem over concerned with things being over-balanced in favor of the town, when in fact it seems scum have the advantage with the 58% win margin.

[alexhans]

I've found that the IRC bot, which automatically balances itself to keep Alexhans being scum 80% of the time, gives setups that I would naively expect are town-favored.

Fixed

[Kmd4390]

My wiki numbers fit these stats pretty closely. 65% of my scum games have been wins while I've only won 38% of my town games. I've been thinking I'm just better as scum than as town, but game balance is probably a factor.