Does voting analysis work?

[Mina]

I often see people look at multiple vote counts, highlight all the players' names in pretty colours, and make sweeping statements along the lines of "there are two scum among V, W, X, Y, and Z" and "at least one of A, B, and C is scum."

That kind of logical analysis always looks impressive, but I never quite get what the principles and assumptions behind it are.

I believe that you can rule out likely partnerships from looking at how players interact with each other, who they protect or distance from, etc. But how do you know that scum will all evenly distribute themselves among lynch mobs? I find that scum do play good cop/bad cop to a certain extent, but often enough to be able to say categorically that there must be scum on a mob?

Do people find voting analysis to be a better tool than gut reads or partnership analysis, or is it just coincidence and confirmation bias?

Also, what exactly is the trick to analyzing lynch mobs? Just assume that scum spread out their votes? Look for the kind of votes that scum are more likely to make (like swing votes away from scum or late wagon-hopping or bussing)?

[Vi]

First: Pretty colors are key. Don't question the colors.

Scum don't evenly distribute themselves among lynch mobs, and all-Town non-lynch wagons are a bit more common than advertised. However, it's usually an achievement if a lynch goes through with no scum on it; it's not expected.

Scum are typically

less

likely to spread their votes out. It's a coin flip as to whether all of the scum are on a given lynch wagon or not during one of the first two Days.

I've found that trying to go too deeply into vote analysis - who's more consistently late on wagons, etc. - doesn't work unless you're trying to catch newbSKs.

Definitely look for the votes that don't make sense or just feel awkward when looking at wagons.

[Elmo]

In my opinion, not really - looking at votes alone strips out most of the useful information in a game. I'm not sure why that's better than just reading the game. It can also be misleading, inthat if everyone's fairly convinced my scumbuddy is a bad lynch, I'm much happier voting for them than if everyone's currently focussed on someone else but willing to switch; the situation's quite different, but the vote count looks the same.

[Zachrulez]

Doesn't work by itself, but I've found it helpful to quantify scumminess on bandwagons. It can be helpful in separating people who are similarly scummy or to help you see a picture that otherwise looks unclear.

[Kataphraktoi]

Sure it works. A big huge analysis post with tons of colors and graphs and whatnots. Its irresistible and if you repeat it enough, it must be true!

On a serious note, i think the idea is when you have a 8 vote wagon and confirm lets say 6 via death or role interaction then out of two remaining one must be scum for x,y,z reasons. If one of those two are obv townie, viola you got a scum. Well hopefully...

[Yosarian2]

There are types of voting analysis that are very effective.

I, personally, do not think that the type of voting analysis that starts with "There were 5 people on person A, and I'm going to assume that there was at least one scum on the wagon and at least one one scum off the wagon.." is at all effective. It makes too many assumptions; I've seen pro-town people lynched without a single scum on the wagon before. And even if those assumptions happen to be right, it doesn't actually give you much information.

[Ythill]

After seeing some of the veterans doing VC analysis in Geezer Mafia, I've started using it and, so far, it's worked quite well.

I don't think using hard and fast general rules (like scum will always spread their votes) is going to help much, but certain situations narrow the likely vote movements from scum. For example, if a big wagon crumbles and there isn't a new one to replace it, the scum on that wagon are very likley to spread their votes out to multiple targets. Or if scum self-hammer, there is probably a buddy on their wagon, otherwise they'd let him hammer. Basically, having all the confirmed alignments color-coded in the VCs increases your ability to see things from the scum perspective, which allows you to make better deductions.

Also, it's very good for townie cred. In the course of a recent SK win, color-coded VC analysis allowed me to find the last mafia by D2, lead an unlikley mislynch to keep him alive, and look townie enough to win in endgame.

[Ellibereth]

Having done it in almost every game I've been in...I think it works very well.
Have had 2 occasions where it completely blown up in my face, the rest have at least reasonably matched up with eventual flips.

Don't really have any concrete statistics that there is almost always scum on a large wagon and some scum off it and other stuff like, but it's been like that almost every time for me in practice so...

[Vi]

Oh right.

If there are multiple scum groups, all-Town wagons are really uncommon.

[The Fonz]

Yes, it does.

[boberz]

SUrely it is just a more formalised, more categorical way of finding linking tells. THere is no difference between finding an obvious busser in language than in the fact he ends up as the L-2 voter on two lynched scumbags.

Bad oversimplified example I know, but I think it makes sense.

[Yosarian2]

To be honest, I might be biased against it because usually when I've seen it, what happens is DGB makes a huge voting analysis post, goes through every person in the game, does fancy color coding and very carefully thought out logic, and then votes for me.

[Elmo]

There are types of voting analysis that are very effective.

What types?

I'm now mildly curious how anyone actually finds this useful. I just cannot see how this analysis would tell you anything new. Maybe if you had a really bad memory, you'd need parts of the game condensed, or something?

[Battousai]

To be honest, I might be biased against it because usually when I've seen it, what happens is DGB makes a huge voting analysis post, goes through every person in the game, does fancy color coding and very carefully thought out logic, and then votes for me.

Maybe you should stop being scum in games with her.

[Vi]

Maybe if you had a really bad memory

Don't hate

It does help to have a macro-level view of events sometimes though.

[Oman]

Isn't the important thing to remember that it's part of a bigger analysis? So yes it is useful but only a branch of your scumhunting, it can't sustain it on it's own.

[Elmo]

Vi: I am like the anti-hate. It just genuinely surprises me. Maybe in like a 24 player game? But I would still rather spend the time rereading..

[Vi]

Vi: I am like the anti-hate. It just genuinely surprises me. Maybe in like a 24 player game? But I would still rather spend the time rereading..

If you're in a 24-player game, you'll probably be grateful for the vote analysis as a tl;dr, especially if you're playing in some kind of tragic comedy of a game (which is pretty much how 24-player games go).

Also, I have hard enough of a time remembering what people said half a page ago.

[Yosarian2]

There are types of voting analysis that are very effective.

What types?

I'm now mildly curious how anyone actually finds this useful. I just cannot see how this analysis would tell you anything new. Maybe if you had a really bad memory, you'd need parts of the game condensed, or something?

Taking a series of votecounts, and analyzing them to see who was on which wagon at what point can be a good way to analyze the game. Like "When the two big wagons were on person A and person B, who was on which wagon, who joined it when; when wagon A faded out and wagon C took over, who joined it then", ect, especally when one of the three wagons was on scum. It's a good way to notice details you might have missed at the time.

Also, on like day 6 of a large game or something, looking at a whole series of votecounts, it will quickly become obvious that certain people were on a series of bad wagons and not on any of the ones that hit scum. It's not necessarily something you'll notice at the time, if all the votes individually seemed reasonable, but it's a good way to look a the bigger picture and look for patterns.

[Elmo]

Neither of those are what I thought the OP meant, though.. namely analysis based solely on who voted who. Otherwise anything in the game falls under "vote analysis". (Fwiw, as stated I'm still not really sold, but whatever.)

[vollkan]

Do people find voting analysis to be a better tool than gut reads or partnership analysis, or is it just coincidence and confirmation bias?

It's the latter.

No town player gets lynched without town on their wagon. So, simply being on a town lynch wagon cannot be considered a meaningful scumtell. Likewise, I see no good reason for thinking that any particular wagoning pattern is scummy. Somebody can be the L-1 vote on three townie lynches and very easily still not be scum. Of course, that L-1 pattern is definitely a good reason to go deeper and analyse their behaviour and reasoning. But I see no reason for treating it as a scumtell in and of itself.

Note: I think it is probably the case that such an L-1 voter has a greater-than-average chance of being scum. However, that only really means that voting analysis is a rough proxy for actually doing detailed, individual-level analysis.

Also, I am always amused by the fact that voting analysis players can stubbornly insist on reducing mafia to a numbers game while, at the same time, they are never able to justify their approach beyond anecdotal evidence of "I've seen it work a lot". (Much as I can't stand gut players, at least when they justify gut play on the basis of anecdotal evidence they are being

consistently

illogical )

Also, I have hard enough of a time remembering what people said half a page ago.

I have the same problem, which is why I use a scoring system for scumtells.

[Shanba]

As a bit of background, I've personally used vote analysis twice in my mafia career (in the sense of looking through a bunch of votecounts, checking off patterns and looking for scum based on those votecounts. The first time was a long time ago, in a friends and enemies game, and I ended up with 2/3 scum. Here's the post. The second time was far more recently in NY 111 where I identify three players as being exceptionally wagonny and 2 as being fairly wagonny and all three being consistent on being on any wagon that was not on one of them. Ironically, the lynchpin in my case thee (that two of them did not join the wagon on the third) ended up being wrong, because he was the only townie in the group. The other 2 were scum. Here's the post. So I can say I've had some reasonable success with it - but this is anecdotal. My justification for using it is below.

The reason I think it can work is that being on a town wagon is always a scumtell, and being on a scum wagon is always a towntell. In as much as scum want townies to be lynched and townies want scum to be lynched, it is probably inevitable that if you look at a player's overall career, they will end up on more scumwagons when they are town than when they are scum. Of course, there's the playstyle aspect - for a player who is a poor scumhunter and a heavy busser it's going to be skewed, but I think that ought to hold for the populace as a whole. Which makes it by definition a scumtell to be on a town wagon, and it's fair to assume that anyone on a town wagon has a higher than average chance of being scum.

That's also the reason I have a problem with this logic:

No town player gets lynched without town on their wagon. So, simply being on a town lynch wagon cannot be considered a meaningful scumtell. Likewise, I see no good reason for thinking that any particular wagoning pattern is scummy. Somebody can be the L-1 vote on three townie lynches and very easily still not be scum. Of course, that L-1 pattern is definitely a good reason to go deeper and analyse their behaviour and reasoning. But I see no reason for treating it as a scumtell in and of itself.

and this logic:

I've seen pro-town people lynched without a single scum on the wagon before.

Just because you have factors that make it less of a scumtell don't make it not a scumtell (and especially in yos' case here, the case against is essentially anecdotal.) The caveat I'd like to add that I've used against certain types of scumtell (notably the anti-town = scummy thing) is that sometimes a scumtell can be so reduced as to be negligible - again I don't think that's the case here, as I think the motivation for scum to lynch a townie is big enough that they'll be mainly trying to lynch townies in pretty much every game. And that's all that's required for this to be a scumtell.

I agree you can be smart about voting analyses - for example, rather than going in with a set of preconceived notions (scum must do this, scum must do that) you simply take all the votecounts and then look for patterns (and being a human, you're exceptionally good at this - actually too good, because we as humans often see patterns that aren't there. But we can afford some mistakes as a town player.) And I think it's always good to back up this kind of analysis with normal behavioural analysis. But I don't think that means we should cut a legitimate source of scumtells from our game entirely.

[Ectomancer]

It works, but understand that a single vote count is not enough, much like a single observation is not enough to determine a planet's orbit.

[Ythill]

ITT vollkan demonstrates that he doesn't know what the OP is actually talking about. True, wagon-position tells are not logical. IME, VC analysis is, and is more objective than what I've seen of your scoring system. Here's one of mine, as an example in a game where Fate was the last scum (ignore my SK hunting, I was the SK).

I think they work because they act as a summary of past game states and so allow a global view of everyone's play and, like I said, they help you to see things from the scum perspective as more alignments are added through cardflips.

[Vi]

Also, I am always amused by the fact that voting analysis players can stubbornly insist on reducing mafia to a numbers game while, at the same time, they are never able to justify their approach beyond anecdotal evidence of "I've seen it work a lot".

(...)
I have the same problem, which is why I use a scoring system for scumtells.

Um...