VCA Stats

[goodmorning]

No gm, that is 8% of total lynches, not just town lynches.

There it is, my troubles return.
There's a reason I'm majoring in econ and not stats.

On an unrelated note, I'm surprised how often BP's get lynched. Seems like it would never be a good idea to lynch a BP without a counter.

Well, given that you can't confirm a BP...

[Lowercase]

Sure, but these were day one. 75% of the time a BP-fakeclaim is counterable even in a goons setup. While it's possible for it to be a fakeclaim, you are more likely to lynch correctly if you target someone else.

A cop might not counter, in this case; but assuming he claims before lylo, the BP-claimer is a dead man walking.

edit: Actually now that I think of it, I guess the BP is more likely to be scum than a random lynch if mafia always fake-claims BP in a goons setup.

[goodmorning]

Could have been anything: deadline, flaker, derphammer, accident, "accident," or maybe they got CC'd by Scum and Scum was more believable.

[PeregrineV]

Thanks for this, very cool.

I'm seeing in Newbie games, if town is lynched day1, you probably have at least one scum on the wagon.

[Cogito Ergo Sum]

I'm seeing in Newbie games, if town is lynched day1, you probably have at least one scum on the wagon.

This is technically true, but given random votes, the odds would be 89% that you'd have a scumbag on a town lynchwagon anyway, so this is also not informative.

[TierShift]

No gm, that is 8% of total lynches, not just town lynches.

There it is, my troubles return.
There's a reason I'm majoring in econ and not stats.

Okay, let me try to explain.

To get the percentage of town lynches without scum on them, you need to divide
the total number of town lynches wthout scum on them
by
total number of town lynches.
Okay?

I cannot find the total number of town lynches or the percentage of town lynches and I took a wrong percentage instead.

[Lowercase]

I cannot find the total number of town lynches or the percentage of town lynches and I took a wrong percentage instead.

I count 50 town lynches and 6 without scum on them. So 88% of town lynches have at least one scum vs 89% expected by chance.

[Catbug]

If you're only looking at "is there scum on this town lynch wagon" then yes, it is worthless, but it's also not vote count

analysis

. Analysis includes looking at many, many other factors regarding the wagon.

[Lowercase]

I guess, but the statistics we have imply that a day one wagon on town in matrix6 is not informative. There may be a relation on later days, i.e. scum are more likely to be on/off wagon both times or sth; but I doubt that is the case and even if it is informative it's not clear how informative it is.

If the D2 wagon is similarly random, then VCA is basically just making arbitrary groupings and divining scum from it.

In larger games there may be some basis to VCA, but I'm not sure why that would be the case.

[Catbug]

VCA is basically just making arbitrary groupings and divining scum from it.

that's not at all what VCA is...

VCA involves looking at timing, reasoning, momentum, placement, associations, and tone of the votes/vote posts, among several other factors, as well as applying those factors to the non-votes in order to get a better picture of what occurred. It's not just "oh hey X voted Y so X must be scum..."

[Catbug]

And if you're saying that town lynch wagons are 89% likely to contain scum, then the actual

vote count analysis

is analyzing scum 89% of the time.

That IS significant.

[Lowercase]

Well sure, the manner in which someone votes is definitely relevant. I was thinking stuff like the scumputer, and methods that take into account mostly wagon participation.

p-edit: What it means is that if I say {A,B,X,Y,Z} were on the wagon so one is probably scum, that is no more useful than saying {A,B,C,D,E} are the first players alphabetically so one is probably scum. Sure, I''d be right most of the time, but I'd just be hiding a meaningless statement behind a veneer of statistical insight.

Btw, I will not lynch outside of {A,B} today; let's not have another day 1.

[Catbug]

lol I understand

[goodmorning]

No gm, that is 8% of total lynches, not just town lynches.

There it is, my troubles return.
There's a reason I'm majoring in econ and not stats.

Okay, let me try to explain.

To get the percentage of town lynches without scum on them, you need to divide
the total number of town lynches wthout scum on them
by
total number of town lynches.
Okay?

I cannot find the total number of town lynches or the percentage of town lynches and I took a wrong percentage instead.

No, I got it, I just meant that I was of course doing something wrong. Total # of Town lynched can be found in the first chart: 140.

So it would be ~12.86% of Town lynches in Newbies having only Town on them, which is to say ~87.14% with at least one Scum. Given that approximately 31.43% of Town lynches in Newbies have both Scum on them, that's about 55.71% with 1 Scum.

[TierShift]

Which is extraordinarily close too random...almost too close.

[PeregrineV]

I'm seeing in Newbie games, if town is lynched day1, you probably have at least one scum on the wagon.

This is technically true, but given random votes, the odds would be 89% that you'd have a scumbag on a town lynchwagon anyway, so this is also not informative.

But I like the solid evidence as opposed to probabilities. Plus, if I am town and on the wagon, I only have 4 or 3 others that I need to narrow it down to.

And then presuming the remaining scum is NOT on the wagon, I have to look between those 2 or 3 OFF the day1 wagon.

I think it will help, but we'll see.

[Scarab]

I'm seeing in Newbie games, if town is lynched day1, you probably have at least one scum on the wagon.

This is technically true, but given random votes, the odds would be 89% that you'd have a scumbag on a town lynchwagon anyway, so this is also not informative.

But I like the solid evidence as opposed to probabilities. Plus, if I am town and on the wagon, I only have 4 or 3 others that I need to narrow it down to.

And then presuming the remaining scum is NOT on the wagon, I have to look between those 2 or 3 OFF the day1 wagon.

I think it will help, but we'll see.

This is very much not what the math is saying :[

The VCAA has shown that you have almost exactly the same chance of finding 1 scum on the wagon as you do of finding 1 scum in any group of 5 people that you choose - you don't gain any benefit from looking on vs off the wagon. If you yourself are town and on the wagon, there's no longer an 89% chance of there being at least 1 scum on the wagon because part of that 89% chance was the possibility that you were scum.

What the numbers are saying is that who does or doesn't vote on lynch wagons is on a large scale entirely random and is completely useless by itself.

[goodmorning]

Newbie ones updated, not much change.

[goodmorning]

Updated, now with minor stats for Large Normals and Opens.

[BROseidon]

VCA involves looking at timing, reasoning, momentum, placement, associations, and tone of the votes/vote posts, among several other factors, as well as applying those factors to the non-votes in order to get a better picture of what occurred. It's not just "oh hey X voted Y so X must be scum..."

This.

The biggest part of VCA is looking at how wagons develop in relation to each other, not how wagons develop in isolation.

[goodmorning]

I don't necessarily disagree. That said, I still feel like that requires too many assumptions, but it's better than the type of VCA that these stats are directed at "please stop"-ing.

[BROseidon]

After a while you can get a pretty good gut as to what the correct assumptions to be making are.

[Psyche]

So I found discovered this thread. It's too bad that goodmorning didn't get as much praise for this as she deserved.

goodmorning, or anyone
are you interested in cooperating on this project, or are you no longer interested in it?

With my coding, this can be done a lot more quickly and comprehensively on a larger amount of data, though there will still be a chunk of manual work involved. If you've quit, I'd at least like some advice on which things you found most worthwhile to do, etc and maybe what you've found so far.

[Psyche]

Without having to read every single vote post (which is as excessively subjective as it is time-consuming), we can still take into account as much as recommended without too much extra work things like vote timing, competing wagons, "momentum", and so forth. And we can even check to see if things like experience, gender, or usual activity make any difference. We can connect voting patterns with players' likelihood to get mislynched and so forth.

One interesting idea for also poking at the more subjective features of voteposts (ie the tone) would be to collect (or just index so that they can be collected on demand) every vote in the data set and program a survey that gives a participant 10 of them randomly picked out to rate on dimensions like tone, depth, and whatever as well as their confidence on these measures. In enough trials (which could be challenging to make happen), we'll have multiple quantized appraisals of every/enough vote post and can take them into account statistically, too.

Lots will be missed by just showing someone a single vote post (ex someone votes in one post and explains in another), but hopefully just focusing on voteposts that result in identical high confidence assessments won't give us a too-skewed understanding of the data set.

But then again, if there are no patterns in the final votecounts of D1, I'm sort of pessimistic about what we'll see in general. Does anyone here with a confident belief about an interesting result that a study of the variables described above might produce?

Surely there are

some

things that scum do more often than town that a study avoidant of subjectivity could identify?

[goodmorning]

I'll think about it.