Open Setup Certification Group

[Ether]

Standard nightless games found by searching for "Nightless" in Little Italy and New York:

• Open 6 (4:8; town win)
  Open 19 (4:8; town win)
  Open 79 (4:7; I think this was the game Farside was talking about; town win)

So I'm really not getting the need to further handicap the scum.

EDIT: Open 79 was 4:7, not 4:8.

[farside22]

Standard nightless games found by searching for "Nightless" in Little Italy and New York:

• Open 6 (4:8; town win)
  Open 19 (4:8; town win)
  Open 79 (4:8; I think this was the game Farside was talking about; town win)

So I'm really not getting the need to further handicap the scum.

Yup it was 4:8 my bad. Still very fun game. I even as shaft stated was accused of people wondering if I was bussing my scum buddies when I was town.
Great stuff.

[Xylthixlm]

Those results surprise me a bit, but okay. More scum are needed.

[Lord Gurgi]

Those results surprise me a bit, but okay. More scum are needed.

Behold the power of the obv-town!

[mith]

As I've said before: three games != statistically significant. Scum clearly need to step up their game, though.

I'm fine with 4:8 (same balance as 3:6, which is definitely preferable to 2:7). 4:7, no.

And you could include 2:4 if you want a 6 player option.

[Lord Gurgi]

Three games is all we have. If three isn't statistically significant, why bother to check the old games at all? Whether three games is a large enough sample size to determine anything is indeed questionable, but three consecutive town wins is not something that happens often.

[mith]

I'm not saying it's meaningless. What I'm saying is that three town wins isn't conclusive evidence that a setup is in the town's favor.

As I said, I'd be fine with it being run. I think it's probably close to 50-50 in practice (EV is only 1/3). I

don't

think it's in the town's favor, though.

[Lord Gurgi]

I'm not saying it's meaningless. What I'm saying is that three town wins isn't conclusive evidence that a setup is in the town's favor.

As I said, I'd be fine with it being run. I think it's probably close to 50-50 in practice (EV is only 1/3). I

don't

think it's in the town's favor, though.

What would you say is significant evidence?

[mith]

That depends on the hypothesis. If the hypothesis is "The town will win about 1/3 of games, as predicted by the EV", probably one or two more games would do. If the hypothesis is "The town is favored", you'd need 6 or 7.

Assuming they're all town wins, of course.

(My statistics is rusty. I don't like statistics.)

[Lord Gurgi]

From what I've gotten from my Statistics AP class, which is albeit thin since the teacher spends most of the time surfing the internet or insulting the class, the hypothesis is irrelevant to the necessary sample size. Just binomial it, right?

[mith]

The binomial distribution depends on the probabilities; if our hypothesis is that the realEV is 1/3, that gives us one distribution (and corresponding range of statistically significant results), while if the hypothesis is that realEV is >1/2, we have another distribution.

In the latter case, the probability of getting three town wins by chance is at least 1/8 (12.5%). Statistical signifcance is a fuzzy concept, but generally 5% or 1% is used.

[Lord Gurgi]

The binomial distribution depends on the probabilities; if our hypothesis is that the realEV is 1/3, that gives us one distribution (and corresponding range of statistically significant results), while if the hypothesis is that realEV is >1/2, we have another distribution.

In the latter case, the probability of getting three town wins by chance is at least 1/8 (12.5%). Statistical signifcance is a fuzzy concept, but generally 5% or 1% is used.

One more win gets to 6.25% which is somewhat close, but of course this is t test, and I don't want to bother with the calculations.

[Thok]

One more win gets to 6.25% which is somewhat close, but of course this is t test, and I don't want to bother with the calculations.

But we don't know that town will win the next game.

(Also, we should really being using a double sided test, since we'd be having this discussion about balance whether town won 3 games or scum won 3 games. That changes the 1% and 5% to .5% and 2.5%)

[Lord Gurgi]

Short version: Peeps will never be satisfied.

I think that the power of a town is understated under the current meta.

[mith]

Right. A sample of 5, all town wins, would get us under the 5% mark, but personally I still wouldn't consider that "conclusive"; if the scum won the next game, we'd be right back above 5% (well above, actually; over 10%). Thus the 6 or 7 I gave above.

Statistics always makes me feel dirty, and this is why.

[edit]That was re: 186.[/edit]

[Xylthixlm]

Ew. Statistics. Let's see if I can remember this, I try to repress it....

To do this correctly, you need a hypothesis and a null hypothesis. Then you calculate the probability of seeing the results you saw if the null hypothesis is true. If it's low enough (usually 5%) you reject the null hypothesis.

The null hypothesis here would be that town wins at most the amount you would expect from random lynches (1/3). The hypothesis is that the town does better. Assuming the null hypothesis, the chance of town winning all three times is 1/27, or 3.7%. That exceeds our 5% significance level, so we reject the null hypothesis; the data is enough to conclude that town probably does better than the random lynch EV in 4:8 nightless.

How much

better is an open question.

[Lord Gurgi]

I hate statistics. We need to make a club.

[Xylthixlm]

My dad is a statistician and my mother is a biologist. I'm a math nerd, but for some reason I hate statistics but like biology. Go figure.

[mith]

I've submitted my PhD thesis in math, and have a degree in chemistry.

Obviously, I hate both statistics

and

biology.

[Ether]

Let's pass 4:8 as it is and worry about this after we have a few more results.

2:4 bugs me a little because I somehow have it lodged in my mind that both 2:4 Nightless and 2:4 Lovers are balanced.

[Korts]

I like Xyl's idea of a variable number of players with a set date for end of sign-ups. And three consecutive town wins doesn't seem conclusive evidence that the setup is inherently in the town's favour.

I'd break that down differently:

6-8 players: 2 scum (2:4 to 2:6)
9-12 players: 3 scum (3:6 to 3:9)
13-16 players: 4 scum (4:9 to 4:12)

I'd go with this over Xyl's version.

[mith]

I guess the main decision to be made, then, is: 3:9 or 4:8 or both? (Well, similar decision for 16 players; 4:12 or 5:11. But they are basically the same question... if we prefer 3:9, we prefer 4:12, and if we prefer 4:8, we prefer 5:11.)

I'm still in favor of 3:9, but wouldn't mind seeing both run.

[shaft.ed]

I guess the main decision to be made, then, is: 3:9 or 4:8 or both? (Well, similar decision for 16 players; 4:12 or 5:11. But they are basically the same question... if we prefer 3:9, we prefer 4:12, and if we prefer 4:8, we prefer 5:11.)

I'm still in favor of 3:9, but wouldn't mind seeing both run.

Seeing as we're already running 4:8 I don't see a need to remove it. I also would prefer having more options not fewer, so maintaining 3:8, 3:9, 4:8, and 4:9 as all viable set ups would not bother me.

[Ether]

I still don't think those other three setups are fair.

I guess by default I'd support a 1:2 ratio and oppose anything lower than that. (I kinda like Guardian's 2:5 thing with the vengekill, though of course that's not standard nightless.)

[Korts]

An idea:

Since we have no solid data of either setup being more balanced than the other, we could have them run as variations of each other, in seperate sign-ups. After the first run we might be able to decide on the issue better.

EDIT: that's basically what shaft.ed said.