Ret’s WAR is way too high

[SantanaTheSmall]

I’ve only tracked 100 games so far, so the sample size is a little small. I will continue to track it for 100 games without ret and 100 games with ret and will make another post then. For now, I’ll show you the numbers I have so far.

Note: the ret games only included games where the ret successfully pulled off a revive. (This is for the independent variable. I still included games where ret didn’t pull off a revive in the control.)

I wanted to track how rets only ability affects the game. All these games happened in Ranked Practice as I’m having some connection issues and don’t want to lose ELO because of that.

Note: I did not include games where anyone on a certain faction left.

I included games where someone left in the late game after the player figured the game was over, however. Without further ado, here are the numbers:

Games with ret: 28 W - 2 L = 93.33% winrate

Games without ret: 42 W - 28 L = 60% winrate

Average Town winrate: 70%

Again, I did not count games where a ret did not pull off their revive. This means that the chances of getting ret is over 30%. Probably close to 35%.
As you can see, if a ret pulls off a revive, it’s pretty much game over. The other factions combined have only a 6.66% chance of winning the game.

That is absurd

Ret is not fun to play as. It’s not fun to play against. It makes the game all around less fun. I will continue tracking these games until I get 100 each, but I suspect the numbers won’t change much.

Please remove this boring and OP role, BMG

[SantanaTheSmall]

I was speaking strictly on Ranked terms.

I also believe ret is okay outside of Ranked as there’s too many lurking variables to change the outcome of the match—specifically the randomized role list.

Still, with a WAR of 33, how is ret possibly not OP in Ranked?

[Brilliand]

Again, I did not count games where a ret did not pull off their revive.

Didn't count those games at all? As in, they don't show up in your statistics?

That's a selection bias. IMO those games should be included as a third category, so we can judge their impact.

[SantanaTheSmall]

It’s not selection bias. Rather, it’s what I chose to measure.

My goal was to measure how the ability affects the game rather than how the existence of a ret affects it.

The reason I did that is because people have already recorded winrates of the simple existence of a ret, so I chose to narrow the scope of said impact. I had no need to measure every game with a ret. I will say from the research of others, if one were to measure the simple existence of a ret, the winrate would certainly be lower than just measuring the ability’s success, but all the data I’ve encountered thus far shows the existence of a ret still increases town’s chance of winning by at least 20%

I might do mayors and jailors next, because I hypothesize that a ret pushing its button affects the game more than a mayor pressing the sun or the jailor pressing an exe.

[MysticMismagius]

It’s not selection bias. Rather, it’s what I chose to measure.

That is selection bias. Whether or not it’s justified is up to you, but know that when you choose to/not to measure certain games you are necessarily introducing selection bias.

[SantanaTheSmall]

It’s not selection bias. Rather, it’s what I chose to measure.

That is selection bias. Whether or not it’s justified is up to you, but know that when you choose to/not to measure certain games you are necessarily introducing selection bias.

Nope, it’s not. There aren’t any sampling errors in my study because I included every possible game where the ret pulled off its ability (when keeping in mind that nobody left the game)—the effect of which is what I wanted to measure. It includes a valid randomized sample of the population within the control (games without rets pulling off their revive) and the variable.

Perhaps I misphrased it. I did not include games where rets don’t pull off their ability in the variable. I included them in the control.

Again, I DID NOT want to measure how the exisistence of a ret affects the game. I wanted to measure how the ret’s ability affects it.

Please tell me how randomization is not achieved when keeping what I’m trying to measure in mind.

I also find it funny that, in EVERY single thread I see when someone does this, people immediately harp on the methodology. Are you going to tell me, from your anecdotal experience, that when rets pull off a revive, town does not have a higher chance of success?

This is silly anyway because, IIRC, there were only two or three games where rets didn’t pull off their revive in the entire 100 games I included, and that’s only because they died D1.

[MysticMismagius]

Oh, my apologies. I thought you merely threw out the games where Ret died N1.

[SantanaTheSmall]

Oh, my apologies. I thought you merely threw out the games where Ret died N1.

No apologies necessary. It was my fault I didn’t make that clear.

[OBeauPeep]

You even admit that your sample size is small, and simply having a hypothesis is not the same as confirming that hypothesis.
The most interesting statistic your report actually provides is that even without Ret, town still wins 6 in 10 games. Ideally, that would be 5 in 10 at most.
Further, the matches you studied could have been decided by any number of things.
Correlation is not causation.

[Brilliand]

The most interesting statistic your report actually provides is that even without Ret, town still wins 6 in 10 games. Ideally, that would be 5 in 10 at most.

There's an argument for the winrate of each faction being proportional to the size of that faction. So:
9/15=60% winrate for town
4/15=27% winrate for mafia (adjust to 32% if we consider the NE to be "pro-evil")
1/15=7% winrate for NK (adjust to 8% if we consider the NE to be "pro-evil")

[OBeauPeep]

Precisely why 9 town is too many to start with.

[kyuss420]

Precisely why 9 town is too many to start with.

Perhaps... in the higher tiers

[SantanaTheSmall]

You even admit that your sample size is small, and simply having a hypothesis is not the same as confirming that hypothesis.
The most interesting statistic your report actually provides is that even without Ret, town still wins 6 in 10 games. Ideally, that would be 5 in 10 at most.
Further, the matches you studied could have been decided by any number of things.
Correlation is not causation.

And I also say I’ll continue to track it. The margin of error here is 10% at the 95% confidence interval. If I want a more respectable number, I’ll simply need another 100.

But, I really don’t think these numbers will change much. If you disagree, I’ll bet you $50 the numbers won’t deviate more than 5% playing another 100 games.

Of course that would be dumb, because there’s no way to ensure that I’m not cheating. Anyway, my point is that these constant criticisms on sample size and methodology are so stupid. I see them literally every time a post like this comes up as if that somehow completely dismisses the findings of the study.

You can say it diminishes them. You can say it’s not strong enough yet. But you cant say it’s invalid.

[Brilliand]

The frequency with with an ability activates is important. As an extreme case, consider the Hex Master's final attack - that ability has a 100% winrate for coven (in games where it's successfully activated). That extreme power is just fine, because getting it to activate is very hard.

Hence, I don't think "if a ret pulls off a revive, it’s pretty much game over" is necessarily a problem in a vacuum. It's the combination of that with "...and the ret almost always pulls off his revive" that's a problem.

Hence it was very important that you included this information:

This is silly anyway because, IIRC, there were only two or three games where rets didn’t pull off their revive in the entire 100 games I included, and that’s only because they died D1.

Properly speaking, that information should be included in your reported data (as part of making your argument that there's a problem here), because it actually matters that very few rets fail to revive someone.

[OBeauPeep]

If I want a more respectable number, I’ll simply need another 100.

No, you won't. Even 200 is not a large enough sample size.

[SantanaTheSmall]

If I want a more respectable number, I’ll simply need another 100.

No, you won't. Even 200 is not a large enough sample size.

That would give me a MoE of 7% with a 95% confidence interval. That’s respectable.

But I do admit my goal is for 400 games to get me under 5%

[SantanaTheSmall]

The frequency with with an ability activates is important. As an extreme case, consider the Hex Master's final attack - that ability has a 100% winrate for coven (in games where it's successfully activated). That extreme power is just fine, because getting it to activate is very hard.

Hence, I don't think "if a ret pulls off a revive, it’s pretty much game over" is necessarily a problem in a vacuum. It's the combination of that with "...and the ret almost always pulls off his revive" that's a problem.

Hence it was very important that you included this information:

This is silly anyway because, IIRC, there were only two or three games where rets didn’t pull off their revive in the entire 100 games I included, and that’s only because they died D1.

Properly speaking, that information should be included in your reported data (as part of making your argument that there's a problem here), because it actually matters that very few rets fail to revive someone.

Ah, that’s a GREAT point. Noted. Unfortunately, because I can’t remember the exact number of rets that didn’t pull off their revive, I’m not gonna fudge the numbers, but I’ll start doing that now.

[Brilliand]

The most interesting statistic your report actually provides is that even without Ret, town still wins 6 in 10 games. Ideally, that would be 5 in 10 at most.

There's an argument for the winrate of each faction being proportional to the size of that faction. So:
9/15=60% winrate for town
4/15=27% winrate for mafia (adjust to 32% if we consider the NE to be "pro-evil")
1/15=7% winrate for NK (adjust to 8% if we consider the NE to be "pro-evil")

The thing about this argument is, well, that it's hot garbage.

I'm not committed to it, but care to explain why you think that?

[SantanaTheSmall]

The argument about proportional winrates is a ulititarian one. I could possibly accept it. The problem is with the ret, the winrates become disproportional. There’s approximately a 33% chance a ret exists in the game, which means 1/3 of the time, the NK’s chance of victory is like 1-2%

[OBeauPeep]

You want the NK to have a 7 or 8% chance at victory in an ideal case. This is low without the Ret being involved.
The Ret does not offset anything because, as your math shows, even if you removed it, the NK should win less than 1 in 10 games.
I don't think your argument is against the Ret itself here.

[SantanaTheSmall]

I said I could possibly accept the proportional argument, but only for the maf. It’s still low for the NK. And it’s absurdly low when a ret exists. This thread is about the ret so I’ll continue making points about it in each of my posts.

The topic of a low NK winrate is for another discussion, but I’ll talk about it in this thread if the ret makes it relevant.

[ICECLIMBERS]

lol imagine saying that mafia wasn’t designed for equal village/scum winrates and expecting for people to take you seriously

[SantanaTheSmall]

lol imagine saying that mafia wasn’t designed for equal village/scum winrates and expecting for people to take you seriously

All the more reason to address roles that dramatically increase Town winrates.

[Brilliand]

lol imagine saying that mafia wasn’t designed for equal village/scum winrates and expecting for people to take you seriously

Didn't think that post would get any agreement.

"This is normal and no one thinks about alternatives" is a pitiful argument.

[ICECLIMBERS]

it’s not a response to any post you made in this thread, but the person saying that proportional winrates are a good thing