Splatfest results - A discussion and comparison

[MeTaGross]

That's weird, and I don't know. But that has nothing to do with this. It even includes the relation between Solo and Team outcomes, which opens up a whole nother can of worms.

If the less popular team didn't have any advantage at all then we would expect an even distribution of wins/team wins. This would mean roughly 25% of splatfests are won 3-0, 50% are won 2-1 by the popular team, and 25% are won 2-1 by the less popular team.

The entire issue of popular teams losing can easily be explained by sample variance with a very small population, with the population being all splatfests, but that is no fun. I think an issue for the popular teams is the repeat battles against your own team. I personally find these battles to be frustrating, so this could cause players on the popular team more tilt. Just a hypothesis with no evidence, but I think it makes sense.

Edit: Just did some math on this. Assuming that there is NO association between the popular team winning or losing there would be equal probabilites of the winning team having all 3, popularity and solo/team wins, or solo and team wins. That means the 25%, 50%, 25% thing from before.

Only using the unique splatfests in NA/OC and EU, so far there have been 3 3-0's, 5 popular 2-1's, and 8 unpopular 2-1's. We would expect 4 3-0's, 8 popular 2-1's, and 4 unpopular 2-1's if there was no association. Based on the actual results and the expected results there is a 14.6% probability that there is no association between team popularity and winning a splatfest.

This means that it's possible for all the popular team loses to be a fluke due to a small sample of splatfests, but it is very unlikely.

*Disclaimer — I didn't include Japan due to more pronounced cultural differences, but I probably shouldn't have done that. Even if I did, it would support the connection between popularity and losing.

 

[MINKUKEL]

If the less popular team didn't have any advantage at all then we would expect an even distribution of wins/team wins. This would mean roughly 25% of splatfests are won 3-0, 50% are won 2-1 by the popular team, and 25% are won 2-1 by the less popular team.

This would imply that there is no relation/overlap/correlation or whatever term you want to use between Solo and Team. There must be, because a player can play both Solo and Team.

You would still expect (if things were truly random) the most popular team and the least popular team to have both Solo and Team roughly half of the time. But as we know, 3-0's are more rare.

This means that it's possible for all the popular team loses to be a fluke due to a small sample of splatfests, but it is very unlikely.

I mean, this is why I entered this kinda stuff in a statistical program, because you can't really say if something's unlikely or not based on just looking at it this way.
Again, splatfest results aren't dichotomous. Using the actual percentages as data instead of just wins/loses, the image is already a bit more nuanced. If I had to guess, there is still some positive effect of being on a less popular team, but you can't actually rule out chance here.

 

[Flammie]

Neither.

That's weird, and I don't know. But that has nothing to do with this. It even includes the relation between Solo and Team outcomes, which opens up a whole nother can of worms.

If there was any reason for the regions to behave differently when it comes to Pearl/Marina wins/loses it wouldn't be arbitraty, but I see no reason to believe that.

I actually forgot that i only calculated the Solo matches and not Teams.
While solo matches are supposed to be more random than others, the Team section shouldn't be this way and be more independent against Solo, but it's not as often as it should be, so you're right, Math-wise this is a disrepancy and shouldn't even exist.

My other theory is that there are saboteurs in Splatfest, joining the other team just to make sure they lose on purpose, both in Team and Solo sections, i've encountered squid baggers and partys in the first game.
But then again that should only make a difference between 1.0% to 0.1% or even less...

How many Splatfest have happened where Solo and Team matches not won for the same team together?

 

[MINKUKEL]

How many Splatfest have happened where Solo and Team matches not won for the same team together?

In the first 25 fests this happened 10 times IIRC.

My other theory is that there are saboteurs in Splatfest, joining the other team just to make sure they lose on purpose,

If that were true, and I'm sure there are people who don't have anything better to do with their life than to do something pathetic like that, that wouldn't explain anything, since those things can happen on both teams. Only if one side is systematically more likely to have trolls on them, this would do anything, and I see no reason to believe that. You COULD argue that the more popular team has more people on that side, thus leading to more trolls to **** things up for the other (less popular) team, but that clearly doesn't rhyme with the results.

And it wouldn't matter anyway. The fact that I have never seen a single squid-partier or troll during Splatfests means the effects of such people are negligible anyway.

 

[MeTaGross]

I mean, this is why I entered this kinda stuff in a statistical program, because you can't really say if something's unlikely or not based on just looking at it this way.

Sorry about that, I didn't realize that you already did some of that. I thought I had read the whole thread already, but I guess not. On that note, did you do any kind of stat test, like a z-test or a t-test? I feel like I did a poor job, so it would be nice to have better results that are up-to-date.

My other theory is that there are saboteurs in Splatfest, joining the other team just to make sure they lose on purpose, both in Team and Solo sections, i've encountered squid baggers and partys in the first game.

I will admit that I did this once in the first game. However, I only did it for a few battles, and I accidentally won a few of those. Doing that made me feel like a bad person and it was incredibly boring, so it would take a real POS to purposely lose more than a few battles, and it would be difficult to do with easy reporting on Splatnet 2. They would also drop to a very low power level, so they would likely have a smaller impact there. I don't see any way that this could have a noticeable effect on the results.

 

[MINKUKEL]

On that note, did you do any kind of stat test, like a z-test or a t-test?

You could do a t-test or a z-test, but only if you group data by win/lose, so by using dichotomous data. And the whole reason I entered this stuff into a datafile was to not use dichotomous data. It'd be like looking at lifespan as a variable and just taking everything under or above 50 and using that as two groups. It only gives part of the picture.

For the info in this thread I just did some correlational stuff. I also looked at some other stuff earlier but I forgot what they are atm.

 

[MindWanderer]

You could do a t-test grouped by more/less popular team, but Minkukel's simple correlations seem more appropriate in this case. As a general rule, always used tests for continuous data when you have continuous data. That said, if we knew how many voters and games were involved, we could do much more accurate tests (z-tests against an assumption of 50% wins).

It's interesting that such small correlations are having such a significant effect on the 50/50 splits, though. It's very suspicious.

 

[MINKUKEL]

It's interesting that such small correlations are having such a significant effect on the 50/50 splits, though. It's very suspicious.

Yeah, it's pretty weird.
I might update my datafile this weekend and show you guys a picture of a scatterplot or something. It looks a bit less suspicious that way, actually.