Boards›Player Behavior

I can't believe this escaped me for so long as a statistician

Z Statistic·5/8/2017, 3:14:57 PM·2 votes·2,571 views

Riot has a loadscreen tip that states "players who abuse their teammates lose more games than those who don't." Riot doesn't try to correct the community inference that "toxicity makes you lose more games."

I mean, what manipulative experiment have you run on this, Riot? Did you systematically fuck with people's games to induce toxicity so you could detect an influence in behavior? Did you hire actors to verbally beat the shit out of teammates to see performance changes? Because if you haven't (because you're not ethically devoid individuals), then this is all just observational experimentation, and the limit of inference is that you can't conclude influence, much less causation, between two correlated variables.

It just smacks as a copout to something that is almost certainly much more nuanced. I'd be more reassured if that nuance was explored in posts since I read Reds a lot, but I just keep seeing posts that operate under the premise that "all toxicity is bad. We need to stamp it out by any means necessary for the good of the game."

84 Comments

FiddlesOnTheRoof5/8/2017, 3:19:19 PM7 votes

Why can't it be that they manually researched x number of games where someone was reported, recorded the win rate of those games, and did the same for games where no report was filed?

ModThe Djinn5/8/2017, 4:38:43 PM5 votes

[{quoted}](name=Z Statistic,realm=NA,application-id=ZGEFLEUQ,discussion-id=QUX6x4pY,comment-id=,timestamp=2017-05-08T15:14:57.980+0000)Riot has a loadscreen tip that states "players who abuse their teammates lose more games than those who don't."

This pre-game tip system is part of an experiment to see if Priming effects can reduce toxicity. You can listen to Riot Lyte talk about it here. Starts around the 17:20 mark.

I mean, what manipulative experiment have you run on this, Riot? Did you systematically fuck with people's games to induce toxicity so you could detect an influence in behavior? Did you hire actors to verbally beat the shit out of teammates to see performance changes?

There are no actors hired, nor are they really fucking with games -- this is no different than the priming effect you get from any number of inputs, and is a necessary experimental step before drawing any sort of conclusions -- you DO have to test them, after all.

...the limit of inference is that you can';t conclude influence, much less causation, between two correlated variables.

I think you underestimate the sheer amount of data Riot collects. We're talking hundreds of thousands of data points for each variable. Sure, some of them can be thrown out -- but if over hundreds of thousands of games a single consistent variable reliably produces a certain result? There's probably something to that. That is indeed scientifically significant. We're not talking a few hundred coffee drinkers surveyed here.

As a statistician, you should understand that the overwhelming correlation there at least merits further study. There's coincidence, certainly, but past a certain point of proving your assertion you have to begin to take a closer look at whether or not there IS some causation effect.

La Belle Sauvage5/8/2017, 5:01:15 PM4 votes

If you're a statistician, then educate all of us what facts they'd need to collect to show causation between toxicity and games lost. Are you stating it's impossible to show causation?

Colgate Gator5/8/2017, 5:02:27 PM4 votes

Man, that's some good shit you're having there.

Joxcab5/8/2017, 3:15:28 PM4 votes

Can I borrow your tinfoil hat? I need to store some dishes.

ModPeriscope5/8/2017, 5:57:47 PM3 votes

So, riot didn't state a causation, and they're supposed to tell people that correlation doesn't mean causation?

Solid idea, but you're clearly biased against riot, with the cursing and all. That bias confounds the statement that you're a statistician.

AMYS GRAVE5/8/2017, 6:27:29 PM2 votes

"players who abuse their teammates lose more games than those who don't."

They posit a statement of correlation, not causation.

and the limit of inference is that you can't conclude influence, much less causation, between two correlated variables.

And they aren't trying to. I don't see the problem here.

AeroWaffle5/8/2017, 4:56:54 PM2 votes

Eh. It's a bit of both.

Does losing make the team more toxic? Yeah. People don't like losing and it puts them on edge to find the blame. Does poor behavior make a team more likely to lose? I also argue that's true, especially since this is a team game. If someone is being an ass to you, you're less likely to help them. And that motivation is likely to harm the team's chances of winning.

Ultimately the whole, "players who abuse their teammates lose more games than those who don't." is likely to just be another effort to convince the players who care about winning to control their behavior.

But I think it doesn't really matter even if being an asshole had no effect on the chances of victory. Playing the game is not just about winning, it's about enjoying yourself. Not being an asshole certainly can't hurt your chances to win, so there's no reason to not remain civil.

Being a dick makes the game that much more unenjoyable for the other players for no real reason other than the one player wanting to be a dick.

Jo0o5/8/2017, 4:12:27 PM1 votes

I'm all for correlation not implying causation, but in this situation, isn't there a pretty clear cause-effect in place? We've all been in games that have devolved into toxic circle-jerks and resulted in losses, after all. Even outside of intentional feeding or AFKing, we've seen games where people either tilt and get toxic or tilt due to toxicity.

Speaking personally, I've observed games lost due to toxicity, losing games won in part thanks to the absence of toxicity (choosing to rally and fight back rather than surrender or stop trying), and I'm sure that I've never witnessed toxicity HELP to win a game. I can't imagine that my experience is unique.

It may not satisfy your criteria of an experimental process, but the correlation of toxicity -> poor game performance is infinitely easier to justify than coffee -> cancer.

So, you may win a hypothetical thought experiment of refuting Riot's stance that toxicity leads to poor game performance, but ultimately that's a very reasonable assertion on their part and I'm not sure what your counterargument is beyond "I'm a statistician and there may be problems with their evidence". Surely you aren't asserting that toxicity HELPS, or that it doesn't matter.

YerroFever5/8/2017, 3:30:38 PM1 votes

Um, that's actually pretty easy to find. Do a regression analysis using the data they already have?

X: Number of accurate reports of harassment/verbal abuse that have been found on a person based on manual review Y: Win rate of that person

They can do further analysis and even go through your win/loss record where you were only reported and compare that with your win/loss record where you weren't reported.

If the correlation is very high, then they can say that there's a correlation between abusive behavior with teammates and win rates and the correlation is that the more abusive you are to teammates, the more likely you are to lose the game?