### Summoners who verbally abuse their team lose 16% more games.

First Riot Post

The Sterg

Senior Member

Quote:
Originally Posted by RiotDerivative
You are correct that correlation != causation, but it can be assumed that this statistic was computed from a population. That is, this statement comes straight from the data and is not an inference about future data.
I saw that figure change by about 5% in one game.

SBSuperman

Senior Member

Quote:
Originally Posted by RiotDerivative
Thanks! I appreciate it. New around here and have to remember I am not talking to stuffy statisticians :-).
Dude, there is a job called "Data Scientist"? That's awesome!!!!!

CoIlo

Member

Quote:
Originally Posted by Blaine Tog
Yeah, but it would be naive to think the data wasn't presented with the intention to have us draw conclusions about it.
This is the key take-away message, folks. The data is presented as a "tip," which means that we are supposed to use the information to increase the quality of the LoL experience. The information itself is only usable if one draws a causal inference, which is the obvious intention based on the presentation of the data in context.

I brought this up with Lyte directly in another thread when he quoted my post and challenged me to identify misleading data presented by Riot. Surprise, surprise--he stopped responding immediately after I served up this example.

Syntax Er0r

Senior Member

The OP's post is why Statisticians are not to be trusted....

RiotDerivative

Data Scientist

Quote:
Originally Posted by Robtard
I'm a stuffy statistician, dont worry about it. i'm finishing up my degree at UoM and yeah, you're definitely right in saying that because they come from the same population isn't a case of correlation vs causation. Being that you just ran the variables against each other within the same pop there is no inference or extrapolation happening. Simply an analysis.

On a side note, as a statistician myself I recently took Calculus 4: Multivariate & Vector calc and am wondering how applicable this stuff is in real world use. I'm assuming that there are programs and everything already in place and that the need to use spherical/polar/cylindrical coordinates in triple integrals isn't ever practiced. How much modeling do you do that is anything beyond possibly calculus 2?
It is actually pretty relevant. Anytime you deal with distributions containing multiple variables, you will be dealing with multiple integrals. I see it everywhere especially in Bayesian statistics. If you do statistical computing, one of the main responsibilities is "optimization." When you have multiple variables in an analysis or simulation, all of the data points form a surface or multidimensional geometric figure that you must optimize on. This is common in statistical computing and machine learning.

The only time I have seen polar coordinates was in seeing a proof on creating random normally distributed values using only uniform random numbers. That was pretty cool.

Eulogistics

Senior Member

If players who abuse their teammates actually lose 16% more games, then the statistic is true. What implication you draw out of that is your own business, but the statistic is true.

EDIT: I've never seen the title "Data Scientist" before. Nice name btw

Fersaken

Member

90% of all stats are made up on the spot.

LitterBug

Senior Member

Quote:
Originally Posted by RiotDerivative
It is actually pretty relevant. Anytime you deal with distributions containing multiple variables, you will be dealing with multiple integrals. I see it everywhere especially in Bayesian statistics. If you do statistical computing, one of the main responsibilities is "optimization." When you have multiple variables in an analysis or simulation, all of the data points form a surface or multidimensional geometric figure that you must optimize on. This is common in statistical computing and machine learning.

The only time I have seen polar coordinates was in seeing a proof on creating random normally distributed values using only uniform random numbers. That was pretty cool.
As an applied math major, most of the "optimization" we have done - is mostly linear optimization. This doesn't require much knowledge of multivariate calc (or at least that in depth).

Lots of people think about the possible combination of points as "forming a surface" but nobody expects you to think of it in multidimensional space (other then to understand which kinds of algorithms you're using to solve the optimization function).

TL;DR: it's good to learn all the multivariate calc/higher math but you don't really NEED it to understand what you're doing (other than to fully understand the concept/proof behind certain things)

Exman33

Senior Member

Quote:
Originally Posted by Eulogistics
If players who abuse their teammates actually lose 16% more games, then the statistic is true. What implication you draw out of that is your own business, but the statistic is true.

EDIT: I've never seen the title "Data Scientist" before. Nice name btw
Trust him, he's a scientist