Statistical Analysis of Tribunal Cases

Comment below rating threshold, click here to show it.

Krispyz

Junior Member

08-09-2012

Any chance that the champion specific stats are skewed by who is free that week that you selected from (assuming your randomized cases were all from the same approximate time)? A person playing a new champ that's free may have more deaths, fewer kills, and/or more reports for either playing poorly or "feeding". And given that the free champs are playing a lot, I wouldn't be surprised if that skewed those particular statistics. Any way of cross-referencing the dates of the cases with what champions were free?


Comment below rating threshold, click here to show it.

Mouse Trench

Senior Member

08-10-2012

Awesome work. Thanks.

Have you run any stats on average kills - deaths or average kill/death ratio by champion?


Comment below rating threshold, click here to show it.

gnfnrf

Senior Member

08-10-2012

Quote:
Originally Posted by Krispyz View Post
Any chance that the champion specific stats are skewed by who is free that week that you selected from (assuming your randomized cases were all from the same approximate time)? A person playing a new champ that's free may have more deaths, fewer kills, and/or more reports for either playing poorly or "feeding". And given that the free champs are playing a lot, I wouldn't be surprised if that skewed those particular statistics. Any way of cross-referencing the dates of the cases with what champions were free?
It's a good thing to check, but it doesn't look like it had much of an effect.

My dataset is mostly from Champion week 34, and a little from week 35. Almost no-one from week 34 turns up in any of the by-champion lists top or bottom 5, and I don't think there are enough week 35 games to matter.

- Ezreal was free in week 34, but I don't think anyone doubts that he's a popular champion, particularly in normals, any week.
- Soraka was fee in week 34 as well, but her placement on the lists is firmly inline with the other pure-support champions.


Comment below rating threshold, click here to show it.

Rathwirt

Senior Member

08-10-2012

This is cool stuff. Thanks.


Comment below rating threshold, click here to show it.

Jayrune

Senior Member

08-10-2012

Most intriguing, can't wait for part 3!


Kudos


Comment below rating threshold, click here to show it.

Basballguy

Senior Member

08-10-2012

what about based upon IP/RP cost. I see Yi up there and he's a 450 IP champ...do they higher costing champs generate less trolls (when not played on a free week)


Comment below rating threshold, click here to show it.

Ilzhahkha

Junior Member

08-10-2012

Statistics! \o/

One of the factors to explain the difference in k/d/a stats between a reported player and his teammates is likely to be a disproportional amount of players that was intentionally feeding among the reported players (or thought to be in the case of pardons). In turn that is a indication that the tribunal is working to remove this even if there is likely to be alot of other factors that provide noise here.

Kinda curious about your part 2, when breaking down the numbers I would guess that the base for some champions would be smaller than prefferable and create a high margin of error? My guess would be that there are a bunch of champions that are reported 30-50 times in your dataset with about half that actually punished.


Comment below rating threshold, click here to show it.

gnfnrf

Senior Member

08-10-2012

Quote:
Originally Posted by Mouse Trench View Post
Have you run any stats on average kills - deaths or average kill/death ratio by champion?
I would like to, but there are so many possible formulae to calculate a single "game quality" score that I can't check them all.

So, for the audience in general, what calculation do you think accurately represents a player's contribution to the game the best?

Kills + Assists - Deaths?

Kills + Assists/2 - Deaths?

Kills/Deaths?

(Kills +Assists/2)/Deaths?

Maybe assists should be divided by 3? Maybe creep score should be added in?

There are too many choices.


Comment below rating threshold, click here to show it.

gnfnrf

Senior Member

08-10-2012

Quote:
Originally Posted by Ilzhahkha View Post
Statistics! \o/

One of the factors to explain the difference in k/d/a stats between a reported player and his teammates is likely to be a disproportional amount of players that was intentionally feeding among the reported players (or thought to be in the case of pardons). In turn that is a indication that the tribunal is working to remove this even if there is likely to be alot of other factors that provide noise here.

Kinda curious about your part 2, when breaking down the numbers I would guess that the base for some champions would be smaller than prefferable and create a high margin of error? My guess would be that there are a bunch of champions that are reported 30-50 times in your dataset with about half that actually punished.
Actually, most champions have far more report representation than that. I included a note about Xerath and Viktor since I only have 71 and 59 cases with them reported. Nearly everyone else is much higher than that, though (the median value is 215).


Comment below rating threshold, click here to show it.

Ilzhahkha

Junior Member

08-10-2012

Quote:
Originally Posted by gnfnrf View Post
I would like to, but there are so many possible formulae to calculate a single "game quality" score that I can't check them all.

So, for the audience in general, what calculation do you think accurately represents a player's contribution to the game the best?

Kills + Assists - Deaths?

Kills + Assists/2 - Deaths?

Kills/Deaths?

(Kills +Assists/2)/Deaths?

Maybe assists should be divided by 3? Maybe creep score should be added in?

There are too many choices.
This would likely be very subjective but cs is only interesting if you do a analysis from position which might be hard to conjure from tribunal data. Otherwise it will be detrimental for support to a major extent and jungle to a lesser extent.

If it is possible to datamine position (jungle should be pretty easy, and there might not be a great need to separate top/mid/carry) you can probably refine k/d/a stats for a game quality formulae by comparing to the average for that position as well. You likely will want something along the lines of:

PLAYERQUALITY = k*(X*"kills"+Y*"assists"-Z*deaths)*("average gamelength"/"gamelength")

Quote:
Originally Posted by gnfnrf View Post
Actually, most champions have far more report representation than that. I included a note about Xerath and Viktor since I only have 71 and 59 cases with them reported. Nearly everyone else is much higher than that, though (the median value is 215).
Not as up to date with my statistics as I would like but shouldn't that be weighted numbers and whats likely to be in your dataset about 46,5% of those?

Awesome work though!