### Statistical Analysis of Tribunal Cases

dannpakk

Member

so we still cant conclude if the tribunal is effective or not. amazing work though

Silverraptor

Senior Member

Dude. I'm pretty sure that a bunch of statistical analysis people are going to ask you to marry them. (Thankfully, I'm not a statistical analysis person, but I know enough about that to really appreciate everything that you've done. Also, this isn't a marriage proposal, but I'd thought I'd give you the heads up )

Mace Wumpus

Junior Member

Can we get this stickied? It's so informative.

Rob Lockster

Senior Member

I would like to mention for the tribunal and voting. I myself vote in tribunal once a day and review 20 cases currently a day. Now, sometimes they are still "building cases" but I always check back on later and finish my cap. I wonder how many trolls vote in tribunal.

Senior Member

you sir deserve some form of public recognition if for nothing more than kicking arse and taking names with numbers, @Lyte, this man deserves a cookie.

gnfnrf

Senior Member

Here's a random tidbit I calculated because it came up in another thread.

61% of cases are punished, but what fraction of reports are on a punished player? Remember, more reports means more likelihood of punishment.

The answer turns out to be 71%. So, 71% of the reports that Riot lets through to the Tribunal are on players who the Tribunal ultimately decides to punish.

Gangplank2010

Senior Member

Thats your hard data you dont use no significance tests.

You only use central tendency statistics.

How do you know your sample reflects and in what amount of degree full tribunal?

gnfnrf

Senior Member

Quote:
Originally Posted by Gangplank2010
Thats your hard data you dont use no significance tests.

You only use central tendency statistics.

How do you know your sample reflects and in what amount of degree full tribunal?

1.) Everything I present as significant has been formally tested for significance, either by standard mathematical projections of p-value, or by monte-carlo simulation with resampling. I don't report the precise significance each time because that is of little interest to a non-statisictal audience.

2.) I only report central tendency statistics, yes, but the math of model building uses a lot more than that.

3.) I know, and I explain, exactly how I gathered the sample, what assumptions I made about it, and what it represents if those assumptions are true or false. It's in "Note 2" and the heading "The Dataset" in the first post.

Gangplank2010

Senior Member

Quote:
Originally Posted by gnfnrf
1.) Everything I present as significant has been formally tested for significance, either by standard mathematical projections of p-value, or by monte-carlo simulation with resampling. I don't report the precise significance each time because that is of little interest to a non-statisictal audience.

2.) I only report central tendency statistics, yes, but the math of model building uses a lot more than that.

3.) I know, and I explain, exactly how I gathered the sample, what assumptions I made about it, and what it represents if those assumptions are true or false. It's in "Note 2" and the heading "The Dataset" in the first post.
I agree that your data is meaningfull only for the subset you used, and from which you somehow (not explained) extracted a sample, between case 5572301 and case 5637015, but not for the universe of tribunal.

gnfnrf

Senior Member

Quote:
Originally Posted by Gangplank2010
I agree that your data is meaningfull only for the subset you used, and from which you somehow (not explained) extracted a sample, between case 5572301 and case 5637015, but not for the universe of tribunal.

I explained in brief my data gathering tools here.

Quote:
Originally Posted by gnfnrf
I didn't plan on it, no. The code isn't designed for general use; it's a couple of very specific non-interactive Python scripts that download and process the data.

The actual statistical calculations are all done in R, and there's no code to share for that; I did it all in interactive mode.
And while I don't think it is meaningful for the universe of Tribunal, I think it is a reasonable snapshot of how Tribunal works today and in the recent past. Once the IP reward is taken away, it will probably be less relevant, but I can gather another sample and run the data analysis again.

But the point is, and the reason I linked you here from the other thread is that I have data. I have 9000 Tribunal cases to back my assertions about the Tribunal. How many do you have?