Awesome work! Looking forward to the rest!

I fear this is just going to bring in certain people who believe that strictly the # of reports you get determines the verdict. Which is what your data says, but not what the data means...if that makes any sense.

It does, and you are correct, if I understand you correctly. I showed that the number of reports in the case is a strong indicator of the verdict. I didn't claim, and I have no evidence with which to claim, that the number of times a player is reported is directly related to the number of reports in the Tribunal case.

Everyone who disagrees with you is a troll, got it.

We just have to ignore the likes of him. He just wants to detail the thread. This is the first thread in a long time that actually has promise in this forum.

I for one can't wait for part 2.

So, theshim provided the Red answer to the odd cases, you should update this before I develop a twitch.

In all seriousness though, this is fairly interesting.

Some other notes:

You may want to point out that riot claims that the number of reports to get the Tribunal, exceeds the number of reports we see in a case. (I'm slightly worried that someone is going to use your data to show that you need very few reports to end up in Trib.)

The increased punish rate for more reports makes me wonder how much this reflects forum posts we've seen for auto-pardoning for cases with only 1 report / very few, vs how many of such people simply accrued few reports due to non-toxicity.

+1 vote op. Data crunchers always in short supply.

Hmmm.... there's a problem there, though.

I can remove the 112 no-longer-a-mystery games from my dataset, certainly. But we don't actually know how the bug works. It's entirely possible that there are duplicate pardon cases as well. The problem is, I have no way to detect and remove them (since they will have a punishment of None regardless). So, if I remove the duplicate punish cases, but leave the duplicate pardon cases, my data will be biased.

I could discard all cases in my dataset from the range in which they occurred, since it seems to be isolated in time. I'm not sure what to do; I'll need to think about it.


This is indeed true. From the perspective of my analysis, we can't draw any conclusions about how often people are reported.

This is something I might look at in part 4 (user requests) but a quick peek at the numbers says that the single report cases have a very different agreement profile than other cases. More than half of single-case pardons are by an overwhelming majority, whereas almost none of the punishes are.

Hmm, I can see how not knowing if there were duplicate pardons would be the case. I think I remember seeing the duplicate cases when they were happening, we kept getting same cases over again (if that was that), and I don't remember pardoning any.
Unfortunately, I'm not a statistically relevant data point by myself.

Here's a thought - what, if anything, is the correlation between Ally to Enemy report ratios and punishment margins? This is one that I'd like to see.