• 1 Post
  • 6 Comments
Joined 1 year ago
cake
Cake day: October 28th, 2023

help-circle





  • The core idea behind Simpson’s paradox is that depending on how you breakup the data, you will get opposite conclusions. The example from the wiki is that the it looks like the university is biased against women in admissions if you look at the university wide data. But if you break up the data by department, then most departments are biased against men.

    The same thing can very easily happen here. The easiest breakdown here is to look at T5 item winrates per hero. So it might be that ex machina has poor overall winrate, when you break it down by hero you find that it’s great on a lot of heroes, but because, say Pudge is popular hero and ex machina sucks for pudge, most of the negative winrate is because of heroes like Pudge.

    I am not saying that’s what’s happening for sure, but you have to check to be sure. You also have to check how ex machina performs when you are ahead at 60 min and when you are behind. Etc.