As part of Pop Ramen’s Unlocked channel, I’ve decided to share my love of Overwatch-flavored math. Overwatch Minus Overwatch is planned to be a twice-per-stage broadcast applying the same nerdy effort that I’ve used in articles previously here.
My debut episode at the start of week 1 went off with one hitch. I screwed up my math. The Binomial model I used to predict all the games fell apart because I realized how screwed up and complicated it was. As such, it was sacked midway through Week 1’s action and this info about the new model will replace it.
The short version is that my original model started with the premise that Overwatch League matches were a series of four trials with a bunch of conditionals thrown in. That’s not true, if you browse the score lines, you can see that. In reality, OWL is best 3-of-5, they just never always play four of them. Whenever there’s a tie, the match is effectively best two-of-three, so that’s how I base my model now. This removes my ability to predict score lines, but I can have something else to do that. This model, however, does allow me to update win chances in the middle of a match, which I tweeted about last week.
So the quick and dirty recap of the new model is to run a 5-trial binomial distribution and a 3-trial one, weigh them appropriately based on the likelihood of ties, and then add up the winning scores for each side. I even have one set up for perusal.
The weighting is that Stage 3 results are given 45%, Stage 2 is given 35%, and Stage 1 is given 20%. The numbers in white are win conditions, and the final totals indicate that the Shock are favored more than 4:1 against hapless Houston.
Next Overwatch Minus Overwatch, I’ll try and not fail at the math.
