UW Doog Bot's Proprietary Season Wins Probability Distribution Calculator® Way Too Early Prediction

Since I guess we are all putting our Ballz out there early with @YellowSnow & @puppylove_sugarsteel because Off Season = Best Season and since I already did some calculating anyways I decided to just pull em' out for all you guys to see.



Just a reminder of last year's accuracy.



Though given, that was post Auburn loss before which, we'd had slightly better odds of going 10-2. Fuck off, and generate your own proprietary calculator that's better.

For a methodology refresher and for Kim's benefit when he plagiarizes this, I split games into the following 7 categories and assign them a statistical percentage likelihood based on some math stuff.

• No Way We? Win - 1%
  
• Highly Unlikely To Win - 16%
  
• Unlikely To Win - 32%
  
• 50/50 To Win - 50%
  
• Likely To Win - 68%
  
• Highly Likely To Win - 84%
  
• No Way We? Don't Win - 99%
  

These are the values I assigned to each game.

• EWU - No Way We? Don't Win - 99%
  
• Cal - Likely To Win - 68%
  
• Hawaii - No Way We? Don't Win - 99%
  
• @BYU - Highly Likely To Win - 84%
  
• USC - Likely To Win - 68%
  
• @Stanford - Likely To Win - 68%
  
• @Arizona - Likely To Win - 68%
  
• Oregon - Highly Likely To Win - 84%
  
• BYE - L
  
• Utah - Likely To Win - 68%
  
• @OSU - No Way We? Don't Win - 99%
  
• BYE - L
  
• @Colorado - No Way We? Don't Win - 99%
  
• WSU - Highly Likely To Win - 84%
  

You might think that the values for say, @Stanford and @Arizona, are different and should be something like 62% and 74% respectively, the point of the methodology isn't to predict the outcome of a specific game but to generate a per game average win % to apply to the whole season for a Win-Loss expected value. For this reason, we can avoid trying to nail down individual % too accurately and assume that across the sample the differences will balance themselves out[(62+74)/2=68]. If you don't get it, just trust me, math.



The TLDR results, Expected Value = 9.88 so we can roughly assume that 10-2 is the most probable regular season outcome.

After assigning our % values we average them out and come up with a 82% average chance of winning each game.

"But Doogie, why do I care what our average is? Doesn't that ignore individual matchups?"

Sure it does, but we aren't predicting one game, or the specific outcomes of any games, we want to figure out what our end of season record is likely to be and an average % will do that accurately enough for our purposes. Even though we might be favored in every game across the whole season we are likely to lose a few games. (This is why undefeated seasons are so fucking special)

So we take that average % and apply it to a binomial distribution and we get this.



So WTF does that mean? Well, it means that 10-2 is indeed the most likely outcome at almost 30%. 11-1 at 25% and 9-3 at 21% being the other most likely outcomes.

"Bot, our schedule is shit and you are telling me that 9-3 is almost as likely as a better record?"



Well, it is and it isn't, if you look at individual outcomes it is a likely outcome BUT if you look at an aggregate of above/below our expected value (9.88) we get a likelihood of winning 10 or more games of 64% and winning 9 or less games of 35%. In other words, take the over, 10-2 or better.

This also means that 9 wins should be viewed as a "floor" of sorts. Less than that and the team is probably under-performing its talent in a consistent manner that is attributable to coaching/development/etc. In other words, we should be talking about who is losing their job.



Later, closer to the season, I will do my regular polling of the masses and update the maff with HH's aggregate prediction. So you will get your chance to make your own predictions with this approach. Who knows, maybe I will feel generous and even do a Duck version for the Refuge since @greenblood is kind of cool and we have a bunch of retard educk poasters seeking asylum and flooding the place with their AIDS over there.

Math gifs