Post by publius on Nov 20, 2017 19:09:53 GMT -5
I like to think of this as a projection and not a prediction...
I use Pablo to project wins and losses (or % chance of winning and losing each match) so that I can have a placeholder for the other 333 teams. This allows me to answer the question for that 1 (each) team; 'If Team A wins X matches for the season, then their approximate (projected) RPI would be Y'. Or; 'how many wins does Team A need to win for an RPI ~ 16?'
Another way to look at this is; 'If Team A wins the expected number of matches (per Pablo) on the season, they will finish with approximately an RPI rank of Y. And if they want a better RPI, they will have to exceed the Pablo expectations.'
What I found, the RPI SOS becomes very predictable early in the season using the Pablo ratings - and once the RPI SOS becomes relatively known, we can quickly start answering that question of how many matches does my team need to win to achieve X goal.
While you think of this as a projection, over the course of this year's Pick the Winners Contest, using Pablo as a predictor was significantly better than using RPI. Picking the exact same contests, Pablo was accurate 73.3% of the time, compared with RPI's 67.0%.