2013 Championship Newton Division

[MikeE]

Originally Posted by MechEng83

Are you suggesting generating a series of random match schedules and then doing the seeding predictions based on that? I get the feeling that this will end up collapsing to just the OPR rank list with enough runs.

Or, are you suggesting using some statistical variation on the OPR values of each team using the current schedule? If so, what variation coefficients would you use? I think this might be a very interesting simulation, but I'm not sure what the valid variation would be. I might need to dig further into the OPR calculations to find some sort of confidence interval.

I think JesseK's expectation is that we would see substantial variation between different schedules, i.e. ranking is more a function of schedule than of OPR "performance".
So yes, while the average would approach OPR, we only play a single schedule which is likely to be helpful to some teams and disadvantage others.

Your other interpretation is more interesting. With some small tweaks the OPR calculation can also be used to generate mean/variance parameters for a maximum likelihood estimate of each team's performance. That's a better basis for real Monte-Carlo simulation, but we'd still need to simulate over a large number of potential schedules.