Posted by jtd79 on 5/11/2016 3:32:00 PM (view original):
Outstanding work Naskin. How long before you think you have enough data points for these to normalize? By the end of season 3?
I'm not exactly sure. I think it's already getting closer. I would imagine some of the top teams will end up around 2200 or so, and some of the worst teams around 800 or so. With the max at 1960 (Northeastern) and lowest at 1061 (St Francis), ELO predicts St Francis would have a 0.5% chance of winning, which I think is too high. Even at 2200/800 it's at .03% chance, but at least that's getting closer to the actual chances they probably have. We could potentially see a team get up to 2400ish territory, or if reigny joins again we may see some 500-600 ELOs :)
It's hard for me to totally predict because all my other sources have bell curves have things that mess with the curves. For instance, if you look at a bell curve of chess players that have played in the last week, it tends to skew higher than 1500, because generally better players play more often (or *are* better because they play more often). The NFL ELOs on fivethirtyeight.com probably never really reach equilibrium, because each season the values regress 1/3 back to the mean (they do this because NFL rosters change so drastically year to year). I tried to avoid doing that with my rating system; if a team truly regresses from one year to the next, you should see that reflected in their ELO once they start having losses.
5/11/2016 4:01 PM (edited)