^ Sportsclubstats doesn't really take into account the "road game to a tough eastern team" thing, though, you know?
Quote:
Originally Posted by mrkajz44
It's impossible to tell which model is more accurate, because at the end of the season they will predict the same thing (whoever is in has a 100% chance).
At this point, percentages are percentages. If Calgary makes it, does that mean sportsclubstats was better because they consistently had Calgary at a higher percentage? No, you'd need thousands of seasons to see which model was more accurate in their prediction. The results at the end of this year will neither validate nor disprove either model.
|
Assume all other things are equal, and assume the Wild end up making the playoffs. If model A consistently has, say, the Wild with an above 50% chance of making the playoffs from January 15 on, and model B has them under 50% until March, then model A was a better predictor this year. You're right that that doesn't mean it will be better next year or that it would be better if they played the same season ten times.
It's not about validating or disproving. Could absolutely be just chance if one works better but I still will be interested to see. I don't know about needing thousands of seasons, though - if you take a predictive model and apply it to, say, the last ten years, and on Jan 1 it predicts the playoff teams with 80% accuracy overall, that sounds like a pretty good model to me. Interestingly, that's what NHL.com claims to have (or something of that sort). Have to be realistic about what's possible here and do the best we can.