View Single Post
04-02-2013, 01:01 AM
Registered User
CanaFan's Avatar
Join Date: Feb 2010
Location: BC
Country: Canada
Posts: 13,331
vCash: 500
Originally Posted by garret9 View Post
Just because there are thousands of variables and split second decisions which act almost in randomness doesn't mean it can't be statistically modelled. Especially when there are particular end goals in mind that are defined to very particular acts.
In sciences like Ecology and Evolutionary Medicine, this is done all the time.
You aren't adding up all the micro-events together in order to determine a macro-model, but instead are starting with the macro-goals/endresults and working backwards.
Correct, though it *can* mean that the model doesn't do much to improve one's ability to either explain the phenomenon or predict future outcomes. Technically every model starts out by trying to be better than random chance as the threshold for "is it useful". So a model that can predict or explain better than a random guess can be considered a useful model (i.e. it can explain at least a nominal amount of variance in the observed outcomes). However it may not be more predictive than observing the event itself (i.e. watching the games) if it isn't a particularly strong model. It sounds like some of these advanced stats can be useful for understanding "what" makes a team successful or not (over time). The problem I see (on these boards at least) is when fans ignore observational evidence and focus solely on advanced stats to argue the merits of certain teams, players, etc. I don't believe any of these stats are predictive enough to replace actually watching games as a method of analyzing and predicting outcomes.

CanaFan is offline   Reply With Quote