Reply to thread

Message: [QUOTE="Cusefan95, post: 4955006, member: 173"] I don’t think that really gives info at a granular enough level to be useful. What I’d be interested in is what percentage of teams favored by a point won, 2 pts won, 3 pts won, etc in a graphical form. Even better if it can take variance into account - if a team is favored by 1 and wins by 20, I don’t think that is a successfully model. I’d want a model which is accurate say 60% of the time at picking the winner AND within something like 4 points of predicted margin. I question if a lot of these models rack up “easy wins” on games like UConn vs UAlbany which drive up percentage of correct predictions, then use anecdotal examples to “sell” the model. I just haven’t seen the proof that games between relatively evenly matched teams get predicted correctly at better than a coin flip. [/QUOTE]