Week 3 has completed; this is historically the decision point: continue to use information from last year to form a basis of historical “reasonableness” OR discard last year’s data and live in the current year.
And the winner is …
Unbiased rankings are posted this morning.
Two important notes:
- Because of the limited number of games, the spread in team strength from the top team to the bottom team is higher than will be seen at the end of the year. Thus, game margin predictions will be more extreme than will be seen on the field (a bias in itself).
- Because of the limited number of games, the overall model is still learning. Accuracy will continue to greatly improve until week 5 rankings are published. After that point, improvements in accuracy definitely wane. Key statistic: the top 5 teams in week 5 win about 86% of championships.
- Week 5 rankings are scheduled to be published on September 26. Mark your calendar.