Re: 2017 Women's D-III NCAA Selection Thread
Although I am definitely not a fan of KRACH, I think you should know that this is not a true statement. Every game played matters in determining KRACH. As a statistician I have issues about the suitability of Bradley-Terry (which is it's real name - "Ken" just adapted it) as a metric for comparing hockey performance over the course of the season, but the ratings don't really depend on the "initial ratings." It is a fairly stable rating, which means it will change slowly from the first published ranking, but subsequent rankings are not based on what the previous (or initial) ratings are. The model can't be computed until the competition graph is connected, which is why it isn't published until mid season, but the actual rating is based on the entirety of the schedule.
KRACH is irrelevant because it leans too heavily on initial ratings.
Although I am definitely not a fan of KRACH, I think you should know that this is not a true statement. Every game played matters in determining KRACH. As a statistician I have issues about the suitability of Bradley-Terry (which is it's real name - "Ken" just adapted it) as a metric for comparing hockey performance over the course of the season, but the ratings don't really depend on the "initial ratings." It is a fairly stable rating, which means it will change slowly from the first published ranking, but subsequent rankings are not based on what the previous (or initial) ratings are. The model can't be computed until the competition graph is connected, which is why it isn't published until mid season, but the actual rating is based on the entirety of the schedule.
Last edited: