We can't use stats alone (Rebuttle)

I received an e-mail about a rebuttle to a previous article

The e-mail claims that it can use stats to predict who will make the tournament. Here are their current predictions if the committee uses the Old RPI vs the New RPI.

They system claims to only have an average of 2 teams incorrect over the past 5 years. Even though the teams are numbered no claim is made for their seeding once selected.

I would ignore the New RPI method. There are 2 reasons for this. The first reason is it does not pass the smell test. There are too many teams in the list that are not at large candidates. The second reason is that the selection committee said that this year will not be a breakthrough for mid-major teams. That would exclude the new RPI from any calculations since there are an abundance of these teams with a great RPI.

Using the old RPI the list is fairly accurate. There are some conference leaders from low majors in the at-large list that would never get an at large bid. If they all win the method will be off by about 2 or 3 as the list currently stands. I have no idea if the teams in the list will change if the conference leaders lose in their respected tourney. I will withhold judgement of this method until it gets closer to tourney time. I will post any clarifications I receive about the method.

I do wonder how long this method will work. There are two assumptions the method makes.

1) The selection committee uses the same method to evaluate teams year to year.
2) The selection committee is consistent in applying the method year to year.

First the selection committee is starting to tinker with the metrics used to evaluate teams. This makes both points above invalid. Second teams are getting wise as to how the RPI is calculated and are starting to change the way they schedule. When teams try to manipulate the stats the stats are no longer useful in being an unbaised evaluator of teams. I predict that the RPI will be rendered useless over the next 10 years which will make computer models useless as well. As for now the method is interesting and valid (the old RPI method) and worth a look. I will definately alert you to the final results.