Wednesday, May 4, 2016

The RPI sucks

It's time for another rant against the RPI.  I've said the things I'm about to say before.  In fact, I encourage you to go back to last April's archives and find the RPI/SoS post there.  I won't reprise the mathematics here, but I'll refer to the end lesson.  The lesson is this:  in building a schedule, the current system emphasizes bad team avoidance more than playing good teams.  It rewards teams more for simply playing below-average teams instead of the bottom of D1, instead of rewarding teams who actually play good teams.

It's fair to say the overall system is biased against mid-majors.  Note that you have to be careful in what you mean about bias, though.  It's not that very few mid-majors get at-large bids.  It's that the system is set up in a way to make earning an at-large bid impossible for mid-majors.  The current system, based on how the RPI is calculated, favors teams who have bad team avoidance, which is something by definition that the mid-majors cannot do.  There is no recourse for a mid-major to prevent this fatal flaw from ruining their resume.

Knowing this, let's brainstorm a new version of the RPI.  This new version should still punish teams who willingly schedule many cupcakes, but it should reward teams who schedule up and attempt to get quality wins, without erasing those gains unfairly.  There's two main ways to do this:

1) Throwing out outliers.  You know in some judging competitions, how the high score and low score get thrown out when there are many judges?  Let's introduce that to the RPI.  For the SoS calculation, throw out the best team and worst team, and then calculate the SoS.  Or even just throw out teams on one end.  Throw out the worst 3 records in a team's non-conference schedule before calculating the non-con SoS.  In fact, this is my favorite solution.

What does this do?  By throwing out the bottom 3 teams on a team's schedule when calculating non-con SoS, you allow everyone a couple of free cupcakes without hurting their computer numbers, which is okay.  You keep the truly horrendous teams from causing extra damage to a good team's profile.  If a team still has several other cupcakes on the schedule, now they're still going to be punished, even more severely when compared to teams with just a couple of cupcakes.  And finally, with less teams factoring into the SoS calculation, the excellent teams will stand out more clearly.

2) Artificial scaling.  For the SoS calculation, put in a floor.  If an opponent has less than a .250 record...bump it up to .250 for the SoS calculation.  Or .300.  Or .200.  Or any number you want.  This would require some research that I don't have time for, but I'm sure an optimal number is out there.  Or you can introduce a sliding scale, where all below .500 records receive a proportional bump in value.  A .200 record becomes .300, a .400 record becomes .450, you get the idea.

What does this do?  It softens the blow that a truly terrible team has on the SoS.  It mitigates the bad team avoidance issue by helping prevent a single bad team from having excessive influence on the SoS.

I'd like to see someone figure out a system incorporating one or both of these ideas.  We need something to lessen the impact of cupcake games on the SoS.