Agree the breakpoints are a weakness. You could go with a system that gives +76 to a win over the top RPI team and +1 to the bottom. But that values that win at 76x versus 6x. And I’m not sure what you would do with the losses. You could do -1 for a loss to #1 RPI and -76 to a loss to #76, but I don’t like taking that much away for losses. The loss is already a missed opportunity for a win.CU77 wrote: ↑Fri Mar 29, 2024 8:16 pmNot to pick on rolldodge specifically (because even the NCAA does this), but groups of 5 or whatever just put in artificial breakpoints: why is beating RPI 41 only half as good as beating RPI 40? Better is a smoothly sliding scale from top to bottom.
And then it should be iterated. Because if you need to modify RPI, shouldn't you then also modify something based on RPI?
This is what systems like KRACH and ColleyMatrix do. They start with raw record, then update based on strength of schedule (which is based on raw record), then update repeatedly until the ratings stabilize.
Another reason for the grouping (other than to match with the selection criteria) is to not take the RPI too seriously. It smoothes out the values within the groupings at the expense of the sharp cutoffs between them. Generally the goal is to assume more rather than less parity between all teams.
Another approach could be to use the actual RPI values so a win over #1 Duke right now would be +0.6864. Versus a win over #76 Hampton at +0.2905. That’s only a 2.64x difference between top and bottom.