I haven't worked this one through yet, I admit it. But I'm tired of the sour taste in my mouth with NCAA basketball. NCAA hockey has a "pairwise" comparison system which attempts to compare all teams in a round robin fashion (this has changed over the years from all teams to "all eligible teams", etc.). The idea is if you're better on the criterion then you get ranked higher. The downside of the hockey system is that it is dominated by the RPI calculation. There may be no way around this.
I may yet run through the numbers but I'm getting too old to want to bother with such things and I'd rather spend my time doing other things.
Current thought, and this is mostly so I can workshop this idea...
Two teams should be compared...
1) Head to head. Complete this and no other comparisons unlike hockey.
2) Record against common opponents.
3) Record against other eligible participants*
4) Computer rank (Ken Pomeroy? Something else, sure as hell not NET)
In this system once you "do better" in one comparison working from the top the other comparisons are discarded. Currently the way I see to game THIS system would be the home/road splits. The way I would counter-game this is to make all road wins worth 1.3 in the head to head and if the head to head comparison is the winner than the winner gets 1.3 points or something. Otherwise getting through on criteria 2-4 would get you 1.0 points.
I've had an idea for a toy to deal with number four. Not a great statistical tool but another Stop tool based on the implied idea of NET. You bin the teams by NET or KenPom then re-rate based upon NET record pools.
===
After you've made all the round-robin comparisons. Sum up all the points, rank order them, tie-break based on head to head within the criteria, failing that, tie-break on computer rank.
Start with all eligible programs with plus or at-500 records. Eliminate one school at a time from the bottom. Worst gets chucked. Recalculate line number 3 each time based on schools remaining plus all autobids. Autobid schools cannot be eliminated.
===
My view, any system that gets implemented for sports, including hockey, will be bullsnarp. Something will be wrong, something will be game-able. I can create a sophisticated likelihood-based model. I could do other things. I'd rather some system rather than the smokeroom hodgepodge that we have right now.
I get this is peak geekery. I just find the whole Sunday BS dumb and have for decades.
I may yet run through the numbers but I'm getting too old to want to bother with such things and I'd rather spend my time doing other things.
Current thought, and this is mostly so I can workshop this idea...
Two teams should be compared...
1) Head to head. Complete this and no other comparisons unlike hockey.
2) Record against common opponents.
3) Record against other eligible participants*
4) Computer rank (Ken Pomeroy? Something else, sure as hell not NET)
In this system once you "do better" in one comparison working from the top the other comparisons are discarded. Currently the way I see to game THIS system would be the home/road splits. The way I would counter-game this is to make all road wins worth 1.3 in the head to head and if the head to head comparison is the winner than the winner gets 1.3 points or something. Otherwise getting through on criteria 2-4 would get you 1.0 points.
I've had an idea for a toy to deal with number four. Not a great statistical tool but another Stop tool based on the implied idea of NET. You bin the teams by NET or KenPom then re-rate based upon NET record pools.
===
After you've made all the round-robin comparisons. Sum up all the points, rank order them, tie-break based on head to head within the criteria, failing that, tie-break on computer rank.
Start with all eligible programs with plus or at-500 records. Eliminate one school at a time from the bottom. Worst gets chucked. Recalculate line number 3 each time based on schools remaining plus all autobids. Autobid schools cannot be eliminated.
===
My view, any system that gets implemented for sports, including hockey, will be bullsnarp. Something will be wrong, something will be game-able. I can create a sophisticated likelihood-based model. I could do other things. I'd rather some system rather than the smokeroom hodgepodge that we have right now.
I get this is peak geekery. I just find the whole Sunday BS dumb and have for decades.
