D3 WBB -> Here's another experiment. Using my inferred Primary Criteria Weights,
how would different single-game results affect your NCAA Resume? I stripped out all the "elbows" and "fudge factors." So we only have weights for WP, SOS, W vRRO, L vRRO. We're more concerned about change from the baseline, not the accuracy or precision of the baseline. Team A resume: 0.500 over 24 games, 0.550 NCAA SOS. Oh, also assume OOWP is constant in the single game. Beat a 12-12 team that is not an RRO: +2.28% Beat a 1-23 team, no RRO: +0.99% Lose to a 12-12 team, no RRO: -2.57% Lose to a 1-23 team, no RRO: -3.86% Team B Resume: 0.833 over 24 games. 0.535 SOS, 3-2 vRRO. Beat a 12-12 team that is not an RRO: +1.23% Lose to a 12-12 team, no RRO: -2.25% Beat a 1-23 team, no RRO: +0.25% Lose to a 1-23 team, no RRO: -3.23% Beat a 23-1 team, RRO: +7.01% Lose to a 23-1 team, RRO: -1.79% It all seems pretty reasonable. I think that's because these weights are treating changes linearly. If I add the "elbows" back into my "amateur inferred weights on primary criteria", we see something different. Team B Resume: 0.833 over 24 games. 0.535 SOS, 3-2 vRRO. Beat a 12-12 team that is not an RRO: +1.19% Lose to a 12-12 team, no RRO: -6.65% Beat a 1-23 team, no RRO: -8.91% Lose to a 1-23 team, no RRO: -10.02% Beat a 23-1 team, RRO: +21.44% Lose to a 23-1 team, RRO: +6.87% Lose to a 19-5 team, RRO: +2.17% (yes I picked a starting SOS close to an "elbow" so the results would flip you to one side or the other) It becomes better to LOSE to a 12-12 nonRRO than to beat a 1-23 team. I know any SOS, WP, or vRRO "elbows" or "thresholds" are likely implicit rather than explicit. I don't know anything for sure, but would have to imagine no committee has something written down like "below 530 SOS is put in a different pile." But implicitly it's HUMAN to create mental buckets. If we don't think that an SOS of 531 vs 529 matters, then we need to enact systems that ensure our human nature doesn't create these mental buckets or heuristics. This isn't to say that a 100% computer model is the answer; just that computers can help overcome human things that we don't want to have an impact on how we value resumes.
0 Comments
|
AuthorI'm Scott! ArchivesCategories |