T-minus 3 weeks and counting until the regular season comes to an end and Conference Championship Week knocks on the door. I’ve finally rid the T100 of the regional column. It was just taking up space with no practical use in reporting the best 100 teams across the land. However, when it comes to comparing and contrasting the expert models, both those that don’t have skin in the game (T100, Massey, IH) and the one that does (NPI), you can be sure a regional stratification of the data will not be ignored any time soon. As a general rule most who read this weekly post know the geographic locations of conferences already, so a numeral defining it is redundant. Regional rankings are vestigial. i.e. A remnant of a structure once functional but having lost its functionality as NCAA bids have given way to the NPI. Not so unlike your appendix or wisdom teeth, really. Though a ruptured appendix or impacted wisdom teeth could sure introduce you to a world of hurt under some dire circumstances, kind of like the NPI might. LOL
For those who want to know how the T100 metrics below forecast a probability for your team to win a match, here is how it’s done: D = Your Team Rating minus Your Opponent Rating (Add 0.5 to the home team before subtracting the ratings.) … P(Win) = 1/(1+e^(-.6(D)))
Consider something seen below on this week’s T100 I’ve never seen before, and if I’m being honest, it kind of blows my mind! Look at the ranked teams from #26 to #29. The absolutely insane MCVL! This Friday & Saturday has to be “Must See Streaming” as Trine plays at Baldwin Wallace and then the next day at Mount Union, both times a “42% DOG.”* Neither Trine nor MU have lost in the MCVL yet, and just two days ago on Friday, 3/14, BW took its first “L” against Wittenberg when the model shown above had it 71% to win at home. Less than 24 hours later Mount Union then defeated Wittenberg as a 61% home favorite. This is the first week all season the T100 hasn’t projected BW as the “MCVL Champ” after going 1 for 2 in forecasting the winner of these two games using the same model. In real time, the other 3 “experts” missed on both. Only worth mentioning because I know T100, Massey & Inside Hitter are all about 7 for 8 in forecasting accuracy this year. Maybe we should start reporting it with an asterisk, excluding any MCVL matches as to make them all look a little better! LOL
I wrote above that Trine is 42% in both matches. Let’s see how to use the formula:
TRINE at BW: D =13.84 -(13.87+0.5) = -0.53, so P(W)=1/ (1+e^ (-.6*(-0.53))) = .4212 which is 42%
TRINE at MU: D =13.84 -(13.90+0.5) = -0.56, so P(W)=1/ (1+e^ (-.6*(-0.56))) = .4168 which is 42%*
*If Trine & MU both win the night before, it bumps to 49%. but should Trine lose to BW, then it will become 34%. This assumes MU defeats Calvin on Friday evening, too. (95% to do so.)
The only projected Conference Champ different this week is the aforementioned Mount Union in the MCVL. However, the musical chairs for the first 5 out continues one part of the saga. The other, of course, is the first NPI which produces the present state of such things. Assuming no “Conference Champion Upsets,” a foolhardy thing to do, what is seen below doesn’t disagree with the NPI’s initial assessment of the at-large candidates. However, be careful not to let that lull you into complacency. That might be the thing most care about right now, and understandably so, but don’t get too comfortable just yet.
I changed up the bracket model to allow every region to have a prospective 5 team bracket. This makes it so any region could be forecasted to be the “4-Teamer.” It isn’t so surprising that either NYU or Southern Virginia are almost 60% to come away with a title, given neither has lost a single match this year. The NPI undervalues Carthage far more than the T100 (3rd to 7th), while it overvalues the Messiah resume, too. (7th to 3rd when the NPI came out last week but now rated 6th by the T100) This is a little surprising, if only because the T100 was in on Messiah as a Top 7 team in America well before anybody else – Just not seeing it in the Top 3 as NPI sees it so far! Lancaster Bible has been surging as of late and could end up with a bye to need just one exceptional match to “sneak” into a quarterfinal. That would be fun for everybody, EXCEPT the opponent that would have been a little shocked by their tenacious “D.” I have little doubt that would be the main reason for such an “upset.”
I have been posting the weekly forecasts below as part of the “Fixings” for two months now and as much as I’d like to tell you I am proud of an 86% accuracy rate over that time, that is neither a reason to be proud nor to hurt my shoulder patting myself on the back. The thing I am most satisfied with is that when averaging these nearly one-thousand favorites’ win probabilities together for that period of time, it is almost the exact same 86%. This makes it plausible the model is “dead nuts” as good as I can make it. Every sport has its own natural variation to deal with. You won’t predict better than 63% winners in MLB. Until this year you couldn’t predict better than 73% winners in the NFL, and Men’s D3 Volleyball happens to be in about an 86% “sweet spot.” It was a few points higher than that just a couple years ago when it rivaled a 90% threshold. What this means is that the parity in the middle of the distribution is coming together. I can attest to that having seen SUNY Poly play yesterday as they moved into the mid 50’s for the first time this year. They look to me to be far more skilled in ball control than any in the 40’s I saw in 2022 & 2023! Kudos to the landscape for improving at such a strong “clip!” It is impressive!
Remember the probabilities offered below are based on the March 16th T100 metrics. By the time many of these games are played, the probabilities will have shifted in real time. In fact, that 86% I mentioned earlier actually bumps to nearer 89% because there will be about 4 matches below that will flip-flop its favorites by the time the first serve is played. In a couple days when I offer an updated Win-Matrix, you are likely to see 92% of the better ranked teams above the diagonal have won matches already played. This is because every time a model incorporates new data it inches its way closer to the absolute truth, never quite being able to say it has arrived. This is why all of the experts, even the NPI, converge nearer each other every week. I just think it unfortunate the target the NPI is approaching is not the same as the other 3. And it’s not just because it is an award-based algorithm rather than a predictive model, either.
Data scientists out there should think about that a little bit. I know the rest of you stopped reading about 5 minutes ago! LOL
