The final T100 rankings for the regular season are not so much a retrospective look at what a team has done throughout the ’24 season as much as it is a prospective tool predicting the expectation for what they will do in the post-season. How else could the only undefeated team in the land end up as the #2 ranked team, overall, and not be considered the favorite to win the National Championship?
For those interested in how the T100 forecasts a future game between two teams, here is how it works: Add .5 to the home team (if there is one), then you subtract the lower T100 number from the higher T100 number to get the difference in points called “d.” Below is a “base formula” to determine chances for any favorite to win a match.
This is the third season doing T100. The name it’s evolved to being known as this year at Frog Jump. When it rated one team better than another, that team won roughly 88% of the time in 2024. This fact, by itself, shouldn’t be confused to mean it is a good model, though. If all D3 Men’s VB teams were to schedule more challenging opponents to play, than they presently do, then that number would shift lower than 88%. If the landscape on the whole were to schedule lighter opponents to produce better win/loss records for themselves, it would go higher than 88%. Should “parity” increase across the whole landscape, the forecast success rate for T100 (or any model) would naturally gravitate lower than 88%, too. However, is it reasonable to expect parity across a whole population to increase, if there are more than 10% new programs joining the ranks in a given year? Reduced parity doesn’t mean the weakest teams are staying weak or getting weaker. Based on my observations, the 100th best team in the landscape is a far more functional team this year than any of the two before it, but parity isn’t increasing just yet because the rate at which the teams in the middle and the upper echelon are improving is relatively at least as good, if not better. As all teams get better at differing rates, parity can either increase or decrease, and I would guess it decreased this year even as I see big improvement by some of the lesser skilled teams. This partly due to so many new programs plus a function of so many better programs having older players using Covid driven eligibility, too. More teams is a good thing to want for the sport, and necessary, before parity commensurate to D1 & D2 volleyball can begin to take place. I think TJ Breshears over at VBelo could confirm his model can’t grab 88% winners in a season, and I know his model is at least as solid as the one I built. D1 has better parity than D3 as seen through the lens of more aggressive non-conference scheduling, at least among the upper two-thirds of their landscape as opposed to ours.
When a D3 Volleyball team produces a 20-10 record, playing an average opponent ranked 30th best in the country, it is certainly a better season than another having gone 22-8 playing teams ranked roughly 50th best in the country. This is the same principle explaining one reason for why the NCAA SOS metric is weak. It is based on a flawed premise of cumulative wins and losses producing a signal of strength when it is only individual wins and losses that can do this. Even as it goes deeper to look at the wins and losses of opponents of their opponents, it doesn’t solve the problem it creates for itself by using the very data which conceived it in the first place. It’s like moving dirt with a pitchfork. Not a very good plan. So how about we solve it by building a pitchfork with twice as many prongs to move the dirt. That will work better, right? Yes, it will, but maybe a shovel is called for, instead!
This is why there is a need for models like the T100 to perform their “magic.” Though it isn’t magic. It is hard work to calibrate it moving as close to the truth as possible! The success rate of the T100 forecasting winners is a metric related most to how teams in this realm build their non-conference schedules. More than it is a testament to any state of parity or lack thereof which exists. Both, far more than it being something to convince anybody how good a model it is or isn’t. For example, in the NFL where teams do not schedule for themselves, and there is arguably some decent parity, the favorite tends to win about 64% of the time. If you play an NFL contest, with 50 to hundred contestants and predict 66% of the winners correctly, you may win it. Should there be 1,000 playing, you’d need to be a few percentage points higher than that if you’d want to come in first! A more aggressive forecasting model with a little luck would be in order.
If substituting d=2 into the formula above, the output is almost exactly a 77% chance for the favorite to win a match. The reason for the T100 being good is because when checking in on the exactly 100 matches this year that had differences between 1.8 & 2.2 (very close to 2), nearly 77 of them retrospectively show the favorite won the game. Likewise, when checking in with the 93 matches with differences between .8 & 1.2 (very close to 1), exactly 59 of the favorites predicted to win the match did precisely that. Check out how 59 out of 93 compares to the probability above when substituting the value d=1, and then we’ll know.
The point of the above is to be perfectly clear how good I think the T100 is, how hard it is to produce using cogent logic as it relates to a game which at times is completely illogical, but mostly as a backdrop to share what follows this last T100 of the regular season, further on down this post, is even better!
Better? Yes, BETTER! As in inching even closer to the truth. The most reliable ranking that wins & losses from almost 1500 matches can produce. (Along with the points and sets which drove them.) Certainly, it should predict what some human experts have to say about such things, even if they are holding on too tight to a list of 5 criteria, each one an off-shoot of wins and losses. Ironically, the best of their criteria is found lower on their list and its weakest found higher up on it.* This is because those below pertain to small groups of individual wins that matter, compared to those higher up associated with cumulative wins and losses that often don’t because being a by-product of how the landscape’s values drive its non-conference schedule building behavior.
Thanks to the folks behind the other 3 models who make so much of their content readily available to create something like the above. I know, regardless of how much better the T100 may or may not become, it will never rival what is found above because of their willingness to transparently report. Now it is getting close to the time to see how it compares to those whose opinion on the subject matters. They being the ones charged with the mission to make, in some cases, what will be really difficult decisions shaping the future possibilities for these teams over the the next 9 days. Thank goodness less than most will have in shaping it for themselves.
* A reminder for the NCAA criteria for decisions regarding ranking its teams:
- D3 Win/Loss %
- Strength of Schedule
- Win/Loss versus Regionally Ranked teams
- Record versus Common Opponents
- Head-to-Head Result
You must be logged in to post a comment.