A couple more concerns I have related to NPI vulnerabilty include …
Pay Me Now or Pay Me Later?
I was reminded recently how injuries can significantly impact wins and losses, all depending on when they happen and for how long the duration, but not for reasons you may think.
Case Study- Part 1: Team A defeats both Team B & Team C while at full strength in January, only to find itself with half its starters injured as they lose to both Team D & Team E in February. Paying attention to Team B & Team C coming together in March where each defeats Team D & Team E in sweeps, both times.
The T100 is a “Pay Me Now” model adjusting for this because it measures the points earned or relinquished in the moment. Based solely on who the teams are when they play. i.e. The probability of its outcome at the time of competition, never to be revisited after the fact. The NPI is a “Pay Me Later” model. It will recalibrate the weighted average points for wins/losses/home/away/bonus against its opponents’ fixed season-long moving average in every iteration. One reason why it isn’t unveiled until much later in the season. For example, a couple years ago Wentworth lost 4 of 7 matches without Laboo available over a 15-day stretch, half of its losses across the whole season. Any direct effect on Wentworth’s standing was unfortunate, but in a “Pay Me Now” construct each loss hurt less and less as they come in a bunch. i.e. The probability of their next match being a win is reduced because the previous loss(es) lowered their metric before being played. However, allowing these to permeate a recalibration for Wentworth’s opponents across a season-long to-date average using NPI, “a Pay Me Later” concept, tells a less true story in a manner humans could never consider it. Humans deciding between two teams that each defeated Wentworth likely wouldn’t think the same about them when their best player wasn’t available for one and was for the other, rightfully so. The NPI doesn’t know the difference because it isn’t discerning. Some might claim that’s exactly the point. I would rebut “Not Discerning” isn’t always a good thing!
Case Study-Part 2: Should B, C, D, & E all be on the bubble and Team A be just off it with a rank of approximately 15th.
This scenario is unfortunate for team A because they did lose those two matches under some difficult circumstances, but it’s just. Where I struggle? If both D & E get the at-large bids instead of B & C by less than 0.8 points, it is nowhere near justifiable. First, there is no statistically significant difference between metrics this close, but mostly because that is equivalent to an approximate ripple effect altering the NPI for the conditions stated above, regarding 2 teams who are decidedly better than the other two. The evidence by sweeping each after the fact is something a “Pay Me Now” system rectifies in the moment, while a “Pay Me Later” system won’t recognize it ever. Ironic that in this closed system of 5 teams, the ones to get the at large bids would both be 1-2, having been lucky to have earned the one win, compared to those not getting at-large bids each having been a more legitimate 2-1. Exactly how impressive would the other wins by D & E have to be for this to happen? Not very much at all as it turns out – Just a little higher win% against a little less quality of opponents.
Though the NPI is not a voting system, I would argue as a valuation system it ought to be free from irrelevant alternatives (Arrow’s Impossibility Theorem). Very little could be more irrelevant than a common opponent of these 4 teams under the conditions described. If its ripple effect were ever to have such a horrendous flip-flop of at-large bid recipients, this would almost be criminal. Not just punishing A which it must, but B & C as accessories having played no part, while rewarding D & E for nothing more than being in the right place at the right time. This is often avoidable in a “Pay Me Now” system than a “Pay Me Later” type which averages team performance over a whole season to then assume an equivalent SOS component for any who played it across a 14-week timeframe.
Coaches never miss a beat to talk about how their teams improve over the course of a season. If that is the accepted norm, and this improvement occurs at various rates of growth due to the varying degrees of good coaching out there, then where is the logic in assuming a win in January over a #1 ranked team should then be treated retrospectively as a win in April over a #11 ranked team. I promise you the online schedule doesn’t go back and change this distinction. LOL
Home & Away isn’t Equal for all Matches? This has nothing to do with the fact home court means a lot more to some than others… Though it does.
When two equal teams play each other on a neutral court, then their chances of winning would have to be 50-50, certainly. Otherwise, one or both of the premises has to be false. When one of these teams visits the other, the home team win-probability goes to roughly 58%. i.e. The marginal win-probability transitioning from away to home playing against an equal opponent is a 16% bump. (From 42% to 58%.) As the disparity in the strengths of these two teams increases, reaching a difference of approximately three points per set (i.e. One team who’d win on average by a score of 25-22), it becomes evident that the marginal win-probability goes from roughly 81% to 89% along the “Away-Neutral-Home” categories. Now being an 8% bump as the advantage for the favorite playing home rather than away. That is half as much as when the opponents were equivalent. By the time you get to a disparity of 6 points per set (Average set score 25-19), marginal win-probability converges to about 2%, going from a 97+% chance to win away up to a 99+% chance to win at home. The idea here being that home vs. away matters 8 times as much when two equal teams are playing each other compared to if they were about 60 ranked spots apart across the landscape That is variable, not constant. When the #1 defeats the #60 away it sure shouldn’t be weighted the same as when they played #2 away. In the first instance it matters not and in the second, a whole lot more.
To have a fixed weight for H & A, not taking any of the above into consideration may have ramifications. Especially for a system which pretends it needs to report its metrics to the nearest 0.001 for little to no good reason I am aware. Most who’d read about the better 2 of 6 teams on a bubble t earn an at-large bid while squeezed within a 0.498 range between 62.001 & 62.499 probably won’t recognize that precision to the nearest .001 doesn’t characterize the accuracy of this NPI metric at all because it lends itself to “numeracy bias.” Something whoever posts these results “COUNTS ON” I’m sure! (Pun Intended)
I’m all for the committee wanting to “reward the institutions that play a significant amount of their matches as the visiting team, while encouraging those that play more home matches to travel.” But not if the risk of doing so isn’t commensurate to the reward, especially as it pertains to scheduling better ranked opponents to play on the road. (The reward only rivals the risk when lucrative win bonuses would occur – Usually defeating a Top 5 program.) The fact, “That the past 3 years of data has shown 55% of all matches are won by the home team, with that rate even higher in non-conference matches,” is less a function of the state of home court advantage, than it is a symptom of scheduling practices for non-conference. Assuming conferences flip-flop Home and Away every year or two (depending on if teams play each other once or twice in a season), I believe the proportion of home winners in conference matches, only, over any two-year period, would better inform any decision about Home & Away weighting. That should be looked at as a function of a degree of parity which exists within any conference, too. The last two years there have been no “upsets” in the MAC regardless of the court played on, and in the UVC there have been many because the teams are so much closer in strength. Based on this I would suggest home court matters far less in the MAC and is a greater influencer in the UVC, for example. This only because it is measured through just wins and losses.
These are some of the things I will take a look at when the NPI is unveiled. If any appear to make it potentially vulnerable in its task to choose the worthiest at-large teams a few weeks later, I’d want to see how it evolves over the month to take note how it attempts to correct itself. Especially over the conference championship week in April, after which I expect its final version will be published. When I looked in on WVB this Fall and saw the coaches 22nd ranked team not in as one of the 64 tournament teams, it was shocking. (There were more than 20 at-large bids available.) That same weekend at a family brunch my cousin shared with me how the #1 & #2 ranked D3 football teams all year in its coaches’ poll, both undefeated, and who played to a single point difference in last year’s National Championship game were put into a bracket to play in an Elite 8 quarterfinal. It is my hope nothing quite so “head-scratching” comes about for MVB later this Spring. And if for some reason it should come to pass, I will want to know the reason sooner than later. That is why I have already considered such things. As a coach you want to have prepared for any situation so that when the time comes to make a choice for your team it’s already been decided. When I coached swimming, it wouldn’t be uncommon for me to carry 3 different 400-Free Relay entry cards in my pocket in case I needed to split relays different ways to improve chances to win a swim meet in the last event. As a writer trying to inform the engaged men’s volleyball public what’s happening, I take preparation no less seriously.