Seen below are the latest published rankings of all expert models as of the last conference match of the season played yesterday. They are placed left to right in order of the algorithms which weigh the accumulation of wins, either directly or indirectly, from least to most. The models used by the NCAA are on the right with its new D3 ranking system called NPI just inside the one that D1MVB begins to employ this year called KPI. It just so happens that they are also placed in order, left to right, by their disagreement for the Midwest teams going from slightly favoring to strongly biased against. No surprise there for a couple reasons. First, the NCAA is enamored by the accumulation of wins regardless of who they are against. The RPI for years was an attempt to dampen it and then the KPI and now the NPI are the latest iterations to attempt to dampen it just a little more. However, when you have a distribution of Midwest D3 Volleyball teams whose median team wins half its games being inarguably stronger than the median of those not from the Midwest who will also win half its games, the accumulation of wins concept is like trying to fit a square peg in a round hole. The RPI sanded those corners just a little and the KPI and NPI took out a file to round them even more, but the problem isn’t the grade of the sandpaper or the size of the file. It’s the fact that the false premise of a square peg in the first place, at least as it pertains to D3 Midwest Volleyball teams.
As you scan the Midwest teams rankings from left to right, don’t lose sight of those from Ohio who tend to play at least half its schedule against teams not of the Midwest. Three times more frequently than those farther West from Indiana, Illinois, and Wisconsin tend to compete against. This is important because just as data proved smoking causes cancer by showing its incidence rate increased with the number of packs per day, those from Ohio tend to have a significantly lesser biased effect on its rank having played a significantly higher portion of teams not of the Midwest.
The depiction below shows that Massey, Inside Hitter, and T100 rank 20 Midwest Conference teams in their top half while those on the right place just 17 & 15, respectively. (Not including Wisconsin Steven’s Point which is ranked left to right: #34, #35, #37, and then #44 by the NPI to #58 by the KPI.) The median team in the Midwest is ranked between #41 & #51 by those three models on the left and above #60 by the NCAA models on the right. The NPI and KPI can’t help itself because it is driven to rank the middle of the Midwest near the middle of the whole landscape.
The anomaly this last weekend was that 14 of 14 top seeded teams from their conferences all secured automatic bids. A less than 1% event considering the historical rate. As you scan the teams to achieve this by focusing in on the blue from left to right, there is more evidence of the accumulation of wins models at work. The #1 seeds all had great win loss records against their conferences, many who play home & away, and then they ran the table for a couple more wins last week. In many cases defeating opponents for a 3rd time to generate 3 times the weight for its NPI rating.
When graphing all 5 ranks as a function of the median Z-Score of their metrics, it is pretty obvious they all generate nearly identical models which differ mostly by their variation. Demonstrating the principles seen above looking at the comparative rankings, take note of the Midwest data points labeled with lighter colors of the NPI and KPI. Nearly 100% of them are systematically underrated by the NCAA based models. (Seen above the best fitting model of each.)
The problem isn’t directly driven by the middle of the Midwest distribution being poorly ranked. It’s that the best of them who systematically are injured by becoming devalued by the NPI because it awards winners of matches with 80% of their opponent’s metric. This produces a compounding ripple effect thereby creating exactly what is seen above.