A funky visual happens when you stratify the rankings of the T100, Massey, & Inside Hitter by way of the CWR (Conference Weighted Rating). The CWR is provided as part of the “Fixings” every Sunday evening when I post a new T100. I did something similar to this concept last year in black & white that made it look like DNA strips, but this year I decided to color code the conferences and put it against a white backdrop. As you scan it from left to right you sure do get an understanding as to how the CWR metrics tell an accurate story for conference strength.
There are all sorts of visual cues that can be determined when you take a similar construct but instead use it to look for relationships across the landscape. The three I chose below to demonstrate are [1] How T1/MAS/IH see the new ODAC conference in the D3MVB landscape [2] A look at the density of teams belonging to the top 4 conferences in the D3MVB landscape [3] The competitive strength of the new programs along the landscape as of right now – 1st & 2nd Year
As a part of “Tuesdays with MORE” posted earlier this week, I mentioned that there’s been a history of Massey and IH rating Midwest teams a little stronger than the T100. i.e. The T100 rating Midwest teams a little weaker than the other two models. (Goes both ways…) I decided to create a visual of this by differentiating Midwest teams within Quintiles of the strongest 125 teams across the landscape. (Each Quintile is a 20% cross section – 25 teams per – of the whole.) The varying shades of red show each team belonging to its Midwest conference, plus Wisconsin Steven’s Point, who I originally thought was the lone independent in the region. Now, it occurs to me Westminster (MO) should probably find its way on to this graphic should I create it again any time in the future. This would make 30 Midwest teams out of the 127 from the NCAA D3 landscape still playing – A little less than a quarter of them. The ribbons below show a density of where they fall on the distribution. Note the “black arrows” showing a transition from the T100 <—> Massey <—> Inside Hitter regarding their Midwest ranked positions. This pattern shows how Massey’s algorithm has a tendency to rate the Midwest better than both IH & T100 at various points along the spectrum, and to a much lesser degree, the IH algorithm ranking this region’s teams, on average, just a smidge stronger than what the T100 claims. These are differences, the degree to which aren’t systematic across every part of the distribution, nor do they have a magnitude elevated large enough to be remotely concerning.
Even if those seen below had been systematic & large enough across the whole spectrum of the D3MVB landscape, there’s absolutely no evidence for any one being closer to the truth than the others to make it the standard! In fact, there is strong evidence against that! Does not matter either way, however, because as I’ve said a number of times before, “None has skin in the game!” i.e. There’s no injury to any region, conference, team or player even had such an assertion been able to be proved.
The same kind of graphic as above with color coded noise on a scatterplot (like the one shown on Tuesdays with MORE a couple days ago) might demonstrate if any model favors specific conference(s), too. i.e. Not only a regional subgroup like that shown above. If there ever were enough evidence for any to be disadvantaged by an algorithmic systemic bias, then it stands to reason there might be some others to have been a beneficiary of the same. Possibly conferences all from the same region, or even a subset of teams all from the same conference, if the sample sizes were large enough. After all, it is the conferences who mandate every team plays each other thus offering the most compelling evidence for team strength, sometimes more than any of the models, themselves. Normally, you’d think three times the match data from out of conference matches would provide more confidence for a team’s strength, but there are times when non-conference play can confound those results, too. Just the other day I invited my wife to come take a look at a match rated nearly 50-50 by the T100, between two teams ranked in the high 30’s. It was being played in the 5th set with a score of 12-10 in the moment and an equal number of points scored on both sides throughout, and then I said to her, “Check out the team’s records on the top.” One was 11-5 and the other was 1-7. Not being mathematically inclined, she just shook her head in disbelief. And then I thought how unfortunate a team would have been this year to have already defeated the 1-7 team rather than the one that was 11-5 so far. That is more confounding than it is clarifying!
With the T100, Massey, and IH algorithms presently entrenched as equivalently truthful strength valuation models for D3MVB, it makes sense they could act as a standard to measure the NPI when it is unveiled to the public. Measuring both the magnitude and any systematic nature of its noise would be a key to either an endorsement or an indictment of its veracity. With the NPI being introduced in a year, when at its best, might have to leave at least 2 of the top 10 home for the dance, I’d hate to see it at anything other than its best.