The ‘E’xtra Edition
Tuesdays with MORE is both figurative and literal, not to mention a play on words with a great NY Times #1 Best Seller by Mitch Albom, which was later made into a movie starring Jack Lemmon and Hank Azaria roughly 25 years ago, “Tuesdays with Morrie.” On the one hand, it represents a new AVCA coaches poll coming out on Tuesday afternoons with a whole lot “more” baked into it. On the other, it is a simplistic & authentic composite ranking of teams across the D3 landscape using the median of four expert sources, i.e. 23 coaches from the AVCA, plus computer algorithm based rankings readily available at Frog-Jump (T100), Inside Hitter, and Massey Ratings. Thus the acronym M.O.R.E. – Median Order Rank by the Experts.
When it comes to valuation of sports teams, I have not only found “Wisdom of the Expert Crowd” to be better than any one individual expert to accurately forecast outcomes, but it has also been the basis for 99th percentile finishes competing with thousands in National Data Science competitions at Kaggle & Nate Silver’s 538, among dozens of local contests, too. So when I claim “MORE is likely the best source for ranking D3 volleyball teams going,” even better than T100, which I independently produce for FrogJump, it “more” than likely is exactly that. MORE is also a valid standard for measuring the quality of any other D3 Men’s Volleyball ranking system precisely because it is a worthy composite for wisdom of an expert crowd. I even utilize it to check the quality of the T100 and others in the public domain.
One ‘E’xtra is the AZR found in the last column above. It is the Average Z-Score Rating derived from the 3 computer models indicated above. Massey has power ratings ranging between 2 and 8, Inside Hitter from –2 to 25, and T100 typically between 2 and 18. Though the distribution of each is slightly skewed from what is considered perfectly normal in a statistics realm, it is still useful to calculate z-scores and then average them to arrive at the AZR. Had these ratings followed a “normal” model, then the expectation for the best 3 teams would be an AZR slightly higher than 2.0. i.e. Meaning more than 2 standard deviations above the mean (Like a 700 on a SAT). Since they don’t, the best teams tend to show an AZR a little less than 2.0 as seen by the best 4 teams from the table above being between 1.76 & 1.88. The AZR metric isn’t ordinal. This makes it more valuable to innumerate strength rather than being associated with uniform ranks from these sources. Precisely the reason why Z-scores are useful. What does all this mean? Consider the Kentucky Derby last May. The horses which finished in the top 3 were all bunched about 1 length from each other, while the horse that came in 4th was nowhere to be seen in the frame at the finish, almost 4 lengths behind the leaders. When it comes to purses and paying off bets, the order of finish is all that matters – Ordinals. When it comes to measuring speed and other characteristics of racing as it pertains to a horses’ ability to run fast and compete well, these things are difference makers in where, when, and how well they will race next time, not what place they finished last time. People who successfully bet on horses consider Beyer Numbers rather than win, place and show counts from recent races. Last Saturday I posted a scatter-plot whose axis was AZR and found 3 distinct “clumps” of teams I referred to as the precious metal tiers, Gold – Silver – Bronze. Sometimes ratings and the z-scores create an illusion of small differences in teams when there really are none, not so unlike 10 decimal places shown on a calculator display that a 7th grader thinks is meaningful enough to write on his paper for fear of losing credit. It is in these times a tiered approach to strength might make more sense. Based on the earlier post from Saturday, I now define the Bronze Tier in the AZR interval [1.00,1.33], a Silver Tier within an AZR interval (1.33, 1.67], and the Gold Tier to be any AZR above 1.67. This is the significance of the colored cells in the AZR column on the MORE table. Should there ever be a Silver above a Gold, a Bronze above a Silver, or a blank white cell above a Bronze, it simply means the ordinals of the experts ranks tell a slightly different story than the actual numerical ratings are describing.
There are times when a single value being used to define a complex concept can be more confounding than helpful. I have found SOS metrics to be the epitome of this. The NCAA has long used a flawed approach using W-L records of opponents (OW) and opponent’s of opponents (OOW) to determine theirs – Did you know there is a team in D3 right now ranked in the lower half of 124 teams whose present W% is better than the aggregate W% of the top 20 teams on today’s AVCA Poll? If that doesn’t tell you all you need to know about that, then I don’t know what could. It looks like the SOS metrics produced by Inside Hitter and Massey are good representations, but I still know too little about their derivation to have clarity on how resistant they are to outliers.
For example, if a team were to play 30 matches, and be 25-5 against the top half of teams in the landscape, then how severely does any SOS metric change when this same team gets to 26-5 by defeating the 100th best team in the landscape on the last day of a season? Particularly if it was only there because of some misunderstood politically motivated allegiance between two AD’s who scheduled it? This scenario is a mindful litmus test to see how SOS metrics behave with regard to outliers. These are some reasons why I don’t report any SOS metrics and I never offer W’s & L’s on the T100. I do, however, like a “picture” of the distribution for ranks of teams played and defeated, together with clarity of the ranks for who beat them.
The picture below can do for a skilled reader what a single metric can’t possibly do for anybody. The graphic is a series of box-plots indicating mean, median, middle 50%, skew and outliers of a distribution of defeated opponents and lays red discs on ranks of teams that beat them. I created an SOS metric last year I called TORA – Typical Opponent Rank – Adjusted. (The geometric mean of a median and the midpoint of the IQR for the ranks of all teams played.) I like it enough to place it next to each team’s name in ( ) because it is resistant to outliers, but will never consider using it as a stand alone value without a visual like the graphic shown below. I guess I could call this parallel box plot graphic “ORD” (Opponent Rank Distributions) since I am having some fun with acronyms in this post? LOL
Since it is Tuesday, with the most recent AVCA Poll out earlier today, now might be an appropriate time to show the AVCA top 20 movement over the first half of the season. Very little noise in the top 5 ranks by the coaches as they are all either undefeated or have just a single loss to one of the others from this elite group. I have boxed in the area to which I refer to demonstrate what little noise means. I am nearly certain two at-large bids will come from either the UVC, CCIW, or CVC, an opinion based only on the hopeful assumption neither Stevens or Wentworth are upset in the MAC or GNAC tournaments next month. Since “Conference Corner” is coming up to take the place of Spotlight next week in the matchup series at Frog-Jump, maybe a preview of the 3 strongest conferences in the land as it relates to the Coaches Poll might be in order. Here is the 7 week movement of all teams ranked by the AVCA this season so far, and then just the AVCA movements from teams in the 3 conferences which I expect to provide both at-large bids when the time comes.
