We had a very good question from a reader. After both UConn 2023 and Houston 2021 won the two most lopsided Value Add Basketball Games we every played (77-39 vs. St. Peter's and 102-50 over Princeton respectively) why didn't they shoot up even higher in our all-time rankings? UConn moved up from 46th to 38th in our all-time rankings, and Houston from 60th to 45th.
First, here are the two teams on this google sheet where we keep all game results and all-time rankings.
We start by calculating something similar to an SRS (Simple Rating System) based on the player cards in the game. Houston's cards rank tied with a bunch of teams for 60th place to start the season due to their cards adding up to a +0, so they are about average for the great teams in the game (meaning more than 20 points better than an average team during a season). UConn ranked as two points better at +2, which made them one of many teams tied for 46th of all teams.
They were playing the two worst (though still) great teams in the game, with St. Peter's in last with a -19 rating and Princeton next to last with a -16 - both still better than a true AVERAGE team but much worse than the average great teams in the game.
Like www.kenpom.com, we start teams with a filler of several games where they were exactly as good as their cards seem to make them. For Pomeroy, he starts them with 10 such "filler game scores" but he replaced one of the 10 with an actual game played. For me, I start them with three "filler game scores" but I never take those three games out of the rating whereas once a real team has played 10 games Pomeroy no longer keeps those filler scores.
So we start off as though UConn had played three games and each time played at a +2 level and Houston had played three games and each been at a +0 ratings.
Since UConn is +2 and St. Peter's is -19 that would be an expected victory margin of 21 points for scoring purposes (only the real data geeks like me would note that would really be in 100 possessions, but there is a reason so stick with me) and with Houston 0 but Princeton a -16 we would expect Houston to win by 16 on a neutral court (if a home court in which die rolls of 36 and 66 can be flipped by the home team, then we expect the home team to win by four more points than the math would say).
If we stopped right there like Pomeroy does then in fact Houston and UConn would have shot way up in the rankings with Houston's incredible 52-point win and UConn's 38-point win. But we don't. Our max formula caps the credit for the win margin at 20 points so it only counts as a 20-point win for both teams and thus just a little more than we expected.
The second thing we do that Pomeroy does not do however is to add 4 points for winning the game and subtract 4 for losing (one reason I keep the margin at the 100 possession rate as referenced two paragraphs earlier).
Therefore UConn is considered to have a +5 for the game (20 point margin win is one less than expected and add 4 to to get to UConn getting credit for winning 3 points more than expected. But expected is based on their +2 rating, so we determine they played like a +5 team that game instead of a +2 team. Therefore they have credit for one +5 game and the three filler games of +2 each average out to a +2.8, which moves them up from the 46th the 38th greatest team ever.
Same credit for Houston - 20 point max win when expected to be a 16 for the game is +4, and then +4 for the win is a +8 for the game (GS is game score) added to three +0 filer games and they move up to a rating of +2.0 and 45th place as just ahead of where UConn was before their win.
Just keep in mind that if you look at the rankings, the second number ranking is how good we actually believe the team is based on the cards, but the first column is the fun "Pomeroy-like" rating that moves up or down with a win.
The biggest disparities right now include two teams with much better results than we will expect as we play more games over time. We believe Colorado's 2021 cards are the 121st best in the game (the second number) but they have pulled off three very close wins to start 3-0 and since I give the 4 points extra credit for winning they have shot all the way to 34th in the fun rankings.
The second two are linked - we believe Mark Aguirre and Terry Cummings 1980 DePaul team has the 106th best cards but they had the biggest upset we've had by beating Kareem's UCLA 1967 team that we believe is 2nd best behind only Bill Walton's UCLA team six years later. Not only did they win, but it was a shockingly easy 86-71 stunner. Look, the UCLA team is much better BUT a team with Aguirre and Cummings that was ranked No. 1 before the early tournament exit certainly is good enough that they could beat UCLA if they caught them in the one in six games that each college team plays well above their head. Consider that the next season Houston beat UCLA in the "Game of the Century," to stay No. 1 ahead of them all year until the national title game, when UCLA improved by 34 points over the first meeting to get revenge 101-69. The dice can duplicate hot streaks in real games and in this case DePaul got hot.
As a result of that and two other upsets, DePaul is now 3-1 and moved up for 106th to 20th in our fun rankings, while UCLA is 2-1 and dropped form 2nd to 28th.
My guess is if I had a rematch UCLA would exact their revenge like they did against Houston.