Pablo

Let's see what Pablo Team rankings has 1) Penn St 12) Lehigh 20) Penn 21) Pitt 29) Lock Haven 41) Bucknell 45) Drexel 58) Edinboro 64) Clarion 74) Franklin&Marshall 78) Bloomsburg It's pretty consistent with your tiers, but a) Lock Haven is distinct from Bucknell/Drexel, and Pablo likes them more b) Clarion is more like Edinboro than F&M I wouldn't put in a separate tier between F&M and Bloomsburg; F&M is a lot closer to Bloomsburg than Lehigh is to Penn St (Pablo has F&M over Bloomsburg 27 - 11; PSU over Lehigh is like 36 - 4)

There are 340 D1 volleyball teams and a lot (maybe not quite as many) softball teams that would be part of this, for example.

My original VBA macros were written back in about 2002, so they get grandfathered in.

4000 variables (wrestlers) with 17000 data points (although they could be separated into probably groups of about 400 and maybe 1700 outcomes and run in serial). I don't know python.

I have been using excel with the built in solver. Structurally, it works well and is straightforward with the built-in functions. Yeah, I'm sure I could program the arrays in whatever app, but the multi-variable, non-linear regression is well-beyond my skills.

One of the things that I can do with Pablo rankings is to compare teams. So what I have done now is to rank the teams, from a dual perspective. A couple of things 1) The first step in the process is to identify a starting lineup for each team. I'm not going to go through every team's lineup to see who the coaches are sending out, so I just pick them on my own. I have a few guidelines I use a) The starting point is the wrester at each weight who had the most matches. In most cases, that's all you need to look at b) Sometimes, when there are multiple wrestlers with lots of matches, it can be more subtle. In that case, what I generally do is to choose the better wrestler (by Pablo rating). This means that if there is a redshirt who has lots of matches who is rated higher than a starter, I'm going with the redshirt. Close enough. And how many matches does a wrestler need? My general guideline was maybe to be within 5 matches of the guy with the most. In the end, I am trying to put the best team on the mat. 2) OK, with the team, the way I evaluate them is to do a round-robin comparison of every team vs every other team. There are 79 DI wrestling teams, so each has 78 matches. The rankings just show who has the most wins over those 78 matches. 3) Scoring: by using the rankings before the NCAA tournament and the outcomes at the tournament, I have been able to create a model to predict the outcomes of a match given a Pablo rating difference. Therefore, there is a prediction of how likely each wrestler is to win, and how likely are the different outcome (dec, MD, TF, fall) in each situation. Thus, I can determine the expected number of dual points given the rating difference. Note that this takes into account the probability of either wrestler winning (the basis of Pablo rankings) and the value of that win. In each dual, I compare each weight and figure out the expected outcomes for the two teams at that weight (note that "expected outcome" refers to the average over a large number of trials). The team that scores more wins the dual. 4) Tie-breakers: In those cases where teams have the same number of wins over the round-robin, the higher ranked team is the one that won the head-to-head. In case of a 3-way tie not decided by head-to-head (all three are 1-1), I used the biggest total point differential among those three teams. That being said, here are the team rankings. I've included the average match score difference just for reference (it isn't used for ranking), remember, there are 78 dual matches. This is based on final (post NCAAs) Pablo ratings. Rank Team Wins Avg Score Diff 1. Penn St 78 43.0 2. Nebraska 77 31.3 3. Iowa 76 29.8 4. Oklahoma St 75 28.0 5. Ohio St 74 28.0 6. Minnesota 73 25.4 7. Northern Iowa 72 24.6 8. Illinois 71 20.9 9. Cornell 70 19.3 10. NC St 69 18.4 11. Virginia Tech 68 17.8 12. Lehigh 67 17.6 13. Iowa St 66 17.1 14. South Dakota St 65 16.8 15. Michigan 64 14.3 16. Rutgers 63 14.3 17. North Carolina 62 12.6 18. Missouri 61 10.5 19. Oklahoma 60 10.8 20. Pennsylvania 58 9.9 21. Pittsburgh 58 9.8 22. Little Rock 57 9.8 23. Stanford 56 7.4 24. Purdue 56 9.7 25. Indiana 54 9.3 26. Oregon St 53 8.8 27. Arizona St 51 6.5 27. Army 51 7.1 29. Lock Haven 50 6.4 30. Princeton 49 5.9 31. Maryland 48 5.4 32. Virginia 47 5.3 33. Wyoming 46 4.5 34. West Virginia 46 4.2 35. Northern Colorado 44 3.9 36. Navy 43 3.5 37. North Dakota St 42 1.6 38. Northwestern 41 1.7 39. Utah Valley 40 1.0 40. Cal Poly 39 0.8 41. Bucknell 38 0.0 42. Columbia 37 -1.0 43. Central Michigan 36 -1.2 44. Binghamton 35 -1.5 45. Drexel 34 -2.0 46. Rider 33 -3.2 47. Appalachian St 32 -4.1 48. Cal Baptist 31 -4.3 49. George Mason 30 -4.9 50. Campbell 28 -5.6 51. Wisconsin 28 -5.7 52. Ohio University 28 -7.5 53. Chattanooga 26 -7.7 54. The Citadel 25 -8.4 55. Michigan St 24 -9.2 56. Bellarmine 23 -9.6 57. American 22 -10.9 58. Edinboro 20 -11.0 59. SIU Edwardsville 20 -11.4 60. CSU Bakersfield 19 -10.4 61. Northern Illinois 19 -11.6 62. Hofstra 17 -11.8 63. Air Force 16 -12.6 64. Clarion 15 -12.8 65. Harvard 14 -13.8 66. Brown 13 -14.3 67. Cleveland St 12 -15.8 68. LIU 11 -16.5 69. Kent St 10 -18.5 70. Buffalo 9 -20.1 71. VMI 8 -22.0 72. Gardner-Webb 7 -22.9 73. Duke 6 -22.7 74. Franklin & Marshall 5 -24.1 75. Davidson 4 -26.2 76. Morgan St 3 -27.0 77. Sacred Heart 2 -29.3 78. Bloomsburg 1 -37.9 79. Presbyterian 0 -38.9

I don't l know how much farther it can go, but in order to improve upon what I've done, I could go in and do a lot more testing of actual predictive parameters. However, the challenge in that is that every time I tweak something to rerun the calculation, it takes like 24 hours to try it again. I've used match-pair data (when two wrestlers face each other twice) to get a pretty good idea of the model, but there are other modeling aspects that can be tweaked. But I'm happy that it's able to run comparable to seeding.

You have to look at it match by match to compare them, because that is how Pablo determines their ratings. Pablo ratings are based on who you wrestle, who you beat, who you lose to, and by how much. It's a match-by-match assessment. It's all about the specifics. Records matter, to the extent that beating an opponent is a good outcome and is a better predictor of success than losing is, but who you beat also matters. So if you ask, "Why is X ranked higher than Y" you have to look at the match-by-match outcomes to see where they got their value. " For instance did Pablo consider Olympic gold? " I don't have the ability to compare freestyle and folkstyle outcomes, and I wouldn't even try.

Um, Pablo did just outperform the current seeding process. Not by a lot (5 matches), but it did.

I'm not going to go through match by match, but I will note that Pablo had Steveson rated ahead of Hendrickson before the NCAAs, so that was already established (although I will note, Pablo gave Hendrickson a much better chance of winning than anyone else was giving him). His win in the championship caught him up significantly, but not enough to overcome the difference.

Fortunately for Robinson, I have the NCAA results, which are weighted higher (more recent) and those losses in Cliff Keene are missing.

Of course that is silly. You can't just compare records, and Pablo is not merely a comparison of records. Quality of the competition and the domination of your opponents also matters. Based on that one outcome, what we would expect is that if they were to wrestle again, that Hendrickson has about a 65% chance of winning. Certainly an advantage, but it's not huge. Moreover, that is only one match, and Pablo is taking the other outcomes into account. Look through the rankings, there are tons of instances where wrestlers with way worse w/l records are ranked higher. At 285, Hayden Flipovich is 7-16 and ranked 10 spots higher than Connor Barket from Duke, who is 21-10. It's common. Pablo looks at the full body of work (that it has, at least).

That's wild. For some reason, a lot of his early season matches aren't there. I thought I had gotten everything from NC State, but it looks like I missed some. Even his Keene results are missing (the only loss I have is the ACC champs). This is likely because I initially pulled Cliff Keene results off Flo and it wasn't from Track. But still I thought I had it. Oh well, until there is a convenient way to get scores, this will happen.

The difference between them is negligible - basically, Pablo is saying that it's a toss up, and I think that is a pretty fair assessment based on the outcome.

You also didn't even read the first post.