Formula based rankings

I couldn't sleep in today and was looking through andegre's (wrestlestat) and setonhallpirate's (britt malinsky) rankings (dual impact index, DPI). DPI is interesting because they rank guys at their place amongst the 76 D1 starters and also add in redshirts/backups. Both use formulas to rank individuals and teams and they're an interesting aside to compare with the other rankings. Theoretically this would remove bias but also the "eye test". Here is how NDSU starters and some interesting others rank:
125- Rodriguez- 7th in both
125- Paul Bianchi- 56 in WrestleStat, 17 overall (#3 among all redshirts/backups) in DPI
133- Sykora- 71 in WrestleStat, 26 overall in DPI (23 among current D1 starters)
133- Landeros, who started for NDSU in semester 1, would rank 72nd in D1 starters per DPI
141- Nein- 121 overall in WrestleStat and #71 in DPI among starters
141- Nathan Smith- 178 in WrestleStat, would be #60 among D1 starters
149- Friedman- 120 in WrestleStat #48 starter in DPI
149- Gliva- 81 in WrestleStat, would be #48 starter in DPI, just ahead of Friedman
157- Ream, 22 in WrestleStat, 12 in DPI among starters (Barnes is no longer a 157, so I bump Clay up a spot), 15 overall in DPI
165- Fogarty- 36 in WrestleStat, 23 in DPI among starters, 30 overall in DPI
165- Clingenpeel- 155 in WrestleStat, would be #66 in D1 starters in DPI
174- Nielson- 156 in WrestleStat, 47 in DPI among starters
174- Urbach- 130 in WrestleStat, 47 in DPI among starters just behind Nielson's rating
174- Popp- Would be #57 among D1 starters in DPI
184- McNutt- 77 in WrestleStat, 49 in DPI among starters
184- Paxton, 145 in WrestleStat, has a crazy high DPI and would be #11 among D1 starters if he had enough matches to qualify.
197- Eaton- 104 in WrestleStat, 48 in DPI among starters
HWT- Tynan- 58 in WrestleStat, 19 in DPI among starters, 20 overall in DPI
HWT- Stibral- 102 in WrestleStat, 55 among D1 starters in DPI

Dual meet team- NDSU- #42 IN WrestleStat, #25 IN DPI

Tournament team- NDSU- #26 in WrestleStat, #27 in DPI

Overall roster component, NDSU- #29 in DPI

Conclusions: 1. WrestleStat appears to take far longer to correct to the actual ranking compared to DPI. Sykora and Tynan are great examples of that, and they are a big reason why NDSU's ranking is so low in WrestleStat. I still love using WrestleStat and all of the awesome features it has, but it is worth remembering when you look at their rankings. 2. NDSU has a lot of "average" D1 starters in Friedman/Gliva, Nielson/Urbach, McNutt, and Eaton all ranking between 47-49 in their respective weight classes among D1 starters in DPI. The good news with that is Eaton and Nielson are only rFr and Urbach and McNutt are rSo, so they should theoretically have plenty of time to climb the ranks. 3. NDSU has one "hole" and that is at 141, where they were forced to start a true Freshman from Bismarck because of an injury to Bengston (who likely would be in the 20's).

As you probably guessed...I'm bored today.

http://www.wrestlestat.com/team/49/n...-state/profile

https://setonhallpirate.wordpress.co...-impact-index/

I hadn't seen the setonhallpirate version before, so thanks for that. As a nerd that has messed around with my own formula based system, the DPI system is a lot more like what I would expect a computerized system to be. The number 1 team in my system is often not the team that I want to have there. Somehow wrestlestat is 10-for-10 in having consensus #1 picks, while DPI has Joey McKenna ahead of Dean Heil, Bo Nickal ahead of Gabe Dean, and Connor Medbery ahead of Kyle Snyder. We should test which system predicts wins and losses better, top-to-bottom, at the NCAA's, although we might need to be really, really bored for that.

Quote:

---

Originally Posted by RadioBison

I hadn't seen the setonhallpirate version before, so thanks for that. As a nerd that has messed around with my own formula based system, the DPI system is a lot more like what I would expect a computerized system to be. The number 1 team in my system is often not the team that I want to have there. Somehow wrestlestat is 10-for-10 in having consensus #1 picks, while DPI has Joey McKenna ahead of Dean Heil, Bo Nickal ahead of Gabe Dean, and Connor Medbery ahead of Kyle Snyder. We should test which system predicts wins and losses better, top-to-bottom, at the NCAA's, although we might need to be really, really bored for that.

---

Is DPI taking in consideration what Kyle Snyder is also spending half his season beating the World's best in freestyle along with wrestling collegiately? Pretty amazing what he's doing.


Sent from my iPhone using Tapatalk

Quote:

---

Originally Posted by Christopher Moen

Is DPI taking in consideration what Kyle Snyder is also spending half his season beating the World's best in freestyle along with wrestling collegiately? Pretty amazing what he's doing.


Sent from my iPhone using Tapatalk

---

That's just it. The DPI probably doesn't consider his NCAA championship, much less his World and Olympic titles. But after crunching all of the margin of victory numbers, that's the kind of results that computer models spit out sometimes.

Quote:

---

Originally Posted by RadioBison

That's just it. The DPI probably doesn't consider his NCAA championship, much less his World and Olympic titles. But after crunching all of the margin of victory numbers, that's the kind of results that computer models spit out sometimes.

---

Another amazing thing about Kyle is that he is not a "big" heavyweight. He's built kind of in the same mode as Kurt Angle. He's a bit taller and has better physique than what Kurt had in college (Kurt had an amazing physique change between 1994-1995 that "raised" a few eyebrows).

For what it's worth, I completed my comparison of all of the ranking systems over the NCAA tournament. I looked at the seeds, InterMat, Flo, TheOpenMat, WIN, Track, WrestleStat and SetonHallPirate to see which one predicted the most matches correctly. The seeds and WIN were virtually tied at the top at 77.2%. The two formula based systems were at the bottom with WrestleStat at 74.5% and SetonHallPirate at 75%. However, with only 539 to 640 matches to work with, depending on the system, I can't even say with any real confidence that the 77.2% for WIN is any better than the 74.5% for WrestleStat. Maybe I'll get there if I run it for a few more years. Stay tuned...