ESPN Computer Releases Its New Top 25 Rankings - Story by Andrew Holleran • 5h ago
- Thread starter RCBeta79
- Start date
Another day, another OSU loss (now 10 of their last 11)... and again, little downward movement. Kenpom drops from 35 to 39, Bart drops from 43 to 45, and NET drops from 41 to 46.
Models just love their blowouts of crap teams earlier in the year.
Models just love their blowouts of crap teams earlier in the year.
OSU still holding in at 29 in the ESPN BPI, lol.... apparently a more powerful team than Kansas St and Miami
So, now OSU has a 21 point home loss to MSU to factor in.
Bart dropped them from 46 to 54 with the loss, after projecting a 59% likelihood of winning that game. Kenpom dropped them to 50. NET already had them at 48, and that will almost certainly drop into the mid-upper 50s.
BPI may drop them all the way from 30 to (gasp) 31 for this!
Bart dropped them from 46 to 54 with the loss, after projecting a 59% likelihood of winning that game. Kenpom dropped them to 50. NET already had them at 48, and that will almost certainly drop into the mid-upper 50s.
BPI may drop them all the way from 30 to (gasp) 31 for this!
Nevada is #4 in the RPI. Rutgers is #59, right behind Hofstra, Vanderbilt, and St Louis. If you thought BPI was bad..
I don’t understand this obsession with RPI. It is absolute garbage.
It's because it's entirely rooted on W/L... and winning is ultimately is all that really matters. At the end of the day, most people don't really care how well a losing team played - the story going into the second weekend of the tournament isn't how efficient that losing 5-seed was, but that the 12-seed won two games to make it into the Sweet 16.
This is the real resistance behind predictive models - they ignore "winning games" in favor of "playing well". IMO, there's a place for both, because they are meant for different things.
Yes - RPI is a flawed system too but it was never intended for seeding. Nevada, at the end of the day, is a 6 loss team that’s in the hunt for at large. Hofstra and St Louis have the same record as us. As Chop said - selection is about winning and resume quality. These other teams would fall short on the latter metric - lack of quality wins.
Vandy played a really tough schedule (I have no idea what blended SOS says but I guarantee you their median opponent was much much tougher than ours). Non-conference - @ Temple, Memphis, St Mary’s (neutral - in Cali), Pitt, NC State, @ VCU, Southern Miss (they are 23-4), Fresno (neutral - in Cali). In conf - 2 games vs Alabama and Tennessee. 13-12 to be in the 60s for them doesn’t seem as ridiculous when you think about that schedule. I’m not sure there are 60+ teams that wouldn’t have 12 losses with their schedule.
Ohio State is 11-14 - and we’re talking about a ranking of 36 - not the 60s. It’s just crazy. A team that’s 3-11 in a not so dominant BIG. They didn’t even play any decent mid-majors to demonstrate that they would actually dominate them. All 6 cupcakes they played have losing records. They played 5 teams with a pulse in non-conference and went 2-3 in those games. Ok great but since then they’ve beaten 3 real teams in 14 attempts. They don’t belong in the top anything because they don’t win.
Right but RPI isn’t good at what it’s doing. I completely understand the draw of W/L only models, but just LOOK at the RPI results. Even Ohio State’s rating is more off in the RPI, like I get that people think they are overrated in these more predictive systems but none of you people really think they are as bad as #151 (especially for the whole season, maybe they are playing that bad today).
Like,
pros:
Its win/loss only
Cons:
Nevada (!) is #4
San Diego st is #5
North Texas is #21
Utah St is #22
Oral Roberts is #26
People bitch endlessly about the few teams that look like outliers in the NET. In the RPI like every team is an outlier.
Yes, let’s prop up a system that has saint louis, Vanderbilt, Hofstra, Sam Houston, drake, vcu, liberty, Louisiana, Kent state, southern miss, Boise st, oral roberts, Utah state, and north Texas all in front of Rutgers because by golly doesn’t Ohio states NET ranking just look a little too high?
Yeah, RPI's definitely got significant flaws - but as the only W/L model out there, it's attractive to people who value wins over efficiency. NET should really be an amalgam of the efficiency-based models and a W/L model, imo - but it seems to hew much more closely to the efficiency models.
Is Ohio State as bad as 151? I don't think so - but any model that doesn't have them in the bottom 3-4 in the B1G is not really reflective of reality. Bart and Sagarin have them 10th, Kenpom and NET have them 9th... and all are buoyed by their signature win over us, which came with an asterisk. It wouldn't be that surprising if they didn't win another game this year.
I fully understand the draw of W/L models but from looking at the results of my own I'm starting to believe that the issue is that there is simply not enough information in a season to properly separate the strength of conferences just from wins and losses.
I have the W/L only model that I am posting in my Trueskill thread, with these current rankings for the Big Ten:
1. #2 Purdue (23-3)
2. #19 Illinois (17-7)
3. #20 Indiana (18-7)
4. #25 Michigan St (16-9)
5. #30 Wisconsin (14-10)
6. #35 Iowa (16-9)
7. #37 Maryland (17-8)
8. #39 Northwestern (18-7)
9. #42 Rutgers (16-9)
10. #56 Michigan (14-11)
11. #61 Penn St (14-11)
12. #69 Ohio St (11-14)
13. #85 Nebraska (12-14)
14. #155 Minnesota (7-16)
These look pretty reasonable to me but there's a huge catch: it uses starting ratings that are based on where a team finished the previous year. This really helps get these ratings looking reasonable but it seems inappropriate for something like selection tournament teams. If I remove the starting rankings and start from the assumption that everyone is equal it devolves into this:
1. #3 Purdue (23-3)
2. #26 Indiana (18-7)
3. #39 Illinois (17-7)
4. #44 Northwestern (18-7)
5. #47 Michigan St (16-9)
6. #49 Maryland (17-8)
7. #64 Rutgers (16-9)
8. #72 Iowa (16-9)
9. #74 Wisconsin (14-10)
10. #99 Penn St (14-11)
11. #116 Michigan (14-11)
12. #135 Nebraska (12-14)
13. #175 Ohio St (11-14)
14. #279 Minnesota (7-16)
Whoops. Now they are basically absurd. And you start seeing some really obvious anomalies in the top 50 like with the RPI:
FL Atlantic all the way up at #4 (#49 with starting ratings.. which is arguably too low)
Col Charleston #13 (#59 with starting ratings)
Oral Roberts #16 (#70 w/ starting)
Southern Miss #19 (#110 w/ starting)
North Texas #20 (#46 w/ starting)
Kent #28 (#65 w/ starting)
Liberty #38 (#94 w/ starting)
Drake #41 (#66 w/ starting)
Utah Valley #43 (#82 w/ starting)
Louisiana #45 (#101 w/ starting)
Sam Houston St #46 (#89 w/ starting)
A lot of these teams are the same ones that look stupidly high in the RPI -- so maybe I am being too harsh on it. The issue is less it's methodology than it is the task is impossible.
This is about the long and short of it. There are 360+ teams and ~30 games for each, and 18-20 of those games are played only against a small 10-16 team subset of teams. And the quality of those 360+ teams is not close to consistent. Finding a universal methodology that truly works is impossible given the tiny sample sizes involved. If the 360 teams each played each other 10 times, then you could develop a pretty strong model that accurately ranked them against each other - but we don't have that.
Still, there has to be some way to value wins while also determining the quality of those wins, which is better than the tools we have today. It's just a matter of iterating and improving year to year.
I still think things like WAB and SOR are the best in this regard, but nothing is perfect
I agree SOR is probably better at evaluating resumes. You should note that WAB thinks pretty highly of the MWC teams too. SOR seems to have their number a bit more for whatever reason.
To be fair, I think your probably discounting the value of good mid major teams a bit. Kent State and Oral Roberts will probably be competitive 13 seeds. RPI knows the difference between them and UMass Lowell whose down near OSU,
So, after an absolute drubbing by Iowa (who were up by 28 with 5:30 left), Ohio State apparently remains 36 in BPI? They've now lost 7 in a row and 12 of their last 13.
36 - ESPN BPI
53 - Sagarin
58 - Kenpom
58 - NET
67 - Bart
106 - ESPN SOR
128 - Bart WAB
156 - RPI
Interestingly, WAB/SOR are closer to RPI for Ohio State than they are to the efficiency-driven models. No idea what BPI is looking at, though.
36 - ESPN BPI
53 - Sagarin
58 - Kenpom
58 - NET
67 - Bart
106 - ESPN SOR
128 - Bart WAB
156 - RPI
Interestingly, WAB/SOR are closer to RPI for Ohio State than they are to the efficiency-driven models. No idea what BPI is looking at, though.
