How many games does it take before a power rating with no pre-season bias is trusted? Right now most teams have played three games and if you look at our All-Division Power Rating, it makes sense in a lot of places. But not in others. And there are a lot of big discrepancies with our Bayesian Updated Pre-season rankings for week 3.
By looking at the first (unbiased) Strength power rating, we might get an idea of where some of these teams really rank. Or, we might see where a small sample is misleading us.
Before we look at the discrepancies, let's see where they agree. For the top 25 teams, there are exactly two agreements: #1 Georgia and #15 Clemson
- #1 Georgia: The Bulldogs ended last season #1, started the pre-season #1, our updates have kept them #1, and our first Strength power rating has them #1. This is good evidence that Georgia should be considered the #1 team in the country right now, and they are almost unanimously considered to be just that. They've won games 49-3, 33-0, and 48-7, which is why they've held their #1 spot in our Bayesian update, and the 49-3 win over Oregon—who has won their other two games 70-14 and 41-20—gives a good idea of why Strength has them #1 over a 3-game sample.
Verdict: Georgia is the best team in the country until proven otherwise. - #15 Clemson: The Tigers interestingly ended last year #15. We upgraded them to #9 in our pre-season projection, and since they've moved up to #8, down to #16, and up to #15 as we updated those ratings based on weekly results. This happens to match where the Tigers sit in our early Strength power rating. The Tigers are 3-0 but it was week 2 vs. Furman that knocked them down in our Bayesian ratings; they won just 35-12 which doesn't cut it against such a low-rated opponent. Is that what holds them back in Strength early on? No, actually that's their highest rated game in Strength, which considers their 48-20 win over Louisiana Tech to be their weak link. But however you slice it, both ratings agree.
Verdict: The Tigers are probably overrated at #5 in the AP poll.
Now we'll scan the top teams in both ratings, looking for teams that are rated a lot higher in one than the other. We assume that the Bayesian updates currently provide a more realistic picture, but that where they err, the Strength power rating will show the direction of the error.
11 teams that rank much lower in Strength:
- Texas A&M (60 spots lower): Bayesian #25; Strength #85. The Aggies started the season at #18, and mostly due to the loss to Appalachian State, have dropped to #25. The win over Miami helped, but not much as Miami was ranked #28 prior to that win. But this downgrade is nothing compared to their early Strength ranking of #85! Strength doesn't respect the win over Sam Houston State at all, ranking the Bearkats #248 out of #769 teams, when they are normally a strong FCS team. Strength also has Miami at #66.
Verdict: We think #25 is much more realistic that #85; Strength needs time to sort out the Aggies' SOS. But it does raise doubts that A&M is a top 25 team. - BYU (42 spots lower): Bayesian: #18; Strength: #60. BYU started pretty high (#13) and was #9 after the Baylor win, but losing badly to Oregon dropped them to #18. Strength only has them at #60 though, being unimpressed at all with the overtime win at Baylor (ranked #61). Which one is right?
Verdict: The AP has BYU at #19 and Baylor at #17, so again it's probable Strength not getting with the SOS program after just three games. - Minnesota (38 spots lower): Bayesian #13; Strength; #51. This one is a bit of a mystery, since the Gophers have moved up a lot each of the last two weeks, from #31 to #24 to #13, in the Bayesian update, winning 38-0, 62-10, and 49-7. Normally unbiased power ratings love blowouts, but Strength ranks those opponents #216, #301, and #241! Resulting in a pedestrian #51 ranking. The AP isn't impressed so far with wins over New Mexico State, Western Illinois, and Colorado, leaving the Gophers in "others receiving votes"
Verdict: Overrated by Bayesian, underrated by Strength, the two shall eventually meet. Probably in the top 25. - Michigan State (25 spots lower): Bayesian: #20; Strength: #45. Michigan State started the season at #12 and has drifted lower ever since, with only the 52-0 win over Akron exceeding expectations. The Strength power rating sees all their performances as roughly consistent. This makes us think they have further to drift down from #20.
Verdict: We started the Spartans too high at #12 and still more correction is needed. - Utah (24 spots lower): Bayesian: #9; Strength: #33. The first of two Pac-12 teams in a row, both of whom lost their first game then came back strong, Utah fell from a starting #8 to #11, jumped back to #6, then fell to #9. It's telling that their one "up" game was a 73-7 blowout, with two games being disappointed in their result. Strength might not have a good handle in their opponents yet but #33 seems more reasonable than #9, or even the #13 that they've climbed back to in the AP.
Verdict: Shockingly, Strength is the voice of reason re: a 67 point blowout. Utes might be headed for a fall. - Oregon (23 spots lower): Bayesian: #8; Strength: #31. Oregon famously got beat down 49-3 by Georgia, but bounced back royally, beating Eastern Washington 70-14 and BYU 41-20. They bounced from #26 to #36 to #11 to #8. Clearly the jump from #36 to #11 was a bit much, but Strength keeps them in the 30s. That first game is a drag when it counts for 1/3 of your schedule so far. Their subsequent games will probably look more like their last two performances than the opener.
Verdict: Maybe not top ten, but with more games diluting the opener, their Strength will rise. - NC State (21 spots lower): Bayesian: #11; Strength: #32. These come down to Strength not having SOS worked out usually, but here it's hard to tell. Are East Carolina and Texas Tech better than we think? The Wolfpack haven't moved much from their #10 start, but in Strength their game ratings are very consistent, suggesting #32 isn't far off, either.
Verdict: Both ratings show consistency, so we'll stick with our pre-season judgement that this is a near-top ten team, until proven otherwise. - Notre Dame (19 spots lower): Bayesian: #16; Strength: #35. The Irish are just 1-2 with one unimpressive win, so it's surprising they haven't fallen farther from their #5 start. The AP pollsters disposed of them at 0-2, of course, so they aren't much help. They definitely need to fall farther than #16 but #35 is another case where a really bad game—probably unlikely to repeat—weighs them down. In some ways their Strength rating can be seen as a relief, as it's not unreasonable that they'd rate even lower at this point.
Verdict: Somewhere in the middle. Borderline top 25. - Oklahoma (17 spots lower): Bayesian: #5; Strength: #22. The Sooners often tend to be ranked high and then hit the skids (see last year) and their current #5 ranking is almost all due to being pre-season #6 and not having a bad game yet. Strength assumes nothing, so the Sooners have to earn their ranking, and #22 is all they get. But Strength also ranks Nebraska #153 right now, which is pretty silly. Meanwhile Oklahoma hasn't done anything to prove they're #5, either.
Verdict: This actually gives a pretty decent range for where Oklahoma might end up. - Cincinnati (16 spots lower): Bayesian: #12; Strength: #28. The opening loss to Arkansas probably made it clear Cincy wasn't a top ten team, so their rebound from #16 to #8 by beating Kennesaw State 63-10 was an overreach by our Bayesian update. This is one where we think Strength is getting it right.
Verdict: Overrated and needing further correction in our Bayesian updates. - Michigan (12 spots lower): Bayesian: #4; Strength: #16. Here's an interesting one: Michigan started #4 and has remained their through 3 updates. Strength is not impressed by the Wolverines' schedule, despite 51-7, 56-10, and 59-0 blowouts over the #169, #246, and #256 teams. The latter two are FBS teams unlikely to remain ranked that low, so we have an SOS issue here, and probably way underrates Michigan's defense at #55.
Verdict: Michigan is the #4 team until proven otherwise; certainly nothing they've done has been less than a top 5 performance.
Now the converse: Team rated a lot higher in Strength than their current Bayesian Update ranking:
- James Madison (59 spots higher): Bayesian: #65; Strength: #6. Ok, the Dukes are 2-0, and a pure, unbiased power rating like Strength isn't going to handle two games very well, especially two blowouts. On the other hand, James Madison started the season at #84 and after just two updates is at #65, suggesting they could go a lot further. Adding another game will help converge these values; they play Appalachian State next and Strength says they win by 24, Bayesian says they lose by 6.
Verdict: They're quite a bit better than #65, but nowhere near #6. - Kansas (47 spots higher): Bayesian: #58; Strength: #11. Just how good might Kansas be? Already up from a pre-season #80 to #58 after three wins, Strength says watch out, they're a borderline top ten team! Ok, let's take a step back and wait for more games, and for Strength to work out how good or how weak their schedule has been so far.
Verdict: That said, it seems apparent that #58 is far too low. It will be very interesting to see if KU continues their climb. - USC (32 spots higher): Bayesian: #34; Strength: #2. Now here's a team to watch. All pre-season the Trojans were the big question: how good would they be? We didn't buy in, putting them at #60 to start. And all they've done is outperform each week, jumping to #51 to #44 to #34. When that kind of consistent update jump happens, you can tell they're nowhere near the target, which Strength says is #2 in the nation! Strength is strangely enamored with their first win over Rice, but their other wins are solid, too. They're clearly a top 25 team and underrated by our Bayesian power rating still. We doubt they're #2 but they might be pretty close. AP has them #7 and that doesn't seem unreasonable, but we still demand more evidence.
Verdict: Look for their Bayesian update ranking to move toward the Strength rating faster than the other way around. - LSU (25 spots higher): Bayesian: #33; Strength: #7. This one is a bit of a surprise, as we didn't see LSU doing anything remarkable, and while they've moved up from pre-season #38, it's all from their recent win over Mississippi State. But Strength has FSU as the #20 team which makes that 1-point loss excusable and has the Bulldogs at #10, making that win a huge one.
Verdict: We think Strength has some issues to sort out here. #33 seems more than reasonable with FSU at #29, and Mississippi State at #23, rather than where Strength puts them early on. - Washington (25 spots higher): Bayesian: #37; Strength: #12. Washington sort of came out of nowhere with the Michigan State win as far as we're concerned; we started the Huskies at #59 and the last two games—52-6 over Portland State and 39-28 over Michigan State—propelled them to #37. Strength sees all three of their games as solid performances and rewards them with a #12 ranking. That's even as Strength has the Spartans only #45.
Verdict: Strength seems to be onto something here, and clearly the Huskies were underrated and haven't caught up yet. - Iowa State (24 spots higher): Bayesian: #38; Strength: #14. Iowa State was supposed to be good last year and finished 7-6. This year they lost nearly all their talent, and we downgraded them to a pre-season #46. They've upgraded themselves to a respectable #38 but according to Strength they're #14. Is that justified at all? Two 30 point blowouts over bad teams and a narrow road win over Iowa (#47 Strength) doesn't seem like it adds up to #14. Against Iowa they could hardly score (understandable) but they did hold Iowa to 7 points (also understandable). This proves nothing.
Verdict: We need to see a lot more before considering the Cyclones a top 25 team. - Maryland (21 spots higher): Bayesian: #42; Strength: #21. Is Maryland a top 25 team? That is the question. After starting at #49 they fell to #54 before two solid games put them at #42. Unbiased Strength suggests they're underrated. Maybe it comes down to where you rank SMU, their latest victim (34-27). Our Bayesian rating puts SMU at #43; Strength has the Mustangs at #23. In both cases Maryland is exactly one spot higher.
Verdict: Who knows. We can't sort out SMU, and Maryland is always a mystery early on. Based on historical performance though they'll hit the skids at any point now, so our Bayesian ranking looks more reasonable. - Central Florida (20 spots higher): Bayesian: #45; Strength: #25. UCF slipped from a starting point at #53 to #59, then made it all back and more with a 40-14 win over Florida Atlantic. That seems an unlikely win to hang your hat on.
Verdict: Nothing suggests UCF is a top 25 team right now, especially the home loss to Louisville. More games will probably make this clear to Strength. - SMU (20 spots higher): Bayesian: #43; Strength: #23. We could just punt and say "see Maryland" as SMU ranks one spot below the Terps in both rankings after losing to them, 34-27. But the odd thing about SMU's Bayesian ranking is, that's not the game that sunk it. The Mustangs started at #29 and a 48-10 win at North Texas boosted them all the way to #22. But their 45-16 win over Lamar was about 35 points shy of what we needed to see to maintain that level, and they fell to #45. The Maryland loss actually moved them up 2 notches.
Verdict: Our Bayesian ranking seems unreliable for jumping up and down so much, but Strength is too early to trust. We punt like we did for Maryland. - Mississippi State (13 spots higher): Bayesian: #23; Strength: #10. The Bulldogs started at #16 and pretty much treaded water until their upset loss to LSU. Somehow, though, Strength didn't get the memo and still has them at #10. How so? Well for one thing, Strength has LSU at #7. And Mississippi State's other foes (Memphis, Arizona) are rated well enough (#64, #70) that the 20+ point wins over them are scored pretty well. But one thing we see is their opponents scoring 23, 17, and 31 points. Not a single game in single digits even after just three games is pretty suspect.
Verdict: Our Bayesian update system says they've declined from #16, and we think that's more likely than an upgrade. - Syracuse (11 spots higher): Bayesian: #30; Strength: #19. The Orange are a rare team that has moved up on every Bayesian update, from #47 to #35 to #32 to #30. That slowing down makes it seem like they're getting pretty close to the actual target. The 31-7 win over Louisville did the heavy lifting and now a win over Purdue only gives them a slight boost.
Verdict: They're pretty close to fair ranking now at #30.
Comments