2008 NFL Week 9 Beatpaths Rankings

Rank Team Notes Last Week BeatPower

1

(Bye) Carolina rises to #1 on the bye, mostly because Washington’s beatwins don’t look as valuable anymore.

2

100.0

(23/23 – 0/23)

2

(Beat GB) Tennessee adds another beatwin to their collection and climbs to #2 with a tough overtime victory.

3

100.0

(20/20 – 0/20)

3

(Beat BUF) The biggest leap of the week belongs to the NY Jets, mostly due to their beatwin over Arizona. But beating Buffalo in Buffalo isn’t a small feat, either.

18

100.0

(23/23 – 0/23)

4

(Beat STL) Arizona holds steady after the expected win over St. Louis.

4

90.0

(22/25 – 2/25)

5

(Beat WAS) Pittsburgh gets a beatwin over Washington, which should help them stay in the top tier.

7

91.7

(15/18 – 0/18)

6

(Lost to PIT) Washington sinks to just below Pittsburgh on the loss, but they were falling anyway due to the weaker beatpaths beneath them.

1

90.7

(22/27 – 0/27)

7

(Lost to NYG) Cleveland protects Dallas from their beatloss to the NY Giants, but their beatwins are looking weaker too, so they sink a little more.

5

80.4

(21/28 – 4/28)

8

(Beat KC) Tampa Bay’s loss to Denver keeps them from a beatwin over Kansas City, but they hardly need it.

10

79.2

(19/24 – 5/24)

9

(Beat OAK) Atlanta breaks into the top 10 – I can’t remember the last time a pro team has been as dominated as Oakland was.

11

72.0

(17/25 – 6/25)

10

(Lost to TEN) No big penalty for losing to Tennessee, and their beatpaths are looking a bit stronger. Green Bay rises on the win. They didn’t look bad against Tennessee, either.

13

66.7

(16/24 – 8/24)

11

(Beat CLE) Baltimore’s beatwin over Miami is actually more valuable than their victory over Cleveland.

14

84.4

(13/16 – 2/16)

12

(Beat DET) Chicago holds steady on the win over Detroit.

12

70.8

(16/24 – 6/24)

13

(Beat NE) Indianapolis with a major jump in the rankings after defeating New England.

20

59.3

(15/27 – 10/27)

14

(Beat HOU) Minnesota is also helped by the beatwin over Miami.

22

52.2

(12/23 – 11/23)

15

(Lost to MIN) Houston also manages to hold on to their beatwin over Miami, which helps them rise too.

24

46.0

(11/25 – 13/25)

16

(Beat DEN) Miami’s so good that they already have an alternate beatpath to Denver through Buffalo.

21

40.4

(10/26 – 15/26)

17

(Beat DAL) I agree that it looks weird that the NY Giants aren’t higher than this. But we can’t discount the Cleveland loss entirely – for it to be this demanding, it also means just as intensely that the Giants really should have beaten them.

15

77.8

(5/9 – 0/9)

18

(Lost to BAL) Cleveland with an ambiguous season so far, holding steady on the loss to Baltimore.

19

38.9

(1/9 – 3/9)

19

(Beat SEA) Philadelphia has really been bouncing around, at the whim of other graph dynamics in a confusing season. I think this week in particular is good reason to consider an approach that will break more beatpath segments, like the iterative method. Maybe doktarr can explain the method again?

6

72.2

(4/9 – 0/9)

20

(Lost to NYJ) Buffalo seems to be following a similar path as Denver – a fast start, but then a major slump mostly due to turnovers. Buffalo loses at home and slips a few spots.

16

36.0

(9/25 – 16/25)

21

(Lost to IND) New England falls just short in the game of the century… ok not really. New England slips a couple of spots on the loss.

17

30.6

(4/18 – 11/18)

22

(Beat JAC) Cincinnati still doesn’t have any beatwins but all of their beatlosses are already ranked, so there is room for them to rise.

31

8.3

(0/12 – 10/12)

23

(Lost to CIN) Since MIA->BUF->JAC exists in the beatpath graph, Jacksonville takes a major tumble this week.

8

28.0

(6/25 – 17/25)

24

(Lost to TB) Kan…zus…. City is the highest ranking team in the AFC West this week.

25

7.1

(0/14 – 12/14)

25

(Lost to MIA) Denver loses at home basically due to Cutler. The defense actually played decently and has been making improvements since early this season, despite the injuries, but Denver’s had an awful turnover streak lately.

9

24.0

(5/25 – 18/25)

26

(Bye) New Orleans slips on the bye due to graph dynamics.

23

17.4

(4/23 – 19/23)

27

(Lost to PHI) Seattle stays mired near the bottom of the graph. Holmgren is all lame duck and just doesn’t seem to know what to do with them anymore. They’re decimated by injuries but it just seems like they’re trying to get through the season to forget about it afterwards.

26

7.1

(1/21 – 19/21)

28

(Bye) San Diego roughly holds steady on the bye.

27

6.8

(1/22 – 20/22)

29

(Lost to ARI) Kurt Warner comes back to St. Louis and dominates them. That really was an amazing couple of years, back in the last eon.

28

0.0

(0/21 – 21/21)

30

(Lost to ATL) Three first downs? Ten yards passing?? And, they’re ranked fourth in the AFC West, but not by much.

30

2.3

(0/22 – 21/22)

31

(Bye) San Francisco roughly holds steady on the bye.

29

6.0

(1/25 – 23/25)

32

(Lost to CHI) Detroit still seems to be playing better – #32 doesn’t seem like their rightful place. Not when Oakland is stinking like they did this week.

32

0.0

(0/24 – 24/24)

14 Responses to 2008 NFL Week 9 Beatpaths Rankings

  1. The MOOSE says:

    My graphs are up: http://www.twomuffin.com/BeatPaths.htm

    The three graphs are wholly in disagreement this week. This gives us a chance to see what assumptions the graphs make and which one seems to make the most sense. Personally, I don’t like seeing CAR and NYJ at the top. Iterative still sells TEN short, marking them as #9, but otherwise looks reasonable. Weighted gives TEN their due, but puts TB beside them which doesn’t feel right, while NYG falls to #8.

  2. Tom says:

    Volatility on the graph shot up again this week because of teams like NY Jets, Eagles, Jaguars, and Broncos esp. Colts, Texans, and Vikings didn’t help much either.

    Volatility:
    Week 9 = 138
    Week 8 = 108
    Week 7 = 90
    Week 6 = 128
    Week 5 = 124
    Week 4 = 152
    Week 3 = 202

    You mentioned the iterative method–does this reduce ranking volatility? How?

    It seems like graphs have certain choke points upon which major systemic swings are possible, as opposed to minor incremental steps. It would be interesting if there were some way to determine these “determinant games.”

  3. The MOOSE says:

    Tom: If you’re interested, run the numbers for Volatility on the Iterative method. I think we’re assuming that it will be smaller, but we haven’t actually tested for it. In order to know for sure, we’ll actually have to run the numbers for previous seasons as well since any one year can be anomolous.

    This may be an interesting statistic to include on my pages.

  4. Tom says:

    Also, I thought it would be interesting to revisit the beatpower “confidence” ranking of picks for Week 9, based on my comment here. If such a confidence ranking makes sense, the incorrect picks should be clustered toward the bottom of the ranking.

    Arizona-St. Louis: 87.0 (87.0-0) CORRECT
    Philadelphia-Seattle: 86.3 (95.7-9.4) CORRECT
    Jacksonville-Cincinnati: 79.2 (79.2-0) WRONG
    Tampa Bay-Kansas City: 58.9 (62.5-3.6) CORRECT
    Denver-Miami: 55.5 (78.6-23.1) WRONG
    Atlanta-Oakland: 51.4 (56.2-4.8) CORRECT
    Tennessee-Green Bay: 50.0 (100.0-50.0) CORRECT
    Chicago-Detroit: 47.5 (47.5-0) CORRECT
    New England-Indianapolis: 35.9 (75.0-39.1) WRONG
    Baltimore-Cleveland: 15.0 (50.0-35.0) CORRECT
    Buffalo-NY Jets: 9.7 (72.2 – 62.5) WRONG
    Minnesota-Houston: 9.6 (26.3-16.7) CORRECT
    Washington-Pittsburgh: 4.2 (95.5-91.3) WRONG
    Dallas-NY Giants: 2.1 (83.3-81.2) WRONG

    Top half of the graph has 2 wrong predictions out of 7. Bottom half of the graph has 4 wrong predictions out of 7. While not perfect, I would say that this might serve as a partial confirmation, based on one case only, that ranking according to beatpower “confidence” is a legitimate approach.

    Also, this may help identify some of those games that would be major systemic shifts if the pick went wrong. Both Jacksonville and Denver took major tumbles when they lost, despite a high confidence prediction.

  5. Tom says:

    MOOSE:

    The numbers for Week 9 on your iterative method are:
    0+5+5+2+2+2+2+3+3+2+7+3+7+0+14+3+14+6+2+2+5+9+14+3+15+1+5+1+3+2+3+2 = 147

    So, it’s actually more volatile this week than the standard method on BeatPaths. The top third of your graph is pretty stable, but big movers like the NY Jets, Bengals, Jaguars, and Broncos really push up the volatility numbers. Colts, Bills, and Packers also were significant.

    Now, I haven’t gone back through your rankings for every week (in part because it’s a pain to do–your page doesn’t list the change in ranking from one week to the next, so I have to flip back and forth to do the rank comparison, PITA), so it may in fact be that the iterative method is *overall* more stable than the standard graph.

    Maybe this is just a week of team-defining games and significant upsets, and there may be no way of getting around that.

    Also, a more stable graph is not necessarily a good thing–a graph may be resistant to change when it shouldn’t be. Greater ranking stability at the cost of less accurate representation of relative team strength is a bad thing.

    It’s not clear to me when the rankings should be getting more stable, or if they ever will in practice. In the beginning they should be pretty volatile. But because teams play division rivals twice, and opponents outside their division only once, the opportunity to define ones self against the field more broadly is limited. Finally, the late season games have a strange bias as strong teams that have clinches playoff positions suddenly lose to middling or bad teams because they’re benching their starters. So unless we endogenize factors like that, I’m not 100% sure that rank stability is an achievable goal, no matter what the method–standard, iterative, or weighted.

  6. The MOOSE says:

    One thing I wonder is if we should use a standard deviation-like method to compute volatility. Calculating such a number would tell us about how much any given team has moved for that week individually, which is an easier number to wrap your head around than a sum. Conversely, we could simply compute an average.

    I can look into having a “last week” column in my rankings. The trouble right now is that would force a redesign in my code as all of my web pages are created automatically after processing the data, and it only processes the selected week without tracking the previous week. I could find a way around this most likely. Right now I am working on a behind-the-scenes redesign of the site which will make it easier to update so I can provide more content.

    In terms of stability, I think it all comes down to the simple fact that there are only 16 games in a season which isn’t enough of a sample size to get reliable data. We have to acknowledge that this is a shortcoming of the approach and that looking instead at at basketball, hocket, or baseball season would provide more reasonable results one quarter way through a season than a full season of football will. In that respect, I hope to expand to include these other sports, once I can find a reliable input and time frame for reporting graphs.

  7. doktarr says:

    Iterative still sells TEN short, marking them as #9, but otherwise looks reasonable.

    I think we should make a distinction between the beatpath-based rankings and the rankings where your algorithm has to infer relative ranking from beat wins. In the case of Tennessee, they obviously have no beatlosses, but can be put anywhere from 1st to 12th, due to there being 11 teams they don’t have a beatpath to. It’s the same 11 teams in the standard method.

    I was wrong about Denver not dropping in the iterative approach. The combination of Buffalo, Jacksonville, and Denver losing to lower-ranked teams in the same week sends that whole segment crashing downward.

  8. The MOOSE says:

    True enough, but regardless, the Iterative method claims that none of TEN’s opponents have been strong. For what it’s worth, the Standard method agrees for the most part, ranking TEN #7. In all three versions, TEN’s best win is considered to be over GB, and their ranking is directly influenced by it. In fact, the whole thing boils down to the chain: TEN->GB->IND->MIN. CAR is considered a good team in all three methods and MIN defeated CAR, however they lose credit for the victory in the Standard and Iterative methods. The margin of victory of the win has kept it alive in the Weighted method so far, though the red arrow makes it appear it is hanging on by a thread.

    So as we’ve been saying up until now, let’s reserve judgement on TEN until they play someone of quality. We said the same thing of NE last year until they finally beat DAL and IND.

  9. JT says:

    Perhaps some blending of the standard and iterative method would produce better results. I’m not exactly sure what that would entail, but maybe if the idea gets out there someone will come up with something that works. It might work better if we had another method or two.

  10. The MOOSE says:

    Tom: Your beatpower confidence idea is much like how I did my superbowl preview article last year. The numbers basically said NE was one of, if not the most favored teams in history, for all the good it did them. When I have more time I’d like to run these numbers on my rankings again and see if there’s any consistency in what we see. Maybe there’s a threshold “confidence” number out there where we can be … say … greater than 75% sure a specific team will win. However, I suspect that any game that fits under this category will be following conventional wisdom.

    JT: I would be interested in some sort of aggregate ranking, but I’m extremely reluctant to include the weighted method in this because it includes score, so we would have to include more methods. Other than the BeatFluke method used on this site two years ago, I don’t think we have any other possibilities. I think it’d be helpful if we had a forum where we could bat some ideas around. More discussion would be helpful in expanding our ideas.

  11. ThunderThumbs says:

    Yeah, I’ve had the redesign half-completed for a little while now. I’ve been keeping wordpress but I’m considering switching to drupal. What would be nice about this is that I could give folks like you (MOOSE) and doktarr accounts to post on the page too. Unless you’re attached to the separate server I’d like to at least offer the capability to post more things collaboratively in one place. That has better forum capabilities, too. Although I’m not opposed to also just putting a phpdb thing up alongside the new design if that’s preferred. The only thing about phpdb is that when there’s not a lot of participation (like more than 30 posts a day) it can always kind of look like a graveyard to me.

  12. JT says:

    I agree that the score weighted method should be kept separate from the methods that do not include score. I’d forgotten about the beatfluke method though. The idea came to me as I was looking at some of the sites that aggregate polling data, and how they weight and combine poll information. The challenge would be having enough of a variety of methods to make it worthwhile. But how many ways are there to resolve loops that we haven’t covered already?

  13. ThunderThumbs says:

    There’s one “graph method” (see my nomenclature post) variance from a discussion that Kenneth and I were having about three years ago, I still have it flagged in my archives. So far for graph methods we have Standard, Beatfluke, and Iterative. And I think Score Weighted is technically a graph method too.

    The confusion is that I think they’re presented in a way that also uses MOOSE’s “power method”. I haven’t yet pulled those graph methods into the system on this end to see how they look with alternate power methods.

    End goal is to have a form here that everyone can play with dynamically. Pick a graph method, pick a power method, see the resultant graph/rankings. I’ve got the pieces in place to enable that but I kind of want to overhaul my codebase a bit too.

  14. Kenneth says:

    Overall, I think it would be useful to get the graphs under one package. Maybe it’s just me, but I find it hard to compare the different graphs to each other because they all use different display programs.

    Oh, and I want to go onboard as being excited for the redesign. Uh, excited!

Leave a Reply

Your email address will not be published. Required fields are marked *