This week, the winning team outscored the losing team by more than twenty points on average. It was also a bad week for picks, although the upsets tended to be the closer victories, with the exception of beatpaths picking Chicago. Beatpaths was 7-6 last week, while IT and the official ITB picks were 8-5. According to the beatpaths, the best guesses for fluke games so far this season are HOU->SF (although not by much), SF->ARI, BUF->NYJ, NYJ->NE, CHI->PIT, PIT->MIN, HOU->CIN, and SEA->JAC. All of these game outcomes will contradict this week’s rankings. It’s possible that throwing out fewer games can still lead to a valid ranking, but the current system is saying that these games the ones that are most reasonable to throw out. The BeatPicks were 4-1 this week, or 15-3 for the season.
Cincinnati helps their case, Arizona helps theirs even more. Atlanta gets hurt, San Francisco gets hurt even more. We’ll list all the beatloops after the jump. Here’s the Week 7 Beatpath Graph:
Loop: ARI=>HOU=>SF=>ARI
Loop: BUF=>NYJ=>NE=>BUF
Loop: GB=>CHI=>PIT=>MIN=>GB
Loop: JAC=>HOU=>CIN=>CHI=>SEA=>JAC
Couple of divisions have a clear “hierarchy”:
NFC East: Giants, Cowboys, Eagles (losing to the Raiders will do that you), Washington
NFC South: Saints, Falcons, Panthers, Bucs
NFC West: Arizona, 49ers, Seahawks, Rams
AFC West: Denver, SD, Raiders, Chiefs
The only one that’s “out-of-order” (a different order in the actual standings) is the NFC East, where the Eagles (due to beating WAS last night) is ahead of Dallas (due to losing to the Giants).
The NFC North is interesting. The other three teams have a direct path to the Lions, but none to each other right now.
@Paul
If the Eagles beat the Giants this coming Sunday, they will be liberated from their loss to Oakland. Nevertheless, they will still probably be ranked lower than Dallas, due to having only remaining beatwins over mediocre teams (Carolina, Kansas, Washington, and Tampa). (Interestingly, ESPN records the line right now as favoring the Eagles by 3, which is surprising to me.)
Continuing the discussion from last week’s thread,
http://beatpaths.com/2009/10/20/2009-nfl-week-6-beatpath-rankings/#comments
JT talked about using beatpower to break loops. This would be very similar to a method that I outlined last year here.
http://beatpaths.com/2008/12/23/beatloop-removal-approaches/#comments
I opined that if we looked at every beatloop (hundreds to thousands) and used that data to break loops, we could break the least amount of beatwins and thus retain the most data. Think of it as a iterative on steroids.
Well, the more I think about it, I’m siding more with TT on beatloop removal and sticking with the standard method. What my proposal and the iterative does is look at the teams that have the most involvement in loops. This is always going to be a bad team that beat a good team. When that happens, you are guaranteed to get a bunch of loops.
So the iterative calls the game “flukey” and breaks only that game out of the beatloop. So in the final results, the beatgraph is basically saying, “That game never happened”
But it did happen. It feels like the iterative method is saying, “On paper Team A should never beat Team B…oh, they did? Well, just break that game, but leave the rest the same.” What is the point of teams playing the games if in the case the result doesn’t match what is expected, it is just thrown away. For example:
A->B->C and A->D->C and A->E->C
Would have a beatgraph of A on top, B D and E in the middle and C on the bottom. C then beats A. Iterative would then remove C->A and the graph looks identical. It’s like that game never happens. Doesn’t pass the smell test for me. The game did happen. Standard method results in a graph that has nothing. And well, looking ONLY at wins and losses, I feel it’s the right answer.
In the second thread I linked, I made an argument exactly opposite to the one I just made (comment #7). Sure, A’s record is now 3-1 and C is 1-3. Just looking at that, everyone probably agrees that A is better than C. But when looking at the teams that they beat and lost to, I now agree that it is unfortunately ambiguous who is the better team.
Boga
Boga, the reason I run multiple methods is because while your argument that the game did happen is true, it is also true that upsets happen. The more upsets that happen, the more of the graph that gets removed. The iterative method allows us to retain more data and does it in a logical way. As you mention in your example, A has a better record than C. A also has 3 separate direct paths to C where C only has their direct win to A.
In a league where there are a lot more games and teams play each other multiple times, “fluke” games are washed away because a season series is more likely to tell us who is a better team. Unfortunately in football, aside from division games a team only plays any other team at most once. As a result upsets have a much larger influence. The iterative and weighted methods are designed to reduce the strength of upsets in different ways, giving us a greater overall picture of a league with a limited number of games.
@boga
I agree. Removing only the “flukey” games as if they never happened is wrong. However, that does not mean they should have more impact than any other game.
In my humble opinion, the Standard method and that Iterative method both have shortcomings. They err at opposite ends of the spectrum. My Iterative method (which I believe is the same one Moose uses) tries to split the difference. It tries to give the same weight to every game.
For instance, say A has a beatpath to B (A->C->B). If B beats A, the beatpath disappears. Both methods agree on this.
Now imagine A has a second, longer, beatpath to B, such as A->D->E->B. Again, both methods agree. B->A still removes one beatpath; the longer remains.
But what if both beatpaths are the same length? This is where the two methods disagree. Standard suddenly removes BOTH beatpaths, despite there being just as much evidence that A is better than B as there was in the previous scenario. On the other hand, the iterative you refer to leaves both intact. You have adequately explained why that is wrong.
My iterative does exactly the same as in the second scenario; it removes one A to B beatpath and leaves one intact. Since there is no way to say which beatpath to remove, it takes half of each. Thus the ‘flukey’ game still has the exact same impact as it would under any other circumstances.
What do you think?