Massive changes this week! Check out Philadelphia.
As we have more data, some of the longer beatloops got shrunk – and there are some unwitting beneficiaries like Denver – the JAC->DEN->TB segment is a pretty key part of the graph right now. As a Denver fan, it’s easy to be down on Denver right now, but their low-scoring victory over Tampa Bay apparently means… something.
No beatloop graph right now, but I might be able to put it together on Wednesday.
TT or Moose — is there a way for one of y’all to list someplace all of the beatloops before y’all start breaking them up. All I see right on y’alls sites now are the ones that are first broken, and not the ones that exist before some of the smaller ones were killed (if I’m reading them right). I’m guessing that you do list them all at some point in y’alls code, so the data should be there. If that is too much, maybe just every beatpath possible for this week only?
It amazes me how differently we can display the same data. Your graph make PHI look like the best team in the league while mine lists them as “third tier”. The more I see of this new stretched out graphing style, the more I like it though. Seeing MIA that high on the graph is unsettling to me.
Boga: There actually is no need for me to list all BeatLoops in my code. Doing so would increase processing time by a huge amount and doesn’t produce any “useful” data from the perspective of our methodology. It’s much quicker to first list all 2-team loops, then only list 3-team loops that survived the 2-team step. Then only list the 4-team loops that survive the 3-team step, and so on.
Just on a whim, I constructed a really large loop:
STL → WAS → PHI → PIT → JAC → DEN → TB → ATL → GB → IND → MIN → NO → SD → OAK → NYJ → ARI → SF → SEA → STL
If I cared to I could create a lot more. In tree structures like this, paths grow exponentially and there could be a hundred or more different loops. By the end of the season, we could probably make thousands. The problem is that large loops just tell us that amongst all these teams there is ambiguity. Our goal from there is to remove this ambiguity.
Think of a case where there is a season split, say between teams A and B. Add to this a team C that has defeated both of these teams, and team D that has lost to both, but D has defeated C. This gives you five loops.
A->B->A
C->A->D->C
C->B->D->C
C->A->B->D->C
C->B->A->D->C
What useful information do the two longest loops give you? If you can offer a convincing argument for some value that this would give us, I will consider looking at these loops.
To answer your second question, there are simply too many paths to list out. If you look at the “Out” it tells you how many distinct paths there are below them. There are 203 paths alone that start with WAS, and another 130 that start with ARI. Overall, there are 1130 unique paths this week. This is the great advantage of the tree structure graphs that we draw, as they can represent all of these different paths in an easy to look at format.
However, unlike the BeatLoops, I do have to view each path in order to count it (I do not store them though). If you wanted this information for some kind of research, let’s talk about what your goals are and we can see about getting the data to you.
Why does New York lose out in the NYG > WAS > CLE > NYG beatloop?
Miami sitting above the rest of the AFC East, while having the worst record of the bunch, makes me scratch my head. I wonder how that’s going to resolve itself.
Jeff: If you mean in the Standard method, how do you figure NYG loses out on that loop? All three links are lost. If you’re referring to the Iterative method, NYG actually “wins” the loop as the NYG -> PIT -> CLE -> NYG loop causes the CLE -> NYG link to get dropped first which saves NYG’s wins over both PIT and WAS.
Reggy: Yeah things get a little screwy in the name of keeping the graph simple. All it really means is that the only teams that still have remaining wins over MIA are high ranking, and MIA doesn’t have any wins of their own. The alternative is to place MIA much lower where we’d expect they might belong, but it would leave a long line, making it harder to see the direct relationships they have.
I’m guessing that Jeff is asking: why does New York end up being the lowest of the three teams in that beatloop?
The answer, Jeff, has two parts:
1) Losing the Washington and Pittsburgh beatwins caused New York to not be left with any high quality beatwins.
2) Where a team like New York (which has no beatlosses but only beatwins over weak teams) is displayed, heightwise, is fairly arbitrary in TT’s graph. This is the mirror image of the reason why Miami is very high up the graph – they have only losses, but only to good teams.
If you think that the Giants are underrated by this measure (and I would agree) then you should check out the iterative rankings, which put the Giants as the #1 overall team:
http://www.twomuffin.com/archive/iterative/2008/I_2008_8_NO.htm
Moose: Fair enough, I now understand this. I’ve got an idea, not sure if it has been thought outloud yet (searching through comments has been time consuming, so I stuck to posts with many comments, but I easily could have missed an important thread in the past), but I’ll gather my thoughts and make it coherent soon. I’ve been busy today.
Reggy: Realize that MIA doesn’t have any beatpaths, and that NYJ, BUF and ENG also don’t have any beatlosses. So technically, Mia can be anywhere between where they are now to even below DET. And the other three can be anyplace even above PHI. But the assumption that the graphiz package makes is also an assumption that I like. Be optimistic with teams losses and pessimistic with team wins. Thus we view MIA as since their worst loss is only to ARI and BAL, the best they can be is below them, and worst anywhere below that. One can prove they belong where they are, but not so much anywhere below. Now Moose has tweaked his graph to try to place these teams more where the rankings show, but strictly looking only at who beat who (and not delving into who beat all the other teams, etc — which the rankings do), you can’t prove that somebody like CHI is better than MIA, but you can say that since MIA only loss to BAL and ARI, they are just one tier less than them. Same goes for the rest of that division. NE is only as good as the teams it has beat, and the teams they have beat has also been beat by many many other teams (following the chain). I like both the standard map and Moose’s standard ranking weighted map, but push comes to shove, I’ll stick with the map on beatpaths to settle disputes.
Doesn’t when the victory happen factor in? As each week goes by should a victory or loss fade in value, and more recent games have more weight? Therefore the NYG victory over PIT would move them back up.
By the current method, what would have to happen for the Giants to regain the victories over WAS and PIT. If PIT beats WAS on Monday what will that do, or visa versa seeing as they’ve beaten both. Or what would have to happen for the loss to CLE to no longer bring them so far down.
Boga: I’m not sure if I follow your explanation of how I tweak the Standard graph. If you look at both you’ll see that every link is exactly the same, the only difference is which tier a team is placed on. As you say, this is determined by the rankings. However, I don’t understand what you mean by I strictly look at who beat whom but not delving into who beat the other teams.
In fact, the vertical placement of a team on my graph is directly dependant on their rank and score, which in turn is derived by looking at that team’s relationship between all teams. In this week’s case, MIA is set to be a “Tier 10″ team because they lost to a “Tier 9″ team, BAL. BAL however, was determined to be a Tier 9 team not because of a direct loss, but because they have a lower score and ranking than NYJ which is also a Tier 9 team.
I am also confused by your reference to my “standard ranking weighted map”. My Standard graph and Weighted graph are separate entities. The Standard graph is exactly the same as the one posted here (with the exception of vertical placement), while the Weighted graph is entirely different. To which were you referring?
Scrub: No, timing on the games doesn’t matter for a couple reasons. First is that the original goal of the site is to come up with these rankings using no data other than who won and who lost. The Weighted method is the only one which goes beyond this and uses the final score as well, but this is accepted only because the final score is the only data needed to determine who won the game. If we were to allow later wins to outwiegh earlier wins, division foes who split season series would automatically go to whoever was lucky enough to get the second game. This isn’t good because it’s likely that each team won their home game which was more likely the factor than the second winner improving in the time between the first game. The second issue is at the end of the season when playoff teams are resting their starters. We don’t want the last graph of the season to have a bad team on top because they got to knock off a top seed who wasn’t playing to win.
For NYG to regain their win over WAS, they will need to defeat them a second time when they play again. This will give their win double strength, and the ability to survive the loop breaking. To regain their win over PIT, they will need CLE to upset PIT when they play their rematch. This will cause a 2-team BeatLoop between them, and drop that link from the loop. It would let CLE->NYG->PIT reemerge on the graph, though either and possibly both links could get looped away in larger loops.
In the Standard way of removing loops, you can only regain a BeatWin if you can play that team again or it is possible to create a loop smaller than the already existing one. Since both of NYG’s wins are in 3-team loops, they can only be broken by 2-team loops which by definition are season splits.
Reggy, although Miami’s overall record is the worst in the AFC East, they actually have the BEST intra-division record in the AFC East.
Note: I am a Dolphins fan, but still not sure what to make of them. I figure they’ll finish up around 7-9 this year.
The loss to CLE really hurt the NYG in these rankings, but the schedule is hurting them somewhat as well. Before that loss in week 6 their wins vs teams other than WAS were against apparently weaker teams (STL, CIN, SEA) so when they lost to CLE and created the beatloop they didn’t have any other strong wins to hold them up.
One of the things I’ve noticed after watching this site for the past few seasons is that because of the way the NFL schedule is set up, some wins or losses are much more likely to cause beatloops than others, and thus make it easier for a team to overcome an earlier loss, or to regain an earlier win that was removed because of a beatloop. Games between division members from early in the season are most likely to have changes in the graph, because the teams play each other twice in the season and 2 team beatloops are so possible and tough to overcome. The same is true of reinforced wins of a season sweep. Loose twice to a team and it’s really difficult to overcome.
On the other hand, the two games that a team has that are based on the previous year’s standings, the so called “strength of schedule” games, are more likely to stay on the graph than other games. Loops get created when a set of teams all play each other and have conflicting results. Given that all the teams in one division play all the members of two other divisions, one in each conference, that can make a tangled mess of teams and loops. It’s those two games that are outside of this common set of opponents that seem to make very stable paths. If I can find the time, it would be interesting to look at if those observations really hold up.
MOOSE: Don’t forget a tie, which is also part of the definition of a 2-team loop, but rarely happens. Also, just a web page setup issue, I was flipping through the weeks on your standard graph and, not realizing that I’d gotten to the current week, tried to go to the next page. That’s when I got a 404 not found error. Not a big deal, but should be easily fixable.
JT: Yeah, I’ve been a little lazy as far as that goes right now. You actually can use the previous year button to do the same thing beyond 2005. It just saves me a little bit of work when I add the next week to not have to go and fix the previous week’s link. All of these web pages are automatically generated by my program. Thanks for the report though, I’ll come up with something to deal with the extra links.
I guess it depends on how you look at a tie on whether or not you count it as a 2-team loop. The way I look at it, a 2-team loop means A->B->A. This is a loop because you can draw a graph with the win links between the teams. For a tie, neither team has defeated the other, and no win links exist to get destroyed.
While either way, the effect is just like the two teams didn’t play each other at all (with the exception of the weighted method where score matters). Only in the case of a season split do the BeatWins exist to get resolved by the BeatLoop portion of the program.