So, you go to the playoffs with the seedings you have, not the ones you wish you had. Here’s how the seedings actually work out:
AFC: SD, BAL, IND, NE, NYJ, KC
NFC: CHI, NO, PHI, SEA, DAL, NYG
For the wildcard round, both the beatfluke and non-beatfluke variant project the same outcomes, although in one case the NYJ/NE game is really close.
Wildcard round:
Indianapolis hosts Kansas City and wins
New England hosts the Jets and wins
Philadelphia hosts the Giants and wins
Seattle hosts Dallas and wins
For the divisional round, the projections differ depending on method. So we’ll split them out here. First, the beatflukes variant:
Divisional round:
San Diego hosts New England and wins
Baltimore hosts Indianapolis and loses
Chicago hosts Seattle and wins
New Orleans hosts Philadelphia and loses
Conference round:
San Diego hosts Indianapolis and loses
Chicago hosts Philadelphia and loses
Super Bowl:
Indianapolis (#1) defeats Philadelphia (#4)
Final rankings:
IND,SD,BAL,PHI,DEN,NE,CIN,PIT,CAR,NO,CHI,BUF,CLE,NYJ,GB,MIA,KC,DAL,
ARI,SF,SEA,MIN,TEN,NYG,STL,DET,ATL,TB,WAS,HOU,JAC,OAK
And, the normal non-beatflukes variant:
Divisional round:
San Diego hosts New England and wins
Baltimore hosts Indianapolis and loses
Chicago hosts Seattle and loses
New Orleans hosts Philadelphia and loses
Conference round:
San Diego hosts Indianapolis and loses
Philadelphia hosts Seattle and wins
Super Bowl:
Indianapolis (#1) defeats Philadelphia (#7)
Final rankings:
IND,SD,BAL,NE,NYJ,CAR,PHI,SEA,DEN,NO,CHI,TEN,KC,PIT,BUF,STL,CIN,
GB,DAL,NYG,ARI,MIA,SF,CLE,MIN,ATL,WAS,TB,HOU,DET,JAC,OAK
In all these scenarios, it looks kind of like Carolina got screwed.
I’m sure you don’t need more work, but I’m curious what would have happened in the graph if Jacksonville had won that final game. Or was there some other event that caused the gigantic drop in the graph?
Anyway, love the site! Glad you were able to put it up.
Looks like (using beatflukes) Jacksonville would be ranked #20, just behind Kansas City.
Thanks, glad you enjoy it!
Long-time follower of the site — thanks for keeping it up. This season hasn’t felt like it gave the best showing for beatpaths though. So many seeming random results and so many teams changing personalities once or even twice through the season has made the ecology very unstable week to week. It really has highlighted what I now think is the biggest weakness of the beatpath algorithm as it is currently — it’s so deterministic.
Really what we want to establish is confidence levels. If two nearly matched teams would split 10 games while a dominant one would take 9/10 games against a far inferior team, when we observe a game result it might be a 1-in-2 result or a 1-10 result and beatpaths can’t tell the difference. It simply hopes that enough other information is (or will become) available to loop away or fluke away the unlikelier outcomes. Unfortunately the other games may be just as fluky. Given enough games, this isn’t a problem. However 256 games is way too few data points that a good percentage of seasons will end up way off. Still, it’s been great following your work. Perhaps when I get more time I’ll see if I can come up with a workable way to quantify the (un)certainty of results.
Anyway, that was a sidebar =). The question I wanted to ask was whether some combination of the 2048 possible playoff results would allow each of the playoff teams to achieve the #1 spot if they win the Superbowl. Or put another way, what’s the best ranking each team can possibly achieve in the remaining games? Though it worked out nice for the Steelers last year (at least with the beatflukes tiebreaker), I suspect it won’t work for a majority of the teams in the field right now. What do you think?
(I realize it might be a pain to run out 2048 scenarios so I won’t blame you if you don’t want to do it =).)
I’m probably not so interested in stuff like confidence levels and probablities, just because I imagine it will make my brain hurt, and because I like the “purity” of considering just how far I can get with wins alone. But you’re right about the lack of data points – it’s also why college football is even harder, and why I think college basketball and pro baseball/basketball could be really interesting (although college basketball would probably make my macbook explode).
I think part of what exacerbated the appearance of instability was my decision to go with beatflukes this season. Beatflukes is slightly more accurate, but accuracy of picks was never really the point here, and the tradeoff is some more extreme shifts in the power rankings. I will probably go back to vanilla next season. I also have a bunch of other tiebreaker variants (where you translate the beatpath graph to power rankings) to try out.
It appears extremely unlikely that any of the NFC teams will end up ranked #1. I’ve tried a few different scenarios and I haven’t come up with a way that that will happen yet.
One of these times I’ll figure out the full season record and see how it compares with last season. Here’s King Kaufman’s regular season panel o’ experts results:
Expert W-L Pct.
1. Charles Robinson, Yahoo 161-95 .629
1. Mike Golic, ESPN 161-95 .629
3. Adriana Sage, EroticModelPicks 157-99 .613
4. Eric Allen, ESPN 156-100 .609
5. Yahoo Users 155-101 .605
5. Ron Jaworski, ESPN 155-101 .605
7. Cris Carter, Yahoo 154-102 .602
8. Buster, Coin Annual 153-103 .598
9. Sean Salisbury, ESPN 151-105 .590
9. Mark Schlereth, ESPN 151-105 .590
11. Merril Hoge, ESPN 149-106 .584
11. Peter King, Sports Illustrated 149-107 .582
13. King Kaufman, Salon 146-110 .570
14. Joe Theismann, ESPN 144-96 .600
15. Chris Mortensen, ESPN 139-117 .543
15. Vinnie Iyer, Sporting News 139-117 .543
While as a DEN fan, I find the Broncos final placement on the graph hopeful. As a realist, I think it overstates the effect of their early win over NE and their very close win over CIN at the end. I’m not certain how one would correct for that. The teams which beat DEN, especially the NFC West teams seem like they really were more “fluke” type wins. I wonder if a STOMPS system, where one first accounts for games with high victory margins, e.g. 10+ point victories, and then adds in less decisive games would help.
Now that is an interesting idea. Like, what if I just threw out all games with victory margins of three points or less. Or handled beatloops of close-margin games first… Hrm. A road fraught with peril that is.
“Miles of code to write before you sleep” as someones .sig says, but I presume you do this because you enjoy it….
Well, sorta.
I look forward to the graphs turning out, but I do find myself resisting tearing apart the code again. Most people here don’t know this, but I wrote the main functionality while I had a REALLY bad cold and it did strange things to my brain. I think differently when I’m sick – I can think really deeply because I can only think of one thing at a time. I look at the code while I’m healthy and all multi-tasky and it ties my brain in knots. Weird recursive functions all over the place.