It appears the one clear candidate for a junk game in Week 17 was CAR@NO. Starters were rested, it was a party, the crowd was cheering all the way through even though the home team lost, etc. So here is probably a better representation of the end of the season, by refiguring the rankings with the CAR@NO game removed entirely.
Vanilla (no beatflukes):
Rankings: IND,BAL,SD,NE,NYJ,PHI,DEN,NO,CHI,TEN,SEA,KC,PIT,BUF,CAR,DAL,STL,GB,
CIN,NYG,ARI,MIA,SF,MIN,CLE,ATL,WAS,HOU,TB,DET,JAC,OAK
Playoffs:
AFC: IND, BAL, SD, NE, NYJ, DEN
NFC: PHI, NO, CHI, SEA, CAR, DAL
Projected SB Winner (actual playoff teams): IND over PHI (same game results as previous non-beatfluke projection)
Beatflukes:
Rankings:
SD,IND,BAL,DEN,NE,CIN,PIT,NO,PHI,CHI,BUF,CAR,CLE,NYJ,GB,MIA,KC,ARI,
SF,SEA,MIN,TEN,DAL,NYG,STL,DET,ATL,TB,WAS,HOU,JAC,OAK
Playoffs:
AFC: SD, IND, BAL, NE, DEN, CIN
NFC: NO, PHI, CHI, ARI, CAR, GB
Projected SB Winner (actual playoff teams): IND over NO
Note that the results would be different if NO had actually won their last game over Carolina… in that case, NO would have actually been ranked #6 in the power rankings (both variants) to start the playoffs, the highest ranking for an NFC team in weeks.
Comparing JAX and PIT’s final ranking, is it fair to say that beatlosses to bad teams are more important in the final rankings than beatwins to good teams? Or is losing 3 times to the bottom 4 teams enough to sink any rank?
Continuing with discussion from the last post, I’d be interested in some other changes, too. I have a thought on beatflukes where instead of considering a fluke based on one alternate path, you require two non-overlapping paths before you label a game a fluke. Or does that sound like too much? (I imagine it would make calculating beatloops more difficult).
In another vein, I know you said the code was weird. I was wondering if you were thinking of releasing it. I’m a programmer (mostly perl), and I wouldn’t be surprised if there were other programmers here too who might be interested. Just curious.
There was that long comment I left about the HOU/JAC situation where it made me wonder if separate alternate beatpaths would be a good idea, too.
I actually just need to rewrite the code because I think I can make it a lot simpler. When I wrote it I knew nothing about directed graphs or topological sorts, and so I very much implemented it the long way around.
As for releasing the code, I’ve been torn on that. On the one hand it would be fun to make it a community effort. On the other hand, when I first wrote the code I was motivated by thoughts of “hey this could be really cool if I could make a column out of it or get someone like USA Today interested in the multi-colored graphs if I made them pretty enough”. So there has been the distant hope of dollar signs in the back of my head. I’m probably not ready to let go of that thought entirely. If I rewrite the code, and hook it up to some web utilities to make it useful year round, then maybe I could at least make some money off of google ads or something. (I don’t really even know what my readership is at this point – maybe I’d already be able to get a buck or two a month off of ads.
) Anyway, as soon as I bring those thoughts into the equation, then it makes releasing the code a dicier issue.
re #1: I don’t think beatlosses to bad teams are more or less powerful than beatwins to good teams. When you figure in strength of schedule, PIT just apparently played a better season than JAC did.
Having a win differential of more than one, though, does make a big difference to the graph. Like a 2-0 season sweep, or a 3-1 season series. No matter how many three-team beatloops there are, it will only remove one of the victories… that still makes me uncomfortable sometimes, but so far every other approach I’ve tried has had worse tradeoffs.
Despite that, JAC is still ranked low if I ignore one of the HOU/JAC victories.
re: Code, that’s cool. I can’t blame you for wanting to keep something you’ve worked on, and profit if an opportunity rises.