NFL Week 12 Game Results

Update: Boooring. Neither the Sunday night nor Monday night games have an effect on the visible graph.

Afternoon games:

Here’s the beatpath graph for the afternoon games. Seattle’s win enables them to finally shake their beatloss to Washington. Washington looks pretty lonely up there, don’t they? Washington should sink quite a bit in the rankings this week since they aren’t propped up by Seattle anymore, and Seattle will rise. That’s pretty much the big story of the week – so far, most other rankings movement is pretty minor (although I will probably switch tiebreaking algorithms again for next week’s rankings).

Early games:

7 Responses to NFL Week 12 Game Results

  1. Gerry says:

    Love Beatpaths, think it adds insight. But like every system, there are always holes which is why one should never rely on just one. In this case, Washington. Way too high.

    But the system is what the system is.

  2. Gerry says:

    Heck, every year the standings have someone ‘better’ than everyone knows they are. No system is perfect.

  3. ThunderThumbs says:

    Well, the vertical placement in the graph doesn’t always correspond to where the team would be in the rankings. But you’re right – the only reason Washington has been so high has been because Seattle has been propping them up. We all “know” that Seattle is better than Washington, but at the end of the day it doesn’t matter because a win is a win is a win. So in that sense it’s perfectly accurate. Washington has to be ahead of Seattle.

    But, since Seattle just beat the Giants (who beat Washington), that link is finally broken. Check out the new graph for the afternoon games – Washington doesn’t even have any beatwins anymore. They’re good enough that their only beatloss is to Denver, but you’ll see them drop hugely in this week’s rankings.

  4. ThunderThumbs says:

    All that said, I agree that it shouldn’t be relied on by itself. Earlier this week I found that any beatpaths system is reliably in the 58-62% range for picks, which is pretty good but not as good as the bookie’s 66% rate. But remember that this system pays no attention to home field advantage. By introducing an arbitary rule to flip the pick if the home team were three or less slots below the visiting team, I got one algorithm variant to go up to 65%. I think if someone used this as a reference, and then adjusted their views to account for things like home field advantage, momentum, injuries, and matchups, they could do pretty well.

    This year the vanilla variant is already outperforming some of the ESPN pickers. So yes, the key is to use it for entertainment or in combination with other systems. Or just admire the pretty pictures. 🙂

  5. thebobster says:

    You mentioned that the beatpaths system gets about 60% for picks. Is that at the current point in the season? One of the obvious characteristics about beatpaths is that it starts out with extremely high variability (i.e. after week one 16 teams are tied for first place and 16 teams are tied for last place). Do you notice that accuracy improving as the sample size (i.e. number of games into the season) increases?

    I wonder what would happen if you retroactively used the last 5 games of last season so you always had the previous 16 games played… would that nudge your accuracy above 60%?

    Just wondering…

  6. ThunderThumbs says:

    Yeah, that’s the current point of the season since week 1. As of today’s games we’re at 61.1% when I tiebreak on raw beatpower. The actual beatpaths games (where the teams have an actual beatpath relationship before the game is played) are at 62.9% .

    The first couple weeks of the season were 9-7 and 8-8, so it has improved since then.

    I’ve thought about somehow keeping beatpath data from the previous season, but I haven’t been sure of the best way to do it. Week 17 isn’t always a great week to use anyway… and I can’t really weight certain weeks higher than others. I’ll probably experiment with it anyway.

    But, one thing is that in cases of ties like in Week 1, the system tends to rely heavily on the power ranking for the previous week (end of previous season), so that probably helps a bit. It isn’t great because it still means that all sixteen winners will be ranked ahead of all sixteen losers (for every variant I have except for one), but it still helps protect against it being totally haphazard.

    This week is 9-6 so far. Seems like most of my variants had picked NE=>KC (I probably would have picked different because of home field advantage), OAK=>MIA, NYJ=>NO, TB=>CHI, WAS=>SD, and SF=>TEN. SF=>TEN was the only true beatpath upset, though. I probably would also have picked SD over WAS just for momentum reasons, but it was dicey.

    One of my variants also had HOU over STL but that relationship depended a lot on which variant I used.

    I’ll probably have to anoint these various tiebreaking procedures with some snazzy names and just include a bunch of them in the power rankings…

    The three beatpath wins were JAC=>ARI, DEN=>DAL, and CIN=>BAL. IND=>PIT is also a beatpath pick.

  7. ThunderThumbs says:

    Interestingly, if I switch to the “beatfluke” version of generating beatpaths (using the same beatpower tiebreaker), things change a shade. That one picked NO=>NYJ correctly. 2005 is at 61.7%, beatpath performance at 61.2% Also, CHI=>TB became a beatpath upset (because TB temporarily had a beatpath to WAS due to the system determining that their loss to SF was a fluke loss).

Leave a Reply

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