Pick Confidence – Week 16

Tom here with the “confidence” rankings of this week’s picks, based on comparing the BeatPower scores of each match-up:

(out of 100)
BeatPower comparison
(predicted winner – predicted loser)
NY Jets-Seattle 93.9 97.9-4.0 WRONG
Tampa Bay-San Diego 70.5 88.0-17.5 WRONG
Indianapolis-Jacksonville 63.2 71.2-8.0 CORRECT
Denver-Buffalo 60.2 81.6-21.4 WRONG
New Orleans-Detroit 41.7 41.7-0.0 CORRECT
Houston-Oakland 40.0 47.5-7.5 WRONG
Chicago-Green Bay 36.6 74.1-37.5 CORRECT
Miami-Kansas City 33.4 35.4-2.0 CORRECT
Dallas-Baltimore 28 92.0-64.0 WRONG
San Francisco-St. Louis 27.1 27.1-0.0 CORRECT
Cleveland-Cincinnati 26.4 36.4-10.0 WRONG
Philadelphia-Redskins 25.0 75.0-50.0 WRONG
Arizona-New England 9.4 59.4-50.0 WRONG
Tennessee-Pittsburgh 5.2 95.7-90.5 CORRECT
Carolina-NY Giants 5.2 92.0-86.8 WRONG
Minnesota-Atlanta 5.2 78.3-73.1 WRONG

The NY Jets matchup with the Seahawks looks about as solid a pick as you can get. Even so, the NY Jets haven’t won any of their games on the West Coast this season, losing to San Diego, Oakland, and San Francisco. We’ll see if they can overcome this weakness.

Another interesting match that could go the other way: Arizona vs. New England. Arizona has the edge, but New England needs it and Arizona doesn’t. (As a minor aside, a friend of mine has about $5,000 riding on this game. He’s one a dozen people left in a rather large football pool. Why he chose the Pats to win, out of all the possible picks this week, is beyond me.)

22 Responses to Pick Confidence – Week 16

  1. Kenneth says:

    I’m pretty sure that PIT-TEN could decide who gets the #1 seed. Seems like they might play the first stringers for that.

  2. Tom says:

    Yeah, playing for homefield advantage, which is not yet decided. Just realized that.

  3. Eddo says:

    My guess is that your friend picked New England in his survivor pool because he’s used any other viable teams this week. Generally, survivor pools only let you use each team once.

  4. Tom says:


    Well, given the final score, it looks like it was a good pick to have made. I guess that’s what happens when desert teams play in 4-6″ of snow. Funny thing is, even if it were a home game for Arizona, they’d still be playing in a blizzard.

  5. ThunderThumbs says:

    Good lord. Every late game was an upset. What a crazy season this has been. If this had been the first season I’ve done beatpaths I might be more discouraged. But it’s as if there were some freak occurrences early in the season that the graphs have been recovering from ever since.

  6. Kenneth says:

    Snow or not, I don’t think it mattered. Arizona looked horrible. I know we don’t know the playoff matchup yet, but I don’t know if there’s a line high enough you could give me to take the Cardinals in the wild card game.

    I mean, within reason–I guess if you gave me 35 points, I’d take it.

  7. doktarr says:

    Crazy week; lots of upsets.

    Iterative did better, although it still wasn’t a banner week. Iterative correctly called Washington over Philadelphia and Baltimore over Dallas. It also almost certainly gets the Giants over Carolina right – there’s no beatpath, but the Giants have a beatpath to every team that Carolina has a beatpath to, plus six more. So, 8-7 in stead of 5-10. Better, but still not good.

    I would not be totally shocked if the Cardinals beat the Falcons at home in the first round of the playoffs. They looked utterly terrible out there in the snow, but I still think they have the potential to abuse the Atlanta secondary. I’d pick Atlanta, but it might not be a blowout.

  8. doktarr says:

    Tom, what sort of pick’em league was your friend in? In my eliminator league I can only pick each team once, so next week I only have 16 teams left to choose from. Thankfully Atlanta still has something to play for (division title, a first round bye, and a home game in stead of two road games), so I’m probably taking them at home over St. Louis.

    New England was also given the largest point spread of any game this week. It looks like the oddsmakers DID know something we didn’t know. Setting aside all our discussions of various algorithms, we should remember that beatpaths is fundamentally descriptive and not predictive. If we wanted it to be predictive, we would at the very least add some element that weights more recent games more heavily.

    Of course, the oddsmakers also assigned the biggest spread of the week to New England over Miami in week 3, and Washington over St. Louis in week 6. So it’s no guarantee. In my eliminator league, I like to look at a variety of sources around the ‘net, and take a pick that is a consensus strong pick by every measure. I’ve gone against a clear consensus pick four times, and two of them are the two upsets I mentioned. That makes me feel pretty smart. But then I remember that I picked Minnesota over Detroit in week 6 and San Diego over Kansas City in week 10, and I realize that I’m a little lucky too.

  9. Tom says:

    Doktarr, apparently he can pick any game, any team, any number of times–no constraints. He could have done just as well picking against Detroit throughout the whole season, if he wanted a low-risk way of staying in the pool.

    You’re right about the descriptive-predictive difference. I guess it’s just a decision you have to make about parsimony vs. complexity in building a model. It’s fun to see how much simple beatpaths can explain, but this week it seems we’ve run up against the limits of the explanatory power of such a small data set (wins-losses). That said, we didn’t do too much worse than the folks over at Accuscore (they went 7-8) with their hundreds of variables and thousands of simulations with genetic algorithms, etc.

  10. doktarr says:

    I don’t think we really ran into the limits of the explanatory power of such a small data set this week. I can make two completely different arguments why I don’t think that is the case:

    1) Parallel beatpaths went 5-10; this is the same record as “point spread favorite”. The oddsmakers picked New England, but they also picked Pittsburgh. The oddsmakers consider a huge number of variables, and they didn’t do any better.

    The issue here is simply that the results of one week, good or bad, do not validate or invalidate an approach. There’s way too much noise/variation to think that way.

    2) Iterative beatpaths went 8-7 using the exact same data set. As noted above, though, it’s just one week.

  11. ThunderThumbs says:

    Yah, I’d agree with doktarr… fundamentally descriptive, fun to play with predictiveness on top of that. And I think the current algorithm could be better – in addition to still being intrigued about Iterative (and just needing to implement it as an algorithm choice here), I’m also wanting to take the time to look at Boga’s suggestion of just looking at all beatloops at once.

    The main thing that bugs me, though, isn’t the CLE->NYG scenario, at least not so much as how one really strong beatwin can “prop up” a team that is otherwise very weak. It looks very likely that the NYJ are going to be ranked #1 this week, since the NFC East and NFC South took such a tumble. I know there’s a possible beatloop that would get rid of NYJ->TEN but other beatloops are getting resolved first.

    But overall I think it’s about sample size. NBA is probably an even better fit for this approach. The lower the teams/games ratio, the better I think.

  12. doktarr says:

    Actually, I’m very confident that the Jets will lose the TEN beatpath this week in iterative, and this is, in fact, a good example of the strength of iterative. It’s basically the Giants’ situation in reverse, and the same arguments apply.

    The reason NYJ=>TEN is so “hardy” in the parallel approach is that it is a “strength of schedule” game. Since both teams beat their other common opponents (KC and Cincy), the smallest beatloop it can be a part of is a four-team loop, e.g. NYJ=>TEN=>IND=>NE=>NYJ. But the Jets managed at least a split in all their division matchups, so those are off the board, too.

    Nothing surprising so far, but here’s the point. No matter how anomalous the NYJ=>TEN game was, it can only get wiped out in parallel if it is part of a loop at least as small as the largest loop containing an unresolved Jets loss. This is the issue with parallel; not only can a single game have a big impact, but the size of that impact is strongly influenced by where that game happens to fit in a team’s schedule.

    The Jets take care of one of their losses themselves – the split with New England. The other five losses (DEN, SD, OAK, SEA, ARI) are ALL part of three-team beatloops with Miami alone. In other words, that ONE Miami win can save the Jets from FIVE losses. It’s the Giants/Browns game in reverse. The Buffalo sweep could also loop away four of the five, although Buffalo lost to San Fransisco.

    In iterative, the three wins over Buffalo and Miami can only erase three losses, total. You end up with SF retaining .8 of a win over the Jets, and OAK, DEN, SD, and SEA retaining .3 of a win over the Jets. That adds to 2, which makes sense because 5 losses – 3 wins = 2 losses left over. Moose hasn’t recomputed the rankings yet, so it’s possible that after all is said and done, the Jets manage to loop away the 2 loss residue and hold onto their top spot, but I really doubt it. In all likelihood, they will shed those losses, but at the cost of their TEN win, leaving Arizona or Miami as their strongest remaining wins.

    Incidentally, losing to Miami in week 17 will cause SF=>NYJ to re-emerge, and could cost the Jets their top spot in parallel as well.

  13. doktarr says:

    I forgot about the NYJ=>ARI=>SF=>NYJ and NYJ=>ARI=>SEA=>NYJ loops, which messes up my iterative math a bit. The iteration has to go through several stages and gets a little hairy, but I think the SEA=>NYJ game (the only Jets loss in all three loops) gets wiped out, and the other four Jets losses are held at .25 strength each, after the three team beatloops are considered.

    This fully exhausts the Arizona, Buffalo (x2), and Miami wins, though. The only remaining hope for the Jets to hold onto the TEN win is the NYJ=>CIN=>JAX=>DEN path, which could allow them to hold onto the TEN win at .25 strength. Actually, I think it will. Damn you, Brett Favre!

  14. doktarr says:

    OK, I did it out on paper this morning (yes, I was trying to handle 11 3-team beatloops involving 9 teams in my head). Assuming I haven’t forgotten any other 3-team beatloops involving the Jets, the iterations go:

    Stage 1: Reduce every loop by .2. Games removed: NYJ=>MIA.

    Stage 2: Reduce every (remaining) loop by .2 (again). Games removed: SEA=>NYJ.

    Stage 3: Reduce every loop by .1333. Games removed: NYJ=>BUF (both of them). NYJ losses that are no longer in a loop: SD=>NYJ, OAK=>NYJ, and DEN=>NYJ, all with a weight of .2666 (4/15).

    Stage 4: Reduce the remaining loop by .0667. Games removed: NYJ=>ARI. NYJ losses that are no longer in a loop: SF=>NYJ, with a weight of .2.

    So… I was pretty close with my .25 back-of-the-forehead calculation. Bottom line is that the Jets retain losses with a total weight of 1 (3 strength .267 losses, and 1 strength .2 loss) that may come into play in larger beatloops. Again, I think the NYJ=>CIN=>JAX=>DEN path will share some of the load, such that the Jets will just barely hold onto their TEN win, with a weight of something like .1.

    And if Miami beats them next week, their whole structure comes crashing down, and they’re left with wins over the Rams, Chiefs, and Bengals. Happy Hanukkah.

  15. JT says:

    Earlier this season I was looking at the scheduling, and was going to look at the 4 types of games on the schedules for each team:

    – In division (6 games)
    – In conference division pairings (4 games)
    – Other conference division pairings (4 games)
    – In conference strength of schedule (2 games)

    The idea I had was to look at which types of games were more likely to be involved in beatloops. I never got around to it, but there were some things that looked pretty obvious. The most common beatloops (due to the algorithm perhaps) are 2 and 3 team loops which involve a set of teams all playing each other. In division games seem more likely to be involved in beatloops as there are 12 games involving just 4 teams, and they make up 37.5% of a single teams season. The strength of schedule games, on the other hand, seem less likely to be involved in beatloops since those are against teams that your division rivals do not have in common with you, and I think it would take a beatloop of at least 4 teams to wipe out one of these games. Since the loops of 2 or 3 teams are removed first, they are likely to break a potential 4 team loop up before the processing gets that far.

    It would be interesting to run the numbers on the types of games by scheduling type, and which get removed in beatloops, maybe if I get annoyed with in-laws these next few weeks…

  16. The MOOSE says:

    Haha, you can relax now Doktarr, my graphs are up. NYJ did hang on to their win over TEN. SD jumped to the #1 spot in Weighted.

  17. doktarr says:

    …which tells you what you need to know about weighted.

    A win by Miami would “solve” the problem and stamp out the Jets win.

    Cincinnati jumping up was a real shocker to me, but then I looked at their season, and realized that the loss to the Browns in Ryan Fitzpatrick’s first game is actually their only loss to a truly bad team. Aside from that, they’ve lost to Pittsburgh (x2), Baltimore (x2), Tennessee, Indy, Dallas, Houston, and both New York teams. Houston is the worst of that bunch and they’re not terrible. Pair that with a win over Washington and a tie against Philly, and maybe the Bengals are better than we think. They might be an average team that only looks terrible because of a tough schedule.

    See, this is why beatpaths are cool.

  18. The MOOSE says:

    Unfortunately I have to root for NYJ to keep our graphs messed up because they’re NE’s ticket into the playoffs. I seriously hate rooting for NYJ and DAL, but have had to in consecutive weeks due to playoff implications. DAL failed me, but BAL can still lose this week. I’d much rather have NYJ win and have NE win the division than BAL lose and NE get the 6th seed though.

  19. boga says:

    Another idea to look at the “predicitiveness” of the beatgraphs would be comparing it to when beatgraphs predicts a different winner than who vegas pegs as the favorite, and then seeing who is right more often.


  20. ThunderThumbs says:

    Here’s one thing about how crazy a week this was. Each week when the victories are figured into the new system, and a new beatpath rankings is figured, I can see how that new ranking would have matched up against the games. Against this new ranking, this week’s results were still only 11-5. It’s as if the system is saying there were five “true” upsets – ATL->MIN, SD->TB, SEA->NYJ, BUF->DEN, and OAK->HOU.

  21. doktarr says:

    That seems about right, although as usual I think we should distinguish between “path picks” and “ranking picks”.

  22. […] Tom here with the “confidence” rankings of this week’s picks, based on comparing the BeatPower scores of each match-up. The picks took a beating last week, but nevertheless, the high confidence picks still did twice as well as the low confidence picks, even on a bad week overall. Here are this week’s: […]

Leave a Reply

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