Game Predictions Week 10

Game probabilities for week 10 NFL games are listed below. The probabilities are based on an efficiency win model explained here and here. The model considers offensive and defensive efficiency stats including running, passing, sacks, turnover rates, and penalty rates. Team stats are adjusted for previous opponent strength. Games in which the model disagrees with consensus favorites are highlighted in red.


  • Spread The Love
  • Digg This Post
  • Tweet This Post
  • Stumble This Post
  • Submit This Post To Delicious
  • Submit This Post To Reddit
  • Submit This Post To Mixx

8 Responses to “Game Predictions Week 10”

  1. Anonymous says:

    How has the model been against consensus?

  2. Anonymous says:

    BBNFLStats: 58-24 (71%)
    Consensus: 52-28 (65%)

  3. Anonymous says:

    CAR 0.62 Loss
    BUF 0.61 Push (ATS)
    PIT 0.86 Loss (ATS)
    KC 0.53 Loss
    TEN 0.61 Loss
    GB 0.66 Win!
    WAS 0.58 Loss
    NO 0.81 Loss

    Whoa Nellie!!!

  4. Brian Burke says:

    Anonymous-I think you misunderstand. These are not ATS predictions.

    But true, there were a large number of upsets in the early games.

  5. Anonymous says:

    Mr. Burke,

    I understand that they are straight up win predictions, but keep in mind, it's not that big of a deal to pick a winner when the spread is 14 or 15 points (as were many of your earlier season picks. Think NE or Indianapolis.)

    Same with Pitts -10 today.

    So judging your efforts ATS is the only legitimate way to adjudicate how good your model is.

    Keep in mind, teams that win, cover the spread more often than not.

    Plus, there are ALWAYS upsets in the NFL. I thought the idea of a mathematical model was to use stats to see through conventional wisdom.

    Having said that, anyone can have a bad week. And you really haven't had a bad week. Just a bad Sunday @ 1:00pm. :)

  6. Brian Burke says:

    Anon-You rightly point out the limits of mathematical models, their purpose, and their advantages. We also agree that predicting mismatch games is not difficult, and therefore any prediction system, mathematical or otherwise, should be compared somehow to a benchmark.

    But mapping my model's probabilities onto estimated point spreads is something I strongly caution against. The model I use is a logistic regression specifically constructed to predict winners, whether they win by 1 or 21 or 51.

    To me the simplest and fairest way of judging a straight-up system is to convert the consensus Vegas favorites into SU favorites, and compare its record with my own.

    I don't pat myself on the back because it can predict at 70% accuracy. I think it's a good model because it's mathematically and statistically very valid. Plus, over the long-term, it's as good or better than any other system of picking winners, including Vegas consensus favorites and other models.

    Even if I accepted your method of judging the model, I would disagree with your assertions on some games. Take BUF 0.61 over MIA. The model estimated MIA would have a fair chance of winning, but BUF likely had the upper hand. That's exactly what happened. I see why you say it's an ATS push, but I'm not sure how the model could have done any better.

    The KC game is another example. By your method a 0.53 win probability equates to a -1 spread for KC. The closing line on the game was -3, so your method would imply an ATS win for my model, not a loss. We could argue all day about which games are "ATS wins" or losses by subjectively converting probabilities into points, but there's no argument if we just convert point spreads into favorites.

    Here's how I would interpret the model's probabilities: The model made the Steelers heavy favorites in the PIT-CLE game. That doesn't mean it's necessarily going to be a blow-out. PIT was able to barely outplay CLE despite falling far behind early. Part of the 0.86 win-probability accounts for just that type of scenario.

    But I agree, 3-5 is not a good stretch for a 3 hr period. Plus it looks like I'm 2-2 in the 4:00 games, losing the single game on which I disagreed with the consensus. Still, even after my worst week of the year, I'm ahead of the consensus.

    It seems about half the comments I get here are from people who ask for spread predictions, or who map my probabilities onto estimated spreads on their own. As a non-gambler, spreads don't excite me very much, but I recognize a lot of people are interested in ATS predictions and I may do that in the future.

    Thanks for keeping me honest!

  7. Anonymous says:

    Mr. Burke,

    Thank you for your clear response.

    I want to reiterate that generally in the NFL, ats winners are Straight Up winners as well.

    For example this week, just picking the SU winners, you would have gone a WHOPPING

    121-1-1 ATS.

    So, I would continue to work on your model that just picks winners.

    And I believer, with nine weeks of data under your belt, that you're about to experience some really suberb results!

    Thanks again for your explanation AND for your openeness to other points of view.

  8. Anonymous says:

    Sorry, that's TWELVE-ONE (one push)

    121-1 is what you're GOING to go! :)

Leave a Reply

Note: Only a member of this blog may post a comment.