Conference Championship Predictions

This weekend AFC defensive powerhouses Baltimore and Pittsburgh match up, and NFC underdogs Philadelphia and Arizona square off.

The game probabilities are based on team performance for all games since week 9, with the exception of week 17 when some teams played at less than full strength. This includes playoff games to date.









PwinGAMEPwin
0.40 PHI at ARI 0.60
0.33 BAL at PIT 0.67



Probabilities based on the complete regular season would be:

BAL at PIT, 0.33 to 0.67
PHI at ARI, 0.67 to 0.33

Why is Arizona favored over Philadelphia when weeks 1-8 are thrown out? The Eagles racked up gaudy stats early in the season despite not having the wins to show for it. Eventually, their luck started to even out and they squeaked into the playoffs. But throwing out their best statistical weeks really hurts them. Plus, Arizona's last two wins against quality opponents were convincing, improving both their stats and their average opponent strength significantly.

  • 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

11 Responses to “Conference Championship Predictions”

  1. Daniel says:

    Looking at your post about the luckiest teams (and actual games where the Eagles lost at the end of games) its surprising to think that a 12.1 win team (Eagles) would be road underdogs to a 8.2 win team especially win the 8.2 win team is from the NFC West.

    Concerning the Ravens/Steelers, we saw the Giants lose to Dallas twice in the regular season before finally solving the riddle at Dallas in the playoffs. The Ravens were so close both times and I wouldn't be surprised at all if they won. They have Superbowl experience, great leadership (Ray Lewis), and a good defense. That is a recipe for victory.

  2. KiranR says:

    I must admit, that both these estimated win probabilities seem "out of the ordinary". In a rudimentary model I built (using the netire season's stats, but not including any playoff stats), I show Philly a 4-point favorite, and Pittsburgh a 3-point favorite. I am not sure how that translates to a win-probability, or how that lines up with Vegas odds, but your model does seem to be suggesting rather different outcomes.

    Your model also seems to suggest high-degree of sensitivity to time period analyzed (last 8 games versus last 16 games).

    Also, I am curious, does your model suggest anything about socres? In other words, does a 0.67/0.33 win-likelihood in the Pit/Bal game versus a 0.55/0.45 likelihood for that same game imply different scores, or is it possible to have different win-likelihoods associated with the same score(s).

    Thanks,
    Kiran

  3. Tarr says:

    I don't mean to sound excessively harsh, but your choice to throw out weeks 1-8 without any strong supporting analysis works against the otherwise strong statistical backing your methods have. Rather than simply pick an arbitrary cutoff, you should do a wide study and find how far back you have to go before games start to lose their predictive value into the future. This would probably lead you to a set of weights which would decay as you move farther back in time.

  4. Brian Burke says:

    Tarr-I don't disagree, and I did do some of what you suggest. It takes varying numbers of games to settle depending on the stat. The turnover stats take a longer time than the running stats for example. There is just a higher baseline rate to consider.

    But in no case does it take more than 9 games to stabilize. Additionally, the prediction model's general accuracy is no better in week 14 than week 6. The model's level of confidence might appear slightly higher using fewer weeks, but not much. There are diminishing returns with each additional week of data. So although it's a little arbitrary, I think throwing out weeks 1-8 is fairly safe.

    But I do provide the full-season probabilities as well, in case you feel better about those.

    Also, my model might be the only one that favors Arizona. One reason might be the exclusion of week 17. That was the week of the Cardinals' worst game and the Eagles' best. Intuitively, I think the PHI game "counted" and ARI's game was a throw-away, but I have to be even-handed.

    What we can conclude is that both the Eagles and Cardinals are inconsistent, and any prediction should have pretty low confidence.

  5. Jason says:

    I think throwing out week 17 is why the Cardinals seem favored. That means the last regular season game for the Eagles is the Redskins, arguably their worst game of the season. Why not throw that one out too, since technically the Cardinals didn’t play a "real" game the rest of the regular season after locking up the division?

    To be even-handed, you should include week 17, not arbitrarily throw it out.

  6. Tarr says:

    I agree you're probably right about being able to exclude weeks 1-8 without losing much, if any, predictive power. But I think the most accurate approach would probably be some sort of declining set of weights, probably resembling an exponential or sigmoid curve.

    "Also, my model might be the only one that favors Arizona. One reason might be the exclusion of week 17. That was the week of the Cardinals' worst game and the Eagles' best. Intuitively, I think the PHI game "counted" and ARI's game was a throw-away, but I have to be even-handed."

    Intuitively, I agree with this as well. The Cardinals were on cruise control for a while at the end of the season, while every Eagles game counted.

  7. Anonymous says:

    Brian, why don't you calculate the win probabilities for playoff games using the win correlations that you've already calculated for playoff games and good vs good matchups?

  8. Anonymous says:

    I'd like to throw my boat in with declining weights - counting week 9 100% but week 8 0% seems very arbitrary and, though maybe safe, takes I think something away from both the objective backing and rationale and the elegance of the model.

  9. Brian Burke says:

    I'm not sure the sample sizes are large enough to just base the coefficients on good v good and playoff games. For example, having a good running game would actually appear to be bad. That's where the data is pointing right now, but I'd be a lot happier with more evidence before basing the model on it.

    I hear everyone about the weeks to include/exclude and decaying weights. I don't disagree, but it would take a month just to tweak it to find out what the decay rate should be, and even if it would make much of a difference at all.

    For now, you can weight my predictions as desired. One is based on the most recent 10 games (9 for PIT because of their bye) minus week 17, and the other is based on the entire regular season minus week 17.

    The bottom line to me is: PIT and BAL are very similar--almost identical--except that PIT has an even more incredible pass defense. PIT has HFA. Edge-PIT.

    PHI and ARI are extremely sensitive to which weeks you keep and which you throw out. That tells me they are both hot and cold teams, and any prediction should have a relatively low confidence level. The consensus favors PHI, partially because they were the "hotter" team going into the playoffs. But ARI had very little to play for the last 3 weeks of the season, and still look just as strong as PHI, plus have HFA. So it's somewhere between 60/40 and 40/60--a toss up.

  10. Anonymous says:

    Hello fellow NFL Fans, first time posting here, but long time NFL Fan and resercher of stats for over 25 years. I've followed many sites and anything that involves stats over the years trying to learn something new.

    Just wanted to make a comment on the model. If you've followed "football outsiders" with their dvoa and wieghted dvoa, well they look into those 2 to determine which was a better indicator for predicting the NFL Playoffs.

    And reach the conclusions that dvoa which covers the complete 16 game season was a better indicator than the weighted dvoa.

    Having researched stats for years myself, I've found the same results.

    There's a stronger correlation to all 16 games as to how teams perform in the postseason.

    However, I will say the past few years hasn't been.

    Should be very interesting to see if it retuns to the norm with Philly spanking Zona, I suspect it will.

    By-the-way my numbers tell me Ravens better than Pitt and likely to win the game, if not for all the injuries.

  11. Anonymous says:

    Well Jake took a bit of a bath on predicting these 2 games. I would like to add that I suspect that the stronger correlation for using 16 games rather than 9 is probably due to simply having a larger sample size.

    In predicting Baseball playoff winners, the intra-divisional record has more significance than the head to head record. Looks like a very similar situation here.

Leave a Reply

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