2012-09-24

Superbowl XLVII Projection, Week 3, Sunday

Houston over Atlanta

Houston remains the strongest team, but Atlanta moves right up behind them, and has the greatest chance among NFC teams of making the Superbowl.

I've added the strength column thanks to a question by Josh from Royals and Randoms. The strength is the Pythagorean Winning Percentage expectation, found from a statistic called PythagenPat, originating in Baseball Prospectus

Each time I run my simulations, the total number of points and games played thus far are used to find the points-per-game (ppg). An exponent, X, is defined as ppg0.287. The 0.287 is actually used for baseball, but I've yet to find a "better" value for football. Then, a team's Pythagorean Winning Percentage expectation is found using their Points For and Points Against, as PFX/(PFX + PAX). A team that has shut out every opponent will have a winning expection of 1.000, and a team with the same number of points for and against will have an expectation of .500, which makes sense.

Finally, I use a tactic from SportsClubStats.com, which is regressing the expectation to the mean. I do this by first subtracting .500, making all values between .500 and -0.500. Then I multiply by the percentage of the season passed so far, 3/16 for most of the teams below. Then the .500 is added back. This means that a team who has had nothing but shutouts for 3 games of the 16 game season would have a strength of .59375, and a team that has scored no points would have .40625. This keeps crazy results early in the season from showing a team with a huge chance of being the champion after 1 dominant game.

Denver has the same Points For and Points Against so far, and so are at exactly .500.

The strengths are used to simulate the games. If team x has strength xp, and team y has strength yp, team x's chances of beating team y are:
xp(1-yp) / (xp(1-yp) + yp(1-xp))
and vice versa for team y beating team x, and it can be seen this will sum to 1.

Then a random number is generated between 0 and 1 and used to determine the winner of the simulated game in a weighted fashion.  I run it at least 100,000,000 times, and The result is the lovely table of percentages below.

TEAMRecordStrengthDivPlyffWCQtrSemiChamp
HOU3-00.57679.23488.53675.50543.09824.29313.103
ATL3-00.57278.63084.60168.26337.63420.47211.157
ARI3-00.56157.84876.19858.89731.62616.7638.956
CHI2-10.55036.81854.55537.79719.62010.0955.279
BAL2-10.54951.60268.12950.41526.87614.0817.212
NYG2-10.54840.10357.00239.64920.40210.4295.427
SEA1-10.53817.27840.74525.97513.0466.5613.352
NE1-20.53422.81537.99024.57312.5096.3483.156
SD2-10.52952.26065.61546.81823.81611.9175.867
MIN2-10.52430.59547.26631.19115.2867.4413.696
BUF2-10.51428.41449.05932.33115.8417.7023.674
GB1-10.51122.21735.30821.82210.3894.9252.383
NYJ2-10.51135.52651.93434.32616.7468.0873.837
SF2-10.51119.48040.04824.85611.7995.6112.712
PIT1-20.50418.98331.74719.2689.2074.3852.048
DEN1-20.50023.06135.23321.35210.0934.7582.205
MIA1-20.49813.24426.22615.2887.1813.3741.557
WAS1-20.49715.04224.52814.2476.5312.9971.410
DET1-20.48910.37018.67110.4994.7562.1570.996
TB1-20.48411.62321.11211.4955.1602.3261.063
DAL2-10.48125.96438.80022.97110.1574.4852.035
CIN2-10.47524.59341.72225.12111.2505.0142.199
NO0-30.4723.5517.6283.8841.6910.7410.329
STL1-20.4645.39512.7536.6632.8491.2270.537
CLE0-30.4634.8239.2364.8342.0990.9190.392
JAC1-20.4607.00421.04211.1244.7712.0850.885
IND1-20.4597.55319.18010.3104.4391.9380.820
PHI2-10.45618.89128.22715.5286.4722.7001.157
OAK1-20.45311.15219.87610.5964.4871.9040.792
KC1-20.45213.52820.95211.2854.7722.0230.841
CAR1-20.4476.19612.5596.2632.5811.0690.450
TEN1-20.4386.20913.5236.8552.8141.1730.472

First to 10% Chance of Superbowl win: Houston (Week 2, Sunday)

First to 80% Chance of Playoffs: Houston (Week 2, Sunday)


[Week 1]
[Week 2, Thursday]
[Week 2, Sunday]
[Week 2, Final]
[Week 3, Thursday]

No comments:

Post a Comment