Sunday, December 9, 2007

Win Probabilities and Expected Wins 2007 - Week 14

The following table lists the probabilities that the home team would win each game given the rushing averages, passing averages (includes sack yards lost and sacks as pass attempts), sack rates, third down conversion rates, penalty yards, and interception and fumble rates (by number of completions and rushing plays) for both teams. A logistic regression model trained on 1996-2006 data computes the probabilities. The final score margin is Home team points - Away team points. The sum of these probabilities of victory form the expected win totals in the second table after the jump. The tables will be updated as more box scores come in.

















Home TeamAway TeamP(Home Team Won) (%)Final Score Margin
WASCHI90.42738
BUFMIA99.945521
DENKC99.975634
SFMIN1.9545-20
NEPIT76.899821
SEAARI99.006821
CINSTL98.47059
DETDAL81.9878-1
GBOAK99.865131
HOUTB65.800214
JAXCAR99.620131
PHINYG27.0608-3
TENSD47.2990-6
NYJCLE18.8732-6


Teams that hould Have Won but Didn't
Detroit Lions (vs. Cowboys), Win Prob = 81.9878% Dallas lost only one of their 3 fumbles. Other than that, Detroit was more efficient passing (6.4 ypp vs 6.0) but less efficient running (4.8 vs 5.4 ypc).




















































TeamActual WinsExpected WinsProjected Win Total
AFC East
BUF7.05.47116.7337
MIA0.04.15505.1138
NE13.010.808513.3027
NYJ3.04.43995.4645
AFC North
BAL4.04.79866.3981
CIN5.06.30747.7629
CLE8.06.24987.6921
PIT9.09.047611.1355
AFC South
HOU6.05.85367.2044
IND10.09.988313.3178
JAX9.08.402010.3409
TEN7.06.25667.7004
AFC West
DEN6.06.58288.1019
KC4.05.40296.6498
OAK4.04.39755.4123
SD8.06.75128.3092
NFC East
DAL12.010.366212.7584
NYG9.06.47947.9746
PHI5.06.85128.4323
WAS6.05.99267.3755
NFC North
CHI5.03.52874.3431
DET6.06.53608.0443
GB11.08.894210.9467
MIN7.07.21908.8850
NFC South
ATL3.05.65587.5410
CAR5.04.75255.8493
NO5.04.98796.6506
TB8.08.501710.4636
NFC West
ARI6.05.33246.5629
STL3.04.83225.9473
SF3.01.97682.4330
SEA9.09.180611.2993


2 comments:

Brian Burke said...

Derek-How do you calculate projected wins? Does it consider wins already under the belt? For example, you've got NE winning only 0.3 more games for the rest of the year. I'd guess that since you've got them at 10.8 current expected wins, you really expect them to win 2.5 more games for a total of 15.5.

Derek said...

Nope, not at all. Keep in mind, there's an average error of 1-2 games, so in reality, New England will win 14-15 games, but they will have outperformed their expectation somewhat.

Each game is assigned an expected win value for each team which is essentially the probability that team won the game given the box score. The expected win total projections you see are the sums of those probabilities project out to a 16-game season.

New England had 2 games that they "shouldn't" have won according to that probability: @IND (based on penalties) and @BAL (based on run, pass stats and other things). This essentially means the New England isn't perfect and is unlikely to go perfect. It's a way of limiting the effects of having one really bad game or one game where you keep running up the score *ahem*.

Most of the time, recovered fumbles are the difference between winning and losing (though teams are still penalized for it). For Chicago, it's been entirely Devin Hester. Other times, it's the timing of the turnovers. Interception rates have some dependence on situation, but it's still random when these incidents occur. If it's at your own 10 yard line, bad things will happen. If it's at their 30 yard line, then perhaps not.

Offseason plans to rename these stats more clearly are pending...