Tuesday, September 18, 2007

Projected Final Standings - Week 2, 2007

Using the projected win probabilities from the logistic regression power rankings, I estimated which games each team would win on the rest of their respective schedules. And then I summed the projected wins with the actual wins so far to get their projected win totals. The teams are sorted by division.



















































Projected Final Standings
TeamProjected Win Total
AFC East
BUF2
MIA5
NE16
NYJ2
AFC North
BAL9
CIN6
CLE8
PIT14
AFC South
HOU15
IND14
JAX8
TEN6
AFC West
DEN14
KC7
OAK3
SD3
NFC East
DAL14
NYG3
PHI9
WAS11
NFC North
CHI5
DET10
GB8
MIN10
NFC South
ATL4
CAR9
NO0
TB12
NFC West
ARI11
STL4
SF3
SEA11

4 comments:

Anonymous said...

To make it more accurate you should add a stipulation in the prediction formula to try to remove the outcome of 16 and 0 wins. Since the possibility of either is so remote it needs to be better factored, looking over these predictions.

Derek said...

My policy is to not manually adjust the results. I'll take out a Hail Mary pass because it's statistical noise, but I'm not going to mess with the probability model itself because something is really unlikely. Human interpretation of the model is as important as the numbers. Of course I don't think the Pats are going undefeated. But based on their stats right now, they're considered more likely to win each of the remaining games than their respective opponents each week. Luck, injuries, special teams, etc. will play a role in the actual outcomes.

Brian Burke said...

My first models, roughly similar to these, were so crude they predicted the Chargers would win 17 games last year after week 2 last year!

They didn't make it to 17, but they did have the best record in the league.

Also, are you adjusting for strength of opponent in the season win model?

Derek said...

I'm waiting for another week before I start adjusting for opponents. After this weekend, though, I will post the results based on opponent-adjusted stats in addition to the unadjusted stats. Shouldn't be terribly time-consuming.