Week 14 DVOA Ratings

December 14, 2010, 4:57 pm ET

175

Photo: USA Today Sports Images

by Aaron Schatz

Remember a few weeks ago, when teams were all bunched together and seemed like the NFL had more parity than it had seen in years? For the most part, that's still true, but on each end of the DVOA ratings you'll find two teams pulling away from the rest of the league.

The Patriots continued to rise after their 36-7 beatdown of the Chicago Bears, and they are even more impressive if we look at weighted DVOA rather than full-season DVOA. In the weighted DVOA formula, the largest percentage change in weight of a game is when that game goes from being 12 weeks ago to 13 weeks ago. The Patriots have only two losses this year, and one of them was 13 weeks ago. Their weighted DVOA is now at 42.5%, and only one team has ever finished the season with a higher weighted DVOA: the 2007 Patriots, of course, with 42.9% weighted DVOA. The difference? That team cooled down in the second half of the season. This current team is playing better than ever.

Right behind the Patriots are the Pittsburgh Steelers, with a big gap between those two teams and the rest of the league. Yes, the Steelers have weaknesses, especially on the offensive line, but they've played very well for the last few weeks, including 59.3% for Sunday's game against Cincinnati. The Patriots have 11 games with DVOA over 20%, the Steelers have 10, and no other team has more than eight.

Oh... and only one team has zero. That would be the worst team of 2010, the Carolina Panthers. Well, they are one of the worst teams of 2010. The Panthers and Arizona Cardinals have been trading the bottom spot back and forth for weeks, but there are now ten whole percentage points between those teams and the rest of the league.

Scroll down past our main tables for a look at some interesting time-based splits for teams in 2010.

* * * * *

These are the Football Outsiders team efficiency ratings through 14 weeks of 2010, measured by our proprietary Defense-adjusted Value Over Average (DVOA) system that breaks down every single play and compares a team's performance to the league average based on situation in order to determine value over average. (Explained further here.)

OFFENSE and DEFENSE DVOA are adjusted to consider all fumbles, kept or lost, as equal value. SPECIAL TEAMS DVOA is adjusted for type of stadium (warm, cold, dome, Denver) and week of season. WEIGHTED DVOA represents an attempt to figure out how a team is playing right now, as opposed to over the season as a whole, by making recent games more important than earlier games. As always, positive numbers represent more points so DEFENSE is better when it is NEGATIVE.

To save people some time, please use the following format for all complaints:

<team> is clearly ranked <too high/too low> because <reason unrelated to DVOA>. <subjective ranking system> is way better than this. <unrelated team-supporting or -denigrating comment, preferably with poor spelling and/or chat-acceptable spelling>

	TEAM	TOTAL DVOA	LAST WEEK	WEIGHTED DVOA	RANK	W-L	OFFENSE DVOA	OFF. RANK	DEFENSE DVOA	DEF. RANK	S.T. DVOA	S.T. RANK
1	NE	38.3%	1	42.5%	1	11-2	48.7%	1	12.1%	27	1.7%	11
2	PIT	35.2%	2	35.5%	2	10-3	12.9%	9	-19.1%	1	3.2%	8
3	PHI	25.1%	3	25.8%	5	9-4	24.2%	3	0.5%	14	1.4%	13
4	NYG	22.8%	5	26.9%	4	9-4	12.3%	10	-15.4%	2	-4.9%	31
5	BAL	22.0%	6	27.3%	3	9-4	10.0%	13	-6.5%	6	5.5%	2
6	SD	21.6%	8	25.2%	6	7-6	19.8%	4	-12.0%	3	-10.2%	32
7	GB	19.6%	4	17.5%	7	8-5	13.5%	8	-9.0%	4	-2.9%	28
8	ATL	17.8%	7	15.3%	9	11-2	16.9%	5	3.2%	16	4.1%	7
9	NO	15.1%	9	16.6%	8	10-3	14.7%	7	-2.8%	10	-2.4%	26
10	TEN	10.7%	14	8.8%	11	5-8	1.0%	17	-4.6%	8	5.1%	3
11	NYJ	10.4%	10	3.7%	14	9-4	-0.1%	18	-6.1%	7	4.4%	6
12	MIA	10.2%	11	9.9%	10	7-6	6.1%	15	-3.4%	9	0.6%	19
13	IND	5.0%	15	2.8%	15	7-6	15.2%	6	6.4%	19	-3.9%	30
14	KC	3.9%	13	-0.4%	16	8-5	11.8%	11	5.6%	18	-2.3%	25
15	HOU	2.9%	16	3.7%	13	5-8	26.9%	2	21.4%	30	-2.6%	27
16	CLE	1.9%	12	3.9%	12	5-8	-2.0%	21	-1.3%	11	2.6%	9
	TEAM	TOTAL DVOA	LAST WEEK	WEIGHTED DVOA	RANK	W-L	OFFENSE DVOA	OFF. RANK	DEFENSE DVOA	DEF. RANK	S.T. DVOA	S.T. RANK
17	TB	-0.8%	17	-0.4%	17	8-5	7.5%	14	7.2%	23	-1.2%	21
18	CHI	-4.6%	18	-3.1%	20	9-4	-16.5%	30	-6.8%	5	5.1%	4
19	JAC	-6.3%	19	-3.0%	19	8-5	11.4%	12	22.3%	31	4.6%	5
20	DET	-8.3%	25	-8.7%	22	3-10	-3.3%	23	6.5%	20	1.5%	12
21	SF	-9.1%	20	-6.5%	21	5-8	-6.8%	26	0.5%	13	-1.8%	22
22	MIN	-10.2%	21	-9.6%	24	5-8	-10.4%	27	0.4%	12	0.6%	18
23	OAK	-10.3%	23	-1.3%	18	6-7	-5.6%	25	3.9%	17	-0.8%	20
24	CIN	-11.4%	22	-16.2%	26	2-11	-1.1%	20	7.1%	22	-3.1%	29
25	BUF	-13.4%	28	-9.1%	23	3-10	-2.5%	22	11.6%	25	0.7%	15
26	DAL	-13.5%	26	-14.7%	25	4-9	-0.2%	19	14.2%	28	0.9%	14
27	WAS	-18.7%	29	-20.5%	29	5-8	-4.7%	24	12.1%	26	-1.8%	23
28	STL	-19.0%	27	-17.2%	27	6-7	-13.0%	28	6.6%	21	0.6%	17
29	DEN	-19.3%	24	-19.6%	28	3-10	5.3%	16	22.4%	32	-2.3%	24
30	SEA	-24.8%	30	-31.8%	30	6-7	-13.1%	29	19.1%	29	7.5%	1
31	ARI	-34.8%	32	-31.8%	31	3-10	-29.8%	31	7.4%	24	2.4%	10
32	CAR	-36.1%	31	-34.4%	32	1-12	-34.0%	32	2.8%	15	0.7%	16

NON-ADJUSTED TOTAL DVOA does not include the adjustments for opponent strength or the adjustments for weather and altitude in special teams, and only penalizes offenses for lost fumbles rather than all fumbles.
ESTIMATED WINS uses a statistic known as "Forest Index" that emphasizes consistency as well as DVOA in the most important specific situations: red zone defense, first quarter offense, and performance in the second half when the score is close. It then projects a number of wins adjusted to a league-average schedule and a league-average rate of recovering fumbles. Teams that have had their bye week are projected as if they had played one game per week.
PAST SCHEDULE lists average DVOA of opponents played this season, ranked from hardest schedule (#1, most positive) to easiest schedule (#32, most negative). It is not adjusted for which games are home or road.
FUTURE SCHEDULE lists average DVOA of opponents still left to play this season, ranked from hardest schedule (#1, most positive) to easiest schedule (#32, most negative). It is not adjusted for which games are home or road.
VARIANCE measures the statistical variance of the team's weekly DVOA performance. Teams are ranked from most consistent (#1, lowest variance) to least consistent (#32, highest variance).

	TEAM	TOTAL DVOA	W-L	NON-ADJ TOT VOA	ESTIM. WINS	RANK	PAST SCHED	RANK	FUTURE SCHED	RANK	VAR.	RANK
1	NE	38.3%	11-2	32.9%	10.6	1	5.3%	10	5.5%	11	18.6%	19
2	PIT	35.2%	10-3	31.8%	9.9	2	7.0%	4	-7.9%	26	12.9%	14
3	PHI	25.1%	9-4	28.6%	9.5	4	0.0%	19	-0.3%	17	10.4%	10
4	NYG	22.8%	9-4	27.5%	9.6	3	-7.1%	29	8.7%	6	24.0%	25
5	BAL	22.0%	9-4	14.6%	9.2	5	5.4%	9	1.9%	15	4.8%	1
6	SD	21.6%	7-6	24.8%	8.5	7	-4.6%	25	-13.3%	30	25.7%	28
7	GB	19.6%	8-5	21.7%	8.5	8	-2.5%	23	18.8%	2	15.6%	17
8	ATL	17.8%	11-2	17.8%	8.8	6	0.5%	18	-15.3%	31	5.5%	2
9	NO	15.1%	10-3	20.4%	8.2	10	-10.8%	31	13.0%	5	9.2%	7
10	TEN	10.7%	5-8	6.5%	7.3	12	3.7%	14	3.9%	14	25.9%	29
11	NYJ	10.4%	9-4	13.7%	7.5	11	6.2%	6	5.7%	9	14.8%	16
12	MIA	10.2%	7-6	7.7%	8.2	9	7.6%	3	5.5%	10	11.4%	11
13	IND	5.0%	7-6	3.8%	7.3	13	4.5%	11	-2.0%	20	8.5%	4
14	KC	3.9%	8-5	11.4%	6.9	14	-6.5%	27	-6.2%	25	25.1%	26
15	HOU	2.9%	5-8	-0.5%	6.5	17	6.0%	7	-5.0%	23	21.0%	22
16	CLE	1.9%	5-8	2.4%	6.6	16	6.5%	5	15.3%	3	18.5%	18
	TEAM	TOTAL DVOA	W-L	NON-ADJ TOT VOA	ESTIM. WINS	RANK	PAST SCHED	RANK	FUTURE SCHED	RANK	VAR.	RANK
17	TB	-0.8%	8-5	2.1%	6.9	15	-4.3%	24	-6.0%	24	11.6%	12
18	CHI	-4.6%	9-4	-6.2%	5.9	19	-1.3%	21	6.6%	7	26.0%	30
19	JAC	-6.3%	8-5	-9.2%	5.8	20	3.7%	15	-3.6%	22	20.3%	21
20	DET	-8.3%	3-10	-7.6%	5.5	22	4.0%	13	-0.3%	18	7.8%	3
21	SF	-9.1%	5-8	-4.6%	6.0	18	-6.8%	28	-10.7%	29	25.4%	27
22	MIN	-10.2%	5-8	-12.9%	5.8	21	3.3%	16	4.1%	13	9.2%	8
23	OAK	-10.3%	6-7	-5.1%	5.0	25	-0.6%	20	-3.5%	21	37.7%	32
24	CIN	-11.4%	2-11	-18.1%	4.8	28	10.8%	1	15.2%	4	9.1%	6
25	BUF	-13.4%	3-10	-17.1%	5.1	24	7.8%	2	19.6%	1	14.0%	15
26	DAL	-13.5%	4-9	-15.9%	5.2	23	5.9%	8	-9.4%	27	23.8%	24
27	WAS	-18.7%	5-8	-18.0%	4.7	29	4.2%	12	1.0%	16	8.6%	5
28	STL	-19.0%	6-7	-11.6%	4.9	27	-11.0%	32	-10.0%	28	12.4%	13
29	DEN	-19.3%	3-10	-18.3%	4.6	30	-2.0%	22	4.7%	12	22.2%	23
30	SEA	-24.8%	6-7	-21.5%	4.9	26	-8.7%	30	-0.7%	19	26.1%	31
31	ARI	-34.8%	3-10	-27.0%	3.3	31	-6.1%	26	-19.6%	32	19.3%	20
32	CAR	-36.1%	1-12	-37.8%	2.9	32	1.9%	17	6.1%	8	10.2%	9

Best and Worst DVOA Ever Watch

BEST OFFENSIVE DVOA AFTER WEEK 14				WORST OFFENSIVE DVOA AFTER WEEK 14				WORST DEFENSIVE DVOA AFTER WEEK 14				WORST SPECIAL TEAMS AFTER WEEK 14
2007	NE	50.6%	x	2005	SF	-48.8%	x	2008	DET	25.6%	x	2010	SD	-10.2%
2010	NE	48.7%	x	1997	NO	-43.8%	x	2008	STL	25.3%	x	2000	BUF	-10.1%
2002	KC	41.8%	x	2002	HOU	-43.4%	x	2004	MIN	22.5%	x	1995	PHI	-9.8%
2004	IND	39.0%	x	2007	SF	-36.6%	x	2010	DEN	22.4%	x	1997	SEA	-9.1%
1998	DEN	35.6%	x	1999	ARI	-36.4%	x	2010	JAC	22.3%	x	2008	MIN	-9.0%
1995	DAL	35.2%	x	2004	CHI	-35.8%	x	2004	SF	22.0%	x	1997	STL	-8.9%
2004	KC	33.8%	x	2006	OAK	-35.6%	x	2002	ARI	21.4%	x	1998	OAK	-8.8%
1993	SF	33.0%	x	2010	CAR	-34.0%	x	2002	CIN	21.4%	x	2002	CIN	-8.8%
2009	NO	31.1%	x	2000	CLE	-33.0%	x	2010	HOU	21.4%	x	1996	NYJ	-8.5%
2006	SD	30.9%	x	1993	TB	-32.9%	x	2004	STL	21.3%	x	2009	GB	-8.4%

The 2010 Patriots still haven't passed the 2007 Patriots for the best offense of the DVOA Era if we measure things as of the current week, but their 48.7% rating is enough to pass the 2007 Patriots if the current Pats can keep things near this level for three more games. Also, congratulations to the Houston Texans for getting out of the 2010 defensive basement. They passed the Broncos and the possibly playoff-bound Jaguars this week. I don't know if Indianapolis is healthy enough to beat Jacksonville this week in the de facto AFC South championship game, but I have a feeling that a Patriots-Jaguars or Chargers-Jaguars playoff game would be a bloodbath.

More on Schedule Strength

Last week, I wrote a big thing about alternate ways for us to measure schedule strength in the future. I appreciate all the suggestions in the comments. However, I do want to point out that we can't really create a context-dependent schedule strength measure that changes based on the quality of the team which has that schedule. That makes it a lot harder to explain schedule strength in articles we write off-site, and it would be strange to use such a metric when we rank schedule strength before the season starts. If we do change the way we measure schedule strength, we're going to use the same format for all teams so we can compare teams on neutral ground.

Good Times, Bad Times

From Twitter user @azmanz: Can you run DVOA for Dallas since Wade Phillips left? Would that make Philly's win more impressive?

Yes, it probably does make Philadelphia's win more impressive, because the Cowboys have played better since they changed head coaches. It's interesting to note where they have played better. Here are the Cowboys' splits between Weeks 1-9 and Weeks 10-14.

	Pass Off	Rk	Run Off	Rk	Pass Def	Rk	Run Def	Rk	All Off	Rk	All Def	Rk	Spec Tms	Rk	Total	Rk
Wade Phillips Era (Weeks 1-9)	5.6%	22	-4.4%	25	44.5%	32	-7.9%	11	-4.1%	23	18.7%	28	-0.7%	19	-23.6%	29
Jason Garrett Era (Weeks 10-14)	31.8%	7	-7.2%	25	9.7%	19	4.0%	23	5.8%	16	7.6%	25	3.6%	8	1.7%	17

For these tables -- and all the tables below here -- the Rk is based on where that rating would Rk this week, not where that team would have Rked during the time period in question.

The run defense has actually declined with Garrett as head coach, but the pass offense and pass defense have both improved substantially. I'm not sure of the reasons for the pass defense to improve -- the pass rush seems to be better, and perhaps the cornerbacks are healthier now than they were early in the season. You might say it makes sense for the pass offense to improve when your offensive coordinator becomes head coach, but then again, he was calling the plays during the first half of the season too. Even stranger, that massive improvement in the passing game has come with Jon Kitna under center instead of Tony Romo. Romo still has a higher DVOA than Kitna on the season, because Kitna's worst games came when Phillips was still the head coach.

Here are some more interesting splits based on player injuries and suspensions:

	Pass Off	Rk	Run Off	Rk	Pass Def	Rk	Run Def	Rk	All Off	Rk	All Def	Rk	Spec Tms	Rk	Total	Rk
PIT w/o Big Ben (Weeks 1-4)	-9.6%	28	10.3%	6	-24.3%	1	-37.5%	1	-4.9%	24	-29.5%	1	3.3%	8	27.9%	2
PIT w Big Ben (Weeks 6-14)	54.8%	2	-9.8%	28	-5.3%	5	-29.9%	1	19.8%	4	-13.8%	3	3.2%	8	36.8%	2

The Pittsburgh offense, as expected has gotten much better since Roethlisberger's return. But the Steelers haven't run roughshod over the league because that improvement was accompanied by a similar (although unconnected and smaller) decline in pass defense. Even more surprising is the drop in the running game's effectiveness. A better quarterback is supposed to move safeties back and create room for the running game, but that doesn't seem to be happening with the Steelers.

	Pass Def	Rk	Run Def	Rk	All Def	Rk
NYJ all games	8.6%	16	-25.0%	2	-6.1%	8
NYJ w/Revis active (no Weeks 3-4)	3.2%	10	-28.2%	2	-11.0%	4

	Pass Off	Rk	Run Off	Rk	All Off	Rk
NE before Moss trade	67.0%	1	28.7%	1	44.1%	1
NE since Moss trade	76.9%	1	27.4%	1	50.6%	1

	Pass Off	Rk	Run Off	Rk	All Off	Rk
CHI all games	-10.6%	29	-9.1%	27	-16.5%	30
CHI w/Cutler (Weeks 1-3, 6-13, and 1H of Week 4)	0.3%	26	-10.4%	27	-12.3%	29

Football Outisders+ Perks & Features image

Comments

175 comments, Last at 20 Dec 2010, 10:02am

#126 by TomC // Dec 15, 2010 - 11:33am

Other than the Bears (who are clearly ranked too low because Ditka cures cancer, etc.), the two teams that jump out at me are Tennessee and Miami. Every time I watch either one of those teams, I think "how have they won any games this season?", even if they're winning the game I'm watching. I guess one answer for Tennessee is "really high variance"---i.e., maybe I've missed all their good games. But Miami's variance is pretty low, so my eyeballs are way out of sync with DVOA on that one.

#146 by Thomas_beardown // Dec 15, 2010 - 4:25pm

The only Miami game I've seen was against the Bears, and I assume their passing attack is a lot better with QBs other than Thigpen.

#151 by Sergio // Dec 15, 2010 - 10:27pm

Not really.

Thigpen was OK against the Titans. The problem isn't the QB: it's the OC. Quite frankly, he makes me pine for Kippy Brown, who was a shudder-inducer by himself.

And, against the Bears, the problem was exacerbated by a non-existing OL.

Maybe I should rephrase: the problem *is* the QB, but there's not much difference between Henne and Thigpen, at least not in results. They're both average QBs.

-- Go Phins!

#131 by Jeff Fogle // Dec 15, 2010 - 12:19pm

"However, I do want to point out that we can't really create a context-dependent schedule strength measure that changes based on the quality of the team which has that schedule."

Yes, you can.

"That makes it a lot harder to explain schedule strength in articles we write off-site,"

Are you analysts or article writers? You're NOT going to pursue something that leads to a deeper understanding because it might be harder to explain in articles? You're purposely choosing to NOT pursue a deeper understanding for that reason?

"and it would be strange to use such a metric when we rank schedule strength before the season starts."

Well, it would be impossible to use such a metric before the season starts. You can use the standard "projection" you're using now for that. It's not like preseason strength of schedule estimates are sacred. It's a cross-your-fingers-and-hope-for-the-best projection anyway. They don't matter at all once actual games are being played and the actual schedule strength is being defined.

*Use your standard method for preseason projections.

*Develop more intense methods for "strength" (caliber of teams faced) and "difficulty" (additional nuances created by game placement within sequences of divisional games, road trips, bye weeks, etc..) that can be used as the season unfolds.

*Say "we rate so-and-so's schedule as the fifth strongest based on who they played, but only the 10th most difficult because they've played more home games so far" or something like that.

That's not enough to satisfy the demands of off-site articles? There's such a demand from less stat-minded fans for the "inside baseball" of Klingon projection formulas that you have to find a kindergarten Klingon for off-site article readers or you're not going to make changes?

People want insights. Pursue insights. Reality has a way of explaining itself once it's discovered...

#134 by ammek // Dec 15, 2010 - 12:47pm

Shouldn't there be an 'AJ Smith mistakes revisited, part two' option: NYJ-NO?

#147 by DeltaWhiskey // Dec 15, 2010 - 4:50pm

Greeting, my apologies for the hiatus. Mother died and so have been out of pocket. Anyhow, here are this weeks breakdown

TOT DVOA:
AVG: 0.68%
SD: 19.03%

ELITE: NONE
GOOD: NE PIT PHI NYG BAL SD
AVG: GB ATL NO TEN NYJ MIA IND KC HOU CLE TB CHI JAC DET SF MIN OAK CIN BUF DAL
BAD: WAS STL DEN SEA ARI CAR
HORRID: NONE

COMMENTS: Suprised the NE has not separated themselves yet. Also surprised by the absence of no "HORRID" teams...a lot of parity.

WEIGHTED DVOA:
AVG: 1.16%
SD: 1962%

ELITE: NE
GOOD: PIT BAL NYG PHI SD
AVG: GB NO ATL MIA TEN CLE HOU NYJ IND KC TB OAK JAC CHI SF DET BUF MIN DAL CIN STL
BAD: DEN WAS SEA ARI CAR
HORRID: NONE

OFF DVOA:
AVG: 3.60%
SD: 16.33%

ELITE: NE
GOOD: HOU PHI
AVG: SD ATL IND NO GB PIT NYG KC JAC BAL TB MIA DEN TEN NYJ DAL CIN CLE BUF DET WAS OAK SF MIN
BAD: STL SEA CHI
HORRID: ARI CAL

DEF DVOA:
AVG: 3.32%
SD: 10.54%

ELITE: NONE
GOOD: PIT NYG SD GB
AVG: CHI BAL NYJ TEN MIA NO CLE MIN SF PHI CAR ATL OAK KC IND DET STL CIN TB ARI BUF WAS NE
BAD: DAL SEA HOU JAC DEN
HORRID: NONE

S.T. DVOA:
AVG: 0.41%
SD: 3.66%

ELITE: SEA
GOOD: BAL TEN CHI JAC NYJ ATL
AVG: PIT CLE ARI NE DET PHI DAL BUF CAR STL MIN MIA OAK TB SF WAS DEN KC NO HOU GB CIN
BAD: IND NYG
HORRID: SD

For those who've hated the broad average range, it's only gotten worse. This suggests to me that "any given SUnday may be more true than ever.

#149 by Eddo // Dec 15, 2010 - 5:01pm

Sorry to hear about your mother, DeltaWhiskey. My condolences to you.

------

As for your clusters, have you thought about splitting "AVG" into "AVG+" and "AVG-"? That way, each group would only contain one standard deviation's worth of teams.

#155 by DeltaWhiskey // Dec 16, 2010 - 8:58am

Thanks.

I've been giving thought to how to split up the AVG group; however, I haven't come up w/ a way that makes me happy. The problem with the split your suggesting separates CLE from TB. Moreover it would cluster CLE with the likes of GB and ATL and cluster TB with the likes of CIN, BUF and DAL.

What I am considering is some way to split the AVG group into three clusters, perhaps AVG-, AVG, and AVG+. One option is to take the mean and SD of the AVG group and break them out again. With Week 14 data you get the following for TOT DVOA:

Mean: 0.48%
SD: 10.81%

Yielding
AVG+: GB ATL NO
AVG: TEN NYJ MIA IND KC HOU CLE TB CHI JAC DET SF MIN OAK
AVG-: CIN BUF DAL

I'm still left with a rahter large (13 team) AVG group, but I think breaking out on top at NO passes the eyeball test as does the break betweeen OAK and CIN.

#159 by nat // Dec 16, 2010 - 9:38am

My condolences to you and your family.

As for splitting the average group, have you considered using 1 SD bands that are centered on multiples of SD rather than bounded by multiples of SD? That lets you have an "Average" group centered on the true average (which you seem to want, judging from your CLE/TB comment) while having all bands equally wide (which followers of your analysis seem to want).

The AVG band would range from -0.5 SD to +0.5 SD. GOOD would go from +0.5 SD to +1.5 SD. ELITE would start at +1.5 SD. You could add ELITE+ and ELITE++ bands in the unlikely event a team managed that level of performance, or you could give the 0.5 to 1.5 band a name between AVG and GOOD, such as HIGH-AVG or AVG+.

#160 by DeltaWhiskey // Dec 16, 2010 - 9:48am

Thanks.

I've considered making the type of split you've described and looking at the data over the past few weeks this might work. What I've liked over the long haul is that while the AVG group has been cumbersomely large, the break from AVG to the others has made sense looking at the numbers and also subjective sense. You are also correct about my thoughts on centering at true average.

Perhaps a 1 SD band around AVG (i.e. +/- .5 SD) and then a 1.5 SD bracket keeping ELITE and HORRID at the 2 SD mark?

If I have a chance later today, I'll play with it more.

#163 by DeltaWhiskey // Dec 16, 2010 - 10:56am

Found a little time: Here's how it breaks out:

ELITE: > +1.5SD
GOOD: +0.5SD TO +1.5SD
AVG: -0.5SD TO +0.5SD
BAD: -0.5SD TO -1.5SD
HORRID: < -1.5SD

ELITE: NE PIT
GOOD: PHI NYG BAL SD GB ATL NO TEN NYJ MIA
AVG: IND KC HOU CLE TB CHI JAC DET
BAD: SF MIN OAK CIN BUF DAL WAS STL DEN SEA
HORRID: ARI CAR

COMMENTS:
1. I don't like that both the BAD and the GOOD groups are larger than the AVG group
2. The differences at the break from AVG to BAD is 0.80%. This is a hard break to defend. However the difference between JAC to DET is 2.00% which would be a little more defensible...perhaps. So if DET can muster a little more suckitude this could break well.
3. The differences at the breaks from GOOD to ELITE and BAD to HORRID are each at least 10.00%. I can live with this.
4. T-tests comparing the groupings to one another all reach significance. Not the most valid way to test for differences, should have done ANOVA, but lack the capabilities here. Nonetheless, at least some indication that these groups may be are different.

#165 by nat // Dec 16, 2010 - 11:05am

I don't like that both the BAD and the GOOD groups are larger than the AVG group

For me, it's not so much a like/don't like issue. I'm mildly surprised, though. It's not a large effect and it's only one sample, so I wouldn't worry about it.

I don't think it's a problem with the bands you've chosen. I think it's how the teams are actually distributed in strength. This year's "parity" is more about there being a shortage of really good and really bad teams, not about there being an unusually large number of dead-average ones.

#169 by chemical burn // Dec 16, 2010 - 5:15pm

Condolences. Love that you post this though and are developing it. I was actually just commenting to a friend that my favorite stuff on FO this year is actually just being posted in the comments (by a guy named deltawhiskey...)

#171 by DeltaWhiskey // Dec 17, 2010 - 12:33pm

Thanks for the kind thoughts and props. I need to do some testing to see who this shakes out.
Applying this model to the 2009 TOTDVOA:

AVG 0.66%
SD 22.51%

ELITE: NONE
GOOD: BAL GB NE PHI DAL NO MIN IND NYJ PIT SD
AVG: DEN ARI CAR HOU MIA NYG ATL CIN SF WAS TEN JAC BUF
BAD: CHI CLE TB KC SEA OAK
HORRID: STL DET

Comments:

1. Looking at it through this lense, at least we can now say that that BAL, while the "best" DVOA team, was not ELITE...that argument is over now.
2. T-tests again suggest each group is different.
3. OAK to STL and BUF to CHI difference >10.0%. Other break, not so good. Clustering this way has its merits, I'm wondering if looking for large break points near these arbitrary breaks more sense. For example, the GOOD group has a %5.00 difference at NO to MIN, which certainly suggests a greater break than the 1.70% break at SD to DEN. Although, now subjectivity creeps in.

#173 by Thomas_beardown // Dec 17, 2010 - 4:51pm

Yeah, I don't understand why you are looking for large breaks. Is there any reason NFL teams should have large gaps in quality and not be a smooth slope?

#174 by Arkaein // Dec 17, 2010 - 7:57pm

The answer I'd give is that if the purpose of this is putting NFL teams into descriptive categories then those categories work best when the teams within a category are more alike than teams between any two categories. If the best AVG team is much more like the worst GOOD team then the worst (or even average) AVG team, then the benefit of using simple classifications becomes a bit dubious.

When DeltaWhiskey first started doing his weekly categories based on standard deviation I suggested that clustering methods might produce better categories. The simplest clustering method would simply be to split the pool of teams at the largest DVOA gap between any two adjacent teams until a stopping criteria is met. This criteria could be either a certain number of clusters or a minimum DVOA split threshold.

Now, you're right in that there's no inherent reason that there should be large gaps in ordered DVOA, except for the fact that the population of NFL teams is quite small. This single-variable clustering technique wouldn't work, for example, if we wanted to categorize all people in the US by height, because the distribution would be too smooth.

I do think it is a valid method for the NFL though. The groupings in most cases would be have more internal similarities than using standard deviation. There is the problem that the labels, and maybe the number of labels used could change from week to week as the clusters change shapes, but this would reflect the fact that the clusters would be driven by the nature of the data.

#175 by DeltaWhiskey // Dec 20, 2010 - 10:02am

Exactly, and much better said than I could have.

Thanks.

#170 by Andrew Potter // Dec 16, 2010 - 6:13pm

Thanks massively for working this out and posting it. I've been in the group wanting to see this particular breakdown for a few weeks. Generally speaking, it passes my eyeball test - very few surprises and those that did surprise me are marginal.

Plus, you know, it's always enjoyable to see my favourite team listed as elite!

#172 by DeltaWhiskey // Dec 17, 2010 - 12:58pm

I agree, having your team in the ELITE category is always good for a little self-esteem boost. I knew if I messed with this enough I could get PIT into that group.

Glad you like, here or the OFF, DEF, and S.T. under this new format.

OFF DVOA:
AVG: 3.60%
SD: 16.33%

ELITE: NE
GOOD: HOU PHI SD ATL IND NO GB PIT NYG KC
AVG: JAC BAL TB MIA DEN TEN NYJ DAL CIN CLE BUF DET
BAD: WAS OAK SF MIN STL SEA CHI
HORRID: ARI CAL

DEF DVOA:
AVG: 3.32%
SD: 10.54%

ELITE: PIT NYG
GOOD: SD GB CHI BAL NYJ TEN MIA NO
AVG: CLE MIN SF PHI CAR ATL OAK KC IND DET STL CIN TB ARI
BAD: BUF WAS NE DAL SEA
HORRID: HOU JAC DEN

S.T. DVOA:
AVG: 0.41%
SD: 3.66%

ELITE: SEA
GOOD: BAL TEN CHI JAC NYJ ATL PIT CLE ARI
AVG: NE DET PHI DAL BUF CAR STL MIN MIA OAK TB
BAD: SF WAS DEN KC NO HOU GB CIN IND NYG
HORRID: SD

COMMENTS:
1. None of the BAD teams have any category above AVG.
2. RE: TOTDVOA - Each ~7.5% of DVOA equals one win (i.e. based on regression equation for correlation of WINS w/ DVOA, DVOA = 0 yields 8 wins, DVOA 7.5% yields 9 wins, therefore, 7.5% DVOA = one win:
a. Dif between ELITE and GOOD is ~10.0% or 1.5 wins
b. DVOA range of GOOD is ~15% or 2 wins
c. DVOA range of AVG is ~13% or almost 2 wins
d. DVOA range of BAD is ~15% or 2 wins
e. Dif between BAD and HORRID is ~10% or 1.5 wins
3. Applying this to 2009
a. Range of GOOD is ~17% or a little over 2 wins
b. Range of AVG is ~20% or almost 3 wins (this is nice since an average season is usually going to be between 7 and 9 wins)
c. Range of BAD is ~13% or almost 2 wins
d. Difference between BAD and HORRID is ~13% or almost 2 wins.

In my mind, looking at these numbers through this WIN prism provides some validation of the clusters.

#150 by Jerry // Dec 15, 2010 - 6:54pm

Condolences. I hope you were able to enjoy Mr. Polamalu the last couple weeks.

#156 by DeltaWhiskey // Dec 16, 2010 - 8:59am

Thanks. I didn't get to observe Polamalu in action, but certainly was pleased with his contributions. It's my mom's fault I'm a Steeler fan as she grew up near PIT, and while I was raised in TX, we were not allowed to be Cowboy fans.

#164 by Hurt Bones // Dec 16, 2010 - 11:03am

Condolences. How people become fans is always interesting. Years ago I was on the ferry from PEI to Nova Scotia. I was in a conversation with a trucker. We were talking baseball, and he said he grew up in Queens in the 50's. His friends were all split. Some were Giants fans or Dodger fans, and a few were Yankee fans. So he asked his father who he should root for. "Well, I met your mother in Philadelphia." So he became a Phillies fan. Ashburn was his favorite player.

#167 by DeltaWhiskey // Dec 16, 2010 - 11:18am

Thanks.

My son became a Packers fan when Favre led a comeback on T-giving against the Lions. My other son is an Eagles fan b/c he likes eagles. The only rule I had for them when choosing teams was that there would be no fair weather fans.

#166 by Joseph // Dec 16, 2010 - 11:06am

"and while I was raised in TX, we were not allowed to be Cowboy fans."

This is not a bad thing.

Regarding your mother, my condolences. I just found out my mom has a cancerous tumor, but it's small, and in her EAR, of all places. Going to check for others; hopefully not, because she has been pretty healthy, no smoking history, etc.

#168 by DeltaWhiskey // Dec 16, 2010 - 11:19am

Thank you for your thoughts, I hope all checks out well with your mom.

It may not have been a bad thing, but it made things a little tough...character building I guess. Had the Cowboys won 2 Super Bowls, who knows how much more character I would have.