Writers of Pro Football Prospectus 2008

Most Recent FO Features

ButlerMal14.jpg

» Audibles at the Line: Week 7

Three shutouts in one day, and the worst defense in the league nearly makes it four. Tennessee survives overtime against Cleveland. Chicago completes four passes and wins. Not much was clear on this very weird day in a very weird season.

26 Sep 2017

Week 3 DVOA Ratings

by Aaron Schatz

The Kansas City Chiefs have taken the league by storm through three weeks, and now sit atop the Football Outsiders DVOA ratings. The Chiefs are No. 1 on offense and also rank among the top dozen teams for both defense and special teams.

The Chiefs on top is not a surprise, but the placement of the other 3-0 team probably is. The Atlanta Falcons are only ninth in DVOA through three games, although their 3-0 start gives them the best Super Bowl odds in the NFC. The problem is that while they have three wins, the DVOA system didn't find any of them to be particularly dominant. Atlanta's highest rating so far is for Week 1's win over Chicago, scoring 38.0% DVOA even though the Bears did end that game within five yards of a game-winning touchdown. However, the Falcons come out at 21.8% for the win over Green Bay and then 3.8% for the win over Detroit that was decided by a single yard. By comparison, Kansas City's lowest-rated game is 37.9% in Week 2, basically the same rating as Atlanta's highest game. The Chiefs' other games scored 68.3% (over the Patriots) and 43.9% (over the Chargers). That's a big difference.

Another surprising team in DVOA so far is Pittsburgh, which ranks second despite Sunday's overtime loss to Chicago. Pittsburgh climbs a spot from last week, although the Steelers' overall DVOA actually goes down. However, it doesn't go down as much as you probably expected, because the Steelers still come out of their Week 3 loss with a higher rating than Chicago had. The reason for this is pretty simple: the two teams were virtually identical in efficiency (Steelers 4.8 yards per play, Bears at 4.9) but the Steelers had two fumbles (both recovered by Chicago) and no Ben Roethlisberger interceptions while the Bears had five fumbles (only one recovered by Pittsburgh) and one Mike Glennon interception. We can add onto that the fact that Chicago's field goal block to end the first half was the kind of play that doesn't show up in DVOA because it's essentially non-predictive. The Bears advanced the ball all the way to the 1-yard line, but it's very unlikely that they will advance the ball to the 1-yard line in that way again this year. And very unlikely that Pittsburgh will allow another team to do so, which is why that play doesn't hurt the Steelers in DVOA any more than any other missed field goal from the same distance.

Pittsburgh had our No. 2 preseason projection behind New England, with a sizeable gap between those two teams and the rest of the league. Combine that with Pittsburgh ranking No. 2 in DVOA after three weeks, and the Steelers have a healthy lead over the rest of the league in DAVE, our rating which combines preseason projection with performance so far. That lead also makes Pittsburgh our current Super Bowl favorites even though the Steelers are a game behind the Chiefs in the AFC standings.

Atlanta, Kansas City, and Pittsburgh are three of the four teams that have a positive DVOA rating in all three games this season. The fourth is a bit of a surprise: Tennessee. Wait, didn't the Titans lose to Oakland in Week 1? Yes, but the Titans now have a higher DVOA than the Raiders for their Week 1 loss because the NFL's play-by-play changes added two forced fumbles in the third quarter, both recovered by Oakland. The Titans are No. 5 overall in DVOA.

The Steelers may have the same record as that other team with the great preseason projection, the Patriots, but they've been far superior so far according to DVOA. New England's single-game DVOA ratings for both their Week 2 and Week 3 wins are close to zero; when you combine that with -48.8% DVOA for Week 1, the Patriots are still stuck down at No. 23 in our ratings. That preseason forecast means that the Patriots are still near the top of the league in DAVE and thus also our Super Bowl odds, but the Patriots need to address their defense soon if they want to be the dominant team we all thought we were getting this season. The Patriots offense is No. 5 but the defense is dead last... by a lot. The Patriots defense is currently at 35.4% DVOA. (Remember, positive defense is actually worse.) The No. 31 Saints are at 20.1% on defense. The gap between the Patriots and the rest of the league is larger than the gap between the Saints and No. 21 Cleveland. Playing the Saints and Chiefs offenses means the Patriots will look better once the opponent adjustments kick in, but the Texans offense isn't exactly a juggernaut most of the time.

Returning our attention to the top of the league: In between the Chiefs/Steelers and the Titans are the top two NFC teams in DVOA so far. The Detroit Lions, at No. 4, don't take much of a hit from that close loss to Atlanta. And the shocking Los Angeles Rams are No. 3 in DVOA after that exciting (?!), high-scoring (?!?!) Thursday night win over San Francisco that was more dominant than the final score of 41-39 would suggest. The Rams gained 7.0 yards per play to 5.7 for San Francisco, and ended up with 36.9% DVOA, moving them up from tenth to third. The Rams are so far No. 8 in the league on both offense and defense. Even stranger is how that offensive rating breaks down: No. 1 passing the ball and No. 26 running the ball. The technical term for this, and I say this as the man who has been driving the Los Angeles Rams bandwagon since April, is "Cuckoo for Cocoa Puffs." Because our preseason forecast had the Rams as an average team instead of the horrible team everyone else thought they would be, the Rams' hot start has made them No. 4 in DAVE and now has them as one of our Super Bowl favorites.

Of course, the response on this is going to be "Well, you don't have the opponent adjustments in yet." And I agree, having the Rams as one of the top five Super Bowl favorites right now seems completely nuts. So what would happen if we put opponent adjustments in? As a little experiment, I went and did that. I ran two additional versions of DVOA for this week. One had opponent adjustments at 30 percent strength, which is theoretically what we might use on the same scale that has us introduce them next week at 40 percent strength. The other version had opponent adjustments at 100 percent strength, but only based on these three weeks of the season.

As you might expect, using a stronger adjustment will move the Rams down: they drop to No. 6 with the mild adjustments and No. 8 with the strong adjustments. But it doesn't necessarily change other teams the way you would expect. For example, despite playing Green Bay, Atlanta actually drops in DVOA once we put in early-season schedule adjustments. New England does move up a bit, but still doesn't climb over zero. No, instead the big climbers if we included early-season opponent adjustments would be Washington and Philadelphia. Washington goes from No. 6 to No. 4 with mild adjustments and No. 2 (behind only Kansas City) with strong adjustments. Philadelphia goes from No. 14 to No. 11 all the way to No. 5. You can see here where we run into problems with doing adjustments at full strength in the early season: essentially, Washington and Philadelphia boost each other upwards significantly because they play each other as well as each one playing another team that gets a big boost from these early-season adjustments (Oakland goes from 18 to 17 to 11, while the New York Giants go from 29 to 27 to 25.)

Note, by the way, that the 100 percent adjustment is not automatically just 3.3x whatever the 30 percent adjustment is, because of the way we use multiple runs of adjusting defenses based on offenses based on defenses, etc.

Team W-L VOA (No
Adjustments)
Rank Adjustments
30% Strength
Rank Adjustments
100% Strength
Rank
KC 3-0 53.4% 1 51.8% 1 58.5% 1
PIT 2-1 39.0% 2 36.7% 2 37.7% 3
LARM 2-1 30.9% 3 26.4% 6 16.9% 8
DET 2-1 30.8% 4 27.0% 5 19.0% 7
TEN 2-1 29.9% 5 30.9% 3 36.5% 4
WAS 2-1 27.6% 6 30.2% 4 39.8% 2
BUF 2-1 24.7% 7 21.8% 8 13.9% 12
JAC 2-1 21.5% 8 23.3% 7 25.6% 6
ATL 3-0 18.6% 9 17.1% 9 12.1% 13
BAL 2-1 11.5% 10 9.0% 14 8.4% 15
NO 1-2 11.0% 11 10.8% 12 15.7% 10
MIN 2-1 8.9% 12 13.5% 10 16.2% 9
DEN 2-1 8.2% 13 9.5% 13 10.0% 14
PHI 2-1 6.9% 14 11.7% 11 28.5% 5
DAL 2-1 5.9% 15 3.1% 15 -0.2% 17
GB 2-1 2.8% 16 2.2% 16 -2.2% 19
Team W-L VOA (No
Adjustments)
Rank Adjustments
30% Strength
Rank Adjustments
100% Strength
Rank
CAR 2-1 2.3% 17 1.1% 18 -0.3% 18
OAK 2-1 -1.8% 18 1.8% 17 13.9% 11
SEA 1-2 -3.3% 19 -5.3% 19 -9.2% 23
TB 1-1 -4.7% 20 -8.4% 22 -8.5% 22
LACH 0-3 -7.8% 21 -8.2% 20 -7.3% 21
HOU 1-2 -10.5% 22 -8.4% 21 1.3% 16
NE 2-1 -12.6% 23 -8.7% 23 -2.3% 20
CIN 0-3 -24.9% 24 -23.3% 24 -23.8% 24
ARI 1-2 -25.8% 25 -27.4% 25 -39.8% 30
NYJ 1-2 -27.6% 26 -28.8% 26 -33.9% 26
MIA 1-1 -32.3% 27 -35.3% 30 -38.6% 28
CHI 1-2 -33.2% 28 -31.6% 28 -34.0% 27
NYG 0-3 -35.4% 29 -31.4% 27 -24.8% 25
SF 0-3 -36.8% 30 -35.2% 29 -39.2% 29
CLE 0-3 -38.2% 31 -37.2% 31 -42.7% 31
IND 1-2 -44.4% 32 -47.3% 32 -57.0% 32

Finally, some notes about the defensive ratings through three weeks. Last week, you might remember, Baltimore had set a new DVOA record for the best defensive rating through the first two weeks of the season. You may not be shocked to learn that the Ravens no longer appear on that all-time best defense list after giving up 44 points to the Jacksonville Jaguars. However, the Ravens are still No. 1 in defensive DVOA, even after that game! This remarkable rating reflects a) just how amazing the Ravens defense was in the first two weeks of the season and b) just how hideous the Baltimore offense was to hand the Jacksonville offense such outstanding field position all day. Baltimore's defensive DVOA for the Jacksonville loss was 35.3%, and their offensive DVOA was -67.9%.

Though Baltimore is still on top of our defensive ratings, Buffalo and Washington are right behind. Buffalo has given up an average of 12.3 points per game, while Washington goes from No. 12 to No. 3 after completely shutting down the Raiders. (That's discussed in this week's Any Given Sunday column.)

The defensive splits below the top teams are also fascinating. The Denver Broncos have had one of the best pass defenses in NFL history for the past two seasons, but their run defense fell to No. 21 in DVOA last year. That was supposed to be a weakness, but the strength of the pass defense was supposed to override it, right? Now, given the way the Broncos shut down the Cowboys, perhaps you would think that the Broncos are back near the top of the league in both pass and run defense. That would be wrong. The Broncos so far this year have the No. 1 run defense DVOA but are only No. 16 against the pass.

But that's not as weird as the split we are seeing from the Jacksonville Jaguars. Did you notice that despite what the Jaguars did to Joe Flacco, the Ravens averaged 5.4 yards per carry on Sunday? Yes, some of that was in garbage time, but the Titans had 5.0 yards per carry against them the week before. Through three games, the Jacksonville Jaguars are No. 1 in the NFL against the pass... and No. 32 against the run.

* * * * *

Once again this season, we have teamed up with EA Sports to bring Football Outsiders-branded player content to Madden 18. This year, our content for Madden Ultimate Team on consoles comes monthly, while our content for Madden Mobile comes weekly. Come back to each Tuesday's DVOA commentary article for a list of players who stood out during the previous weekend's games. Those players will get special Madden Mobile items branded as "Powerline, powered by Football Outsiders," beginning at 11am Eastern on Friday. We will also tweet out images of these players from the @fboutsiders Twitter account on most Fridays.

The Football Outsiders stars for Week 3 are:

  • DE Demarcus Lawrence, DAL: 3 sacks, 6 QB hits.
  • MLB Demario Davis, NYJ: 3 TFL, 2 tackles to prevent conversions on third-and-long. Thirteen combined tackles with an average gain of just 2.7 yards.
  • K Kai'mi Fairbairn, HOU: 4-of-4 on field goals, 6 touchbacks on 8 kickoffs. (Even on the kickoffs that were returned, Patriots started at the 19 and 25.)
  • LG Andy Levitre, ATL: Helped lead Falcons RB to 152 yards on 27 carries with 52 percent success rate. No sacks allowed.
  • LT Taylor Lewan, TEN: Helped lead Titans RB to 169 yards on 27 carries, including 108 yards on 10 carries to the left side. Titans also allowed no sacks or QB hits to Seattle defense

* * * * *

All stats pages should now be updated through Week 3, including snap counts and playoff odds. The FO Premium DVOA database is also updated through Week 3, and tonight or tomorrow the Matchup View will be changed to reflect 2017 early-season splits instead of 2016 full-season splits.

* * * * *

These are the Football Outsiders team efficiency ratings through three weeks of 2017, 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. As always, positive numbers represent more points so DEFENSE is better when it is NEGATIVE.

Please note that there are no opponent adjustments in DVOA until after Week 4. (It's still listed as DVOA instead of VOA because I don't feel like going through and changing all the tables manually.) In addition, our second weekly table which includes schedule strength, variation, and Estimated Wins will appear beginning after Week 4.

DAVE is a formula which combines our preseason projection with current DVOA to get a more accurate forecast of how a team will play the rest of the season. Right now, the preseason projection makes up 65 percent of DAVE for most teams (80 percent for Miami and Tampa Bay).

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
TOTAL
DAVE
RANK W-L OFFENSE
DVOA
OFF.
RANK
DEFENSE
DVOA
DEF.
RANK
S.T.
DVOA
S.T.
RANK
1 KC 53.4% 2 19.7% 2 3-0 43.2% 1 -7.0% 11 3.1% 8
2 PIT 39.0% 3 28.8% 1 2-1 20.4% 6 -19.8% 4 -1.2% 20
3 LARM 30.9% 10 12.3% 4 2-1 15.7% 8 -14.0% 8 1.2% 13
4 DET 30.8% 4 11.9% 5 2-1 3.1% 16 -17.0% 7 10.7% 1
5 TEN 29.9% 8 9.5% 10 2-1 37.2% 2 11.7% 23 4.5% 7
6 WAS 27.6% 15 10.1% 7 2-1 6.3% 12 -25.1% 3 -3.8% 26
7 BUF 24.7% 11 0.9% 17 2-1 -3.2% 20 -27.4% 2 0.6% 15
8 JAC 21.5% 20 0.8% 18 2-1 8.3% 11 -18.7% 5 -5.6% 29
9 ATL 18.6% 9 8.8% 11 3-0 18.7% 7 7.2% 20 7.2% 3
10 BAL 11.5% 1 6.4% 12 2-1 -21.1% 27 -32.4% 1 0.2% 17
11 NO 11.0% 24 -1.9% 20 1-2 30.0% 3 20.1% 31 1.1% 14
12 MIN 8.9% 19 1.2% 16 2-1 26.0% 4 16.9% 29 -0.2% 18
13 DEN 8.2% 12 -1.5% 19 2-1 3.1% 15 -10.7% 9 -5.5% 28
14 PHI 6.9% 16 4.8% 13 2-1 8.4% 10 3.2% 17 1.7% 10
15 DAL 5.9% 18 10.4% 6 2-1 -1.4% 19 2.9% 16 10.1% 2
16 GB 2.8% 14 9.9% 8 2-1 5.7% 14 4.4% 19 1.5% 11
TEAM TOTAL
DVOA
LAST
WEEK
TOTAL
DAVE
RANK W-L OFFENSE
DVOA
OFF.
RANK
DEFENSE
DVOA
DEF.
RANK
S.T.
DVOA
S.T.
RANK
17 CAR 2.3% 5 1.6% 15 2-1 -17.8% 25 -17.3% 6 2.8% 9
18 OAK -1.8% 7 4.3% 14 2-1 10.6% 9 18.1% 30 5.7% 6
19 SEA -3.3% 17 9.6% 9 1-2 2.4% 17 4.0% 18 -1.7% 22
20 TB -4.7% 6 -4.7% 22 1-1 1.9% 18 13.7% 25 7.1% 4
21 LACH -7.8% 13 -4.1% 21 0-3 6.2% 13 -2.2% 14 -16.2% 32
22 HOU -10.5% 23 -10.1% 25 1-2 -14.0% 23 -4.8% 13 -1.3% 21
23 NE -12.6% 22 12.8% 3 2-1 24.0% 5 35.4% 32 -1.1% 19
24 CIN -24.9% 27 -8.2% 24 0-3 -35.2% 31 -8.9% 10 1.5% 12
25 ARI -25.8% 21 -12.0% 26 1-2 -21.0% 26 -4.9% 12 -9.6% 30
26 NYJ -27.6% 30 -21.7% 30 1-2 -16.0% 24 12.1% 24 0.5% 16
27 MIA -32.3% 25 -12.4% 27 1-1 -13.6% 22 16.5% 28 -2.2% 24
28 CHI -33.2% 26 -15.7% 28 1-2 -21.2% 28 8.7% 22 -3.2% 25
29 NYG -35.4% 29 -8.2% 23 0-3 -11.4% 21 14.0% 27 -10.1% 31
30 SF -36.8% 28 -22.1% 31 0-3 -29.5% 30 14.0% 26 6.7% 5
31 CLE -38.2% 31 -25.2% 32 0-3 -25.6% 29 7.3% 21 -5.4% 27
32 IND -44.4% 32 -20.0% 29 1-2 -41.4% 32 1.0% 15 -2.0% 23

Posted by: Aaron Schatz on 26 Sep 2017

86 comments, Last at 03 Oct 2017, 5:11pm by junglejoe_lv

Comments

1
by theslothook :: Tue, 09/26/2017 - 7:34pm

If you are an aspiring sb contender - is it better to be 1st on offense and 32nd on defense or vice versa? What about 1st in Pass Defense 32 in run defense or vice versa? tough choices.

In all seriousness, I never expected the Patriots to be this hideous on defense, but through three games - they've been hideous. It hasn't even been a pass rush issue - instead its just terrible coverage across the board. Before the season, I wondered if this was their strongest secondary since the early 2000s when it was Law, Harrison, and Samuel. Man, what happened?

3
by Tundrapaddy :: Tue, 09/26/2017 - 7:45pm

The old adage that 'offense wins games, defense wins championships' still appears to be largely true, from what I've seen recently.

In their current form, I would not expect the Pats to get to the SB, much less win it. They'll inevitably meet a team with a defense that stymies the O-line and knocks Brady down (like the Ravens always used to).

5
by Cogitus :: Tue, 09/26/2017 - 7:52pm

As a Bills fan, I thought it was pretty hilarious Gilmore got all that money from you guys. He's never been elite, or even good for half his career.

I'm pretty mad you stole Hogan and Gillislee for pennies, though...

12
by Tundrapaddy :: Tue, 09/26/2017 - 8:27pm

I'll thank you to please not assume that I am a Patriots fan, thankyouverymuch!

Good day, sir! I say good day!

14
by blan :: Tue, 09/26/2017 - 8:36pm

I don't think Gilmore has been part of the defensive problems. He looks to me like he's playing pretty well.

22
by RickD :: Tue, 09/26/2017 - 10:07pm

My understanding is Gilmore has been the best player in the secondary. McCourty and Butler are not playing as well as last season and Rowe has injury issues.

The LB corps is the real problem, esp. with Hightower injured. Though the line isn't much better. In any case, Gilmore is the least of their problems.

34
by Cogitus :: Wed, 09/27/2017 - 8:09am

Huh, maybe he's just secondary cancer then

36
by roguerouge :: Wed, 09/27/2017 - 9:39am

I'd agree, with the caveat that defense has greater variation, so it's easier for a D to go on a hot streak and over-perform during the playoffs.

27
by Alternator :: Tue, 09/26/2017 - 11:42pm

Linebackers happened - they're playing terribly with Hightower hurt. Butler and McCourty are also playing worse than last year, though I would assume that's a temporary issue.

If the defense remains abysmal when Hightower is back and (reasonably) healthy, then it's time to panic. Until then, onward to a top-four seed and hopefully to a bye week.

52
by Anon Ymous :: Wed, 09/27/2017 - 11:32pm

Linebackers are only part of the problem, and likely not even the biggest contributor to the unusual amount of chunk plays allowed. You can chalk that up to confusion in the secondary, where zone coverage often ends up with two guys taking the same man and leaving someone else completely uncovered.

Oddly, it hasn't been just Gilmore (though he has had his issues). Since we know that everyone but Gilmore can perform in this defense, the miscommunication should diminish as the season goes on.

83
by Vegas :: Fri, 09/29/2017 - 10:54am

Scott actually wrote about this at length - I want to say it was 3 parts

http://www.footballoutsiders.com/stat-analysis/2016/building-super-bowl-...

2
by Tundrapaddy :: Tue, 09/26/2017 - 7:43pm

"Titans also allowed no sacks or QB hits to Seattle defense."

Well...they allowed no in-bounds and legal QB hits, anyhow.

4
by Cogitus :: Tue, 09/26/2017 - 7:50pm

I think the Colts won last week, didn't they?

11
by theslothook :: Tue, 09/26/2017 - 8:18pm

Yeah, but it was a squeaker over the 31st ranked team. According to D(VOA), we are close to twice as bad as Cleveland. Egads/

18
by D2K :: Tue, 09/26/2017 - 8:50pm

The Colts are suffering from getting absolutely smoked week 1 @the Rams 46-9, while Cleveland has lost by 3, 14 and 3.

29
by Aaron Schatz :: Wed, 09/27/2017 - 12:33am

Oops! My mistake. I'll fix that.

6
by DezBailey :: Tue, 09/26/2017 - 7:53pm

The Week 3 BES Rankings went out this morning - http://besreport.com/week-3-bes-rankings-2017/

DVOA and BES both agree on the Chiefs at No. 1. They also agree in general on the overall occupants of the top 10 except the BES swaps the Titans, Rams and Redskins for three other teams.

One key Week 4 game ...BUF@ATL should be very interesting as the BES has ATL No. 2 and BUF No. 4. Honestly this game is one of those that comes down to homefield which the Falcons have. That arguably gives them an edge. Things would tilt towards Buffalo were it vice versa. I think that's a testament to how impressive they've been....particularly in Week 3 vs Denver. That's certainly supported by DVOA having BUF (7th) ranked higher than ATL (9th). Should be a great game!

Lastly, the BES and DVOA agree on the Ravens still ranking No. 1 in defense even after the 44-7 loss to JAX. It took some steam off the Ravens BES DEF score but not enough for them to be supplanted atop the BES DEF rankings.

85
by DezBailey :: Tue, 10/03/2017 - 11:52am

Week 4 BES Rankings were just published - http://besreport.com/week-4-bes-rankings-2017/

7
by FootballTalker :: Tue, 09/26/2017 - 7:54pm

Seems to me the best way of incorporating opponent adjustments at the start of the season would be to factor in results from the previous season in ever-diminishing proportions each week (as the current season is taken more and more into account). Obviously that wouldn't incorporate personnel and coaching changes from one season to the next, but it seems a lot more likely to produce results which match our perceptions of which teams are good and which ones aren't. For example, the Rams may actually be a good team, and if so, the DVOA numbers will eventually reflect that...but if DVOA treats them like the 2016 Rams for now, that would probably give us a better sense of what the Colts, Redskins and 49ers have accomplished in playing them than simply using DAVE or no/limited opponent adjustments.

15
by Richie :: Tue, 09/26/2017 - 8:38pm

The problem is, we just don't know which bits of information are more valuable.

It's probable that the Rams really aren't one of the 10 best teams in the league. But maybe they are. Using the last few games of 2016 is just a guess.

Likewise, it's probable that the Patriots aren't one of the 10 worst teams in the league. But maybe they really are. If so, then using the last few games of 2016 is just feeding in bad info.

I think it's just a fools errand to try to have any kind of accurate opponent strengths before teams get a few games under their belts. Better just to provide the raw data along with the asterisk of "we don't know how much to weight these performances yet".

Last year, you could have used Carolina and the Jets good performances at the end of 2015 as indicators of their 2016 strength, but you would have been wrong to do so.

17
by D2K :: Tue, 09/26/2017 - 8:45pm

I definitely get the underlying point you're making but just to be clear, you can't use last years games in the data simply because of coaching and personnel changes.

33
by Cogitus :: Wed, 09/27/2017 - 6:11am

This is basically what ELO tries to do, right?

40
by Richie :: Wed, 09/27/2017 - 1:51pm

Yeah, Five Thirty Eight's elo ratings are a cumulative rating that uses information over the history of a franchise. But the only inputs are final score and game location.

41
by JoeyHarringtonsPiano :: Wed, 09/27/2017 - 1:56pm

I remember reading that something similar is used to rank chess players. That makes sense because individuals don't change much over time. For team sports, however, this method of rating can be problematic, since teams, being a collection of multiple individuals change a great deal from season to season.

42
by theslothook :: Wed, 09/27/2017 - 2:58pm

Yeah Elo is particularly illsuited for football for a variety of reasons. You can futz with some of the parameters, like variance and mean - but when the whole environment experiences changes over time, then elo tends to do a pretty poor job overall.

I've been playing around with all kinds of opponent adjustments for a while. The one's that make some sense to me are one's used in finance where part of the equation is the expectation of the future. Here I'm thinking Vegas odds.

If I had had more time, I would have been able to post the pressure article I'd been working on, but alas, work got in the way.

46
by Cogitus :: Wed, 09/27/2017 - 4:59pm

Yeah Elo seems like it would work better with a sport like basketball where there are a lot more games and a lot less players, especially players whose performance isn't a wildly unpredictable variable like football. The more I watch football and read about advanced statistics on this site and others is just how much insane variance there is in this sport, yet it's never part of the conventional conversation about how much luck matters in football.

Interesting. Please post that if you get a chance in some other thread!

47
by theslothook :: Wed, 09/27/2017 - 5:06pm

One problem with basketball is - the regular season IS fundamentally different from the postseason. For one - starter minutes are truncated, but there's also a ton of mixing of lineups for experimentation purposes. This is especially true for the best teams, who typically know where they are slotted and can thus play around with lineups. Someone showed a stat suggesting the most played lineups in the regular season by team are often varied in the playoffs. This, in effect, means the regular season sample is an imperfect capture of the playoff sample.

Fortunately, this doesn't seem to be the case in football where every game is super important.

48
by intel_chris :: Wed, 09/27/2017 - 5:52pm

Yes, ELO was specifically designed to match the rating system used by FIDE the international chess association. The numbers have been tuned a bit, but the idea is to follow the same convention, beat a better player and your score goes up a lot, lose to a worse player and your rating goes down (a lot), beat a worse player or lose to a better player, the score may only change by a little (or not at all).

The idea is that this helps compensate for not having lots of different stats to measure.

DVOA takes a different approach, find a way to turn one game into lots of stats, each play is at least one stat. You could take the DVOA approach to chess also, and count each move as a stat, but we don't have a good way of judging how good a move is (unless you consider the algorithms of chess playing computers a metric--BTW, I believe there are some chess programs that will estimate your rating by doing essentially that, looking at the individual moves you make and compare them to the algorithm and what level of player makes those kinds of moves).

While ELO may not be that good of a metric for a given season, particularly for teams that have had major changes (and which teams don't have at least some major changes?). It might do well for teams that are dynasties (or anti-dynasties). So, if ELO tells you that the Patriots are generally good year-in-and-year-out and that the Browns tend not to be, it does capture something.

Thus, Patriots fans who did not completely freak-out when Brady had "sausage fingers" or this year when the defense hasn't jelled yet, are just reflecting the same thing that ELO tries to quantify. It's the same thing that influenced the FO pre-season projections. One expects the Patriots to play more up to their ceiling than other teams do. That expectation eventually becomes an unconscious bias that influences how we see different teams.

Even fans who are disappointed in their teams have these biases. You see it when a fan derides their team for making the same mistake(s) that it has made in the past. I did the same thing in the SB when the Broncos got slaughtered by the Seahawks. I remembered how previous Bronco teams could get blown out in games that were too big and emotionally loaded and transferred that characteristic to the team on the field even though the players I was remembering had long retired. It wasn't Joe Montana leading the Seahawks, but to my brain it had the same feel.

As the investment banks say, "past performance doesn't guarantee future results", but it is hard to get ones brain to believe that.... ELO just turns that sentiment into numbers.

8
by alvastar :: Tue, 09/26/2017 - 7:55pm

Vikings defense having a rank of 29 seems crazy low... Does DVOA not take into account that soft yardage and meaningless scores late in the 4th quarter doesn't necessarily reflect the skill of the defense? That's all I can think of for this...

13
by Tundrapaddy :: Tue, 09/26/2017 - 8:31pm

I could very much be mistaken, but per recollection, (D)VOA tries to remove some of the 'soft yardage and meaningless scores', but probably does not capture all of it.

Which is probably why the Jaguars have the 11th-ranked offense, for example. Bortles is the kind of 4th-quarter touchdowns (when already trailing by more than 2 scores).

16
by alvastar :: Tue, 09/26/2017 - 8:39pm

That was my recollection as well (that (D)VOA accounted for it), which is why I thought it was puzzling.

And indeed, The Bort has long been formidable at collecting meaningless stats. I'll be curious to see where their season goes as people figure out that stopping their offense might be as simple as selling out to stop Fournette.

38
by iapetus :: Wed, 09/27/2017 - 10:59am

Except, of course, that teams are already selling out to stop the run against the Jaguars. More so than against any other team in the league. That hasn't stopped their offence in two out of three games...

20
by Will Allen :: Tue, 09/26/2017 - 9:40pm

They have tried to adjust for that, but have find it weakens the model. Consistently yielding easy scores while up several scores is predictive of doing so when it is more costly. It's a two game sample, so it isn't much to work on.

43
by alvastar :: Wed, 09/27/2017 - 4:26pm

ah, that makes sense why VOA and DVOA is down on the vikes. also, i find it very odd that giving up such "easy scores" is predictive of a defense doing so in a pivotal situation.

for example, in the saints game, the vikes only allowed 9 points and 201 total yards (and were up by 17) when the saints got the ball with 13 minutes left in the game. on their next two saints possessions, the vikes played a soft shell, therein basically giving up 150 yards (always when up by at least two scores!) and forcing the saints to use 8 minutes of clock on their two drives.

similarly, in the bucs game, the vikes had only allowed 10 points and 197 yards (and were up by 21) when the bucs got the ball with 2 minutes left in the 3rd. on the next 3 bucs drives, the defense again played a shell, giving up exactly 150 yards, always up by 14 points, and forcing the bucs to use about 6 minutes of clock (not as much clock, as two of these drives ended with INTs, which, once again, you would think would help the 29th ranked defense).

i get it if DVOA doesn't negate big gobs of yardage given up in one-score games or doesn't negate scores that are all done in a 70 yard, 90 second drive or the like, but it is curious to me that the first 45 minutes of a football game are so much less predictive than the last 15 minutes in which a defense purposefully (and intelligently, imo) allows an opposing team to complete a 5 or 6 minute drive that concludes with the defending team STILL being up two scores. when you are up by two scores late in a football game, the clock is your friend, and it seems odd that DVOA seems to only value selling out to avoid giving up a single score in a multi-score game.

tldr: i think it's noteworthy and weird that there is no way for DVOA to value "milking the clock" in games where one team is up by two scores throughout a 4th quarter.

44
by LyleNM :: Wed, 09/27/2017 - 4:40pm

tldr: i think it's noteworthy and weird that there is no way for DVOA to value "milking the clock" in games where one team is up by two scores throughout a 4th quarter.

Because, in the end, it's still better if your defense simply forces a stop and you can run out the clock on offense.

65
by alvastar :: Thu, 09/28/2017 - 3:04pm

is it better to GET the stop?

well of course, no one is arguing that.

is it better to SCHEME for forcing a three-and-out or the like, when you are up by three scores in the 4th quarter?

i would say no, and i think that the vast majority of people who have played football would agree. you need to keep the ball ahead of you. a lot of 4 yard gains is very unlikely to sink you at that point. a single 70 yard gain might. case in point; in both games the vikings win probability never dropped below 95% in the fourth quarter.

69
by Eddo :: Thu, 09/28/2017 - 3:36pm

I think it basically comes down to the fact that it's better to have a defense that is more likely to force three-and-outs than it is to allow conversions.

Even when you scheme to play less aggressively when up by multiple scores in the fourth quarter, the defenders on the field are still trying to stop the offense. Except in some very specific end-game scenarios (that come into play only in close games), no defense is actively conceding scores - they are still trying to stop the offense from getting first downs.

Defenses often do choose to decrease the likelihood of forcing turnovers or stuffing plays completely in exchange for lessening the likelihood of 10+ yard plays, but on any given play they are still attempting to minimize yardage.

And one thing Aaron has said over the years is that, in the "down big, during fourth quarter" bucket, the average raw value of a play, from the offense's perspective, is lower. (That is, if in a close game, a five-yard gain on first down is worth X value, in a not-close game, it's worth less than X.) So what could be true is that, compared to other teams who are up three scores in the fourth quarter the Vikings allow more yards per play.

45
by Will Allen :: Wed, 09/27/2017 - 4:44pm

Well, offensively, milking the clock, meaning continuing to get 1st downs, while up several scores, is viewed, by DVOA, as being very, very, valuable. Defensively, giving up big, easy, chunks of yardage, resulting in td drives, is nearly always bad for DVOA.

56
by Aaron Schatz :: Thu, 09/28/2017 - 11:29am

Hi! We're the 2016 Atlanta Falcons!

I can't believe I'm still getting questions about "garbage time" and DVOA after Super Bowl LI.

58
by nat :: Thu, 09/28/2017 - 12:19pm

Yup, that's the correct example to shoot down garbage time complaints. I totally get your exasperation.

Still, it would be worth looking at the DVOA excluding all (not just garbage time) fourth quarters every now and then, wouldn't it? While I get it that four quarters of data is better than three for official DVOA purposes, the fourth quarter is the one which will often feature teams playing for something other than the usual net expected value of the next score. That probably introduces some extra noise there.

There's little harm and much to be gained by using DVOA's excellent ability to do analysis of splits to look at each team's contrast between fourth quarter and non-fourth quarter play.

Does that sound like it could be an interesting application of DVOA to you?

59
by milo :: Thu, 09/28/2017 - 12:25pm

Aaron's probably too polite to respond here. For a low price subscription to premium stats, you can wallow in just those stats you suggest Aaron should be compiling.

61
by nat :: Thu, 09/28/2017 - 1:43pm

Jeez. Such venom. What gives? [edit] You allowed just six minutes before you stepped in to take offense on Aaron's behalf, where I doubt any offense was there to be taken. Double Jeez.

I was praising FO's stats and pointing out a great way they could be used to at least partially deal with a reader's question. No one's saying Aaron has to do anything.

Who kicked your puppy today?

68
by milo :: Thu, 09/28/2017 - 3:33pm

I apologize for offending you.
Let me try again.
You wrote: "There's little harm and much to be gained by using DVOA's excellent ability to do analysis of splits to look at each team's contrast between fourth quarter and non-fourth quarter play.
Does that sound like it could be an interesting application of DVOA to you?"

I was responding that what you thought would be good and interesting already exists:

Below are just 4 of the pages available in the Premium Database listed by title and with the column headings. Available for offense and defense. There are more pages available.

Ranking Teams By Quarter or Half
Team First Qtr Rank Second Qtr Rank Third Qtr Rank Fourth Qtr/OT Rank First Half Rank Second Half Rank Late & Close Rank Total Rank

Ranking Teams By Down and Distance
Team 1st Down Rank 2nd & Short Rank 2nd & Mid Rank 2nd & Long Rank All 2nd Rank 3rd & Short Rank 3rd & Mid Rank 3rd & Long Rank All 3rd Rank All Plays Rank

Ranking Teams By Downs and Play
Team 1st Pass Rk 1st Rush Rk 1st All Rk 2nd Pass Rk 2nd Rush Rk 2nd All Rk 3rd/
4th
Pass Rk 3rd/
4th
Rush Rk 3rd/
4th
All Rk All Downs
Pass Rk All Downs
Rush Rk All Downs
DVOA Rk

Ranking Teams By Score Gap
Team Losing Big Rank Tie/Losing Small Rank Winning Small Rank Winning Big Rank Late & Close Rank Total Rank

I would post screen shots but it's not allowed in these comments. Also, the tables presented in the Premium Database are much easier to copy and paste into a spreadsheet than the normal tables(with two rows of headers) if you want to play around with the data.

70
by Eddo :: Thu, 09/28/2017 - 3:38pm

To be honest, milo, I didn't read any venom into your words. You were a little blunt, maybe, but I think both you and nat have been pretty civil in this thread.

77
by nat :: Thu, 09/28/2017 - 4:35pm

Okay. I accept the apology. You weren't all that offensive. Just surprisingly snippy and quick to the draw. No worries. I can see you didn't mean to be.

I guess sometimes the right answer is "that's a good reason to buy premium stats". Even after all these years, I have no clear feeling of which questions or stats ideas Aaron will look into if asked, and which you should only see if you have the premium package.

I do see the conundrum Aaron faces. He's got these recurrent "garbage time" discussions that so obviously annoy him. He has data that would either put them to bed or would give a clear idea of how small the effect is. He has limited time and a need to keep premium stats premium.

Thanks for the info on the premium package. That's some good stuff.

60
by Will Allen :: Thu, 09/28/2017 - 12:42pm

I think if people had a sense of how DVOA would have scored differently for the Vikings, if those pretty easy 2nd half touchdowns they have yielded had instead resulted in 30 yard field goals, with a couple turnovers on downs tossed in, they'd have a better grasp of it.

Me? I'm just an Unfrozen Caveman Vikings fan. Your ways frighten and confuse me, so I just chant to myself, "Giving up touchdown BAD!!!"

71
by alvastar :: Thu, 09/28/2017 - 3:44pm

Hi Aaron!

So first of all, I find it amusing that you are using a single occurrence of a historically-unprecedented 25 point comeback as the only point of analysis to explain why you disagree. Arguing that it is an invalid tactic to play defense to avoid a quick score when you are up by multiple scores late because the Patriots came back from such a situation once is like arguing that it is an invalid tactic to avoid critiquing white nationalists in a presidential election because Trump won once. NFL teams had a record of 2,545-4-2 after being up by 25 points in a game through that point. Sometimes crazy things just happen! Using these crazy things as evidence of a greater point is a bit silly.

More importantly, you missed my central point. My argument was that when a defense REMAINS AHEAD by multiple scores at ALL times and give up the majority of yardage AND points AFTER THIS IS THE CASE AND WHILE IT REMAINS THE CASE, that these points and yards may not be predictive of the actual quality of the defense.

Moreover, the Patriots drives did not take 5 or 6 minutes like I clearly specified; they averaged 4 minutes once they got down 28-3. In fact, if they did take 5 or 6 minutes like I specified, the Falcons would have won, as the Pats would have run out of time. Even though, once again, I find using one of the most statistically unlikely games to ever occur as a case model for what is predictive to be a curious practice.

Does that make sense? Sorry if it came across as snarky, but honestly your comment did not exactly treat me with the greatest amount of respect!

72
by Will Allen :: Thu, 09/28/2017 - 4:13pm

To be clear, the Vikings gave up two third quarter tds to the Bucs. One was 46 yards, in about 3.5 minutes, and the other was 75 yards in about 2.5 minutes. That isn't really milking the clock. They were just bad against the Steelers.

73
by Will Allen :: Thu, 09/28/2017 - 4:13pm

To be clear, the Vikings gave up two third quarter tds to the Bucs. One was 46 yards, in about 3.5 minutes, and the other was 75 yards in about 2.5 minutes. That isn't really milking the clock. They were just bad against the Steelers.

75
by theslothook :: Thu, 09/28/2017 - 4:15pm

The point that you are missing is - the occurrences of garbage time points still is predictive. This might be due to sloppy modeling, but more likely it is not. And the question is always to what degree. Should we discount it by 100 percent? Ie - all yards given up in garbage time(poorly defined to begin with) or should we discount it by 10,15, what number?

The point is - while there are plenty of games where a defense gives up yards and wins, there are also enough games where yielding such yards costs you the game. I can name several off the top of my head. I think its best to acknowledge this fact rather than bemoan it as miss measurement.

Btw - when should we consider something garbage time because as far as I can tell, no one has come up with a proper rigorous definition.

9
by Al Hirt Hologram :: Tue, 09/26/2017 - 7:59pm

I don't have anything concrete to base this on, but this season feels like one where the home teams go 1-3 in the wildcard round.

10
by milo :: Tue, 09/26/2017 - 8:02pm

86 pass offense and 26 rush offense = 2 total offense? (NO)

21
by Eddo :: Tue, 09/26/2017 - 10:02pm

Huh? The Saints have +52.1% offensive passing DVOA, -1.9% offensive rushing DVOA, for an overall offensive DVOA of +30.0%.

35
by milo :: Wed, 09/27/2017 - 9:21am

Looking at numbers for week 3 game only.

19
by Will Allen :: Tue, 09/26/2017 - 9:32pm

Funny to see the Vikings slotted at 4 on offense, 29 on defense. They have played soft 2nd half, especially 4th quarter, defense, once they built big leads. I wouldn't be surprised to see their overall rank end somewhere around 10ish, but with the respective offensive and defensive ranks moving in opposite directions.

23
by intel_chris :: Tue, 09/26/2017 - 10:09pm

There was some discussion last week about using "conventional wisdom" to pre-load the database. IMHO, that's exactly what DAVE does--the conventional wisdom in this case being FO's pre-season predictions. That seams as good as any. I'm assuming it is a subjective measure, but it probably includes information about how good the team was last year, but adjusted for what the FO writers are significant adjustments such as new coaches and players. As a subjective measure, it somewhat goes against the grain of FO, but it does temper the swings one would get from small sample size of not doing anything, and fortunately, we have the raw VOA numbers to compare DAVE to. You might prefer some other subjective (or even objective) measure, but to my mind DAVE mostly achieves its goals and is arguably reasonably consistent with the FO philosopy.

It was interesting to see opponent adjustments figured in and what effect they have, despite hurting the team of whom I'm a fan ranking-wise. It reminded me of the experiments Aaron did years ago when figuring out how much to weight opponent adjustments and whether to iterate until they converged. At this point in the season, it is hard to know how good teams "really" are, simply not enough data. So, slowly adding the adjustments in makes sense.

The main point is that we have a variety of different estimations of each teams' strength: VOA, DAVE, opponent adjustments. Different perspectives give us more to consider. It is good to temper our own rabid fandom with a variety of reality checks. Expecting one number to be tell-all is wanting too much.

24
by stinkubus :: Tue, 09/26/2017 - 10:21pm

In this week's "Sign of the Apocalypse" the Rams have the second best SB odds in the NFC. They also have better odds than NE.

25
by ChrisLong :: Tue, 09/26/2017 - 10:34pm

It would be interesting to see which early season statistics are the best predictors of continued success. That is, which statistics at this point have the best correlation with their final DVOA values. Or, in another correlative approach, are the correlations of your (D)VOA now with final DVOA stronger if you are at the very top or very bottom after week 3? Perhaps this is best left until opponent adjustments are fully incorporated, but it would help us start to get a handle on which conclusions we can jump to early on in the season.

26
by young curmudgeon :: Tue, 09/26/2017 - 11:26pm

Pittsburgh is clearly ranked too high because they are almost physically painful to watch. My own fleeting impressions as a fan are way better than this. If the Steelers are truly the second best team in the league, I truly feel sorry for fans of the others, because I haven't seen much good football from the team I support, so those other fans must really be suffering. The Steelers don't look well-coached and, with the exception of Antonio Brown, they don't look particularly skilled or athletic. And the annual "don't bother to show up for a game against an inferior opponent" has gotten very old.

81
by The Ninjalectual :: Fri, 09/29/2017 - 12:26am

This is by far the best comment anyone has made about Pittsburgh all week, and perhaps much longer.

82
by young curmudgeon :: Fri, 09/29/2017 - 10:44am

Thank you. I'll be here all season. Don't forget to tip your waitress.

28
by stinkubus :: Tue, 09/26/2017 - 11:47pm

While I'm also a Steeler fan who is befuddled by their high ranking, there isn't a team in the league that doesn't have those type of games at least once a season.

30
by snw-o :: Wed, 09/27/2017 - 12:42am

"Titans also allowed no sacks or QB hits to Seattle defense" LOL

Sorry, read comment above

31
by snw-o :: Wed, 09/27/2017 - 12:45am

To comment one more time in a row...what IS the rule about late hits at sideline? If a player is clearly headed out of bounds but has not yet touched down, is he fair game? [I thought that MM's foot had not come down when he was hit.] How do you prevent QBs from running toward the sideline to cause defenders to slow down, then turning upfield?

51
by El Muneco :: Wed, 09/27/2017 - 9:01pm

The situation you describe happened at least once the past few years. Ball carrier angled towards the sideline, defender pulled up while ball carrier still had both feet in bounds. Ball carrier turned upfield and tiptoed for quite a few more yards.

Sherman's hit on Mariota was legitimately late - Mariota's whole body was over the white - but Mariota's second-to-last step had been in bounds. It wouuld have been tough for him to cut hard enough to keep the last step in bounds, and their gameplan emphasizes Mariota avoiding contact even more than Seattle protects Wilson. But I've seen guys do it in the past.

Basically, coming up in support at the exact time Sherman did, he had no good options. He needed to be a step sooner or a step later to be able to make a good play.

And given how fractious the game had been to that point, it was pretty obvious, right or wrong, which way he was going to choose.

53
by Dan :: Thu, 09/28/2017 - 12:09am

Looked to me like a good play by Sherman. Mariota was heading downfield until he got a yard away from Sherman, and the hit came before he set foot out of bounds. If the runner had been DeMarco Murray instead of Marcus Mariota then they play probably would've gotten a very different reaction.

54
by Dan :: Thu, 09/28/2017 - 12:30am

To add more detail:

The LB Bobby Wagner first gets a hand on Mariota when he is at the SEA 40 yard line; Sherman is 9 yards away at the 31.

Wagner keeps a hold of Mariota, but Mariota gets the corner and continues downfield as he angles towards the sideline. Mariota's eyes are on Sherman as he steps with his right foot at the 37.3 yard line. Sherman is 3 yards away crossing the 34.

Mariota's next step with his left foot goes forwards another yard, to the 36, and lands in bounds. Sherman at that point is at the 35 with his shoulder lowered on his way into the hit.

Only then does Mariota stop trying to advance the ball. He pulls up at the 36 and brings his right foot over to the right to land out of bounds. Sherman's right shoulder hits Mariota's left shoulder before Mariota's right foot has reached the ground.

So, it looks to me like a clean hit. If Mariota wanted to avoid the hit then he should've gone out of bounds at the 37 instead of trying to pick up the extra yard.

32
by justanothersteve :: Wed, 09/27/2017 - 12:59am

The Packers being completely average across the board comes across pretty accurately. Some good streaks, some bad streaks, mixed in with a few amazing Rodgers' theatrics. They'd be completely meh if not for the occasional 'holy crap did he just do that' play from Rodgers.

50
by DisplacedPackerFan :: Wed, 09/27/2017 - 7:12pm

Yeah that agrees with what I've seen watching the games. There is one thing that feels very odd though. When you look at the Team Offense page GB has the 5th most efficient running game in the league at 4.3% DVOA. I actually don't feel that value is too far off despite some of the awful runs, they have been better than in the recent past at short yardage running, and of course Rodgers has added some good value on scrambles. And yes the ordinal ranking is fairly meaningless since they are closer to 32nd than they are to 1st (KC has been really good at running the ball). So I get it but it still seemed wrong to see. But yeah being pretty much league average across the board, feels right.

80
by The Ninjalectual :: Thu, 09/28/2017 - 11:59pm

Occasional? Seems to me he has one to two of those a game on average

37
by drobviousso :: Wed, 09/27/2017 - 10:34am

The Steeler's placement doesn't surprise me too much, because after long years of reading FO, the concepts behind these stats is starting to penetrate even my dense skull. Everything non-predictive that can cause you to lose a game is happening to the Steelers. Everyone but AB looks like garbage on the offense. Well, ok, but this is measuring only predictive stuff, and it includes AB.

The blocked kick, fumble recover rates, etc - those matter. A lot. And they are the product of skill. But they our outside of what is being measured.

And AB? Well, he might not have found the end zone last week. But he did, and has been doing for years, the kind of thing that makes you confident he's going to be able to carry an offense again.

Another way to consider it. The Bears and Steelers went to overtime. That's a sure a sign as anything that the two teams had outcomes so close you can call it a draw for these purposes (but *NOT* for playoff seeding purposes). For the Bears to get there - every bounce had to go their way. For the Steelers to get there - every 'f up had to happen. Bears + good luck = Steelers + bad luck. Remove the luck, and the Steelers are a good team (even if my eyes didn't see it last week)

Tis tells you that, in the small-ball unexciting plays, the Steelers were doing a lot of thing really well. They don't jump out at you, they don't have a lot of emotional load. But its those 'junk plays' that provide the best signal for how a team is going to perform next week and the week after. Hell, I don't know how I can look at that defensive play and say they dominated, but ignore those two scores off super short fields (ie - non predictive scenarios), and the score looks mighty different.

If the Steeler's could get their head out of their rear end, they'd have smoked the Bears. Seems like the Steelers are one of the streakiest teams as far as this goes. Every year they have a funk. Every year they an injury to Ben and rush him back too early. And every year they pull their heads out of their rear, everything clicks, and they look like a juggernaut (outside of Foxborough). But these metrics don't measure any of that. They are blind to the Steelers amazing play-down-to-the-opponent skill. They are blind to Ben's propensity to miss time.

39
by Ambientdonkey :: Wed, 09/27/2017 - 11:19am

The Bears missed a field goal, blew 4 easy points on the blocked kick return, Jordan Howard didn't really fumble, and Tarik Cohen didn't step out of bounds. I'm pretty sure the Bears didn't need everything to go their way to achieve victory.

49
by intel_chris :: Wed, 09/27/2017 - 6:25pm

Bravo. I've thought and felt the same thing for years. And if you read enough about the genesis of DVOA, you can see how that is baked into the formula--intentionally. DVOA basically was designed to reward teams that consistently improve their relative field position (and prevent the other team from doing the same). This is certainly one thing which helps a team win. All other things being even, if every change of possession you end up a little closer to the scoring goal line and the opposing teams ends up farther, then eventually you will score and they won't.

That discounts "fluke" plays a bit. DVOA tries to incorporate them to some extent and long plays get more DVOA value than short plays, but not exactly in the same way that points on the scoreboard do. If I recall correctly, DVOA also tries to take into consideration position on the field and time on the clock, and uses some measurements that try to normalize those plays so that they can be compared to what an average team might do. Still, the metric is intentionally not the same as the points on the scoreboard nor the W/L column, because DVOA wants an independent measure of quality. Again, that is intentional, so that DVOA can say that the losing team actually played better football than the winning team, and outside of non-repeatable events, probably should have won the game (in the sense that if the teams played a dozen games on a neutral field, the losing team could have been expect to win more games if it had a better DVOA value).

Much of the problem has to do with quantization. The scoreboard is a quantitized metric, so is the W/L column, so is DVOA, and even the traditional stats like yards and sacks are quantitized. DVOA may have more data points (it has at least one for each play), but those data points are still discrete and not continuous.

More importantly the DVOA metrics intentionally are not influenced by whether the play scored or not or whether the team won or not (at least as far as I know). If you want to know which team one, or who scored more points, those stats are available. If you want to know, who played better football (at least in teams of winning the field position game), DVOA gets you much closer to that answer.

If you want to know both, you need to look at all the relevant stats and weigh their importance. That stat is likely to be subjective, as there isn't a well-established objective measurement proven to return that result, "which, as they say, is why we play the game".

57
by Aaron Schatz :: Thu, 09/28/2017 - 11:31am

Whether the play scored does matter. There's a bonus for getting a touchdown because that last yard is harder to gain than all the other yards.

78
by intel_chris :: Thu, 09/28/2017 - 6:10pm

Thanks for explaining that. I've tried to read all your articles on what goes into DVOA and you clearly have put a lot of thought into it. It will never match perfectly to the scoreboard or the W/L column, because you want it to be MORE than that. However, it is also clear that you haven't lost sight of the fact that the point of football is to win the game, so you do try to capture all that is relevant (and repeatable) towards winning the game.

And your point about the last yard being the hardest, certainly rings true. No Chicago fan is going to forget that for a while. Or ask a Seahawks fan from a couple of years ago (and that's only the last example I can recall where a play at the goal line at the end of the game decided the SB winner). I would be willing to be that the probability of scoring does not go up by very much (if at all) as teams get closer to the goal line within say the last 5 yards. Have you ever studied that? I know you have for longer distances, as I recall reading that here as part of the early rational of DVOA.

55
by nat :: Thu, 09/28/2017 - 10:53am

I just notice an interesting (and quite weird) tidbit about KC and NE.

KC leads NE in offensive (D)VOA by 43.2% to 24.0%. They've been that much more efficient per play than the Patriots, even though the Patriots offense has been quite good.

And yet, here are the "big three" drive stats of Yds/drive (getting yards), TOs/drive (protecting the ball), and DSR (moving the chains):

tm Yds/Dr TOs/Dr DSR
KC 34.21, 0.29, .725
NE 37.94, 0.28, .758

New England seems to have been marginally better than KC on those important drive stats, instead of KC having the sizeable lead we'd expect.

KC is also near the bottom of the league in plays per drive.

What's up with that? Any ideas? I know VOA is per play not per drive, so I don't think it's a matter of VOA being broken. But how did KC squander so much per play efficiency?

62
by greybeard :: Thu, 09/28/2017 - 1:58pm

KC has played 155 snaps versus 196 snaps for NE. They also have fewer fumbles. KC also has far larger number of penalties and DVOA may be underrating them as outliers while drive stats don't.
A consistent drive performance would have higher DVOA than a boom and bust drive mix even though they may have the same averages. I don't know if that is at play here though.

63
by nat :: Thu, 09/28/2017 - 2:37pm

Snap count: these are all rate stats. Probably not relevant.
Fumbles: Patriots 5. Chiefs 4. I think they've each lost one. That could account for some of it, but not much.
Penalties: That could be some of it. For example, the Patriots gained two first downs via 12-man penalties, which I think DVOA just ignores. So more yards/drive and better DSR in the drive stats for the Patriots without adding to their DVOA.

Thanks for the ideas. I bet there's more to it. But that's a good start.

64
by Travis :: Thu, 09/28/2017 - 2:56pm

One of Kansas City's 4 fumbles was a muffed punt that wouldn't count against the offense; another was a bad snap on what eventually was an incompletion (no idea how DVOA scores this). All 5 of New England's came on normal offensive plays.

Kansas City's also had 3 touchdowns on drives that started across midfield, which might account for some of the difference in drive length.

66
by nat :: Thu, 09/28/2017 - 3:20pm

OK, I missed that. Three fumbles.

Avoiding fumbles is definitely part of it. Or, to put it differently, in part by recovering four instead of three fumbles, the Patriots made their drive stats look better than their VOA.

I doubt the three short drives have much of an effect. Their average length for a TD drive is very close. Good thing to check, though.

I appreciate the help.

67
by Eddo :: Thu, 09/28/2017 - 3:29pm

Number of plays matters, in that DVOA is more of a per-play metric than a per-drive one. If two teams both average 80 yards per drive, but one does it in 20 plays each drive, while the other does it in 10, it's seems reasonable the one that averages 8 yards per play instead of 4 would come out with a higher DVOA.

74
by nat :: Thu, 09/28/2017 - 4:14pm

I understand that VOA is per play.

I'm trying to get at this: what is KC doing (or not doing) that causes their higher per play value to not result in better yardage per drive, more first downs per drive, more points per drive, etc.

Fooling around with the PFR play finder, I think I have it.

Third Down.

Despite excellent first and second down stats, KC is a bit below average in third down conversions and near the bottom in third down yards per play. NE, by comparison, is above average in conversions and near the top in yards per play.

There's a lot of leverage on third down.

It bodes well for KC. Getting uncharacteristic results on third down doesn't usually hold up over time. We should there expect their drive stats to improve over the season.

As for the Patriots, their results don't look so out of whack. Maybe their drive stats will drop over the season. But maybe not. They look about right.

76
by dmstorm22 :: Thu, 09/28/2017 - 4:19pm

Chiefs are gaining more than 10 y/p on 1st down I believe.

Man, if their 3rd down offense gets better this could be real scary.

79
by Richie :: Thu, 09/28/2017 - 8:22pm

"Man, if their 3rd down offense gets better this could be real scary."

That's exactly what I was thinking.

84
by Eddo :: Fri, 09/29/2017 - 11:54am

Nice work with the additional research. That makes a lot of sense.

86
by junglejoe_lv :: Tue, 10/03/2017 - 5:11pm

I love the Adjustments Table!!!
Please include for relevant weeks and every year!

Thanks.