NFL Week 7 Statistical Power Rankings and Predicting Against the Spread
Readers who have followed my work closely know that I have a particular soft spot for statistical analysis when it comes to football. Obviously, this needs to be tempered with film work and analysis, but where metrics can demonstrate a larger point or test a claim, I’m partial to it.
That reasonability doesn’t extend to these power rankings, which makes them about as sensible as any other set of power rankings.
In this case, it involves the use of statistical analysis that has served me well in the past in terms of predicting games, as well as some other statistics I’ve found useful to talk about. All of these statistics are adjusted for opponent, which is to say they take into account what the opponent has done, what the opponent’s opponents have done and so on until the iterations don’t change the data.
It’s also a useful jumping off point for analysis, because we can take a look at what individual parts of the game any team is excelling or struggling at in order to figure out how to attack or fix them. I’ll be using some statistics derived from research I’ve done in the past and will compare them to other power rankings throughout the internet-o-sphere.
The first set of rankings to look at is simply point-differential per game. There’s not much to explain here, so it’s simple enough. The rankings:
Point Differential | |
---|---|
Rank | Team |
1 | San Diego Chargers |
2 | Baltimore Ravens |
3 | Denver Broncos |
4 | Philadelphia Eagles |
5 | Seattle Seahawks |
6 | Dallas Cowboys |
7 | Green Bay Packers |
8 | Detroit Lions |
9 | New England Patriots |
10 | Indianapolis Colts |
11 | Kansas City Chiefs |
12 | Cincinnati Bengals |
13 | Arizona Cardinals |
14 | Houston Texans |
15 | Cleveland Browns |
16 | Buffalo Bills |
17 | Miami Dolphins |
18 | Chicago Bears |
19 | San Francisco 49ers |
20 | New York Giants |
21 | Carolina Panthers |
22 | New Orleans Saints |
23 | Pittsburgh Steelers |
24 | St. Louis Rams |
25 | Washington Redskins |
26 | Minnesota Vikings |
27 | Tennessee Titans |
28 | Atlanta Falcons |
29 | New York Jets |
30 | Oakland Raiders |
31 | Tampa Bay Buccaneers |
32 | Jacksonville Jaguars |
Raw rank order doesn’t change much as a result of strength-of-schedule adjustments. San Diego’s penalty due to schedule doesn’t move them at all, as they’ve been dominant enough against their competition to earn top billing regardless. The biggest benefactor is Seattle, whose schedule moves them up from tenth in the raw point differential calculations to fifth. Other than that, Buffalo benefits by moving from 19th to 16th. The New York Giants lose the most ground because of the adjustments, from 17th to 20th, but not many other teams are affected.
The next set of rankings are related to point differential, but instead of looking at it per game, we look at it per second. Originally used to examine true run-pass ratios, “game scripts” allows us to look at who held a lead and for how long in individual games. It’s a good measure of dominance, and also does a very good job of reducing the impact of garbage time points, while still punishing teams who were forced to come back from behind—which is good if you’re trying to predict future success, as they aren’t sustainable (which the Bills should have known before handing Ryan Fitzpatrick to a $59 million contract after five consecutive close games, including several comebacks from behind).
Basically, if two teams win 35-0, the team that took care of business in the first quarter with four touchdowns is better than the team that racked up 14 garbage time points against backups (though in fairness, the difference isn’t much). Similarly, a team that wins 28-7 where they didn’t allow a fourth-quarter point isn’t much different (though still better) than a team that won 28-21 when all three touchdowns from the opposition came in the final minutes and the game was never in doubt. Once you find out the game script, you can adjust for opponent and create similar rankings.
Game Scripts | |
---|---|
Rank | Team |
1 | Indianapolis Colts |
2 | Denver Broncos |
3 | San Diego Chargers |
4 | Seattle Seahawks |
5 | San Francisco 49ers |
6 | Philadelphia Eagles |
7 | Detroit Lions |
8 | Baltimore Ravens |
9 | Arizona Cardinals |
10 | Cincinnati Bengals |
11 | Kansas City Chiefs |
12 | New England Patriots |
13 | Green Bay Packers |
14 | Houston Texans |
15 | Miami Dolphins |
16 | Buffalo Bills |
17 | Chicago Bears |
18 | Dallas Cowboys |
19 | Washington Redskins |
20 | New York Giants |
21 | Tennessee Titans |
22 | Atlanta Falcons |
23 | Cleveland Browns |
24 | New York Jets |
25 | Carolina Panthers |
26 | Pittsburgh Steelers |
27 | Oakland Raiders |
28 | Jacksonville Jaguars |
29 | St. Louis Rams |
30 | New Orleans Saints |
31 | Minnesota Vikings |
32 | Tampa Bay Buccaneers |
It looks like last year’s comeback kids, the Indianapolis Colts (who actually finished with a negative game script last year because of the comebacks they had to pull off, despite the 11-5 record) lead the way. This year so far, it doesn’t look like they have to come back as much as they did earlier.
The opponent adjustment for game script ended up mattering a lot more than for point differential, with Arizona (18th to 9th), Buffalo (23rd to 16th), San Francisco (10th to 5th), New York Giants (25th to 20th) and Tennessee (26th to 21st) all experience major bumps. At the other end of the spectrum were even larger penalties. Pittsburgh (14th to 26th), Atlanta (12th to 22nd, in part thanks to Baltimore’s thwapping of Tampa Bay right after Atlanta did the same) and New Orleans (21st to 30th) all experienced massive drops.
We can also look at Drive Success Rate, which is more useful when it comes to separating the contributions of the offense, defense and special teams than it is at outright ranking, but has been a favorite team statistic of mine for some time now. It measures only a team’s abilities to get or prevent first downs. It doesn’t care about “third down efficiency” so much as how likely a team is to get a new set of downs or a touchdown with every opportunity, and it does that by adding up the total number of first downs a team produces, adds the touchdowns, then divides by the total number of drives and first downs (as every first down produces a new opportunity to create a first down, adding to both the numerator and the denominator).
As an example, a team who drives down the field from their own 5 to their opponent’s 15 before making a field goal, and grabbing six first downs on the way, ends up with a DSR on that drive of 0.857. If they had scored a touchdown, they would have a DSR of 1.000. On the other hand, if that same team started at the 50 and only drove 35 yards before attempting a field goal, they would at best have a DSR of .750, perhaps even as low as .500.
What’s really nice about the statistics is that it does a very good job of quieting the effect of field position. That is to say an offense ranked highly in DSR can have a difficult time putting points on the board because their defense keeps putting them in a terrible spot (or vice versa). It automatically incorporates turnovers (each turnover ends an opportunity to create a first down, obviously) and punishes teams for taking field goals (or alternatively, rewards defenses for forcing them).
I’m not too crazy about the differential as it’s not predictive, but it still does a good job of figuring out if the team can move the chains, whether or not they are generally successful and whether or not luck is playing an outsized role in their performances so far (if a team’s record is better than their DSR, they may be “too lucky” to count on in the future. If it’s the other way around, their wins may regress upwards). The biggest major issue with it is that teams that tend to rely on explosive plays will get unfairly discounted by it because getting a first down on a ten-yard play is counted the same as a first down on a 75-yard play.
So, the power rankings by Drive Success Rate differential (offense minus defense), adjusted for opponent:
Drive Success Rate | |
---|---|
Rank | Team |
1 | Dallas Cowboys |
2 | San Diego Chargers |
3 | New Orleans Saints |
4 | San Francisco 49ers |
5 | Indianapolis Colts |
6 | Carolina Panthers |
7 | Baltimore Ravens |
8 | Kansas City Chiefs |
9 | New York Giants |
10 | Chicago Bears |
11 | Denver Broncos |
12 | Cleveland Browns |
13 | Green Bay Packers |
14 | Cincinnati Bengals |
15 | Tampa Bay Buccaneers |
16 | New England Patriots |
17 | Seattle Seahawks |
18 | Atlanta Falcons |
19 | Miami Dolphins |
20 | Tennessee Titans |
21 | Washington Redskins |
22 | Oakland Raiders |
23 | St. Louis Rams |
24 | Pittsburgh Steelers |
25 | Minnesota Vikings |
26 | Houston Texans |
27 | Philadelphia Eagles |
28 | Detroit Lions |
29 | Buffalo Bills |
30 | Arizona Cardinals |
31 | New York Jets |
32 | Jacksonville Jaguars |
There are some massive differences here between the points-oriented rankings and this one. The average difference in rankings is actually eight ranks, which is massive. Still, there’s something to be said about measuring things differently if only to look at what might be the causal factors in a team’s success (special teams? Luck?).
The biggest differences between DSR and point differential may surprise you, although given how often some of these teams rely on explosive plays, it is perhaps not entirely unpredictable. The biggest penalties go to the Eagles (6th to 27th), the Lions (7th to 28th), the Cardinals (13th to 30th), the Seahawks (5th to 17th) and the Texans (14th to 26th). The biggest boons are for the Saints (22nd to 3rd), the Buccaneers (31st to 15th), the Panthers (21st to 6th), the 49ers (19th to 4th) and the Giants (20th to 9th).
Truthfully, the “true value” of a team likely lies somewhere in the middle of point differential and drive success rate, with some heavy emphasis on the former over the latter.
Ignoring either and simply looking at play-by-play data, I’ve constructed my own efficiency metric that was built on eight variables found to be predictive in the 2013 season. That may not be a robust sample size and it was crudely developed, but I think it provides more “truth” than any of the metrics above, even if it isn’t as good as FO’s DVOA or AFA’s Efficiency scores.
Homebrew Efficiency | |
---|---|
Rank | Team |
1 | Denver Broncos |
2 | Green Bay Packers |
3 | Cincinnati Bengals |
4 | Dallas Cowboys |
5 | Indianapolis Colts |
6 | San Diego Chargers |
7 | Seattle Seahawks |
8 | Baltimore Ravens |
9 | New England Patriots |
10 | Chicago Bears |
11 | Detroit Lions |
12 | New York Giants |
13 | Kansas City Chiefs |
14 | Buffalo Bills |
15 | San Francisco 49ers |
16 | Miami Dolphins |
17 | Houston Texans |
18 | Philadelphia Eagles |
19 | Washington Redskins |
20 | Carolina Panthers |
21 | Atlanta Falcons |
22 | Tennessee Titans |
23 | New Orleans Saints |
24 | Pittsburgh Steelers |
25 | St. Louis Rams |
26 | Cleveland Browns |
27 | Arizona Cardinals |
28 | Jacksonville Jaguars |
29 | Oakland Raiders |
30 | New York Jets |
31 | Minnesota Vikings |
32 | Tampa Bay Buccaneers |
I’m hesitant of endorsing any metric that gives the Packers that much credit, but there you go. The variables, with different weights of course, are completion rate, net yards per attempt, passing touchdown rate, running touchdown rate, fumble rate, interception rate, yards per carry and run success rate.
As you can imagine, the bottom of the list will change over time because they’re all polluted by multiple quarterbacks for each team. As the season progresses, their performances should smooth out. That said, there are defensive ranks in there as well, so it’s not as if every rookie quarterbacked team is out of the woods. Compare these to other statistical rankings to get a feel for teams from a play-by-play perspective.
Holistic Efficiency Rankings | ||||
---|---|---|---|---|
Avg | Home | AFA | FO | Team |
1.0 | 1 | 1 | 1 | Denver Broncos |
3.3 | 3 | 2 | 5 | Cincinnati Bengals |
4.3 | 7 | 3 | 3 | Seattle Seahawks |
5.0 | 2 | 9 | 4 | Green Bay Packers |
6.7 | 4 | 6 | 10 | Dallas Cowboys |
7.0 | 8 | 11 | 2 | Baltimore Ravens |
8.0 | 6 | 10 | 8 | San Diego Chargers |
8.3 | 11 | 7 | 7 | Detroit Lions |
11.0 | 5 | 15 | 13 | Indianapolis Colts |
11.7 | 15 | 5 | 15 | San Francisco 49ers |
12.0 | 9 | 16 | 11 | New England Patriots |
12.3 | 16 | 4 | 17 | Miami Dolphins |
13.7 | 13 | 14 | 14 | Kansas City Chiefs |
14.3 | 26 | 8 | 9 | Cleveland Browns |
14.7 | 10 | 18 | 16 | Chicago Bears |
16.0 | 18 | 24 | 6 | Philadelphia Eagles |
18.0 | 19 | 12 | 23 | Washington Redskins |
18.3 | 12 | 22 | 21 | New York Giants |
18.3 | 14 | 23 | 18 | Buffalo Bills |
19.7 | 20 | 17 | 22 | Carolina Panthers |
20.3 | 23 | 13 | 25 | New Orleans Saints |
20.3 | 21 | 28 | 12 | Atlanta Falcons |
20.7 | 17 | 21 | 24 | Houston Texans |
21.3 | 24 | 20 | 20 | Pittsburgh Steelers |
22.3 | 22 | 19 | 26 | Tennessee Titans |
24.0 | 27 | 26 | 19 | Arizona Cardinals |
27.3 | 25 | 30 | 27 | St. Louis Rams |
27.7 | 30 | 25 | 28 | New York Jets |
29.0 | 29 | 29 | 29 | Oakland Raiders |
29.3 | 31 | 27 | 30 | Minnesota Vikings |
30.0 | 28 | 31 | 31 | Jacksonville Jaguars |
32.0 | 32 | 32 | 32 | Tampa Bay Buccaneers |
Be sure to check out Football Outsiders’ explanation for their rankings here (what’s up with Dallas?) and Advanced Football Analytics’ explanation here (they don’t explain Miami but do go into the Cleveland Browns).
I won’t go into much detail on my rankings (I suspect my model is overvaluing completions and undervaluing net yards, which is why Cleveland is ranked so low despite their decent success rate), but the differences are interesting enough to look into. The Seahawks are only ranked seventh in this model, in part because they’ve been allowing successful runs, even if they allow a low yards per carry.
Detroit loses ground in my model against the other two models because they are more likely to value net yards in the passing game than my model does on the offense, and the inconsistency from Stafford has a bigger inconsistency hit. My model overvalues Green Bay relative to the other ones for exactly the opposite reason, where Green Bay’s net yards are mediocre, but they’re consistency in gaining yards are pretty decent. My model favors the Colts and that feels like a red herring. Their defense against the pass is the worst in the NFL after adjusting for opponent, but their ability to prevent passing touchdowns so far has been unsustainably high—so they get a boost in my rankings that they probably shouldn’t.
All of these can be translated into a point differential, which can then be measured against Vegas. But first, let’s look at subjective power rankings from ESPN, NFL.com, Bleacher Report, CBS, NBC, the Associated Press, SB Nation and Yahoo (SI hasn’t released power rankings for this week as of this writing).
Subjective Media Rankings | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|
Team | ESPN | NFL.com | B/R | CBS | Yahoo | Fox | NBC | AP | SBN | Avg |
Denver Broncos | 1 | 1 | 4 | 1 | 1 | 2 | 3 | 1 | 1 | 1.67 |
San Diego Chargers | 2 | 2 | 2 | 2 | 2 | 1 | 1 | 2 | 3 | 1.89 |
Dallas Cowboys | 3 | 4 | 1 | 4 | 4 | 4 | 2 | 3 | 2 | 3.00 |
Philadelphia Eagles | 4 | 3 | 6 | 5 | 3 | 3 | 5 | 4 | 4 | 4.11 |
Seattle Seahawks | 5 | 5 | 3 | 6 | 5 | 5 | 6 | 5 | 7 | 5.22 |
Arizona Cardinals | 6 | 8 | 12 | 3 | 8 | 7 | 4 | 8 | 5 | 6.78 |
Indianapolis Colts | 8 | 7 | 5 | 9 | 7 | 6 | 8 | 7 | 10 | 7.44 |
San Francisco 49ers | 7 | 9 | 13 | 7 | 10 | 9 | 11 | 6 | 6 | 8.67 |
Green Bay Packers | 9 | 6 | 11 | 10 | 9 | 11 | 7 | 9 | 8 | 8.89 |
Cincinnati Bengals | 11 | 12 | 8 | 8 | 6 | 13 | 9 | 11 | 9 | 9.67 |
New England Patriots | 10 | 11 | 7 | 11 | 11 | 12 | 12 | 10 | 12 | 10.67 |
Baltimore Ravens | 12 | 10 | 9 | 12 | 12 | 10 | 10 | 12 | 11 | 10.89 |
Detroit Lions | 13 | 13 | 10 | 13 | 13 | 8 | 14 | 13 | 13 | 12.22 |
Carolina Panthers | 14 | 15 | 14 | 16 | 14 | 14 | 13 | 14 | 17 | 14.56 |
Cleveland Browns | 15 | 14 | 17 | 17 | 15 | 15 | 15 | 15 | 14 | 15.22 |
Chicago Bears | 16 | 17 | 16 | 18 | 16 | 18 | 18 | 16 | 15 | 16.67 |
Kansas City Chiefs | 18 | 16 | 18 | 15 | 17 | 16 | 20 | 17 | 18 | 17.22 |
New York Giants | 20 | 18 | 20 | 14 | 23 | 19 | 22 | 21 | 16 | 19.22 |
New Orleans Saints | 16 | 23 | 15 | 20 | 19 | 22 | 17 | 20 | 22 | 19.33 |
Houston Texans | 19 | 20 | 21 | 19 | 20 | 21 | 21 | 18 | 19 | 19.78 |
Buffalo Bills | 21 | 21 | 22 | 22 | 18 | 20 | 16 | 19 | 21 | 20.00 |
Pittsburgh Steelers | 22 | 19 | 24 | 21 | 21 | 24 | 19 | 22 | 20 | 21.33 |
Miami Dolphins | 23 | 22 | 23 | 24 | 22 | 17 | 23 | 23 | 23 | 22.22 |
Atlanta Falcons | 24 | 24 | 19 | 23 | 25 | 25 | 24 | 24 | 24 | 23.56 |
Minnesota Vikings | 27 | 25 | 26 | 25 | 24 | 23 | 27 | 26 | 25 | 25.33 |
Tennessee Titans | 25 | 28 | 29 | 26 | 28 | 29 | 26 | 25 | 26 | 26.89 |
New York Jets | 29 | 29 | 25 | 29 | 26 | 26 | 25 | 29 | 27 | 27.22 |
St. Louis Rams | 26 | 26 | 28 | 27 | 27 | 27 | 28 | 27 | 31 | 27.44 |
Washington Redskins | 28 | 27 | 27 | 30 | 29 | 28 | 30 | 28 | 30 | 28.56 |
Tampa Bay Buccaneers | 30 | 30 | 32 | 28 | 30 | 30 | 29 | 30 | 28 | 29.67 |
Oakland Raiders | 31 | 31 | 30 | 31 | 31 | 31 | 32 | 31 | 29 | 30.78 |
Jacksonville Jaguars | 32 | 32 | 31 | 32 | 32 | 32 | 31 | 32 | 32 | 31.78 |
Those rankings shouldn’t surprise you. Media rankings tend to do a few things: incorporate recency bias (often far too much), “win” bias (that is, instead of looking at quality of play, look at the record) and marginalize strength of schedule to a small degree. Of those above, the AP has the smallest difference from the group, while the Bleacher Report’s is the largest difference. That’s to be expected, there are a few rankings composed by individuals (B/R, CBS, NBC, Yahoo, NFL.com) and a few that are polls of reporters (AP, ESPN, SBN). For the most part, the individual rankings diverged far more than the group rankings.
The numbers people seem to disagree heavily when it comes to the Arizona Cardinals, San Diego Chargers, Philadelphia Eagles (except FO), Miami Dolphins (though that’s mostly AFA) and the Washington Redskins (who have an extremely underrated defense at the moment). There’s also significant disagreement on the Bengals.
Anyway, the numbers can be converted into spread outcome (NOTE: this is NOT how Advanced Football Analytics or Football Outsiders has argued is a good way to use their data, nor is it their official predictions against the spread—merely my recalculation of their probabilities into spread outcomes. Also, this isn’t gambling advice, etc.), so we can look at what Vegas thinks against the three efficiency systems.
Predictions Against the Spread | ||||
---|---|---|---|---|
Game | Vegas Prediction | AFA Prediction | FO Prediction | Home Prediction |
NYJ @ NWE | NWE by 10 | NWE by 6 | NWE by 7.7 | NWE by 9.1 |
ATL @ BAL | BAL by 7 | BAL by 9.4 | BAL by 6.2 | BAL by 7 |
TEN @ WAS | WAS by 6 | WAS by 3.5 | WAS by 4.7 | WAS by 3.6 |
SEA @ STL | SEA by 7 | SEA by 6.4 | SEA by 4.3 | SEA 2.4 |
CLE @ JAX | CLE by 6 | CLE by 4.9 | CLE by 5.2 | JAX by 2.3 |
CIN @ IND | IND by 3 | CIN by 2.4 | IND by 1.2 | IND by 2.4 |
MIN @ BUF | BUF by 6 | BUF by 4.5 | BUF by 7.9 | BUF by 8.6 |
MIA @ CHI | CHI by 3.5 | MIA by 0.5 | CHI by 3.6 | CHI by 4.4 |
NOR @ DET | DET by 2.5 | DET by 5 | DET by 6.7 | DET by 6.4 |
CAR @ GNB | GNB by 7 | GNB by 5 | GNB by 7.4 | GNB by 8.7 |
KAN @ SDG | SDG by 4 | SDG by 5 | SDG by 4.1 | SDG by 5 |
ARI @ OAK | ARI by 3.5 | OAK by 1.5 | ARI by 1.0 | OAK by 2.3 |
NYG @ DAL | DAL by 6.5 | DAL by 7.5 | DAL by 5.3 | DAL by 5.6 |
SFO @ DEN | DEN by 6.5 | DEN by 6.5 | DEN by 8.1 | DEN by 8.2 |
HOU @ PIT | PIT by 3.5 | PIT by 3.5 | PIT by 4.3 | PIT by 0.7 |
There you go. Denver is the best team in the league, though we’re probably wrong about that. Tampa Bay is the worst team in the league, something that I can almost guarantee you isn’t true when January rolls around.