website statistics
For the final team efficiency rankings of 2014, I thought it might be instructive to break down some of the process that goes into making the sausage.  AFA uses a different secret sauce than other analytics rankings sites, and if you're interested in the math behind the process, Brian has posted a detailed example of how this model spews it outs rankings here and here.

While I won't be doing the same computational legwork, there are elements within the model that reveal some general trends about the types of teams that AFA sees favorably (or unfavorably).  You don't have to agree with the final outputs, but I suspect that the small sample size of a 16 game season coupled with football's naturally high attrition rate prevents us from ever generating an ideally predictive model year after year.

We'll kick off the final look with a breakdown of a particularly unusual season-long ranking, as well as a look at the biggest changes from the beginning of the season.


The Black Sheep

Looking at a pair of controversial AFC East teams.

- Despite what other metrics and the eye test would suggest, AFA has put the Miami Dolphins ahead of the New England Patriots all season, including in the final rankings.  Miami barely eeked things out in the end, with a 0.61 Gross Winning Percentage (GWP) compared to the Patriots' 0.60 mark.  If the Pats had not treated their regular-season finale, it seems plausible that they may have passed the Fins at the finish line, especially when seeing their ugly offensive numbers.

Nevertheless, there was not one week since the rankings came out after Week 3 that the longtime AFC East bullies finished ahead of Joe Philbin's Toon Squad.  It's especially odd when seeing that AFA's own Expected Points Added (EPA) metric portrayed New England as possessing a significantly better offense, while also holding the edge on defense due to Miami's late-season collapse on that side of the ball:



The answer may lie in the fact that Miami appears to stand out particularly in rushing offense.  The Dolphins finished the season with the NFL's highest success rate at 51.6 percent.  New England's best stat was actually also rushing success rate—they checked in at third overall at 45.6 percent—but otherwise, the Patriots were surprisingly average in just about everything else.

It's an interesting trend over the past two years, as the top-ranked rushing offense has exceeded ostensible expectations all season long.  In 2013, Chip Kelly's Eagles were a top-five fixture in the rankings, despite sitting 3-5 at one point and not truly heating up until the end of the season.  Philly finished fifth at the end of last season's rankings, the same place where Miami ended up this season, spurred by a 49 percent rushing success rate that was four percent higher than the second-place team.

Considering that running the ball is definitely less mathematically favorable than passing it, the first thought should be one of alarm—is this site being hypocritical to its own research?  Correlating each team's ranking in run success rate with their overall ranking in GWP, DVOA and SRS—three metrics of team strength—here's what we get:



To be clear, this is far from perfect methodology.  It's only looking at one season, after all, so it would be more helpful to look at all of the available data going back to 1999 before drawing any actual conclusions.  Moreover, it's not as though all three team ranking measurements incorporate that run success rate into their formulations, which would be necessary to really draw definitive conclusions.

Still, it's telling that the AFA trendline differs noticeably from either DVOA or SRS, which are virtually indistinguishable from each other.  The correlation coefficient for the blue AFA line is 0.34, while both DVOA and SRS check in right around 0.17.

Even the stronger AFA correlation doesn't go so far as to say that run success rate should be a primary determinant in the team's ranking, so that alone can't explain why the Dolphins were ranked so high the whole season.  But it does represent a part of the formula, and based on these findings, it appears that AFA places more weight on the running game than at least those two other measures.  Again, this doesn't mean that one methodology is particularly better or worse than the others, but looks like the most prominent way in which this site diverges from its advanced metric cousins.


The Poles

No, not the people.  This is a look at the teams that constituted the top and bottom of the rankings.


- The Denver Broncos have been the site's top-ranked team since Week 6, and despite some concerns surrounding an ostensibly dead-armed Peyton Manning, the Broncos are the runaway top team in the final rankings.  Denver's 0.77 GWP, which would translate to an average of 12.3 wins per season, puts them almost exactly one full win better than the next best team in the league.

However, in finishing with a GWP of 0.71, the Seattle Seahawks became the first non-Denver team to crack the 70 percent barrier all season.  Indeed, stats like weighted DVOA suggest that the Seahawks have been the league's "hottest" team recently; Seattle's +95 point differential since Week 12, when Bobby Wagner and Kam Chancellor simultaneously returned to the defense, is nearly 30 points better than any other team over that span.

Right now, it appears exceedingly likely that the Super Bowl will consist of some combination of Seattle or Green Bay from the NFC and Denver or New England from the AFC.  AFA does not give any of the other eight playoff teams more than a five percent chance of reaching the Super Bowl; Football Outsiders and FiveThirtyEight give similarly heavy odds to the four-team hegemony.

If those four teams comprise the likeliest contenders, perhaps a more accurate division would be Seattle and then everyone else.  EPA doesn't always agree with these rankings, as we've seen throughout the season, but dividing the season into eight game halves, we see that the Seahawks' have been far and away the league's best team based on cumulative EPA in the second half:



EPA would suggest that the Seahawks and Packers are the two best teams in the league, though New England's second-half offensive EPA would probably have exceeded Seattle's if the Patriots had not rested starters in Week 17.  But while Green Bay's gaudy second-half offensive EPA looks impressive, the complexion of today's NFL suggests that having a premier defense may be more valuable.  Whereas nine teams finished with a cumulative offensive EPA over 100 this year (including all Big 4 teams), just four teams had a negative defensive EPA (with Seattle being the only Big 4 team to reach that criteria).

Combined with holding home-field advantage in the inferior conference, the Seahawks are probably the statistical favorite to repeat as champs.  The model may see Denver as the best team based on season-long aggregation, but based on conference competition and second-half form, the numbers all point to Seattle.


- On the other end of the spectrum, the Oakland Raiders finished dead last in the rankings, despite a late-season surge.  In actuality, the Raiders might as well share the Sacko Award with the Tampa Bay Buccaneers and Jacksonville Jaguars, however, as that trio stood out as a whole different kind of bad this season.

The terrible troika all finished with GWP's that would suggest a win total between 4 to 4.5 wins.  The 29th-ranked Atlanta Falcons had a GWP that was eight percentage points better than 30th-ranked Tampa, resulting in a win projection of 5.8.  In other words, all three performed roughly 1.5 to 2 wins worse than even the fourth-worst team in the league, an astounding number when one considers how little typically separates the teams at the bottom of the standings.

When digging for common denominators, all three failed most prominently in the passing game.  Jacksonville and Oakland were the two worst teams in the league in terms of net yards per attempt (NY/A), while Tampa finished fourth-worst.  The Jags and Raiders were also the two worst teams in terms of EPA/P and success rate, a byproduct of rookie struggles from Blake Bortles and Derek Carr.

Given that these were also the bottom three teams in last year's final rankings, it's legitimate to wonder whether these franchises are essentially dogs chasing their own tails in a circle.  All three teams switched quarterbacks from last year, while Tampa and Oakland made coaching changes.  Instability has been a constant with these franchises—since 2000, the trio have combined for 16 head coaching changes and 43 starting quarterbacks.

While finding a franchise quarterback is always priority No. 1 for franchises that don't have one, it's possible to win big without one.  The signal-callers of the last three Super Bowl champs finished 15th, 23rd and ninth in success rate.  On the other hand, blowing top-10 draft picks on a routine basis will destroy a franchise unlike anything else.  Here is the wildly depressing list of Pro Bowlers the three teams have combined to draft over the past 10 seasons:



Yep, that's seven total—between all three.  Khalil Mack probably deserves to be on that list, but that doesn't make the overall picture look much prettier.  Jacksonville and Oakland have now gone a combined 14 consecutive drafts without selecting a Pro Bowler.  Given that Pro Bowl rosters have swelled so that there are roughly as many players as fans in attendance at those games, that distinction is even more ignominious.

Such sustained incompetence is essentially the purpose of the relegation system in European soccer, but alas, no such salvation exists for this trio.  At some point, it might be worth an article to explore how these eras compare to other sustained stretches of failure in NFL history.


- In terms of movement from this year's initial rankings, here were the biggest risers and fallers from the initial post-Week 3 rankings:



This isn't necessarily surprising data—though in hindsight, how the Titans were ranked fifth is beyond explanation—as there is "filler" data used at the start.  In other words, three games of perfectly league-average performance are used as a counterbalance to whatever outliers may be weighing a team's stat in one direction or the other.  One of those average counterbalances is removed each week until, after six weeks, we get the first unadjusted rankings.

Even with those balances, it's clear that one-game outliers are still going to alter the picture significantly, and this is just a quantifiable illustration of that truth.  There are always methods to improve the model; hopefully this final look shed some light on the methodology that goes into those neat little tables at the end of these.


Here are the final team efficiency rankings for the 2014 season.  As always, observations, questions and snide remarks are welcome in the comments section.


RANK TEAM LAST WK GWP Opp GWP O RANK D RANK
1 DEN 1 0.77 0.49 4 1
2 SEA 2 0.72 0.49 6 3
3 GB 3 0.69 0.49 1 12
4 IND 6 0.62 0.49 5 17
5 MIA 4 0.61 0.50 9 11
6 NE 5 0.60 0.52 8 14
7 DAL 7 0.59 0.47 3 24
8 BAL 8 0.59 0.49 10 7
9 PIT 10 0.58 0.48 2 28
10 BUF 14 0.58 0.52 26 2
11 DET 9 0.56 0.49 18 4
12 KC 11 0.55 0.51 20 8
13 CAR 16 0.55 0.49 17 9
14 NO 13 0.54 0.48 7 29
15 CIN 12 0.53 0.52 15 13
16 HOU 17 0.52 0.47 16 10
17 CLE 15 0.51 0.48 21 6
18 PHI 18 0.51 0.50 11 21
19 SF 19 0.51 0.50 24 5
20 NYG 20 0.47 0.49 13 23
21 ARI 21 0.44 0.51 23 19
22 SD 23 0.44 0.51 22 20
23 MIN 24 0.43 0.50 25 15
24 WAS 25 0.40 0.48 12 30
25 NYJ 29 0.40 0.54 29 22
26 TEN 22 0.39 0.50 27 18
27 STL 26 0.39 0.50 28 16
28 CHI 28 0.37 0.52 19 31
29 ATL 27 0.36 0.49 14 32
30 TB 31 0.28 0.50 30 25
31 JAC 30 0.27 0.52 32 26
32 OAK 32 0.25 0.55 31 27

TEAM OPASS ORUNSR% OINT% OFUM% DPASS DRUNSR% DINT% PENRATE
ATL 6.9 35 2.4 1.2 7.6 58 2.8 0.41
ARI 6.4 34 2.1 1.4 6.8 62 3.1 0.35
BAL 6.7 41 2.2 1.3 6.2 66 1.8 0.42
BUF 5.8 40 2.2 1.8 5.4 62 3.4 0.50
CHI 5.8 45 3.1 1.5 7.2 57 2.6 0.50
CAR 6.0 44 2.2 2.2 6.1 62 2.5 0.37
CIN 6.5 40 3.4 1.9 6.2 54 3.3 0.39
CLE 6.5 37 3.2 1.5 5.8 54 3.6 0.44
DAL 7.4 43 2.3 2.1 6.9 62 3.2 0.42
DEN 7.5 42 2.5 1.3 5.3 65 2.8 0.50
DET 6.2 40 2.0 1.7 5.8 64 3.4 0.49
GB 7.5 41 1.1 1.8 6.0 53 3.2 0.38
HOU 6.5 39 2.7 1.9 5.9 56 3.2 0.39
IND 7.1 44 2.4 2.2 6.3 57 2.2 0.40
JAC 4.8 36 3.2 1.3 6.6 60 1.1 0.28
KC 5.9 42 1.2 2.1 5.5 53 1.1 0.34
MIA 5.8 52 2.0 2.2 6.2 60 2.6 0.31
MIN 5.7 43 3.5 0.8 6.2 56 2.4 0.42
NE 6.5 46 1.5 1.0 6.2 57 2.8 0.51
NO 6.9 45 2.6 1.6 6.9 57 2.2 0.32
NYG 6.7 38 2.3 1.7 6.8 59 3.3 0.43
NYJ 5.4 43 3.0 2.3 6.4 59 1.1 0.46
OAK 5.0 34 2.5 1.9 6.8 58 1.7 0.46
PHI 6.7 42 3.4 2.0 6.6 61 2.0 0.44
PIT 7.4 42 1.6 1.3 7.0 61 2.0 0.41
SD 6.7 34 3.1 1.3 6.2 54 1.3 0.48
SF 5.7 42 2.0 1.8 6.0 62 4.2 0.47
SEA 6.6 49 1.5 2.4 5.4 64 2.6 0.52
STL 6.0 38 3.1 2.5 6.6 65 2.4 0.58
TB 5.7 34 3.8 2.5 6.8 62 2.5 0.47
TEN 6.1 38 3.1 2.1 6.5 58 2.2 0.48
WAS 6.7 41 3.3 2.6 7.1 63 1.3 0.57
Avg 6.3 41 2.5 1.8 6.3 59 2.5 0.43