Things look bad for Chip Kelly and his Eagles. Fans are calling for his head; discord is rampant in the locker room; Kelly seems to be linked to every college job not nailed down or on fire; and the Eagles have found themselves on the receiving end of consecutive beatdowns so humiliating that mayor-turned-governor-turned-commentator Ed Rendell — a man last seen in public supplication outside a McDonald’s, begging for a McRib — was so ashamed that he hid his head in a bag. If Kelly is the NFL’s mad scientist, the lab is on fire.
In the short term, Kelly is probably safe. He has two years left on his contract after 2015, and his newfound authority over Philadelphia’s roster — wrestled away from Howie Roseman in a power struggle in January — makes him a little more entrenched than your typical head coach. But the Eagles have also been one of the NFL’s most disappointing teams, by both the eye test and fancier metrics. So while Kelly the personnel czar still has some rope, we can entertain ourselves with a thought exercise wondering just how often such a disastrous season traditionally leads to a coaching change.
To measure just what it takes for a coach to get fired in the NFL, I trained a classification model on data for every NFL coach since the 1970 AFL-NFL merger, looking for the factors that predict whether he’ll return the following season. The best model, in terms of having the lowest “out-of-bag error,” accounts for how long a coach has been with his team, the team’s outlook going into the season, and how many playoff wins the coach has recorded during his tenure with the team. Coaches are typically given more leeway early, with firings peaking three to six years into their tenures, and unsurprisingly, playoff wins buy more job security. We used our Elo ratings as a proxy for team outlook; whatever you think of its predictive capabilities, Elo is a fantastic gauge of team perception and expectations.
According to that historical rubric, no current coach6 should be feeling the heat more than Kelly, whose indicators have traditionally led to termination a shade over half the time (through 11 games):
| ELO CHANGE | ||||||
|---|---|---|---|---|---|---|
| COACH | TEAM | SEASONS | PLAYOFF WINS | BEFORE 2015 | DURING 2015 | CHANCE OF FIRING |
| Chip Kelly | PHI | 3 | 0 | +144 | -86 | 52% |
| Gus Bradley | JAX | 3 | 0 | +8 | -18 | 51 |
| Marvin Lewis | CIN | 13 | 0 | +184 | +101 | 40 |
| Jason Garrett | DAL | 6 | 1 | +141 | -117 | 35 |
| Sean Payton | NO | 9 | 6 | +94 | -76 | 19 |
| Mike Pettine | CLE | 2 | 0 | +32 | -83 | 19 |
| Lovie Smith | TB | 2 | 0 | -59 | +45 | 17 |
| Jeff Fisher | STL | 4 | 0 | +124 | -56 | 13 |
| Tom Coughlin | NYG | 12 | 8 | +81 | +21 | 7 |
| Chuck Pagano | IND | 4 | 3 | +173 | -31 | 6 |
| Mike Zimmer | MIN | 2 | 0 | +4 | +99 | 5 |
| Mike McCoy | SD | 3 | 1 | +32 | -105 | 5 |
| Jim Caldwell | DET | 2 | 0 | +66 | -64 | 5 |
| Mike Tomlin | PIT | 9 | 5 | -6 | +2 | 4 |
| Ron Rivera | CAR | 5 | 1 | +145 | +154 | 3 |
| Mike McCarthy | GB | 10 | 7 | +136 | +20 | 2 |
| Andy Reid | KC | 3 | 0 | +185 | +84 | 1 |
| John Harbaugh | BAL | 8 | 10 | +117 | -74 | 1 |
| Bill Belichick | NE | 16 | 21 | +164 | +76 | 1 |
| Jim Tomsula | SF | 1 | 0 | — | -103 | <1 |
| Pete Carroll | SEA | 6 | 7 | +276 | -37 | <1 |
| Rex Ryan | BUF | 1 | 0 | — | -22 | <1 |
| Dan Quinn | ATL | 1 | 0 | — | +6 | <1 |
| Bill O’Brien | HOU | 2 | 0 | +120 | +10 | <1 |
| Todd Bowles | NYJ | 1 | 0 | — | +34 | <1 |
| Jack Del Rio | OAK | 1 | 0 | — | +37 | <1 |
| John Fox | CHI | 1 | 0 | — | +50 | <1 |
| Jay Gruden | WAS | 2 | 0 | -28 | +55 | <1 |
| Gary Kubiak | DEN | 1 | 0 | — | +66 | <1 |
| Bruce Arians | ARI | 3 | 0 | +120 | +125 | <1 |
The idea here is to measure what sort of team performance, somewhat devoid of context, would get a typical coach fired. You’ll notice in the table that Marvin Lewis — a coach pretty unlikely to be fired — is third on this list; that’s mainly because he’s been with the Bengals for 13 years and has zero playoff wins. The real world context is that while that is a broadly undesirable outcome, the Bengals being the Bengals, the team has more than exceeded the existential target of being Not The Browns. (You can see the lasting improvement that Lewis has brought to Cincinnati in the “before 2015″ column, which shows how much the team’s Elo rating improved between the time he got the job and the beginning of this season.) So Marv, like Chip, is probably fine, for reasons that are very different but just as amusing.
As for Kelly, he’s in his third year of his tenure with a club, after winning zero playoff games in Years 1 and 2 — a scenario that has traditionally been the death zone for NFL coaches. Coaches in that predicament must show progress to keep their jobs: Roughly two-thirds of surviving coaches improved their team’s Elo rating (relative to preseason expectations) in Year 3, while nearly three-quarters of those fired oversaw an Elo decline from the preseason. So the Eagles’ 86.4-point Elo drop this season doesn’t look good for Kelly — since 1970, only 15 of the 128 coaches in Kelly’s position (Year 3 with a team, no previous playoff victories) oversaw a bigger drop-off in Elo through 11 games of the schedule, and two-thirds of them were fired before the following season began.
| COACH | YEAR | TEAM | CHANCE OF FIRING |
|---|---|---|---|
| Herman Edwards | 2007 | KC | 47% |
| Dave Campo | 2001 | DAL | 39 |
| Sam Wyche | 1994 | TB | 30 |
| John Mazur | 1971 | NE | 29 |
| John Mackovic | 1985 | KC | 28 |
| Chan Gailey | 2011 | BUF | 28 |
| Jim Hanifan | 1983 | CRD | 26 |
| Dick Jauron | 2008 | BUF | 25 |
| Neill Armstrong | 1980 | CHI | 22 |
| Dave McGinnis | 2002 | ARI | 21 |
Sometimes coaches still survive odds like Kelly’s. This table shows the coaches whose jobs were in the most jeopardy 11 games into a season yet went on to keep the job. But if Kelly does manage to stick with Philly beyond this season, he would set a new mark for unlikely job retention.
Kelly’s trajectory wasn’t always so negative. After Philly beat Dallas in early November to make their record 4-4, there was only about a 5 percent chance that Kelly would be fired according to the model. But the Eagles’ current three-game losing streak has caused his probability of being fired to skyrocket:
There’s still time for Kelly to turn things around. For one, the average coach who was fired since 1970 was assigned a 76 percent probability by the model at this stage of the season, so Kelly, at 52 percent, has some room left to fall. Second, despite their abiding awfulness, the Eagles somehow have a 15 percent chance of winning the NFC East, which would put a little shine on the turd. Although one of the most similar coaching seasons to Kelly’s7 resulted in Wade Phillips being fired by the Denver Broncos in 1994, other similar years belonged to coaches who held on for another year (Bruce Coslet with the 1993 Jets) or even unexpectedly went to a Super Bowl a year later (Jim Fassel with the 2000 Giants).
However the season plays out, Kelly will probably hang onto his job — particularly given the personnel control matter we mentioned earlier. But based on what typically gets coaches fired, Kelly should be thankful that he’ll (probably) remain employed when the smoke clears on this garbage fire, because historical precedent says it should pretty much be a coin flip whether he gets swept out with the ashes.