It may not seem that difficult to guess which coaches' seats are hottest in the NFL from season to season. But in a sports universe becoming increasingly reliant on data, why should we even have to guess?
There's a machine learning model for that, too.
We taught our model using Pro Football Reference's data on all NFL head coaching tenures since 1979. Unsurprisingly, the most important factor is the team's win-loss record from the present season, but there are a number of other factors that interact with each other within the model:
The degree of decline or improvement from the previous season -- this is almost as important in our model as this year's record. The fact that GMs and owners are much more likely to pull the trigger after a team fails to meet expectations is hardly surprisingly, but it's important to note that a coach who goes from 3-13 to 5-11 is less likely to be fired than a coach who slips from 9-7 to 5-11. Regressing has a bigger impact than overall record.
The next most important factor is the coach's long-term record with the team, represented by his rolling average winning percentage for the past two to five seasons.
Playoff success matters. The model looks at how long it has been since a coach made the playoffs, and the number of seasons since a playoff victory. The recency of a playoff appearance appears slightly more important than that of a playoff win.
Did the current GM hire you? The thinking here is that a general manager (or whoever the hiring and firing authority is) is more likely to cut ties with his predecessor's coaching hire. Coaches have a 25 percent chance of getting fired when working with a GM who didn't hire him, as opposed to 18 percent when the current GM is the one who did.
Tenure has an impact on the model, but not in a straightforward way. Whether you've been the coach for a long time or you just got there, the data still relies on a combination of other factors to determine if your seat should be hotter.
I also included Super Bowl appearances and victories as factors, but they turned out to be the least important among all the factors. Coaches with a Super Bowl appearance or victory on their resume aren't given much extra slack.
Applying this data to last year's group of coaches, Mike McCoy (Chargers), Jeff Fisher (Rams), Gus Bradley (Jaguars), Chip Kelly (49ers) and John Fox (Bears) were seen as the five coaches most likely to be fired heading into the season. Only Fox survived. Fisher and Bradley did not last the season.
For 2017, let's project what would happen if every coach endured a poor season -- say a 4-12 record. Who's most likely to be shown the door in that scenario?
Chuck Pagano, Indianapolis Colts
Record with team: 49-31 in five seasons, 3-3 in playoffs (last appearance: 2014)
Bad 2017 firing probability: 76.0 percent
Pagano will officially have six seasons under his belt after 2017 (closer to five if you don't count the 12 games he missed in 2012 due to cancer treatments). A 4-12 season would leave him with a mediocre long-term record, and he also would have gone three seasons without making the playoffs. Interestingly, Pagano's firing probability takes a hit because GM Ryan Grigson was replaced in the offseason. Pagano and Grigson endured a well-publicized power struggle, which Pagano ultimately won, but the data doesn't think the presence of new Colts GM Chris Ballard necessarily improves Pagano's chances for sustainability.