Commentary

You want winners? Here's a formula

According to a Ph.D. (and Patriots fan), math can help derive successful coaching hires

Originally Published: December 30, 2009
By Seth Wickersham | ESPN The Magazine

George Halas, Vince Lombardi and Bill BelichickAP Photo, Getty ImagesGMs, take note: Perhaps science provide a formula for determining the next (l-r) George Halas (kneeling in hat), Vince Lombardi or Bill Belichick.
During the past 10 years, a few NFL teams and some outside researchers have attempted to discover a quantifiable method for finding a great coach.

The Philadelphia Eagles are believed to be the first team to commission a study, a 14-page report in 1999 that led them to hiring Andy Reid. The late Johnnie Cochran -- yes, him -- in 2002 did a paper about minority coaches. The San Francisco 49ers completed a study in 2005.

And two researchers at New York University completed an 82-pager a few years ago. The studies took into account age, race, experience, background and dozens of other variables that could serve as predictors in determining who will be the next great coach.

Considering we're about to enter the NFL's hiring-and-firing season -- the Washington Redskins, Oakland Raiders, Buffalo Bills and Cleveland Browns, among others, could be looking for new coaches -- you'd think those studies would be in high demand. (I delve into all the studies in the next issue of The Magazine, and the next great coach, according to the data, will shock a lot of people.)

But I left out one study. It's the most intriguing, complex one, in fact: a statistical formula that, in theory, could predict success for head coaches.

In 2006, a University of Massachusetts computer science Ph.D. student named Andrew Fast studied statistical models that showed how social surroundings impact individuals. Fast essentially asked: Do birds of a feather flock together?

He used models called relational probability trees -- RPTs -- to produce data about different industries, like securities trading. In the middle of finishing his dissertation Fast, a New England Patriots fan, took a monthlong detour to analyze if coming from a coaching tree was a true predictor of a coach's success. Read more about Fast's study here.

He determined a list of 15 "championship coaches," including guys such as George Halas, Vince Lombardi, Hank Stram and Joe Gibbs, and looked at how many protégés they had, and the success rate of those offspring.

Fast easily found the answer: Yes, successful coaches produce successful coaches -- as studies done by the 49ers and NYU also determined. In fact, the NYU research showed that of the past 84 coaches hired, those who hailed from one of four trees -- Bill Walsh, Marty Schottenheimer, Bill Parcells and Jimmy Johnson -- had a 52 percent winning rate, as opposed to 44 percent for the others.

That question answered, Fast set out to design a huge, complicated algorithm that, if functional, would have had GMs drooling. Fast is good at algorithms. He once created one that predicted the 2007 playoff teams better than the (cough, cough) ESPN experts. (Not that he's bragging.)

The coaching algorithm weighed 19 variables in a coach's background -- if he's from a tree, if he's an offensive or defensive specialist, his level of experience, to name a few. In theory, it would compute the chances of the coach's team making the playoffs.

"My hope was that GMs would be calling me to learn about it," Fast says.

Fast analyzed the data of not only every head coach but also every assistant since 1920. (Seriously -- he actually did.) In the process, he unearthed some interesting stuff.

1. Good coaches usually have fewer mentors than lousy coaches. The best, in fact, have between one and two bosses during their careers. The theory is, if a coach is good, his bosses want him around.

As a predictor, that bodes well for someone like Josh McDaniels, who until coaching the Denver Broncos this year had worked only under Bill Belichick. But it's not good news for those like Kansas City Chiefs coach Todd Haley, who's at his fifth team in nine years.

"Instead of 'paying dues' the most talented assistant coaches tend to stay with a single mentor until they are promoted to a head coaching position," Fast wrote.

2. Coaching staffs matter. Fast devised an RPT to predict a team's chances of making the playoffs based on the quality of the assistants. He found that when a coach hires a staff comprised of members whose teams made the playoffs at least 34 percent of their careers, the probability of making the postseason in any given year is 55 percent. If the assistants' number is less than 34 percent, the chance of the team making the playoffs drops to 40 percent. So the more the assistants have won, the more they'll win.

Fast's research was startling for those who believe that numbers and data can serve as predictors. Of course, few NFL teams actually do.

The Atlanta Falcons commissioned a study a few years ago, before hiring Mike Smith, to see if there were statistics that the great coaches shared. But owner Arthur Blank disregarded it because he didn't believe it could accurately forecast a good coach.

And those who blow whistles for a living believe that some data matters, such as being from a successful coaching tree. But that's about it.

"I think the studies are good," says San Diego Chargers coach Norv Turner. "But I do believe the key is the players that you've got. The best coaches have good players."

Fast diagrammed the model. It's a vast, boxy thing with lines connecting each piece of data. If someone tracked your path as you walked around a shopping mall, this model is what it would look like.

But then, Fast hit a block.

He knew he had to include data that showed a coach's personality intangibles. For instance, was the coach a disciplinarian or player-friendly? Decisive or dithering? Honest or dishonest? These sorts of indefinite qualities often make or break a coach.

League execs who believe in data, like Eagles president Joe Banner, tried and failed to quantify these sorts of intangibles.

Fast was thisclose … but then he gave up. After all, he had to, you know, graduate.

"I couldn't connect this study to my dissertation [on improving algorithms for Bayasian networks]," says Fast, now 30 and working as a research scientist in Charlottesville, Va.. "So I never really finished it."

I couldn't believe it. After all, this is the NFL! Forget the dissertation!

"Who says I wasn't ever going to finish it?" Fast says.

"I just haven't finished it yet. "

Nobody's waiting. Only the Redskins, Raiders, Bills, Browns …

Seth Wickersham is a senior writer for ESPN The Magazine and a columnist for ESPN.com.

Seth Wickersham | email

ESPN The Magazine senior writer
Seth Wickersham joined ESPN The Magazine after graduating from the University of Missouri. Although he primarily covers the NFL, his assignments also have taken him to the Athens Olympics, the World Series, the NCAA tournament and the NHL and NBA playoffs.