New York Giants head coach Joe Judge got in trouble with the self-professed smartest guys online when he failed to demonstrate proper deference to the Analytics God. Here were Judge’s comments in a press conference earlier this week.
"Analytics is just a tool. It’s nice to look at the numbers and how they go through the flow of the game, but the analytics change based on the opponent, based on who you have available for the game, and how the flow of the game is going, too. You can look at a stat sheet all ya want. I promise ya if Excel was gonna win football games, Bill Gates would be killing it right now. But you've got to take those numbers as a tool and go ahead and factor in how your team's playing at the time and how the opponent is as well. You’ve got to measure your strengths and weaknesses against the opponent every time and then also in terms of the flow of the game.”
Judge’s tone and suggestion that analytics can be expressed by a simple Excel one-sheet have been ridiculed by many who profess to be devotees to football analytics, whatever that means (more on that below). Football statistics and metrics are attractive for a variety of reasons. One of the appeals for those who like to invoke “The Numbers” is that it offers the allure of a superior intelligence without having to do the work of province a warrant to defend the claim. And when capital-A “Analytics” is used as the cover, then the individual gets hide find the protection of the mob. “Joe Judge doesn’t believe in analytics, what an idiot” is the perfect sentiment for those whose deepest thoughts come in 280 characters or less.
As an aside, those familiar with my work know that I value analytics. It was the addition of basketball analytics into my handicapping toolbox that helped conclude the 2010 NCAA Tournament on a 20-4 ATS run after the first day of the Big Dance. Sabermetrics has been a foundational aspect of my handicapping of MLB regarding the respective starting pitchers since Day One in the field. Expected goals (xG) play a critical role in my handicapping of soccer and the NHL. Analytics plays a role in my football handicapping, but I consider much of that work so fundamentally flawed that I take many of their predictive numbers with a large grain of salt. I am not anti-analytics. I am anti-lazy thinking (and making claims without warrants).
Interestingly, the criticism of Judge’s comments exposes some assumptions many (not all) in the football analytics community make that would probably not withstand scrutiny if put under a microscope. Many football analytics adherents seem to advocate that there is a One Truth exposed by analytics as The Way. “The numbers say go for it on fourth-and-one.” Well, where did those numbers come from? League-wide data? What years? Does it include the pandemic season without fans in the stands? How big is the sample size? How big should the sample size be? Does that league-wide data treat Derrick Henry’s fourth-and-one numbers as equally relevant to those numbers for, say, Theo Riddick?
Is Judge wrong when he argues that “the analytics change based on the opponent, based on who you have available for the game and how the flow of the game is going, too”? Do the football analytics folks really want to suggest that there is no statistical difference between Saquon Barkley rushing the football on 4th-and-one versus Gary Brightwell, their sixth-round pick from Arizona? Is there no statistical difference between attempting a fourth-and-down rushing play against the Tampa Bay run defense as opposed to the Detroit run defense?
Another assumption many in the analytics community makes is that every statistical moment is the same. Many in the analytics basketball community presume this when defending the use of shooting tons of 3-pointers. They are later surprised when the Houston Rockets missed 27 straight 3-pointers in Game Seven of the 2018 NBA Western Conference Finals. Too many in the quantitative field rigidly support the belief that those 27 misses could have just as easily happened on a Tuesday night in February of the regular season. To suggest otherwise is to expose a fundamental problem with their project: that the numbers may not apply as nicely to the situation at hand. For those interested in nuance and perspective, this revelation is not threatening. For those who have a vested interest — financial or otherwise — this proposition is an existential threat.
It is not a radical idea to offer that basketball teams not good at shooting 3-pointers should shoot less of them (in place of higher percentage shots for their team’s skillsets). This is probably true even after confronting the fact that 3-point baskets offer 50% more value than 2-point baskets. Is it a radical idea by Judge that perhaps the percentages for his team on 4th-and-one may not be as prolific as that of the Kansas City Chiefs?
And is it a radical idea that going for it on fourth-and-one (to continue using this one example since it most often gets deployed by the football analytics folks as if there is capital-t Truth answer to this question that can fit on an Excel sheet) that the answer may change based on field position, game score, and how much time is left in the game?
The other major sports are getting better at appreciating that rather than establishing an Analytics Department to expose the Truth, instead the use of numbers and statistics is fluid that can be exploited for a strategic advantage. All numbers are not created equal because all formulas are not created equal. Some analytics are simply more illuminating. While yards per game offer some value, yards-per-play may offer a more insightful perspective. Just like the NFL has 32 unique scouting departments that make different evaluations, the league will eventually have 32 unique analytics departments that have differing views — and this is before head coaches then interpret that data based on his available personnel, the score of the game, and the moment in the game.
Just using the analytics umbrella does not provide invincibility against potential critique. The audience is not privy to the NextGen formulas used to develop their stats ESPN hawks (in partnership) regarding what a coach should do in a certain situation. The broadcast is not a math class, but it is theater. It’s a smaller narrative within the bigger story.
Many statistical models in football do not put any value on first downs and time of possession. Do they do this because they disagree with many football coaches who find both those aspects of the game critical? Or, do they do it because it is more convenient to ignore those facets of the game? Using Yards-Per-Play as the base unit of efficiency is easier than the messy work of determining how to value the reset of downs offered by generating 10 yards in four plays.
There are many differences between Joe Judge and his critics, but one I would like to close with is this: it is only Judge that risks losing his job if gets a football decision wrong. His critics risk nothing. Many of his critics have a vested financial interest in presenting their criticism since that it is the foundation of their business model. When contemplating going for it on fourth-and-one (to torture this one example), I suspect there would be a quick about-face in opinion if the critic were to lose their job if they got the decision wrong. In fact, I suspect all it would take for many of his critics to demonstrate caution and nuance would be the mere threat of being blocked or unliked if their opinion from the cheap seats turned out to be correct.
Best of luck — Frank.