Analytics show Tyreek Hill among NFL’s most versatile receivers

Let’s say I told you I spent the Kansas City Chiefs’ bye week grinding away at the film, charting the routes run by every targeted receiver on every pass play from the 2019 season. Would you believe me?

If you did, you might be a bit too trusting.

This season, there were about 18,000 passing attempts in the NFL — and with them, 18,000 targets and 18,000 routes run. Let’s say it takes 15 seconds to watch a play and accurately draw the route of the targeted receiver — and remember: we want to be precise about the receiver’s position on the field as they run their route; we need more than a quick, messy line.

That adds up to about 75 hours of work. But even that assumes you can seamlessly transition from play to play and game to game — and that you are a robot who doesn’t need to take breaks for essential activities like meals, sleep... or reading Arrowhead Pride.

Thankfully, there’s another way to get that information.

I’ve previously used the work of Baltimore Ravens data analyst Sarah Mallepalle. While a student at Carnegie Mellon University, she and her colleagues developed a publicly-available tool called next-gen-scrapy to obtain data from (or scrape) Next Gen Stats passing charts. By transforming the charts into coordinate data, they could analyze pass location probability or the expected completion percentage of a given pass.

But the Next Gen Stats site has more to offer than just passing charts. They also have route and carry charts — and next-gen-scrapy was limited to the passing charts. What could we learn if we knew the receiver’s routes on every pass thrown in 2019?

To satisfy that curiosity, I spent some time building on next-gen-scrapy, adapting it to collect and analyze data from NGS route charts. This new code — called next-gen-scrapy-2.0 — is also public, allowing anyone to obtain and analyze the data. Or you can just pull the current data from the .csv files I will keep updating. Both the code and the data are available here.

So what did I find with this new data set?

My first thought was pretty straightforward: we can use the data to see what areas of the field a receiver is most likely to occupy while running their routes.

After lots of recent talk about whether George Kittle is as good as Travis Kelce as a receiving tight end, I thought I’d compare them. Which of them forces the defense to cover a larger area of the field?

You can think of these charts as a probability estimate — that is, how likely a receiver is to occupy a specific area of the field. The dark green areas are where a receiver’s routes never go; the dark red ones represent the areas where their routes most frequently go.

We can see in the second chart that Kittle has two dark red areas just outside the hashes — between 0 and 5 yards. This means a large proportion of Kittle’s routes are in this short area. But Kelce (in the first chart) has no area of the field as dark as Kittle’s, which means his routes are more varied.

Remember, though, that we’re only looking at the areas of the field their routes cover — not the types of routes they are.

This isn’t to say Kittle is a one-trick pony. For a tight end, his routes are quite varied. But he’s no Travis Kelce — at least, not yet.

Speaking of one-trick ponies...

Last year, a certain cornerback declared Chiefs wide receiver Tyreek Hill to be a “return specialist” rather than a fully-capable wide receiver, implying Hill was nothing but a gadget player possessing limited versatility. Can this data tell us if that’s true?

This certainly doesn’t look like the chart of a limited wide receiver! Hill’s routes are quite varied, covering all portions of the field with very little repetition.

But how does he compare to other wide receivers?

We could create a chart for each receiver and inspect them visually. Or we could just calculate the density of their most common region of the field.

If you’re a math-y person, this would be the peak value of the kernel density estimate (KDE) of each receiver’s routes — evaluated on the same size grid for all players so all values summed to the same number.

I restricted this analysis to wide receivers with at least 8 games of charts on NGS site. Here’s what I found:

1. Mike Evans, Buccaneers
2. Kenny Golladay, Lions
3. Tyler Lockett, Seahawks
4. Tyreek Hill, Chiefs
5. D.J. Chark, Jaguars

A high ranking here means low density in their routes — that is, they spread the field more. You can think of this as as the versatility of a wide receiver.

Let’s look at the most versatile wide receiver: Mike Evans.

Like Hill, Evans covers a large potion of the field. There are no red areas — meaning it’s not common for him to run in the same parts of the field.

In contrast, here is the wide receiver in last place — the one with the densest concentration of field area: the Seahawks’ D.K. Metcalf.

If any NFL wide receiver could be called a one-trick pony, Metcalf fits the bill. A large portion of his routes are to the same area of the field; and almost all of them are outside the left hash.

While he has been a solid addition to the Seattle receiving corps, it must make things easier for defenses to know where he’s likely to be going when the ball is snapped — which makes his success this year even more impressive. A quick look at Metcalf’s most recent game confirms our analysis — which is pretty neat!

Without watching a single game, we have determined the most likely area of the field that each receiver will run their routes, how often they run their routes to those areas and which receivers are used to spread the field more.

But what’s really crazy to me is that this data doesn’t even scratch the surface of what NFL teams have available to them. Each team is given the exact coordinates of every player on every moment of every play — not just the receiver who is targeted. By doing a similar analysis on this data, just imagine all the insights you could find! If you want to know where George Kittle is most likely to run from 11 personnel against single-high coverage, you can find out — not by watching film, but by analyzing the data.

But this data has limitations, too. The most glaring is that there is little nuance — and we know football is a nuanced game. We can’t see Sammy Watkins’ footwork. We can’t see Travis Kelce fool a defender with a head fake. Those things absolutely matter — and analytics will likely never be able to tell us about them.

The film will always be vital — and there will always be an army of football experts ready to review it. But what this data can do is something film alone cannot accomplish: sift through tens of thousands of plays in minutes, finding meaningful macro and micro-level trends.

To me, this is proof enough that NFL analytics are here to stay. Any team that is slow to adapt will soon be left in the dust.