Passing Traits Analysis - NFL NextGen Stats

You can officially start panicking. We're already through three weeks of the year, and by next Monday at this time, you'll be looking at the schedule to realize that a quarter of the season is already behind us. It sucks, but it means there are still three more quarters ahead of us! Yay!

We are in 2019, not 1976 (even though the Vikings run-heavy game makes it look like it.) There is no excuse for you not to access the advanced statistics being used in every sport you follow and their importance. Back in the day, it was all about wins and losses, passing yards, and touchdowns scored. It is not that those stats are worthless, but they don't offer enough to the savvy analysts. While football is yet in its infancy in terms of analytics compared to baseball, the evolution the sport has seen lately in those terms is notable.

I'll be tackling NFL's Next Gen Stats on a weekly basis, bringing you the data from the just-completed week's games and highlighting some takeaways you should consider when assessing fantasy players for the upcoming games. In case you're new to the series, or Next Gen Stats altogether, I recommend you read our preseason primer. Now, it's time to get to the data!

The State Of The Passing Game Through Week 3

Back a couple of weeks ago when I introduced the series to you, I mentioned one of the most important concepts to consider when analyzing players: Air Yards. The metrics around it are key to know who is really over-performing or under-performing among receivers and passers, but it doesn't have much to do with rushers. For this last group, which mostly features on the ground, we can look at time, speed, and efficiency metrics.

For the first week's analysis, I opted to go with the receivers (and tight ends) group. For the second one iteration, I turned my attention to running backs, for which we looked at Efficiency and time-related metrics to try and get some insights from the data.

Now the time has come to tackle the most important position at football and probably every other sport out there: quarterbacks.

A key metric for passers is Completion Percentage (COMP%). This is nothing new. Just take all the completed passes from a quarterback and divide them by the total number of passes he has thrown. Keep in mind, though, that it takes two to complete a pass, and that bad receivers (those with low catch rates) will always drag the QB's COMP% down.

We should also remember that not every pass has the same probability of ending in a catch (due to the receivers separation, from the coverage, the position of the quarterback in the field, etc) NFL Next Gen Stats have built a model based on that information that uses completion probability to give each pass an Expected Completion Percentage (xCOMP). This metric is important because it can be directly compared with COMP% to see whether a quarterback is over- or under-achieving. By looking at the difference between COMP% minus xCOMP% we get the quarterback's Completion Percentage Above Expectation (+/-).

Instead of just focusing on Week 3 performances, and as three games are already a good sample in terms of size, I'll be tackling the accrued data from the first three weeks of the season together. Here is where we stand right now:

• NFL Next Gen Stats dataset includes only quarterbacks with at least 15 pass attempts through the first three weeks. You already know there have been multiple injuries around the league, but with most happening to quarterbacks in weeks 1 and 2, we're safe in that we'll be looking at every newcomer to debut or take the reigns this past weekend. There are 48 total QBs in the dataset.
• Between all of them, a grand total of 3,361 passes have been attempted in the season so far, and they have made for a staggering 25,078 yards on completions!
• Getting down to the basics, there are 157 total TD logged in the dataset to 69 interceptions for a 2.27 TD/INT ratio so far.

Who's Hitting (and Missing) the Nail on the Head...

The simplest of things we can do to try and see if a quarterback is performing at a tolerable level is to look at his Completion Percentage (COMP%). If I tell you that someone has completed 30 passes in a game, you could probably be inclined to think he had a good outing. If I complete the information telling you he threw an unsustainable 90 passes as the game went to overtime, maybe you change your opinion on that performance. It is not the same to complete 30-for-30 than 30-for-90, so we are better looking at the rate rather than the raw numbers. Ratios also put everyone on a leveled field, so we don't have to worry about volume and can easily compare Kyle Allen (26 attempts in one game) to Andy Dalton (129 attempts in three).

Let's take a quick look at the best and worst quarterbacks in COMP% through Week 3:

I'm sure you expected some of those results, but I'd be surprised if some of the names in the list (either at the top or the bottom) didn't make you raise your eyebrows. Gardner Minshew second-best!? Cam Newton third-worst!? What the hell is going on here? This is definitely not the NFL I know. Let's dig a little bit deeper:

• Getting rid of the usual suspects first, there is nothing mindboggling at finding Dak Prescott, Matt Ryan, Patrick Mahomes, and Russell Wilson in the top 10. All of them are proved quarterbacks with track records of excellence under their belts. Not saying that finding Prescott first, for example, isn't at least a bit shocking, but it helps you understand while he's on the verge of becoming one of the better-paid players at the position. Of course, you will have no chance of getting your hand in any of those any time soon in your re-draft league if you opted to pass on them.
• At the bottom of the table, we run into the same idea. Bad players keep being bad players, mostly. You were surely waiting for the Dolphins pair to be stuck in the mud, and indeed they are. Ben Roethlisberger had a tough start to the season and now out for the rest of it, his number will remember that low forever (exactly the opposite of Drew Brees, also injured but who will return later this season).
• Two names that have surprised everyone: Gardner Minshew and Kyle Allen. Minshew stepped as the Jaguars starter after Nick Foles went down injured, and he could have locked him in the position for way longer than expected at first. He's about to break the 100-pass mark and still boasts the second-best COMP% among qualifiers. Kyle Allen has only played one game this season backing up Newton, but he looked great for Carolina. Allen completed 19 passes for three touchdowns this past weekend and is a must addition from waivers with Newton expected to miss more time (he's been ruled out for Week 4).
• Speaking of Cam Newton, he's been quite disappointing so far. The arm he injured last season must not feel right at all. Newton is having trouble to complete any passes, and at this point, it very well merits consideration to drop or trade him for something of use before it's too late.
• Kirk Cousins himself is throwing way below his career-average COMP%, although he's attempted just 63 passes in three games. Minnesota looks to be held-bent on the run-game and that gameplan can only work for Cousins if he ramps up his production (a lot). If he doesn't, not only he but every Vikings' receiver upside will be affected, so analyze your team and see if you need to make some moves.

...And Who Should Really Be Doing It (Or Not)?

Now that we know the best and worst completion percentages around the league and who they belong to, it is time to ask the really valuable question: are those numbers for real? Turns out, only in some cases.

We have already introduced the xCOMP% metric. Instead of relying on pure data coming from what a single quarterback is doing on the field with his passes, xCOMP% offers an "expected value" of what he should have accomplished given the situations he has been put into.

As this metric goes hand-in-hand with COMP% and they can be compared easily, I think instead of a table with the most and least overperforming players, we'd be better off looking at a simple plot showing both values for each of the 38 players in the dataset. Players under the green line are performing over the expectations, and player above it have completed fewer passes than history said they should:

Right now you probably are a little surprised, yet again. What does it even mean that Mahomes is close to Brissett, both being under the line? Maybe you think both of them have just been lucky. On the other hand, would you say Fitzpatrick has been just unlucky in his outcomes, being where he's at?

In the next paragraph of takeaways, I'll use what it's called "+/-". It just the difference between xCOMP% - COMP%. A big +/- should indicate luck or overperformance, and a negative +/- exactly the opposite.

Let's look at some particular cases and try to explain them:

• Dak Prescott leads the league in +/- at 9.9. I don't think nobody would say he's been lucky. Not at least that lucky, anyway. Prescott has taken the next step and he's playing as good as ever. He has had the benefit of playing the Dolphins, Redskins, and Giants though, so expect some decline in his numbers.
•  The next three highest +/- values belong to Kyle Allen (9.2), Russell Wilson (8.7), and Drew Brees (8.5). Kyle Allen has only played one game, but he shredded Arizona. Although they don't have the scariest of defenses it is still probable that Allen keeps up his numbers. Small sample, but I'd make Allen a sure addition through waivers if possible. Drew Brees will be kept out for a while, but he's always been a staple of this metric. Don't call it luck, call it talent. He should be re-inserted into lineups as soon as he returns. As for Russell Wilson, I'm personally happy to find him here. Seattle is running one of the most efficient offenses in the league this season. Wilson is not throwing as many passes as other quarterbacks but he's been one of the best on low-volume throwing and is tied at fourth in touchdowns with seven (no interceptions).
• I already told you about Gardner Minshew's fast start. The rookie is completing 73.9% of his passes, but the expectation is only 2.1% lower. This is not a one-game sample. We're talking about three games already and Minshew might be a blessing in disguise for Jacksonville. If you're streaming the position (or even have a hole in deeper leagues), Minshew is someone to get from waivers asap.
• Although Sam Darnold and Luke Falk look virtually the same in COMP% (68.3 to 68.1), their xCOMP% are miles distant. Darnold should have completed 74.3% of his passes, while Falk only 0.1 percentage points more. I don't think Jets fans can't see the day Darnold is back at the helm.
• You know what happens to boom-or-bust quarterbacks: their performances and outcomes widely vary. That's the case of some players with low xCOMP% such as Jameis Winston, Kirk Cousins, and Baker Mayfield. All of them either don't care about throwing bombs, hold onto the ball for lots of time, or make dubious passing decisions that could lead to interceptions. Of the three, only Mayfield's COMP% is lower than his xCOMP%, and even with that, only by 3.2 percentage points.
• As a last note, and getting back to Newton, his +/- could very well be related to his injury woes. He's at 56.2 COMP% but his xCOMP% tells a completely different story sitting at 65.5%. That negative-9.3 difference could be explained by a guy who is hurt throwing passes in uncomfortable ways that otherwise would be easily turned into receptions. In any case, and although we should expect a rebound, I'd strongly advise looking for alternatives at the position if you only have Newton in your roster.

Conclusions

At the end of the day, the only truth here is the very real completion percentage passers are putting together. We can only be sure about what and what has not happened in games. We can use metrics such as xCOMP% to better inform ourselves of what should have happened in a perfect world, but can't deny that if someone misses nine out of ten wide open passes, maybe there is some issue with that player.

We shouldn't make xCOMP% the end-all-be-all of our studies, because that metric lives in the world of what ifs while we want facts. Yes, it can give us a little bit of extra information and make us be a little bit more level-headed in making decisions, but it is what it is. Just expectations.

This will do for this week entry, and as games keep being played and more data gets stored and available for analysis I'll keep coming back with more insights from the NFL Next Gen Stats set of numbers. We still have to cover pretty interesting topics such as time to throw, Air Yards completed and intended, receiver Yards after Catch and then cycle back to our almost forgotten running backs!

More Fantasy Football Analysis