Building Better QB Projections
Editor’s Note: This article is a collaboration between Joshua Lake and Sean Slavin. Find Sean on Twitter @Slavin22 …
Biases and unconscious preferences slip into all of our rankings, into all our opinions on players, but we rarely acknowledge the impact those outside factors have on us. When I sit down to create a tier list at quarterback, I tell myself I will only stick to the numbers. I will rely on the data. But in reality, it’s not that simple. I carry my own baggage into the process. (I may never recover from the Philip Rivers disappointment of 2016. My therapist knows.)
Objective projections based only on math and numbers can guide us and provide a beneficial baseline upon which to attach our narratives and hunches. If we start with the numbers, we can at least begin from a blank slate. In this article, we take a look at a projection model designed to give us that detached, history-based prediction of how each quarterback will look in 2017.
This entire article is the brainchild of Sean Slavin, a man with mathematical intelligence and ability that go exponentially beyond anything I can muster. He pulled the data, ran the numbers, and created the projection system. All the insight is his. I am simply trying to do his work justice by putting words to the information he shared with me. You should absolutely go check him out on Twitter (@Slavin22) and thank him for sharing this with us.
Sean began by collecting data for every quarterback season since the 2000 season. From that data, he created aging curves for every statistical category. For example, a quarterback’s number of attempts per game tends to rise through age 38, but rushing yardage certainly does not follow the same trajectory with age. It was important to create separate age curves for each stat rather than just for PPG, total points, and stats like those, which are actually aggregates of a number of pure stats.
To create aging curves, Sean used a higher-number polynomial system rather than a quadratic equation, which, he taught me, led to much higher r^2 correlation figures and also allowed for complex age curves that can have small dips, spikes, or normal arcs depending on the dictates of the data set. You can see that illustrated below in two versions of the attempts per game aging curve.
Next, Sean created a baseline for every quarterback by averaging their prior seasons, giving more weight to recent seasons. From there, he applied the aging curves to each player’s baseline numbers to create a 2017 projection. (Sean placed minimum and maximum limits on every rate stat projection to eliminate absurd projections based on things like Kevin Hogan’s 13.1 yards per rush and Tom Savage’s touchdown rate of 0%. That kept all rate stat projections from hitting unprecedented levels.)
That was done for eight statistical categories: attempts per game, completion percentage, yards per pass attempt, passing touchdown rate, interception rate, rush attempts, yard per rush attempt, and rushing touchdowns. For example, Tom Brady has a baseline completion percentage of 64.1%, but his 2017 projection is 63.0% after adjusting for his age.
Sean calls this the 2.0 phase of his projection system, and he has plans to explore other factors like coaching staff, surrounding players, offensive line, strength of schedule, and more to see if any reliable data points can be added to his projections in the future. I must say, however, that this current version does a great job at highlighting quarterbacks who might be over- or undervalued if history is a reliable guide.
Want to see this baby in action? Me too!
Rather than use the projections to spit out a complete ranking of quarterbacks for 2017, Sean dug into his data to provide action steps for drafters. (Remember, MFL10 bestball drafts kick off any day now, so many of us will be making choices and picking QBs this month.)
We just needed a starting point. In came Sal. Mr. Stefanile recently shared his way-too-early 2017 QB rankings with us, so Sean began with those to find a few similarly-ranked QB pairs. The idea was to take quarterbacks who are perceived similarly to see if these projections might shed some light on which one we should prefer. As you will see, we took one pair from each tier of quarterback so that you can walk away with something useful whether you draft QBs early or late.
QB1s: Dak Prescott vs Derek Carr
In a debate between two young, rising stars, how do you choose? Both Dak Prescott and Derek Carr will be hot targets in 2017 drafts, and for many of us, there is little to separate the two. For Sean, the choice looks clear.
In the first chart, we see Sean’s projections for the entire 2017 season; there, projected rank is based on the estimated number of games started that’s shown. In the second and third charts, however, the projection shows what each QB’s stats would look like assuming they played in all sixteen. The final chart shows startable quarterback percentage (“SQB%”) data, which you can find for every quarterback by clicking over to our QB Cards.
Notice all the green at the top of each chart? Dak Prescott comes away looking good in the majority of statistical categories, although we know his data is based on only one season compared to Carr’s three.
Verdict: Dak Prescott wins.
QB2s: Philip Rivers vs Colin Kaepernick
Moving to the QB2 tier, Sal has perenially starter Philip Rivers three spots behind mercurial star Colin Kaepernick. We know there’s uncertainty as to Kaepernick’s situation next year — will he be in San Francisco or elsewhere, and will he be starting? But what do his history and age suggest, assuming he starts in 2017?
In news that won’t surprise most of us, Philip Rivers looks like the far better passer, and Kaepernick destroys him on the ground. Interestingly, Sean’s projections spot Rivers a two-point lead in the PPG category, suggesting that Kaepernick is not such a reliable option even when he’s starting.
Verdict: Philip Rivers wins.
QB3s: Tony Romo vs Carson Wentz
In the QB3 tier, do you take an aging veteran with an uncertain situation, or do you trust the second-year quarterback who started hot and flamed out fast? Tony Romo has plenty of supporters, but Sal is not among them so far, ranking him at QB32. I suspect that Romo’s ranking is likely to move significantly once we know where he’ll be playing in 2017. Carson Wentz currently slots into QB26 in Sal’s ranks. All my bloviating behind us, let’s see what Sean’s numbers tell us.
Much like Dak Prescott earlier, Wentz’ numbers are heavily impacted by his 16-game sample size, but they also emphasize just how bad his rookie season was after a few hot weeks to start. Wentz comes in at QB32 in projected points per game, and his TD-to-INT ratio is downright dreadful. If you want to buy into the sophomore Eagle this year, you need to do so with a plausible argument for why he will be dramatically different from his 2016 self. In this matchup, Tony Romo’s only flaw is his likelihood of missing games; but for that, this would be a sweep. At this point in the offseason, the biggest edge we can give Wentz is his guaranteed starting job, which Romo is seeking but doesn’t have yet.
Verdict: Tony Romo wins.
Fliers: Mike Glennon vs Tom Savage
And here we are, the five-dollar bin of discount quarterbacks. This is where 2QB leagues are won and lost, and it’s where Sal specializes, often finding fantasy starters who you could’ve acquired for free. He ranks Mike Glennon and Tom Savage back-to-back at this early point in the offseason, likely giving both reasonable odds at starting games in 2017. Let’s see if Sean’s model also sees them as similar options.
This one is a tough sell, because we only have four games on which to base Tom Savage’s projection. That said, those four games gave us nothing on which to pin hopes that Savage might become a fantasy starter. Glennon also hasn’t looked incredible, but he projects to be a decent low-end fantasy asset if he earns a starting job next year. His history suggests that the volume may be low, but all that can change if he lands a gig in the right offense. Savage, however, offers us almost nothing to like. At least he doesn’t throw many interceptions!
Verdict: Mike Glennon wins.
Please remember to thank Sean (@Slavin22) for sharing his projections with us! I can personally say I learned from his approach and love the idea of projecting players in an objective way that keeps all our biases and narratives off to the side. I look forward to seeing what he comes up with next!