QB Functional Mobility Model 2019
Guest column by Jeremy Rosen and Alexandre Olbrecht
The 2018 draft revealed significant changes in the way NFL teams view quarterbacks. For the first time in league history, four quarterbacks were drafted in the top 10. In addition, the Cleveland Browns' decision to select 6-foot-1 spread quarterback Baker Mayfield with the first pick surprised many analysts. For example, Scouts Inc., which we used to estimate the natural logarithm (ln) of draft position for our model, did not grade Mayfield as a top-10 pick. However, those in the statistics community were less surprised. For instance, we wrote in our FO guest column last year that experience running a traditional pro-style offense in college no longer predicts NFL success. Furthermore, FO's QBASE system gave Mayfield its highest grade of all quarterbacks in the 2018 draft class.
Before we delve into our 2019 projections, we show in Table 1 what our 2018 projections would have been had actual ln (draft position) been available last March. (Credit to last year's commenter BigBlue6790 for his/her request.) In particular, we would have projected Mayfield second, just behind Sam Darnold of the New York Jets. Mayfield outplayed Darnold overall, but Darnold was one of the NFL's best quarterbacks in December. In addition, Kyle Lauletta and Luke Falk are no longer on the list because they were not drafted in the first three rounds. Finally, 95% PI stands for 95 Percent Prediction Interval; each quarterback's career Adjusted Net Yards per Attempt (ANY/A) has a 95 percent chance of ending up in his 95% PI. Because there is a lot of uncertainty with quarterback prospects, these intervals are necessarily wide.
|Table 1: 2018 Predictions with Actual ln(Draft Position)|
|Quarterback||Our ANY/A||95% PI Low||95% PI High|
Predictions for the Class of 2019
In addition to ln(draft position), our model contains ln(run-completion ratio) and rushing yards per (rushing) attempt. A quarterback with a low ln(run-completion ratio) is desirable because he is an accurate pocket passer; a high completion percentage is correlated with a low ln(run-completion ratio). A quarterback with a high number of rushing yards per attempt is also desirable because he can run well when necessary. We consider a quarterback strong in both categories to be functionally mobile (credit to Doug Farrar for his SI.com article on functionally mobile quarterbacks).
Like last year, we have updated our model with the most recent data to make our 2019 predictions. Once again, we do not include the previous year's class in our dataset because first-year performance data is unreliable. However, as we explain in the next section, teams are increasingly valuing functional mobility when evaluating quarterback prospects. Therefore, a model combining functional mobility with projected ln(draft position) may effectively count it twice, thus overvaluing it. For that reason, we are using a college stats-only model this year. In Table 2, we rank the 2019 quarterbacks by their projected ANY/A. These projections are higher on average than in previous years, not necessarily because this class is special, but because NFL teams appear to be seeking out functionally mobile prospects.
|Table 2: 2019 Predictions|
|Quarterback||Stats-Only ANY/A||95% PI Low||95% PI High|
Kyler Murray is this year's wild card. He has our highest projection thanks to a 67.4 percent completion rate and a whopping 7.1 rushing yards per attempt. In addition, he won the 2018 Heisman Trophy, and the Arizona Cardinals may select him first overall. However, Murray is 5-foot-10, and there is little precedent for a quarterback that short. At 5-foot-11, Russell Wilson is the only other sub-6-foot quarterback in our dataset. Furthermore, unlike Wilson, Murray is a dual-threat quarterback, which increases his likelihood of injury. Therefore, despite his sky-high potential, the Cardinals have reason to be wary of going all-in on him.
With the exception of Ryan Finley, whom Scouts Inc. currently projects as a second-round pick, the rest of our projections are mostly in line with those of scouts. Drew Lock completed fewer than 60 percent of his passes, but he made up for it by having this year's lowest ln(run-completion ratio). He also rushed for over 2 yards per attempt. Dwayne Haskins' rushing stats are less impressive, but he completed an exceptional 70.0 percent of his passes. Meanwhile, Finley checks off both boxes our model likes: he has this year's second-lowest ln(run-completion ratio), but he still rushed for almost 2 yards per attempt. Scouts' main concern with him is his lack of physical talent, which earns him the dreaded game manager label. However, we view him as a potential sleeper prospect.
Will Grier's projection is more mixed. He is an accurate pocket passer, but he has this year's fewest rushing yards per attempt. However, it is worth noting that his 1.0 rushing yards per attempt would not have been bad in previous years. Daniel Jones has more rushing yards per attempt than Grier, but his projection is lower because he ran more often and was less accurate. Finally, Jarrett Stidham has this year's second-fewest rushing yards per attempt. That, plus his tendency to run when pressured, is why he has our lowest projection.
NFL Teams Are Increasingly Valuing Functional Mobility
The data shows that NFL teams are drafting functionally mobile quarterbacks higher than before. When we wrote our original paper, we used data from 2000 to 2014 to identify functional mobility as a significant predictor of NFL success. In fact, Table 3 shows that a college stats-only model is more correlated with NFL success than draft position is. In this table, we rank the quarterbacks by their stats-only ANY/A and measure correlation with Pearson's r, which ranges from -1 to 1. In addition, the actual ANY/A values are as they were in 2014. Finally, like last year, we set a minimum ANY/A equal to Jimmy Clausen's 3.40.
|Table 3: 2000-2014 Model Results|
|Quarterback||Stats-Only ANY/A||Draft Pos. ANY/A||Actual ANY/A|
|Correlation w/ Actual ANY/A||0.423||0.357|
|Stats-Only's Correlation w/ DP||0.371|
Since we started making predictions in 2015, functional mobility has remained as effective as ever. However, during that time, NFL teams have improved at drafting quarterbacks. Table 4 shows that from 2015 to 2018, draft position is significantly more correlated with NFL success than it used to be. Furthermore, the correlation between each quarterback's stats-only ANY/A and draft position ANY/A has increased from 0.371 in our training set to 0.575 in our prediction set. In other words, NFL teams have gotten better at drafting quarterbacks in part by increasingly valuing functional mobility.
|Table 4: 2015-2018 Predictions|
|Quarterback||Stats-Only ANY/A||Draft Pos. ANY/A||Actual ANY/A|
|Correlation w/ Actual ANY/A||0.466||0.565|
|Stats-Only's Correlation w/ DP||0.575|
The Cleveland Browns' QB Curse May Be Over at Last
From the Cleveland Browns' inaugural 1999 season through their winless 2017, they started over 20 different quarterbacks. In addition, those they drafted in the first three rounds performed significantly worse (p = 0.039) than our model projected. The Browns' futility even inspired an infamous jersey with each quarterback's name written on it. Then in 2018, to the surprise of many, they drafted Baker Mayfield with the first pick. While we recommended Sam Darnold, who impressed late in the season, QBASE gave its highest projection to Mayfield. Considering how electrifying Mayfield's rookie year was, the Browns' quarterback curse may finally be over.
Jeremy Rosen is a doctoral student of economics who conducted the majority of this research at Georgetown University. Alexandre Olbrecht is an associate professor of economics at Ramapo College of New Jersey and the Executive Director of the Eastern Economic Association. The views in this column are expressly our own and do not represent the views of Georgetown University, Ramapo College, the State of New Jersey, or the Eastern Economic Association.