Skip to main content
Menu
Marcus Mariota Is Projected To Be Better Than Jameis Winston

We’ve heard the debate for so long that its edges have nearly gone dull: Jameis Winston or Marcus Mariota? Who should go first?

If there were a formula for how to choose a No. 1 quarterback in the NFL draft, Tom Brady would have gone No. 1 overall and JaMarcus Russell would have been lucky to be selected 199th. This stuff is hard, especially when you consider the stakes. Missing on a first-round quarterback can set a franchise back for years. Because of the “boom or bust” nature of the position, selecting a quarterback, particularly in the first round, is riskier than selecting a player in any other position.

What, then, to do about Winston versus Mariota this Thursday? Build a model, of course. This year, ESPN’s Production Analytics crew created a QB model to help teams reduce the risk of drafting the wrong quarterback. Like all models, this one had a few outliers, but it would have predicted that Andrew Luck would be the top QB in the last three draft classes, that Russell Wilson would be far better than his third-round grade, and that first-rounders Brandon Weeden and EJ Manuel would be below-average quarterbacks — and of course that’s without using those years of data to fit the model.1 The model’s opinion on this year’s top two: Mariota — not Winston — is the top prospect.

The goal of the model is to predict a player’s Total Quarterback Rating over his first four years in the league, which is generally the length of his rookie contract. The main inputs into the projections were a player’s college stats (adjusted for defenses faced),2 combine/physical measurements, scout grades and play-type frequencies in college. After determining which factors mattered most,3 the model projected a player’s NFL success in four categories — on passing plays, on running plays, how many sacks he’s likely to take, and how many penalties he’s likely to incur — over his first four seasons. These play-type projections and how often each play is expected to occur4 were combined to produce the QBR projections.

Not surprisingly, scout grades5 were the most significant predictor of quarterback success. Scouts have the luxury of knowing things that aren’t completely captured in college stats, and they generally do a good job measuring a quarterback’s passing capabilities. Where the scouts fall short, however, is their evaluation of efficient rushing quarterbacks.

The model found that players who were effective rushers were generally undervalued by scouts. That does not mean that every player who runs for 1,000 yards in college will be a good NFL quarterback; rather, the quarterbacks who are efficient runners have an ability to extend drives that serves them well in the NFL. For example, Luck ran for 150 yards in his final college season but was one of the more effective college rushers, converting a first down on 39 percent of his rushes (excluding sacks). Once in the NFL, Luck has been one of the best scramblers in the league and was the most efficient rushing quarterback in 2013.

The added rushing component is also a major reason that Wilson was projected to be one of the top quarterbacks in his class (47.3 projected QBR) but only the eighth-best quarterback by Scouts Inc.

Winston had a better Scouts Inc. grade, but Mariota’s better rushing stats and combine tests helped him beat out Winston in the model’s approach.

PLAYER FOUR-YEAR PROJECTED TOTAL QBR
Marcus Mariota 64.1
Jameis Winston 60.8
Brett Hundley 39.8
Bryce Petty 24.5
Garrett Grayson 17.0
Rakeem Cato 15.8
Cody Fajardo 12.0
Anthony Boone 10.2
Grant Hedrick 10.1
Blake Sims 9.7
Sean Mannion 9.1
Taylor Heinicke 8.8
Shane Carden 8.3
Brandon Bridge 5.3
Connor Halliday 2.7

The model did not explicitly take into account the sexual assault allegation made against Winston or his off-field transgressions — although those were likely baked into the scout grades, which were a part of the model.

Out of the 67 players evaluated in the 2012 to 2015 draft classes, Mariota was the most efficient rusher on a per-play basis. Excluding sacks, he averaged 9.8 yards per rush attempt in his college career and gained a first down on 41 percent of those plays. Winston, on the other hand, ranked 29th in per-play rushing efficiency and gained a first down on 24 percent of his carries.

Mariota and Winston are each projected to be above-average quarterbacks in their first four years in the league, but they are far from the elite level that the model projected for Luck (79.2 projected QBR) when he came out of college. Their four-year projections are closer to the expectations for Robert Griffin III entering the 2012 draft.

After Mariota and Winston, Brett Hundley is expected to be the third-best quarterback in the class, but there is a clear gap between the top two QBs and the rest of the prospects. Like Mariota, Hundley is athletic and ranks in the top 10 in per-play rushing. Other models look favorably upon Hundley, making many believe he could be the sleeper of the 2015 draft class.

The rest of the 2015 QB class is expected to perform at the level of replacement level QBs.6

No one model can perfectly predict NFL quarterback success, but looking at what numbers have mattered in the past tells us that we will be lucky to have three players come out of this draft class who will have long-term NFL success.

Footnotes

  1. The model is based on 122 quarterbacks from the 2005 to 2011 NFL drafts.

  2. Expected points added (EPA), the backbone of QBR and many other NFL models, was used to measure quarterback success in college. We adjusted for opposing defenses faced to measure a QB’s success in college, and EPA/play was used as an output to project QBR. Because of small sample sizes, stabilization techniques were also applied to obtain less volatile measures of success.

  3. To do this, we used a censored regression, using Bayesian information criteria as the variable selection criteria. Prospects that never played in the NFL or players that performed worse than replacement level were left censored.

  4. A random forest model was used to project NFL play-type frequencies using the same set of predictor variables from the censored regression. Only players that had at least 300 NFL action plays were used to fit this model.

  5. Mel Kiper and Scouts Inc. were used for the grades.

  6. Defined as a player with a QBR of 25 or lower.

Sharon B. Katz is an analytics writer for ESPN’s Stats & Info Group.

Zach Bradshaw is an analytics specialist for ESPN Stats and Information.

Comments