The Minnesota Vikings have passed the quarter-pole and then some, but with a short week, it would be difficult to get a comprehensive statistical review of how the Vikings quarterbacks have performed. Regardless, it’s not as if there is a large enough sample size for any of the involved statistics for rigorous analysis, but enough to get an idea of where the Vikings are and what places might need to be fixed.
More than most other positions, the quarterbacks are difficult to really grab a hold of. Most good quarterback statistics become stable around 200 passing attempts, and no one on the roster is even close to that. What can increase the ability to find the “true” ability of the quarterbacks is to modify for opponent strength.
Teddy Bridgewater has thrown against the New Orleans Saints and the Atlanta Falcons, while Matt Cassel (theoretically not on the team because of his injury) has thrown against the Saints, New England Patriots and St. Louis Rams. Christian Ponder has thrown against the Green Bay Packers. Given their performances against those teams, their performances look as follows (ALL non-proprietary scores are modified for opponent, so they are not the same as the box scores found after the games):
|Opponent-Adjusted Quarterback Statistics|
There are a few statistics that are a bit unfamiliar to some—net yards per attempt (NYA) and adjusted net yards per attempt (ANYA) are the first two. The “net” refers to the inclusion of sack yards as negative yards and sacks as (failed) attempts. So a quarterback who threw for 300 yards on 30 attempts would have 10.0 yards per attempt. But if he was sacked 5 times for 30 yards, then he would have 7.7 net yards per attempt (270 yards divided by 35 attempts).
The “adjusted” refers to providing weights for touchdowns and interceptions. Touchdowns are worth 20 yards and interceptions worth negative 45 (because interceptions end any opportunity to score and give the opponent a better opportunity, where a failed touchdown does not)—numbers derived from rigorous statistical modeling.
If that quarterback threw for one touchdown and no interceptions, his adjusted net yards per attempt would rise to 8.3 (290 yards divided by 35 attempts).
At the moment this gives us some context for evaluating the statistical performances of the quarterbacks, though the film performance is another matter entirely. It makes sense to downplay, in a small part, Teddy Bridgewater’s performance against Atlanta, but even after adjusting, Bridgewater looks favorable.
Including his performance against the Saints, Bridgewater about average despite it being his first start and emergency spot play. He is credited with interceptions he doesn’t have, of course, because the teams he played against don’t often get interceptions. In that sense there is an “ease of opponent” penalty. It will average out over time, but for now it is what it is.
He also happens to have been credited touchdowns he doesn’t have for the opposite reason, so that helps. His adjusted net yards per attempt, when adjusting for opponent, is slightly above average, where his passer rating and efficiency rating are slightly below average.
Of note, Ponder has a negative touchdown percentage. Green Bay’s defense has not given up many passing touchdowns relative to the rest of the league and it may be considered a strong passing defense as a result, but the iterative calculation—adjusting the Green Bay defense for the offenses they played, and adjusting those offenses for the defenses they played and so on—reveals that the Green Bay defense may actually be easier than most to throw touchdowns against. The fact that Ponder didn’t throw any means his touchdown percentage is effectively “negative” even if that’s not a real concept.
The average rookie quarterback over the past eight years threw for 5.6 net yards per attempt and 4.2 adjusted net yards per attempt in their first start, nearly a full standard deviation below an average quarterback. Teddy, on the other hand, in his first start and in significant backup play, threw at league average.
Though he has many games to go until he can be judged a star-worthy quarterback, it bears more good news than bad. Of those whose careers can be judged, the best of those rookie quarterbacks was Matt Ryan. The top of the lists features some busts (Mark Sanchez and Tim Tebow, to name two significant ones), but more good than bad (though there have been very few quarterbacks in those years whose careers can be judged and entered after 2007). The best of those who we have some idea of are Cam Newton and Andrew Luck.
The bottom of the list contains no quarterbacks who have had good careers, with Ryan Lindley, Brandon Weeden, Josh Johnson, Jamarcus Russell and Keith Null.
The best of the worst was Joe Flacco, but there’s significant room between him and another hit on the list, suggesting that even in their first start, good quarterbacks outpeform their rookie counterparts—and Teddy did so with aplomb.
Christian Ponder and Matt Cassel’s opponent-adjusted net yards and adjusted net yards stats are abysmal. No quarterback was so bad last year that they threw as low as 3.5 adjusted net yards per attempt with at least 200 attempts. In 2012, one quarterback managed to do that (Ryan Lindley), but not so in 2011. In 2010, Jimmy Clausen repeated the feat, but was the only one.
The point is that 3.5 adjusted net yards per attempt is so fathomably bad that it’s important to emphasize it. The fact that it’s below Christian Ponder’s by a whole yard is hard to believe, especially as his would have nearly ranked last overall last year, barely edging out Geno Smith and Terrelle Pryor, but still below Brandon Weeden.
After those two on the table is passer rating (noted as “Rate”), which most people are familiar with.
The third unfamiliar statistic is an efficiency rating metric I’ve been experimenting with, but not one I’m entirely comfortable with yet. As it is, think of it as something between adjusted net yards per attempt and passer rating in that it provides some value to completions, but not as much as the traditional passer rating formula—it is designed to be entirely predictive instead of retrodictive; it measures how a quarterback’s past performance can translate into future wins—as opposed to passer rating, which measures only how good a quarterback’s past performance has been. Average is 100, and every standard deviation above or below is worth 15 points in the rating system.
For context, the top quarterbacks, in order, are Peyton Manning, Russell Wilson, Philip Rivers, Aaron Rodgers and Andrew Luck. The bottom quarterbacks/teams are in order from worst to least-worst, Geno Smith, Derek Carr/Matt Schaub, Nick Foles and Tom Brady. Minnesota’s three-quarterback contingent fits between Foles and the Raiders’ quarterbacks if you wanted to include them, but it’s clear that Bridgewater is distinct from the other two based on his limited outings.
Aside: though much has been said of Tom Brady’s decline, it may be a bit much to call him a bottom five quarterback—as his excellent performance last night showed. It should also be noted that a) there are not nearly enough attempts to really call this an effective rating yet and b) like almost all quarterback ratings, it is dependent on team play.
Last night was not enough to wash out the previous four games, some of which were truly bad, but it did move him up from third-to-last to fifth-to-last. The biggest boost comes from the fact that the Cincinnati defense is so strong.
Nevertheless, one could call it another nail in the Brady coffin, if not an entirely credible one. Also relevant, Tom Brady ranks 24th in Adjusted Net Yards per Attempt (28th before the game) and 22nd in passer rating (29th before last night), so it’s not new information, simply surprising.
Along with that are ProFootballFocus grades, which have developed some very mixed reviews from football fans, as well as some data from Advanced Football Analytics and Football Outsiders. WPA/G and EPA/P refer to changes in “game state.”
Those game states are the result of years of analysis of every football game played in Advanced Football Analytics’ database. Every “game state” (down, distance, point differential, time remaining, timeouts remaining, etc.) produces an “expected point total” (i.e. teams will, on average, score this many points in this game state) and “win probability” (teams will, on average, win this percent of the time in the same game state).
Players who contribute to changing the game state (each play changes the game state) get assigned credit for the difference in expected points and win probability. Comeback victories tend to earn enormous win probability added, while a game winning drive will earn the same expected point total as an identical such drive earlier in the game.
Bridgewater has, by himself, has added a third of a win per game. That ranks 6th of 39 qualifying quarterbacks, behind Joe Flacco (who has put together a surprising performance this year, and ranks highly in at least one other comprehensive quarterback rating system), Peyton Manning, Carson Palmer, Drew Stanton and Philip Rivers.
In expected points per play, he ranks third overall, behind Derek Anderson and Andrew Luck (Derek Anderson’s game was very good; 35-49, for 381 yards, three touchdowns, no interceptions for a passer rating of 114.4 and an ESPN Total QBR of 89.3. He ended with 8.6 adjusted net yards per attempt, which would rank third overall over the course of a season).
The last AFA statistic is “success rate” (marked as “SR” on the table) which has different definitions for different groups, but for AFA is just the rate at which a quarterback adds expected points—every play they add a success is counted as a one, while failures are zeros, regardless of the magnitude of the successes and failures. Bridgewater’s 50% ranks 20th of 39, which implies that though he was explosive, he was not necessarily consistent (though I suppose no more consistent or inconsistent than the average quarterback).
Related to that are his Pro Football Focus scores (the only one not adjusted for opponent) and his Football Outsiders DVOA score. Both, though again a product of limited sample, are very good. To look at those rookie quarterbacks again, Teddy ranks 9th of 52 quarterbacks.
Above him are Tim Tebow, Robert Griffin, Cam Newton, Kirk Cousins, Andy Dalton, Joe Flacco and Matt Ryan. The quarterbacks whose careers are still questionable or are reasonably good that are below him are Jake Locker, Geno Smith, E.J. Manuel, Matt McGloin, Andrew Luck, Russell Wilson, Ryan Tannehill, Blake Bortles, Derek Carr and Matthew Stafford. Limiting out those whose careers are undecided, that’s only Matt Stafford.
At the bottom with Stafford are Jamarcus Russell, Brandon Weeden, Troy Smith, Ryan Lindley, Nick Foles, Mike Glennon and Blaine Gabbert.
The known hit rate above Teddy is far better than below Teddy, and that’s a positive (though weakly positive) sign as well.
As for Christian Ponder and Matt Cassel, Ponder ranks in the middle and Cassel ranked a little below that in their first debuts (though Cassel’s debut was not as a rookie). As for right now… not only have both performed worse than Teddy in all of their limited snaps, they’ve been far below league performance standards.
Though their passer rating and efficiency ratings are merely “below average” by about a standard deviation (which would mean ranked at about the 16th percentile of quarterbacks—where Geno Smith ranks now and where E.J. Manuel and Chad Henne were last year).
While Cassel’s DVOA is nearly unbelievably bad, Ponder’s hasn’t been measured yet (Football Outsiders updates at the end of every game week). As for their AFA scores, Ponder and Cassel have contributed negatively to their team.
That’s not an enormous surprise or huge shock; it means that they contributed to plays that have net hurt the teams’ position compared to the average contribution on an average play. The split isn’t even (it is not the case that half the quarterbacks have a negative EP or WP) because coaches will tend to avoid liability for a bad quarterback by turning to a running back or calling conservative plays, which changes the total, but it IS the case that half the quarterbacks will be above or below the 50% line when it comes to success rate.
In WPA/G, Ponder ranks 33rd of 39, while Cassel ranks 29th. In EPA/P, Ponder ranks dead last. Cassel of course fares better, six spots above him. In success rate, they rank 38th and 37th respectively.
This means that though they fail more than they succeed, Cassel’s failures have at least been slightly less impactful per play than Ponder’s. On the other hand, Bridgewater’s limited sample has seen as many failures as successes, but the magnitude of his successes have been far greater than his failures, or indeed most quarterbacks.
Of all the AFA stats, success rate tends to be the most sustainable going forward, followed by EPA/P. WPA is so sensitive to game context, that high leverage plays tend to be emphasized far more than most plays, and it arbitrarily increases the impact of a random sample. Regardless, it does at least tell a better story of what happened than what will happen.
Despite the new offensive system, there hasn’t been a sustainable increase in either the play of Matt Cassel or Christian Ponder, at least according to the statistics. Film reviews and subjective evaluations (like Pro Football Focus) agree, and they all paint a harsh picture.
In particular, Matt Cassel’s inability to avoid interceptions or make completions despite a relative ease of opponent in those areas and Christian Ponder’s inability to create touchdowns or produce yards in spite of the strength of his opponent.
For what it’s worth, Bridgewater’s ability to create yards isn’t an issue, nor are his lack of touchdowns (his opponents are already adept at limiting those) quite yet—he simply needs to be more consistent in creating completions (an average quarterback would have a higher completion rate against these opponents).
Normally, the completion percentage of a quarterback isn’t of much concern to most quarterback statisticians, but there’s some indication that it matters more than most people may have thought, even with an equivalent amount of yards per attempt—creating consistent gains is often more important than explosive gains, though both are critical (and when in conflict, more yards per attempt are often better than creating completions).
Also useful are the splits that differentiate Bridgewater from Cassel and Ponder. Using PFF splits against pressure, Cassel and Ponder rank 36th and 34th respectively (unadjusted for opponent) in adjusted net yards per attempt, of 37 qualifying quarterbacks. Bridgewater ranks 23rd, right above Tom Brady. In passer rating, Bridgewater ranks 17th, right above Cam Newton. Ponder and Cassel rank 33rd and 35th.
In accuracy under pressure (which assumes all drops are completions, but takes away throwaways, throws while hit, spikes and so on), Matt Cassel exceeds Ponder and Bridgewater. Cassel hit his target 60.9 percent of the time under pressure, good for 17th in the league, Bridgewater and Ponder are virtually tied, each hitting their targets under pressure 55.6 percent of the time.
This suggests that Teddy Bridgewater, whose likelihood of hitting a target is just below league average when pressure is involved (in this one sample), is more aggressive about his targeting (he goes further downfield) and more productive (he avoids interceptions under pressure, at least for now). What we know of him on film bears that out.
In terms of average depth of target, we know from watching that Teddy attacks further downfield than Cassel. The statistics bear it out, using Pro Football Focus’ data. Of all quarterbacks with at least 50 attempts, Cassel ranks dead last at a depth of target of 6.4 yards per targeted pass. Last year, he was closer to league average at 8.9 yards.
Ponder was below him in 2013 at 8.8, and he repeated that in his one showing against the Packers (or close to, at 8.9)—a far cry from his conservative 2012 campaign (ranked last in the league at 6.8).
Teddy is between the two, still more conservative than the average quarterback at 7.5, but not as much as Cassel. When just looking at his Atlanta game, his average depth of target was higher at 8.3, somewhat close to, but still below league average.
When he does pass deep (9.8 percent of the time, compared to a league average of 12.1 percent, Cassel’s 8.1 percent or Ponder’s 11.4 percent), he’s been accurate 40 percent of the time (he’s only had five attempts so far, so the math is clean). League average accuracy is 39.6 percent, while Cassel (0 percent) and Ponder (20 percent) fare predictably worse in their small samples.
Helping Bridgewater is the fact that he threw off of play action in his two games more than nearly any other quarterback, with only Russell Wilson, Chad Henne, Nick Foles and Aaron Rodgers with a higher percentage. Bridgewater has done it 30.5 percent of the time, while the rest of the league does it on 27.5 percent of the time—not a huge difference when considering Bridgewater’s low sample.
Interestingly, Bridgewater has been schemed to do it more often than Ponder (20.8 percent) or Cassel (22.2 percent). I suspect it’s a combination of both his general skill at deception, quick footwork and the fact that play-action passing is generally used to protect rookies. Bridgewater’s completion percentage rises 8.8 percent and his yards per attempt increases by 4.4 yards (the second-most of anyone with as many attempts, behind Tony Romo at +4.5).
Bridgewater may be more adept at play action because he is better in a muddy pocket as well. Aside from his clearly superior performance, so far, under pressure, he happens to do well with more time in the pocket than the other quarterbacks. Though on average he and Christian Ponder have spent similar amounts of time in the pocket (averaging 2.85 and 2.81 seconds—Cassel was at 2.63), Bridgewater’s accuracy when the pocket time needed to be extended was significantly higher.
Bridgewater hit league average in completion rating (56 percent) when in the pocket for 2.6 seconds or longer, and did better than the average NFL quarterback in passer rating, per Pro Football Focus. On average, quarterbacks threw for a passer rating of 80.7, while Bridgewater was over ten points higher at 90.8.
Ponder was accurate on 52.2 percent of tries for a passer rating of 35.6, and Cassel had a slightly higher passer rating (36.5) with a lower completion percentage (40).
At any rate, there’s some shallow statistical indication that the Vikings should be happy with what Bridgewater has provided, even when we can acknowledge that the Atlanta Falcons have a poor pass rush—Bridgewater still exceeds their average opposing quarterback, and we know some specific ways he can win. He seems to excel under pressure, and the Vikings can take advantage of that by running more time-consuming or late-developing plays, like play-action shots.
Given his known general ability on the run, his skills mesh together well in that regard.
Further, though he hasn’t pushed downfield as well as other rookie quarterbacks in his class—at least in college—early returns suggest he may want to continue attempting deep shots.
And though the Falcons have been up against Mike Glennon (and Josh McCown) in particularly poor form, they also played against a high-performing Andy Dalton (as unusual as that is), an extremely well-performing Eli Manning and Drew Brees. Further, the statistics don’t change when re-adjusted for further opponents (the defenses that all those quarterbacks played, and the quarterbacks THOSE defenses have played and so on)—Teddy Bridgewater has been genuinely good.
As there are more snaps against more teams, a clearer picture will emerge of Teddy Bridgewater’s first season and his most glaring weaknesses (and reliable strengths) at the end of the year.
In the coming week, there will be more statistical looks at other positions, though not nearly as exhaustive.