"Teddy Bridgewater's numbers are down." "Teddy Bridgewater simply isn't producing." "Teddy Bridgewater isn't playing well, and they don't need him to." "Teddy Bridgewater has simply been disappointing."
You would have to be pretty insulated from media not to have heard some versions of each of those statements. I think a lot of people agree that Bridgewater is statistically pretty lackluster. But when we say that, what do we mean?
More importantly, which statistics matter?
If you're not a fan of using numbers to evaluate players, I already did "eye test" work on the three quarterbacks that have had significant playing time from Teddy's class. You can find it here.
If you're at least interested in the statistical case for (or against) Bridgewater, read on—some of the statistics are more scouting than raw data, anyway.
After a rookie season where Teddy Bridgewater finished strong, there's an unusual amount of disagreement about where the Vikings stand with their young signal-caller. Fans of the Oakland Raiders and Jacksonville Jaguars are pretty happy with their quarterbacks and with good reason—Carr ranks 5th in touchdown percentage while Bortles ranks 8th. Teddy? 32nd.
Statistically, Bridgewater is lackluster compared to his peers. Even Johnny Manziel, who's started four games this year and thrown 159 passes total, has only three fewer total touchdown passes than Bridgewater.
The easiest conclusion, of course, is that Bridgewater simply isn't as good as them. Perhaps he's more valuable a franchise asset than mercurial Manziel, but if he's not as good as them, then it's an issue—not because it reflects bad decisionmaking from Spielman (it would, if true), but because these players will be the peers he competes against for titles (perhaps with a stroke of luck, those players being in the AFC is a big help), along with Russell Wilson, Andrew Luck and so on.
But for the most part, we have to figure out what we're measuring.
Drilling down on important attributes or data often sounds stupid and reductive, but ultimately it's a great focusing tool. So, when we say that the goal of any quarterback is to win games, we sound pedantic but it does allow us to narrow down a wide range of statistics to what they can tell us about winning.
Touchdown rate is awesome—you want your QB to throw completions in the end zone—but if touchdowns are replaceable (either through the run game or by scoring three on every single drive)—then they are not as telling as another potential statistic might be, like the ability to create yards through the air.
There are also two components to figuring out which statistics are the best ones to use when evaluating: statistics that forecast and statistics that explain. Passer rating is fantastic at explaining what happened, in part because of the heavy emphasis it places on interceptions. Those lose games.
But it's not as great as other statistics are at forecasting; interceptions are really low-frequency events that can fluctuate a lot (the best example may be Eli Manning, whose interception rate has oscillated from 2.0 this year, a top-ten figure, to 4.9 percent, which has led the league).
Luckily, someone has done the grunt work to figure out, in-season, which statistics best predict success. There's also been work done to figure out which statistics best predict success for young quarterbacks over the course of a career.
We have some cool data to work with that's not in traditional NFL box scores, too—average depth of target, accuracy rate (as opposed to completion rate), Pro Football Focus grades and so on.
So I decided to expand on the work that Chase Stuart did at Football Perspective by testing the new data in the same way he did, by taking half-sample seasons and testing it against the other half of the season for predicting in-season future performance (performance here means "wins"). Jason Lisk did work predicting the career performance of sophomore-year players, and I attempted to sort of follow it up by testing the same data (full season this time) for young quarterbacks against the best measures of in-season performance two and three years later.
For predicting in-season performance, as determined by Stuart, were ESPN's Total QBR, adjusted net yards per attempt and passer rating. After running similar tests, my data came to the same conclusion—except that two measures of performance beat them all when I added them in: Football Outsiders' DVOA and PFF's grades.
This is pretty surprising—there's been a lot of pushback against Pro Football Focus in the past several years, and against Total QBR for a little bit longer. Both have been under fire for "undergrading" Aaron Rodgers on days when his NFL passer rating was well over 100, while "overgrading" statistically lackluster performances.
All of this has to do with proprietary grading processes that theoretically add context from plays and strip away noise—things like dropped interceptions, receivers providing additional YAC and so on. For all the heat they take, they seem to produce a reasonable measure of quality, even if they don't account for performance.
We also have some tools that are supposed to provide context—accuracy rate allows us to get rid of drops, throwaways, spiked passes and passes batted at the line of scrimmage, and average depth of target (aDOT) allows us to give context to accuracy rate. Normally, people would look at yards per completion or something similar, but given that aDOT and yards per completion don't have a linear relationship, aDOT is a bit better at providing us with a crude understanding of how much a quarterback slings it.
Because accuracy rate and aDOT have an intrinsic relationship—it is harder to be accurate the further you throw it down the field—one can come up with a few different measures that use the two metrics. There's an improved form of YPA if you simply multiply the two, but it doesn't predict wins as well as if you weight the two separately.
By using different weights and figuring out which combination of the two metrics best predicts performance, we can come up with an adjusted yards per attempt that takes into account some degree of system change and isolates the quarterback a little more from the effects of their receivers.
There could be some overlap between that method and DVOA, depending on how they construct the metric, but it's pretty clear that DVOA does an excellent job matching quarterback play to win prediction.
To that end, we have the following tools that help predict in-season performance:
PFF grade was turned into a rate stat, represented as that sample's grade prorated to 500 snaps. If you're familiar with Teddy Bridgewater's overall performance in these various statistical categories, you'll be able to identify the issue that I'll touch on later when it comes to statistically evaluating his performance.
Typically, a quarterback without much hope will produce pretty consistently poor performances in all of these metrics. One good example is Christian Ponder in 2012, who produced the following ranks in each category:
Going through the second-year statistics for other washouts produces pretty consistently similar results:
The only truly ambiguous one is Freeman, who produced such a stark difference between QBR and his orthodox production that he would, by himself, be worthy of a case study. We of course saw the ultimate conclusion of Freeman's fall from grace, and it seems like what impacted him (either mental illness or simply not caring about football) doesn't seem to be a huge risk for the current crop of young passers.
And remember, this group of statistics is designed to measure in-year quality without necessarily having an eye on long-term future (although, of course, many of them are pretty good for that as well).
Aside from Freeman, there is some question about Robert Griffin, who may have had more issues with injury than anything else, though it didn't really take statistics to determine whether or not Griffin was a lost cause; another relatively unique case. Chad Henne is different as well. I wouldn't say he's a unique case, just an example of how everything exists with uncertainty and not everything is open-and-shut.
As for the quarterbacks from those same classes who succeeded (or merely did not bust), there is less uniform agreement—though much stronger overall numbers:
I didn't include Kaepernick or Tebow because their second years were functionally their first, though Kaepernick's numbers were all very good except for the adjusted YPA and Tebow's were all very bad. At any rate, the one quarterback that people seem to disagree on or have a difficult read on—Tannehill—is the one that seems to straddle the 25th-ranking, the worst of the "not bust" group.
No one seems to have called for replacing Tannehill, but whether or not he's a quality quarterback or the answer in the future is up in the air.
On the other hand, Stafford is not a bust but certainly not on the same trajectory as Cam Newton, Andy Dalton and so on—and his ranks are almost identically as bad as Tannehill's while substantially better than the bust group.
It shouldn't be much surprise that the statistics that best predict how a quarterback's team is going to do in-season also provide us with a reasonable proxy for whether or not they'll be good several years down the road. That said, let's look at the young signal-callers from Teddy's class:
It's a confusing story. No one has an unambiguous stamp of approval from the performance indicator statistics, though it's obvious both to these numbers and our eyes that Derek Carr is performing better than the other two, with Teddy mostly following behind and Bortles just behind him (sort of; Teddy beats Bortles in three of the indicators, ties one, and is behind on the other two—and does more in his better indicators than Bortles does in his).
This matches the film work on third down I did for the three passers in terms of how good they are right now, though it is definitely intriguing that Bridgewater is the 6th-best quarterback in the NFL when simply looking at how accurate he is on aimed passes and how far downfield he throws it. The full ranks in this accuracy-adjusted statistic:
|31||Alex D. Smith||6.86|
The big discrepancies between this and traditional yards per attempt are likely due to two factors: drops and throwaways. Bridgewater has the highest percentage of his passes excluded in this analysis because he leads the league in throwaways (a statistic somewhat ameliorated by the fact that he has a somewhat low drop rate), astounding especially given the low number of passes he throws compared to many other passers.
Those drops and throwaways are also probably the reason he beats out Carr and Bortles in a "context" statistic like QBR while matching them in PFF grade—an interesting outcome, given that both sets of grades took a lot of fire for downgrading Bortles' three-touchdown, 250-yard performance this past Sunday (for similar reasons: two dropped interceptions, a fumble, and iffy accuracy).
At any rate, there are other statistics that allow us to predict, to some degree of accuracy, the long-term prospects for quarterbacks. Lisk's research centers around age, and is probably a better way of looking at things than by "year one, year two," etc. The make-or-break year in Lisk's data is 25 years of age, which is some time off for Teddy, but will be next year for Carr and the year after for Bortles.
Regardless, there are some interesting notes:
Most rookies do struggle relative to the league average, at least in most categories. I certainly wouldn't exclude a player just because he threw alot of interceptions as a rookie. Alot of really good quarterbacks struggled with interceptions as rookies, moreso than Harrington. Only 36.7% of the successes were at or below league average in interceptions as rookies. Though I've previously noted that throwing for a high touchdown percentage at a young age is a positive indicator, a below average touchdown rate as a rookie is not a reason to exclude a player, as most of the rookies who struggled simply played on bad teams. 30% of the successes were at or above league average in touchdowns as rookies.
There is one category, though, where the rookies who became successes did pretty good as a group--yards per attempt. 70% were at or above the league average in this category. Even most of those that were below average were not significantly so. Most of the young quarterbacks who became good starters in the league may have struggled in completing a high rate of passes or in throwing interceptions, but they showed an ability to make plays.
That is an indication of rookie-year performance, not second-year performance, but it does tell us what to look for if we want an early indicator. Bridgewater's YPA was slightly above league average as a rookie, while both Carr and Bortles were well below. That means one of three things:
- Carr and/or Bortles occupy the other 30% (a pretty large group, all things considering)
- Carr and/or Bortles won't be successful
- The research from previous eras of football is not useful for the current era of football
In any case, you may be surprised to know that as rookies, Bridgewater's touchdown rate was higher than either Carr's or Bortles, but Carr threw the ball quite a bit more—so none of the three get the benefit from rookie touchdown rates.
As for second-year passers, or 25-year-old passers, Lisk almost exclusively relies on yards per attempt, though he doesn't go into whether or not having a high touchdown rate in one's second year or age-25 year is as positive an indicator as it is for rookies. If the conclusion is similar—that a low rate isn't bad, but a high rate is good, that would align with my own research which shows a strong relationship two years out for those with good touchdown rates (like Carr and Bortles) but provides no information about those with low touchdown rates (like Bridgewater).
If we know that YPA is a good indicator, then it stands to reason that having a good accuracy-adjusted yards per attempt is a good indicator, and my tests showed exactly that—that it is a better indicator than raw YPA. In determining performance for young players two years out, the strongest statistics were PFF grade (prorated to 500 snaps), QBR, DVOA and accuracy-adjusted YPA (too much noise for young passers: passer rating and ANYA). Let's look once more at how busts and successes did in those categories.
Stripping out ANYA and passer rating does make it easier to predict the future because it clears up some of the complications. Freeman's enigma isn't entirely resolved, and one could argue that a "plus" touchdown rate as a positive indicator probably could not compete with the negative indicators but that's a murky case at best.
Let's leave Freeman where he was: a unique case in quarterback development.
There is one success that generally was below average (Stafford), but the rest are either average in the categories presented or in Tannehill's case, just a smidge below it. That brings us to Bridgewater, Carr and Bortles:
A confusing class. Bridgewater is above average in three of the four linearly indicative categories (by linearly indicative, I mean "bad score predicts a bad outcome, while a good score predicts a good outcome") even if he falls well behind in the "green light" category of touchdown rate.
Carr blows the first two indicators out of the water, but has some slightly concerning issues with the second set of two (though he does better in those two categories than the passers who populate the bust category). His prodigious touchdown rate boosts his résumé. The quibble over those second two is probably not enough to cast much doubt on his future prospects.
The distribution of Bortles' statistical characteristics is very Freeman-like, and I don't think I mean it in a bad way. There are two bad indicators (accuracy-adjusted YPA and QBR), one indicator that's sort of bad (DVOA) and a good indicator (PFF score). It would be almost identical to Freeman if one flipped DVOA and accuracy-adjusted YPA, especially with the positive touchdown indicator.
Freeman had the highest ranks among the busts in the leading indicators, and Bortles' ranks are below all the successes, but that high touchdown rate is unlikely to be fluky (or rather, he's a volatile player, much like Stafford, and variance is part of his makeupâmeaning we will have fluky results, but that those are entirely expected).
If you privilege certain kinds of information over others, then you can do the rankings yourselves. Like I mentioned at the top, I did "eye test" work a few weeks ago and came up with the following ranks: 1) Derek Carr, 2) Teddy Bridgewater, 3) Blake Bortles. For Carr and Bridgewater, I was positive we'd see franchise-level play given the traits I had seen from them this year. For Bortles, I was more open-ended.
If you appreciate a combination of the eye test and statistical indicators, then ESPN's Total QBR and PFF's grade are for you; they involve a high degree of subjective evaluation and combine it with statistical measures (though Total QBR has a higher influence of statistical data influencing their outputs). Teddy's average may be highest (marginally) in that analysis, which would produce a rank of 1) Teddy Bridgewater, 2) Derek Carr, 3) Blake Bortles.
If that seems off, that's fine—I don't really have an issue with emphasizing high ranks for young players and de-emphasizing low ones (e.g. ranking highly means more than ranking lowly). If so, Carr beats out Teddy. In either case, all three passers could be said to have a bright future.
If we move by purely statistical predictors of success, like DVOA or accuracy-adjusted YPA, then it gets confusing. Carr ranks 7th in one and 24th the other, while Teddy flips the script; he ranks 20th and 6th in those categories. Though that does mean Teddy ranks higher on average, I think one could reasonably point to Carr's touchdown rate as a reason his statistical case is stronger. Bortles is clearly third in that kind of analysis, probably even after incorporating his touchdown rate.
And remember, there's a sort of rank-order to the value of these predictors. By far, the best predictor were Pro Football Focus grades, and they had a correlation to future winning above 0.5, which is great—although I wouldn't read into it that much, given that the sample size of young quarterbacks who played in the PFF era is pretty low. But if one wanted to look at the four indicators of future performance in order of which ones best predict future performance by weighting the best ones highest, then Carr and Teddy would be virtually tied, with a slight advantage for Carr (again, before incorporating touchdown rate), with Bortles well behind.
For fun, we can look at all the second-year passers and give them points based on how strong the indicators are or were.
|14||Robert Griffin III||87|
Treat the above points and numbers more as a crude demonstration of the concept than any seriously rigorous work towards grading second-year passers. It probably makes sense to add bonus points for good touchdown rates as well as age, neither of which I did.
Still, it seems to fit very well—so well that it feels more like overfitting data than producing sustainable evaluation methods (thank you Chad Henne and Josh Freeman for screwing things up to make this look more believable). The process to come up with these numbers didn't ring any alarm bells from that perspective but nevertheless, it's a rough piece of work that should encourage us more than annoy us when it comes to Teddy.
At the very least, if we bemoan Teddy's numbers, we should at least know which numbers to look at.
He's having a good year, according to the numbers.