Don't Worry About Taking a Quarterback in the First Round

The Vikings are in a less-than-enviable position at the eighth pick in the draft, and it seems like they have three options: trading up to grab a premier signal-caller, grab a defensive player, or trade down and pair their quarterback with another pick.

I don't have that much of a problem with the now-unpopular option of standing pat and grabbing a quarterback.

Chris pointed out, with Phil Mackey's research, that grabbing the fourth quarterback has been a largely unsuccessful proposition, with a seven percent success rate.

With all due respect to Mackey and Chris, I think that's an improper use of the data. Between 1990 and 2010, the fourth quarterback has been taken in the fifth round twice, the fourth round three times, the third round five times and the second round eight times.

Being the fourth-best quarterback in a particular draft doesn't tell you anything about his relative skill level in the same way being the best quarterback in the draft usually does, and that's not completely fair, either—in 1996, the first quarterback was selected in the second round. In 1988 (not in the dataset I looked at), the first quarterback selected was in the third round.

The 1996 second-round selection (Tony Banks) was not particularly good, and bounced around five teams before retiring, accumulating a total of 78 starts, but no playoff appearances as a starter. The 1988 third-round selection was actually pretty good, once he transitioned to full-time punter.

Similarly, the fourth quarterback selected can vary in success by round as well. Between 1990 and 2010, the fourth quarterback selected has been successful only two out of all 21 drafts, but one of those times was as a first-rounder (out of three drafts in the set where the fourth quarterback selected was in the first round).

Generally speaking, the pick a quarterback is selected at has a far better bearing on whether or not they'll be successful than the order of position in which they were selected (or even the round). There are a few ways to look at it, the first of which is not by round, but by Value Groups.

What's interesting is that there is nothing intuitive about a talent grouping of 32 (though in fairness, there's nothing intuitive about any static number), which means that grouping talent into 'rounds' of 32 may make sense mechanistically (there is no other way to give every team a shot at drafting before another team drafts twice), it doesn't make sense from a talent evaluation standpoint.

That's why you'll hear scouts say things like "I don't think there's a top-five talent in the draft," despite the fact that there definitively has to be five "top-five" talents. Scouts will also say things like "there are only fifteen first-rounders in the draft," which also runs afoul of the fact of the draft, where there are necessarily 32 first-rounders (barring a penalty taking picks away).

It really doesn't make sense that the top pick in the draft will be put in the same talent category as the last pick of the first round, because the logarithmic distribution of talent in the draft implies that the difference there is far larger than the difference between the last pick of the first round and the last pick of the second round.

You have to draw the line, somewhere though, if you're going to put the data into bins. Luckily, DraftMetrics (a website that had been immensely helpful, but no longer seems to exist) constructed Value Groups that more naturally judge the talent of a particular pick (the paper is hosted by the Wayback Machine, here). The website broadly determines value through a combination of metrics like total number of starts, postseason honors and "starting velocity," which looks at how long it takes a pick to become a starter.

When applied to individual players, the "value" metric isn't great, but when applied to broad groups of data, it's fantastic. They determined that in terms of performance, the first "round" is really picks 1-13. Groups are as follows:

VG 1: Picks 1-13
VG 2: Picks 14-28
VG 3: Picks 29-48
VG 4: Picks 49-74
VG 5: Picks 75-114
VG 6: Picks 115-200
VG 7: Picks 200+

They did this study over two different sets of multiple years and came to the same conclusions: the Value Groups were a much better predictor of success than rounds.

I'm not convinced of some of their conclusions (that the first pick and the seventh pick are virtually identical in value), but in terms of grouping players, this is very good insofar as it predicts success even if it doesn't predict upside or worth.

Determining value is difficult, and Pro-Football-Reference uses a marker called "Approximate Value," which takes into account post-season honors (like Pro Bowls, All-Pro, etc), starts and various position-specific statistics. It's very good, but at the individual level there are of course outliers (Jeff Saturday's AV was 11 in 2012, for example, while John Sullivan's was 8). Nevertheless, it's the best position-neutral way of comparing players across eras and positions.

Using that, you can also look at value by each pick.

In order to account for spikes in the data (even with 21 data points, there are random spikes), I also used a moving average to smooth some things out.

The bump at the bottom of the third round to the middle of the fourth round could be an area of value to target, incidentally, whereas the bottom of the first (excepting the 32nd pick, which has almost always beaten the 31st and 33rd pick in value) seems to be a pit of value.

Regardless, it makes the most sense to linearize it, which provides you with sort of a draft value chart with a more modest dropoff than the traditional chart, although less so than other, more rigorous methods to produce an AV-specific chart.

All that aside, it's clear that the pick, not the order of a specific position, determines the likelihood of success of a particular player. I've found the same when it comes to quarterbacks. That makes sense; had Andrew Luck decided to declare for the 2011 draft, he would have been the first pick. That wouldn't make Cam Newton any less successful as the second quarterback, nor would Brandon Weeden have been a better player had he been the third QB off the board.

But viewing quarterbacks by what order they pick implies a very high degree of difference in chances of success—moving a QB up or down the order changes the projected probabilities by wild swings of over twenty percent, per Mackey's grades. Moving a passer up and down by a few picks doesn't do that in the slightest.

When determining whether or not to select a quarterback, it is absolutely incomplete to make that determination by comparing equally the QBs drafted in the fifth round to those in the first.

Draft Metrics' "value groups" align with the AV chart quite well, incidentally. Surprisingly, actually, given that different definitions of value were used.

Using a simple metric to produce an initial dataset produces an interesting data set that only requires a little bit of back-end modification. By grading all quarterbacks with at least twenty starts and a career adjusted net yards per attempt (which is a metric where one can lose "yards" from interceptions and sacks, but gain them on touchdowns) higher than the average league ANY/A of their careers, you can get a rough sense of "success."

I modified the list somewhat, so that Drew Bledsoe, Matt Stafford and Joe Flacco were "successes" while Brian Griese and Aaron Brooks were not. This ended up creating 31 successes between 1990 and 2010, with 235 unsuccessful starting quarterbacks. Note that in the past twenty years, that's fewer successes than the number of teams in the league, which of course emphasizes how difficult it is to find a quarterback.

The degree to which they were successful, or whether or not they were successful relative to their expectations (certainly Griese and Brooks were) is irrelevant. Instead, I used the Andy Dalton line proposed by Big Cat Country (or rather a modified version of it) to figure out borderline cases: would I rather have Andy Dalton than the quarterback in question?

All of this to figure out the answer to the larger question: in a generalized situation, would I be OK taking the fourth quarterback overall with the eighth pick? For that to be the question, there are certain natural givens and certain required givens. Natural givens include the fact that for a team to consider selecting the fourth quarterback, three other quarterbacks necessarily have to have been taken.

In the dataset, three quarterbacks had been the third QB taken in the top thirteen picks (Akili Smith, Mark Sanchez, Jay Cutler and Ben Roethlisberger) and two of them (Smith and Sanchez) had been taken in the first seven picks. The 1999 draft is interesting, by the way. The first three picks were Tim Couch, Donovan McNabb and Akili Smith.

In those two drafts, the fourth quarterbacks selected were Daunte Culpepper and Charlie Frye (picked in the third round). That means the only instance of a quarterback selected in the first thirteen picks was a success (before his injury).

More importantly, because value is more stable relative to a universe of all draft-eligible than it is to the universe of the top available quarterbacks, it makes sense to take a chance at a quarterback when their value relative to the class is much, much higher than other players (which is not to say "value relative to other top-five considerations").

Further, It might be fair to assume that the fourth-best quarterback in this draft would be gone by the beginning of the second round. ESPN and CBS have been ranking players in publicly available archives since 2004 and 2005 respectively. Quarterbacks ranked between 20th and 32nd tend to be overdrafted by about eight picks. It's no surprise that quarterback picks are not distributed evenly, like they would presumably between other positions—it's the most important spot on a football team.

Other positions are distributed a bit more randomly:

Which of course suggests that QBs are overvalued, a fact we basically see in practice. If you want your quarterback and he is purely valued around pick 22, you'll need to pick him around pick 8-12 in order to get him. This is further magnified by the fact that CBS and ESPN already grade positions with a little positional value thrown in. They will never have the nation's top long snapper, kicker or punter at the top, and fullbacks will never be given a first-round grade.

There are more than four teams that could be eyeing a quarterback. Passing on one in the first could mean another team grabs that quarterback—be they teams with ambiguous quarterback issues like the Tennessee Titans or St. Louis Rams or teams with clear long-term needs like the Kansas City Chiefs or Arizona Cardinals. Further, scenarios involving trading back just a few picks doesn't preclude any of those teams from trading up to get a first-round quarterback instead of second-round QB.

Quarterbacks selected in the first Value Group (Picks 1-13) have a success rate of 42.4%, while those in the third Value Group, at the top of the second round (Picks 29-48) have a success rate of 14%. That means finding a QB in the first thirteen picks will be better than finding a QB at the top of the second round by increasing the odds three-fold that he will succeed in the NFL. Given that fewer than half of the best QBs selected succeed, this is non-trivial.

For those concerned about success rate by pick by order, QBs selected fourth in the first round (or in the first group, or in another more Vikings-specific pick group of QBs from picks 8-16), the rate is 100%, because there has only been one: Daunte Culpepper.

The 5th quarterback selected in the third Value Group has a success rate of 0% (there has only been one such instance, Drew Stanton). The 5th quarterback, when looked at across both rounds has still been successful 0% of the time in the data set (there were only two in addition to Stanton: Tarvaris Jackson and Cade McNown).

The order+Value Group analysis is limited, though, mostly for the reasons given above on why "order" is a bad way to look at quarterbacks. The second reason, of course, is extremely limited sample size. It is more of a curiosity to combine the two methods than an honest analysis.

Naturally, this doesn't mean that selecting a particular quarterback (say, Derek Carr) earlier rather than later will increase his odds of succeeding. Had the Jets selected Greg McElroy in the first round instead of the seventh, he would be no better (there's an argument that it would have meant more reps, so in a sense, he would strictly be better than he is now, but not meaningfully so).

Even so, there's any number of teams that could look to investing in a first-round quarterback that might be willing to stand pat or trade up to do it. By my count, there are at least six teams who select in the first round after the Vikings do that might be willing to part with a first-round pick (though two of them also happen to select before the Vikings do as well). Of the six teams that could select a QB before the Vikings do at pick eight, one or two of them could take a stab before the Vikings could at pick forty.

As an aside, there isn't a certainty that the three teams slated to pick a quarterback by pick eight will do so. No team has been functionally guaranteed by the media to select a quarterback like Indianapolis in 2012 or Carolina in 2011. Should we give all three teams (who for now are Houston, Jacksonville and Cleveland) and 80% chance to pick a quarterback, that still leaves a 50% chance that the third quarterback will still be available for the Vikings to pick.

There are chances that teams like Oakland, St. Louis and Tampa Bay select a quarterback as well, and these chances are dependent on other picks (which is why the draft is so crazy). So, if ignore that and we arbitrarily assign values for the likelihood of selecting a quarterback (Houston: 75%, St. Louis: 10%, Jacksonville: 90%, Cleveland: 90%, Oakland: 40%, Tampa Bay: 33%) we can find that the likelihood the Vikings have the opportunity to select the fourth quarterback to be nearly 40%. Manipulate those odds and they vary between 30% and 65%.

Should that happen, it is unlikely the Vikings would select a quarterback in the first round at all, but we know that.

Again, ignoring the fact that a QB like Bridgewater dropping would increase the odds of a team selecting a quarterback or trading up to do so, the odds that the Vikings select the third, second or first quarterback are 15%, 2%, or 1%, and can vary between 30/10/1 and 7/1/0.

The key here is that when we talk about "the fourth quarterback selected," we shouldn't compare him to every fourth quarterback selected in drafts, because those don't often analyze the fact that the QB is still being selected (and judged) as a first round talent. It isn't fair to compare Derek Carr (or Blake Bortles or Teddy Bridgewater or Johnny Manziel) to Tee Martin, Danny Kannell, Doug Nussmeier  or Pat Barnes.

No one thinks of those quarterbacks as the prospect that Barnes and Martin were. Instead, better success rate comparisons would be other quarterbacks picked between pick five and pick eleven (success rate: 30%) or, because drafts are closer to auctions with a ceiling on value, picks eight and sixteen (success rate: 50%). These include quarterbacks like Ben Roethlisberger, Daunte Culpepper and Jay Cutler (and less optimistically, Matt Leinart, Cade McNown and Dan McGwire).

Should the fourth passer have the success rate of a quarterback within that range, perhaps around 40%, it would not be completely out of the range of quarterbacks who had been picked in the top three (47% of 21 total quarterbacks).

There are other ways to measure boom-bust rate naturally. The "success rate" criteria implemented is a harsh one. If a quarterback exceeds the average pick value (here measured by AV), then they could be a "boom" while those below that pick value are "busts".

For those who don't want to "reach" on a quarterback, one can look at quarterbacks selected in the fifth to eleventh-pick range to see if they met their pick value. 40% of those quarterbacks met their value (48% for a generalized pick), while quarterbacks in the 8-16 range met their value 66% of the time (again, 48% for a generalized pick).

Of course, not all of those quarterbacks were "reaches," but we can be assured that a high percentage of quarterbacks selected in the first half of the draft are, even in drafts without many exciting quarterbacks (like last year's).

Perhaps the best argument that the Vikings should pick a first-round quarterback, should one of that talent still be available at number eight, is that there is no guarantee that they will be able to pick a top-tier quarterback in a later draft.

Of the 21 drafts in the dataset, 17 of them saw the eighth team pick later in the draft the following year, while 12 of them moved entirely out of the 1-13 Value Group. On average, they moved up seven picks (despite "only" improving their record by two wins). The second Value Group produces a successful quarterback 21% of the time, and the pick range (15-23) produces a successful quarterback 25% of the time. The range of picks from 13 to 17 has produced 0 successful quarterbacks out of two.

It is unsurprising in a league celebrated for its parity that a team with the eighth pick is significantly more likely to move up the rankings than down, and it should be priced in to any consideration for the Vikings if they intend on drafting with the mind set that future drafts can solve present problems.

This is further evidenced through the brimming optimism of Vikings fans. A (probably) good coach like Mike Zimmer will very likely move the needle on the Vikings from a 34.4% win rate to something much higher, even in a strengthened NFC North.

Finally it is important to note that taking a quarterback is the highest priority for the Vikings.

It's true that the defense hurt the team far more than the offense, ranking 29th in points per drive given up against an offense that ranked 15th, even after accounting for special teams touchdowns.

But no player improves a team more than a quarterback, and the Vikings have significant room to improve. When excluding Josh Freeman, the Vikings rank 31st in adjusted net yards per attempt. They also ranked 20th in Drive Success Rate, which measures the rate at which an offense produces first downs and touchdowns and is independent of field-position. Football Outsiders also took into account strength of opposition when they ranked the Vikings offense as 21st in the NFL.

Given the importance of the passing game in the NFL, there's only reason to believe the offense will get worse than their 15th overall ranking in points per drive with how abysmal the passing game has become. It will regress from its relative luck to perform below even fan expectations and flounder. And improving the quarterback position will do absolute wonders for the offense:

Per Pro Football Focus

That offense is stacked. With two elite starters, four high-quality starters and a strong role player, it is far better player for player than other offenses. An offense that ranked top five in drive success rate, top ten in points per drive and top ten in DVOA only had average adjusted net yards per attempt (Carolina) and they have significantly less talent across their offensive roster.

Offenses with decent passing games and less overall talent have been scoring many more points than the Vikings, and the overall talent grade given by Pro Football Focus to the Vikings offense was in the top ten (ninth overall). Yet, they don't produce anything like a top ten offense. San Diego finished second in points per drive, third in offensive DVOA, and second in drive success rate, yet had a worse total offensive grade than the Vikings.

It's not too hard to believe that with average QB play, the offense could be great. Teams that are top five in one unit and average in the other are far more likely to make the playoffs. In the last five years, teams with a top five DVOA unit and an average DVOA unit made the playoffs eight times, while those with just average units made the playoffs once. The Super Bowl attendees have had top five units paired with average units three times, but purely average units no times.

And it only takes one player to do it.

The defense will require a much larger fix. There are question marks at every linebacker position, two defensive line positions, and two (or three, depending on what you think of Jamarca Sanford) secondary positions. No one player will fix that.

Adding an impact player will only improve that marginally, too. Not only is every unit in need of help, every unit needs help in multiple areas. They cannot choose to strengthen the linebacker corps by picking a high-impact linebacker because there are still other linebacker questions to resolve. The same is true of the defensive backs and defensive line.

Even if every "question" was answered with a generous 50% chance of producing an average starter, that leaves three or four positions with severe deficiencies. The most optimistic (and wrong) assumptions (an  80% chance of re-signing Everson Griffen, a 55% chance of drafting an impact CB, a 70% chance that Greenway reproduces his 2010 season, a 60% chance Mauti can fill the MLB position, a 40% chance Hodges can fill the WLB position, a 60% chance the Vikings sign a good NT and a 50% chance the Vikings find a slot cornerback through development, the draft or free agency) produces poor odds (33%) of seven solid starters out of twelve (the nickel is the twelfth).

Average defenses by points per drive (in order, Indianapolis, Philadelphia, Detroit, St. Louis, New York Jets and Tennessee) had about nine solid-to-elite starters on defense.

It takes a much more uniform talent level to succeed on defense than it does on offense, which makes sense: on defense, the chain is only as strong as its weakest link. On offense, the team will be able to pick and choose how and where to attack. If the tight end is not very good at catching passes, they can go three or four-wide. If the slot receivers aren't very good, they can target the outside receivers.

But on defense, having a weakness means being targeted there. Naturally, a good offense will have all the tools to make sure they can exploit those weaknesses, but a few weaknesses will make it easier for any offense to make a dent (inversely, this also underscores the value of a diverse offense—being one dimensional gives you no options and puts you at the mercy of the opposing defense. An average multidimensional offense is better than a good one-dimensional offense for this reason).

The defensive positions are also easier to fix. While quarterbacks have a success rate of 14% in the third Value Group and 6% in the fourth Value Group, linebackers have a success rate of 39% in the third VG and 13% in the fourth one. DEs hit at 25% and 20%, and DBs hit at the same rate. Only defensive tackles, who are the most difficult to capture with AV, have a worse success rate on defense than QBs: 10.7% and 9.8% in the two VGs.

With that, moving down the draft order doesn't hurt as much. QBs drop from 42% to 21% from the first VG to the second (from picks 1-13 to picks 14-28) while LBs (55% to 54%), DEs (50% to 38%), DBs (50% to 34%) and DTs (59% to 37%) all drop to much more acceptable levels. A projected movement to later in the draft does not hurt nearly as much as those positions as they do at QB, especially because quarterback play is so critical to winning in the NFL.

Picking a defensive player early would also potentially moot one of the best skills the Zimmer staff can bring: creating talent out of late-round picks and UDFAs. In the last five years, the Bengals have invested one first-round pick in a defensive player and four in offensive players.

In those five years, the Bengals have had (or traded for) 103.8 units of Approximate Value by year, and invested 61.6 of those into the offense, predictably 60% of the value. Underscoring the point even more is the traditional draft value chart. They invested nearly 10,100 value points over that period according to the chart, but 6800 of it went into the offense.

Over five years, very few teams invest more than 55% in one unit. Despite playing counter to that, Cincinnati leaves the 2013 season with one of the top defenses in the country.

On defense, the Bengals have added about one point of value on average for their picks (the difference between the first pick and the seventh pick, or the difference between the 33rd pick and 65th pick). On offense, the Bengals have added a fifth of a point (the difference between the first pick and the third pick or the difference between the 33rd pick and the 38th pick).

Much of this has to do with Zimmer's input on recognizing talent (folks like Geno Atkins), development (Michael Johnson) and bringing in troubled talent that he can get to stay on course (Adam Jones, Vontaze Burfict). The average draft position of the non-FA defense (assuming all UDFAs were Mr. Irrelevant) has been 125. For the offense, the non-FA crowd starters average as the 43rd pick.

Regardless of the reasons the Bengals were able to do it, the fact is that Zimmer's unique talent (and Vikings' comparative advantage) would go to some waste with a high pick. Norv Turner might be a QB guru, but he's done his best work with high picks (Troy Aikman, Philip Rivers and Alex Smith).

This is a comparative advantage the Vikings can make a lot of, because an investment in the defense is functionally upgrading picks by a round. Functionally, picking an offensive player in the first round and picking a defensive player in the second round may yield results that are equivalent to two first-round picks.

In summary:

1) The order of position picked doesn't matter, the pick value does;
2) QB picks in this situation are successful;
3) Passing on the fourth-best quarterback in the first will likely mean taking the fifth-best one in the second with drastically lower success prospects;
4) Picking a quarterback this year, when the Vikings likely have the best pick value to select a quarterback for some time, is crucial;
5) The QB position is the most important position on the team;
6) Yes, even when the offense was adequate and the defense was terrible;
7) The defense won't be meaningfully improved by an impactful, first-round starter, while the offense will be with a quarterback
8) Mike Zimmer makes picking defensive players late far easier to stomach;
9) Reloading at other positions the next year is substantially easier than doing so at quarterback.

This is not to say that whomever the fourth quarterback will end up being necessarily is worth it. That's a different question, and it's about evaluation. But the team-building and strategic ideas behind taking the fourth quarterback left in the first round are sound.