SITE NEWS: We are moving all of our site and company news into a single blog for Sports-Reference.com. We'll tag all PFR content, so you can quickly and easily find the content you want.

Also, our existing PFR blog rss feed will be redirected to the new site's feed.

Pro-Football-Reference.com ยป Sports Reference

For more from Chase and Jason, check out their work at Football Perspective and The Big Lead.

Brees, Ryan, Rivers, Cutler and Manning (and maybe Brady, too)

Posted by Chase Stuart on March 23, 2009

The biggest story in the NFL these days not involving the NFL draft has to do with the bizarre circus involving Jay Cutler. Many have written about everything from Cutler's production and his psyche to macro thoughts on Bill Belichick disciples; I have nothing to add there. I'd rather take on an impossible task and take a statistical look at how Jay Cutler ranks among other QBs.

And when I say how Jay Cutler ranks, I mean how Jay Cutler 2009, 2010, 2011, 2012, and so on, ranks. And that's why it's an impossible task. I can't predict what will happen in three years. You can't predict what will happen in four years. We won't know how good Cutler is over the next five years until five years from now. It's all a guessing game, but that doesn't mean we can't refine our guessing. There are obvious and not so obvious flaws in the approach I'm about to outline, and I'll do my best to explain them.

What I'm trying to figure out is how valuable is Jay Cutler, the commodty, in March 2009? For example, we know we'd rather have Cutler than Tarvaris Jackson and we'd rather have a 26 year old Peyton Manning than a 26 year old Jay Cutler. But to determine his value -- his trade value, if you're a Broncos fan -- you need to know what he'll do in the future. And as Yogi Berra once said, it's tough to make predictions, especially about the future. Consider:

  • At the end of the 1970 season, Billy Kilmer was 31 years old and had just 11 career wins to his name. In 1970, he threw 6 TDs and 17 INTs. Who would have guessed he would have been one of the best QBs in the NFL over the next half-decade?
  • At the end of the 1995 season, Rich Gannon was 30 years old and had started just four games over the past three years. Who would have guessed that he would make four Pro Bowls over the next seven seasons?
  • Stop me if you've heard this one. QB is drafted 330th overall. He throws 1 TD and 7 INT his rookie season. He throws 1 TD and 3 INT and has an 0-2 record his sophomore season. You probably wouldn't think he'd still be in the league five years after that, but in his seventh season Brian Sipe was named a first team All Pro.

This goes the other way, too, of course.

    Archie Manning struggled early in his career, but the #1 pick was very good at ages 29, 30 and 31. Who would project him to throw 15 TD and 30 INT the rest of his career?
  • Mark Rypien won a Super Bowl and had one of the better seasons in passing history in 1991. He was just 29 years old. He never came close to duplicating that success.
  • Greg Landry was a first round pick who had big years as a runner and passer in 1971 and 1972. He was 26 years old in March 1973, and he may have been the single best QB prospect in the NFL at that time. When you consider his rushing, he'd been more productive than Manning or Bradshaw or Griese, he was younger than Tarkenton and Namath, and he had a better pedigree than Ken Anderson. Landry stuck around for awhile but there were about 15 QBs who outperformed him from that moment on.
  • Steve Bartkowski was a former number 1 pick who made the Pro Bowl in '80 and '81 and had an incredible 22 TD/5 INT ratio in 1983. He was 31 years old, but he threw for fewer than 4,000 yards the rest of his career.

None of this is breaking news; we all know it's difficult to predict one year down the road, let alone five or ten. But it's important to set the stage before we answer the key question: if you could have any QB in the NFL right now, for your favorite team, who would you pick? Can statistics and past history guide us?

There are a ton of factors that you would want to use to predict future QB success, but there are three that seem most prominent: age, past production and draft value. Unfortunately, each of these are complicated variables and require a full description, but first, let me describe how we measure future QB success. If you hate reading the details, skip to the end for the QB list.

First I calculated each QB's adjusted net yards per attempt (passing yards + 20*TDs - 45*INT - sack yards lost) / (passes + sacks) metric. Then I compared that ratio to "replacement level", defined as 75% of the league average. Then I multiplied the difference between the QB's ANY/A and replacement level by the QB's number of pass attempts plus sacks to get a measure of "adjusted yards over replacement." Then, I added to that number every adjusted rushing yard (rushing yards + 20*rushTDs) over four yards per carry. So if a QB had 100 carries, 500 yards and 5 TDs, that would be plus 200 adjusted yards. Whether or not this formula is perfect isn't that important -- for what we're looking for, with hundreds of QBs, something that's generally correct is all we need. We don't need to know specifically if Troy Aikman was better than Jeff Garcia, but just that both were better than Danny Kanell and Bobby Hoying.

Measuring QB success or value in a given year isn't enough, though. We want to know how they'll do for awhile, although we also want immediate success. To grade QB value for the long term, I took 100% of their production in Year N+1 (2009, for Jay Cutler's purposes), 95% of their value in N+2, 90% in N+3, and so on, for 8 seasons. This will thus reward great immediate production and sustained levels of strong play. This weighted measure of eight years of production will be the output variable in our regression formula. Not surprisingly, Manning in March 2000 (the year he turned 24) and Marino in March 1984 (the year he turned 23) come out as the top two scores.

Age:: The effects of age are obviously nonlinear; going from age 23 to 24 is good; going from 35 to 36 is bad. I looked at the top 50 QBs (based on NFL production, using something similar to the metric I used in the Greatest QB Ever series) drafted since 1970, that are no longer active, to determine the general "dropoff" rate for a QB. Here are the results:

21	 12.4
22	 34.6
23	 53.1
24	 68.1
25	 79.9
26	 88.8
27	 94.9
28	 98.5
29	100.0
30	 99.5
31	 97.3
32	 93.7
33	 88.9
34	 83.2
35	 76.8
36	 69.9
37	 62.9
38	 56.0
39	 49.4
40	 43.4
41	 38.3
42	 34.2

This shouldn't be too controversial; it says that QBs peak from ages 28-31, and a QB at 36 is about as good as a QB at age 24. However, we can't just use these numbers as the inputs in the regression formula because we're trying to predict a weighted value of eight years worth of scores. So the value of being 24 corresponds to 100% of the value of being 25, plus 95% of the value of being 26, and so on. Here are those numbers:

21	 12.4	452
22	 34.6	510
23	 53.1	552
24	 68.1	578
25	 79.9	591
26	 88.8	591
27	 94.9	581
28	 98.5	563
29	100.0	537
30	 99.5	505
31	 97.3	469
32	 93.7	430
33	 88.9	391
34	 83.2	352
35	 76.8	315
36	 69.9	282
37	 62.9	254
38	 56.0	225
39	 49.4	196
40	 43.4	172
41	 38.3	151
42	 34.2	132

Therefore, the ideal range of ages where your average QB has his best days ahead of him is somewhere between 24 and 27 years. And that makes a lot of sense. And whereas the table above said being 36 was slightly better than being 24, now being 24 is much better than being 36. And that makes sense, too.

Past Production: I used the same formula to measuring past production as I did for measuring future production; adjusted net yards added over replacement value. However, I didn't want to limit myself to 2008, so I used a weighted average of the last three years. So for Philip Rivers, he gets 3*2008_value plus 2*2007_value plus 2006_value, all divided by six. For JaMarcus Russell, he gets 2*2008_value plus 2007_value, divided by three. For Matt Ryan, he simply gets his 2008 value. This becomes a little unfair for someone like Aaron Rodgers, who is treated like a star in 2008 and a nothing in '07 and '06. That's not an accurate portrayal of what happened, but we can tweak that later.

Draft Value: I used the numbers here to assign a value to each slot in the draft. I feel really comfortable with those values, but there's one big problem: a player's draft value is really important when they're young but not very important when they're old. I made some rough, back of the envelope calculations as to a good multiplier for draft value at each age:

22	1.0
23	1.0
24	0.9
25	0.8
26	0.7
27	0.6
28	0.5
29	0.4
30	0.3
31	0.2
32	0.1
33	0.1
34	0.1
35	0.1
36	0.1
37	0.1
38	0.1
39	0.1
40	0.1

This just quantifies what my gut says, although I have no idea how accurate it really is. Deriving a good formula for this could be a whole separate blog post, so I'm going to put that off for another day. So I simply multiplied each player's draft value (as derived in that post) times their multiplier based on age. There's a problem with this, though: it makes the "draft value" variable for a 31 year old Brett Favre equal to the draft value variable for a 24 year old Tom Brady. That's beacuse Favre's an old guy that used to be a high pick and Brady's a young guy who is a low pick. That's what I ended up doing, but I don't love it. Unfortunately, I'm not sure if there's a better way -- using regression analysis -- to handle this.

If you've made it this far, the payoff is about to come. We now have our input variables -- an age variable, a draft value variable that decreases over time, and a past production variable. The output is a weighted average of performance over the next eight seasons. What's our dataset? Every QB who entered the league since the merger, but excluding all player-seasons since 2003 and any QB born after 1977. Further, I only looked at player-seasons with at least 200 pass attempts.

What's the formula that best fits that curve? It won't mean much to you, but here it is:

-1972 + 4.21*age + 28.6*draft + 2.38*pastproduction

That doesn't mean anything until we apply it to the current list of QBs. Here are the top 25 projected QBs going forward in the NFL as of March 2009, but using only QBs who had 200+ attempts in 2008. The age is how old the player was during the 2008 season.

08Val   Age	AgeV	 DraftV  3YrVal      Proj	 Rk	Name
2157	 29	537	 11.8	  1861	     5049	 1	Drew Brees
1245     23	552	 59.4	  1245	     5009	 2	Matt Ryan
1919	 27	581	 33.5	  1429	     4829	 3	Philip Rivers
1495	 25	591	 34.4	  1169	     4277	 4	Jay Cutler
1501	 32	430	  7.3	  1702	     4096	 5	Peyton Manning
1217	 28	563	  1.5	  1372       3699	 6	Tony Romo
 875	 27	581	 43.9	   648	     3270	 7	Eli Manning
 484	 23	552	 73.2	   288	     3127	 8	JaMarcus Russell
 492	 26	591	 30.1	   727	     3106	 9	Ben Roethlisberger
1422	 25	591	 26.6	   733	     3016	10	Aaron Rodgers
1234	 32	430	  6.4	  1205	     2889	11	Donovan McNabb
 477	 23	552	 36.8	   477	     2538	12	Joe Flacco
 731	 27	581	 19.6	   614	     2495	13	Jason Campbell
1030	 27	581	  9.9	   728	     2489	14	Matt Schaub
1721	 37	254	  0.3	  1305	     2207	15	Kurt Warner
 645	 30	505	  4.3	   801	     2179	16	David Garrard
1477	 32	430	  3.7	   925	     2143	17	Chad Pennington
 663	 25	591	 13.0	   490	     2050	18	Trent Edwards
1013     26	591	  3.4	   495	     1789	19	Matt Cassel
  74	 25	591	  4.6	   429	     1665	20	Derek Anderson
 539	 26	591	 10.2	   302	     1523	21	Kyle Orton
1072	 33	391	  0.3	   752	     1468	22	Jake Delhomme
 553	 24	578	  4.9	   354	     1444	23	Tyler Thigpen
 528	 28	563	  1.5	   410	     1413	24	Shaun Hill
 462	 28	563	  7.0	   277	     1254	25	Seneca Wallace

How does that list look to you? This is my best attempt to, using just objective data, figure out which QBs would have the most value either on the open market or through a trade. All three input variables were highly significant, which means they are all certainly correlated to future production. The R^2 was just 0.35, which isn't very high, but I'm not sure if you can come up with a formula to make it any higher. There's a ton of randomness in future production, and if 35% of it can be predicted through this formula, that's pretty good. Before we conclude, let me throw some general thoughts out there on this list:

1) If we ignore Rodgers' 2006 and 2007, and just use last year for his 3YrVal, that would give him a projected score of 4654, and he'd move into 3rd place on the list. Very interesting. Doing the same analysis with Cassel bumps him just north of 3000, where Rodgers currently is.

2) Joe Flacco's ahead of Kurt Warner. That makes some sense to me. We're basically projecting one or two big years out of Warner versus eight good years out of Flacco.

3) Matt Ryan (along with Rodgers if we tweak his score) comes in too high, I think. Why? My buddy Maurile would recommend using some sort of Bayes' theorem analysis, because being great over a small number of attempts is not as convincing as being great over a large number of attempts. As I've currently structured it, three great years will have the same score as one great year, if you only play one year. Usually, that's not a problem, but when you have the greatest rookie season ever (or you edit Rodgers' career), things get dicey. Suffice it to say, while I love Ryan, I'm not convinced just yet that he's mega elite. I love Ryan as much as anyone, but putting him at #2 scares me a little bit.

4) Look over there at JaMarcus Russell. He's up there over Big Ben! Two of the three factors point Ben's way -- being older actually helps him, since he is entering his prime, and obviously he's already better. The only thing pointing Russell's way is that #1 draft pick status. Is that right or wrong? Here's how every QB drafted #1 overall since the merger (along with Steve Young, a supplemental #1) ranked among QBs in their first two seasons.

                        RookYr  Yr1     Yr2 
JaMarcus Russell	2007	82	24
Alex Smith	        2005	81	26
Eli Manning	        2004	79	10
Carson Palmer	        2003	--	24
David Carr	        2002	84	25
Michael Vick	        2001	29	 5
Tim Couch	        1999	35	27
Peyton Manning	        1998	20	 3
Drew Bledsoe	        1993	25	12
Jeff George	        1990	36	59
Troy Aikman	        1989	81	63
Vinny Testaverde	1987	49	81
John Elway	        1983	74	16
Steve Bartkowski	1975	23	70
Jim Plunkett	        1971	12	68
Terry Bradshaw	        1970	65	21
Steve Young	        1984	--	67

Sure, Russell looks bad. But so did Troy Aikman. And Vinny Testaverde. And Terry Bradshaw. So did Steve Young. John Elway wasn't much better. I have to say, it's very counterintuitive to me to suggest that Russell is going to be better than Roethlisberger, going forward. On the other hand, you could probably have made the same argument about Jim Everett and Troy Aikman in March 1991, too. As for Ben and Russell, I think it's at least arguable that Ben's statistics understate how good he really is. If his 3YrVal was a bit higher, he'd easily be ahead of Russell and Manning.

5) Notice anyone missing? I don't really know what to do with Brady, but his score after 2007 was 5179. Giving him an extra year of age would drop him to 5008, but we can't ignore that he suffered a serious injury. Who knows how he'll come back, and I don't really know where's a good place to put him. If he's healthy, though, he's right there up with Brees. Carson Palmer was right behind Brady in 2007, so if he's healthy, he's got to be in the mix, too. If we age him a year based on his '07 numbers, he'd project at 4471.

Let me close with a subjective list, using the numbers above as my guide (along with the changes I mentioned above). I won't try to be so exact, but I'm going to group the top 25 or so QBs in the NFL into tiers.

Build your team around them (6): Drew Brees, Tom Brady, Peyton Manning, Philip Rivers, Jay Cutler and Matt Ryan. These six QBs are at their own level, IMO, as no one matches past production and youth the way they do. You build your team around any of these guys, and all would be a good bet to be the NFL MVP in 2009, 2010 or 2011.

You're in love (5): Aaron Rodgers, Tony Romo, Eli Manning, Ben Roethlisberger, Carson Palmer. It wouldn't take you more than a second to find fault with any of these guys. That said, you've got to feel like you can go to the playoffs anytime in the next five years with this guy at the helm.

I really, really like you (3): JaMarcus Russell, Donovan McNabb, Joe Flacco. These guys are a half step below the guys above and a half step above the guys below. You're happy with your QB outlook, but you're not going to puff your chest too much. While Russell hasn't received nearly the praise that Flacco has, Russell put up at least equal numbers last season despite playing for a worse team and despite being eight months younger than Flacco. Warner arguably deserves to be in this group, but I don't think he has enough left in the tank.

You like what you've got (7): Jason Campbell, Matt Schaub, Kurt Warner, David Garrard, Chad Pennington, Trent Edwards, Matt Cassel. These guys don't have a lot in common. Warner may be an All Pro next year, but you're going to be looking for a new QB very soon. Garrard and Pennington have had big years while Schaub and Campbell have shown flashes. Edwards and Cassel are inexperienced but have done okay so far. The whole group is a mixed bag, and if this guy is your QB, you can't even guarantee that he'll be in the league in five years. I suppose a healthy Matt Hasselbeck could join this tier, too. Based on just 2008, Warner and Pennington are the obvious class of the group, but both have long histories.

Can I really trade for Jay Cutler? (6): Derek Anderson, Kyle Orton, Jake Delhomme, Tyler Thigpen, Shaun Hill, Seneca Wallace. The less said about these guys, the better.

So what do you think? How would you rank the QBs?

This entry was posted on Monday, March 23rd, 2009 at 7:29 am and is filed under History, Statgeekery. You can follow any responses to this entry through the RSS 2.0 feed. Both comments and pings are currently closed.