Negative Regression Candidates by SQB%
Even before Matt Ryan’s team came up short in Super Bowl 51, sharp fantasy owners pegged him as a regression candidate for 2017. He’s coming off a career year. Expecting elasticity to his traditional norms is a safe and easy bet.
Regression analysis is not binary, though. To borrow from Scott Pianowski of Yahoo, we can’t simply ask if a player will regress, find a believable answer, and “walk away.” The question we must ask is *how much* a player will regress.
Everyone can identify obvious outliers. The game is figuring out what is likely to happen next. You can’t scream “regression” and walk away.
— scott pianowski (@scott_pianowski) September 30, 2015
For Matt Ryan, some combination of factors allowed him to exceed expectations and post MVP numbers in 2016. Next season, some of those factors should still be in place, like Atlanta’s improved receiving arsenal. Others, like continuity in Kyle Shanahan’s offense, will be lost. Ultimately, narrative-driven benefits and/or baggage are difficult to quantify for the purposes of establishing performance baselines. Typically, we must turn to statistics in search of answers to the question of “how much regression.”
Enter Startable Quarterback Percentage, or SQB%. I introduced this measure of quarterback performance in August as a draft analysis tool, and SQB% values are featured prominently in TwoQBs.com’s quarterback cards. Rather than rate passers by their total fantasy points or fantasy points per game, SQB% looks at how often a player finishes in the top-10, in the top-20, or below QB24 on a weekly basis.
Going back to Matt Ryan in 2016 as an example, he finished all of his games as a top-20 quarterback (100%) and was a top-10 quarterback 62.5% of the time. From 2012 to 2015, however, Matty Ice’s top-10 SQB% was 33.3%, his top-20 SQB% was 73.0%, and he busted in 14.3% of his games. Ladies and gentlemen, we have a regression baseline.
The balance of this article will look at other candidates for negative regression based on Startable Quarterback Percentage. Before we get there, I have a few notes and disclaimers:
- Thanks to a fair criticism from a Twitter follower, I’m no longer comfortable with my methodology of counting “bust” weeks. I initially had my reasons for using finishes below QB24 instead of below QB20, but in certain bye-heavy weeks, only 26 quarterbacks take the field as starters. In those weeks, it isn’t fair to call QB25 and QB26 busts while QBs 21-24 get a free pass. With that in mind, I will avoid bust week analysis in this piece and focus only on top-10 and top-20 finishes.
- Sample size and quality of samples are very important when comparing a single season to multiple past seasons. To help identify playing time discrepancies between past seasons and 2016, I have included games played numbers in both tables. Games played doesn’t always tell the whole story, though, so I’ll attempt to point out corner cases where I can.
- Furthermore, for the sake of simplicity, data for the following players was thrown out because of small or unindicative samples: Jimmy Garoppolo, Matt Cassel, Nick Foles, Geno Smith, Mark Sanchez, Matt McGloin, Tony Romo, Mike Glennon, Chase Daniel, Matt Barkley, Landry Jones, Drew Stanton, EJ Manuel, Ryan Mallett, Derek Anderson, Matt Moore, Matt Schaub, Shaun Hill, Tom Savage, Scott Tolzien, Chad Henne, Charlie Whitehurst, and Kellen Clemens.
Top-20 SQB% – Negative Regression
Player # Yrs 2012-5
GP Diff. 2012-5
Tyrod Taylor 3 7.3 15 -7.7 54.5% 86.7% -32.2%
Brock Osweiler 4 4.8 15 -10.3 15.8% 46.7% -30.9%
Matt Ryan 4 15.8 16 -0.3 73.0% 100.0% -27.0%
Colin Kaepernick 4 13.3 11 2.3 58.5% 81.8% -23.3%
Kirk Cousins 4 7.5 16 -8.5 53.3% 75.0% -21.7%
Aaron Rodgers 4 14.3 16 -1.8 78.9% 100.0% -21.1%
Derek Carr 2 16 15 1.0 59.4% 80.0% -20.6%
Sam Bradford 3 12.3 15 -2.7 59.5% 66.7% -7.2%
Tom Brady 4 16 12 4.0 76.6% 83.3% -6.7%
Brian Hoyer 4 7.5 6 1.5 43.3% 50.0% -6.7%
Carson Palmer 4 13.3 15 -1.8 73.6% 80.0% -6.4%
Philip Rivers 4 16 16 0.0 70.3% 75.0% -4.7%
Marcus Mariota 1 11 15 -4.0 63.6% 66.7% -3.1%
Blaine Gabbert 4 5.3 6 -0.8 47.6% 50.0% -2.4%
Blake Bortles 2 15 16 -1.0 66.7% 68.8% -2.1%
Joe Flacco 4 14.5 16 -1.5 67.2% 68.8% -1.6%
Andrew Luck 4 13.8 15 -1.3 85.5% 86.7% -1.2%
Alex Smith 4 14 15 -1.0 66.1% 66.7% -0.6%
Andy Dalton 4 15 16 -1.0 68.3% 68.8% -0.4%
Case Keenum 3 5 10 -5.0 40.0% 40.0% 0.0%
Skipping over the outliers at the top of the list (I’ll get to them soon), Colin Kaepernick is our next biggest regression candidate after Matt Ryan. Kap wasn’t the full time starter in 2015, so his numbers from that season drag down his 2012-2015 averages. Neverthless, Kaepernick was remarkably consistent between 2012 and 2014, posting a top-20 SQB% value of 61.5% or 62.5% every season. I spent much of 2016 waiting for Kaepernick’s fantasy value to crash back down to earth, but my fears were never fully justified. Aside from his bad-weather stink bomb in Chicago, Kap was pretty reliable, but this was only one season in a very quarterback-friendly scheme. Barring an unforeseen reunion with Chip Kelly on a new team, Kaepernick’s top-20 SQB% mark of 81.8% in 2016 shouldn’t be repeated in 2017.
Aaron Rodgers & Tom Brady
Before we get to Kirk Cousins, I want to skip ahead to Aaron Rodgers and Tom Brady. Rodgers finished first and Brady finished fifth in 2016 top-20 SQB%, and like Matt Ryan, they are inherently slated for negative corrections in 2017. The trouble is we know Rodgers and Brady are among the league’s elite behind center. In fact, if we rank all qualified quarterbacks by SQB% for 2012-2015 and throw out Jameis Winston’s one-year sample, Rodgers ranks fourth (78.9%) and Brady ranks sixth (76.6%). They aren’t going anywhere yet, so even with some amount of regression expected from their 2016 statistics, both deserve to be drafted as top-tier starters next year.
Circling back to Kirk Cousins, some might argue he’s not a contender for regression because he’s still improving and rounding into form as an NFL passer. His 75% SQB% could appear repeatable when compared to marks of 100% by Matt Ryan and 81.8% by Colin Kaepernick, but that’s why I brought up Rodgers and Brady for context. Look at their 2012-2015 top-20 SQB% marks. In that span, they averaged just above the 75% mark Cousins posted in 2016. Sure, Cousins could continue to improve, but he’s not the same caliber of quarterback as Rodgers, Brady, Brees, Luck, etc. I peg Cousins as a 60-70% top-20 guy, so his draft price in 2017 will likely be too costly for my tastes.
Drafting Sam Bradford hasn’t been in good taste since the Rams selected him first overall in 2010, and these numbers don’t help his case to fantasy owners in 2017. With that said, he barely had any opportunity for real value between 2012 and 2015, so we shouldn’t read too much into his four-year SQB% average. He very well might be the heir to Alex Smith’s check-down kingdom. Bradford’s 2016 top-20 SQB% of 66.7% is right in line with Smith’s four-year mark of 66.1%. Now that Bradford has a path to regular starting time with a functioning franchise, he could provide sneaky fantasy value for LRQB drafters in 2QB leagues, just like Smith has in recent years.
Going back to the idea of bad samples, the numbers for Tyrod Taylor are a little misleading. He tops the list of regression candidates, but keep in mind that 8 of his 22 games played between 2012 and 2015 occurred before he became a starter. That span includes a season without any action at all (2014). While it’s a smaller sample size, we’ll get a better idea of Tyrod’s baseline by contrasting only his two years as Buffalo’s starting quarterback:
- 2015: 14 GP, 50.0% Top-10, 78.6% Top-20
- 2016: 15 GP, 26.7% Top-10, 86.7% Top-20
- Difference: -1 GP, +23.3% Top-10, -8.1% Top-20
Because we’re only comparing two seasons, it isn’t correct to expect regression from 2016 to 2015 only because one set of data happened first. The truth about Taylor probably falls somewhere in the middle. Thanks to his rushing ability, he’s a stable top-20 producer with occasional upside for top-10 weeks — an ideal QB2 in fantasy.
Brock Osweiler’s dubious sample of pre-2016 data is similar to Taylor’s, but Osweiler’s track record is more concerning overall. Even in an awful 2016 season, SQB% indicates Osweiler actually should have been worse. Things look a little better for Osweiler if I pull the same trick with Tyrod and only compare the last two seasons, but I only have so much lipstick for this pig:
- 2015: 8 GP, 25.0% Top-10, 37.5% Top-20
- 2016: 15 GP, 13.3% Top-10, 46.7% Top-20
- Diff.: -1 GP, +11.7% Top-10, -9.2% Top-20
Perhaps it’s fair to hope for growth after more time as the starting quarterback, but we simply haven’t seen any signs of progress from Osweiler to this point. Ironically, though, there’s nowhere for him to go but up, so he could certainly get better. Negative regression in this case means potential for benching and relegation to backup duty.
Rounding out the top-20 SQB% section, Carson Palmer and Philip Rivers both make the regression list. Consider also their rapid approach of the quarterback age cliff, and you might find yourself staying away in 2017. On the flip side, if the fantasy hive mind turns against these two based on their age and assumed regression, they could plummet in ADP and become reasonable draft values. Statistically, I expect a negative correction, but we still might want to draft them. It’s a similar situation as described above with Aaron Rodgers and Tom Brady, only Palmer and Rivers don’t have the same sorts of ceilings.
Top-10 SQB% – Negative Regression
Player # Yrs 2012-5
GP Diff. 2012-5
Matt Ryan 4 15.8 16 -0.25 33.30% 62.50% -29.20%
Blaine Gabbert 4 5.3 6 -0.75 4.80% 33.30% -28.50%
Derek Carr 2 16 15 1 25.00% 46.70% -21.70%
Colin Kaepernick 4 13.3 11 2.25 34.00% 54.50% -20.50%
Brian Hoyer 4 7.5 6 1.5 16.70% 33.30% -16.60%
Ryan Tannehill 4 16 13 3 26.60% 38.50% -11.90%
Aaron Rodgers 4 14.3 16 -1.75 57.90% 68.80% -10.90%
Matthew Stafford 4 16 16 0 40.60% 50.00% -9.40%
Ben Roethlisberger 4 14.3 14 0.25 35.10% 42.90% -7.80%
Andrew Luck 4 13.8 15 -1.25 52.70% 60.00% -7.30%
Drew Brees 4 15.8 16 -0.25 55.60% 62.50% -6.90%
Tom Brady 4 16 12 4 51.60% 58.30% -6.70%
Case Keenum 3 5 10 -5 13.30% 20.00% -6.70%
Jameis Winston 1 16 16 0 18.80% 25.00% -6.20%
Alex Smith 4 14 15 -1 28.60% 33.30% -4.70%
Brock Osweiler 4 4.8 15 -10.25 10.50% 13.30% -2.80%
Carson Palmer 4 13.3 15 -1.75 32.10% 33.30% -1.20%
Kirk Cousins 4 7.5 16 -8.5 43.30% 43.80% -0.50%
Derek Carr made both lists with steep regression figures around -20%, but he made similar leaps between his rookie year and second year. Can he continue to get better on this trajectory? I doubt it. If Carr somehow pulls it off, he’d enter the same elite tier of quarterbacks I denied Kirk Cousins access to a few paragraphs ago. Furthermore, Carr had the benefit of an extremely easy schedule in 2016, which inflated his number of top finishes. The comparison to Kirk Cousins is apt. Neither is an exceptional talent, but both rank in the top portion of fantasy’s quarterback middle class.
Despite qualifying for positive regression in the top-20 finishes department, Ryan Tannehill overperformed by top-10 SQB% standards. When the metrics pull in different directions like this, I typically chalk the results up to variance. Tannehill only played 13 games this season, to boot, so there’s a decent chance his SQB% values would have leveled out toward career norms with an extra three games played. For example, if he had played three more games and finished each in the QB11 to QB20 range, he would have ended the season with five top-10 weeks and 11 top-20 weeks, compared to four-year averages of 4.25 top-10 weeks and 11.25 top-20 weeks. Tannehill is who we thought he was.
Matthew Stafford & Ben Roethlisberger
Matthew Stafford and Ben Roethlisberger are less polarizing examples of the same effect we saw with Tannehill. They overperformed versus top-10 SQB% baselines, but finished slightly below their four-year averages in top-20 SQB%.
Despite losing Calvin Johnson in the off-season, Stafford posted his best top-10 SQB% value since 2012. Jim Bob Cooter deserves some of the credit, but like Derek Carr, Stafford was the beneficiary of a cream-puff schedule this year. Most drafters won’t remember that when prepping for 2017 drafts, so look for his ADP to rise back into the QB7-QB12 range. Unless the Lions add some offensive firepower over the off-season, I won’t pay up for Stafford in the top half of that range. He has a case as QB10 or later, but based primarily on his Matt Ryan-esque reliability to play full seasons, rather than an assumption of a 50% top-10 SQB%. His four-year average forecasts six or seven top-10 finishes per 16 games, not eight or more.
On the other hand, Roethlisberger turned in a 2016 effort eerily similar to his 2015 campaign, with both seasons rating much better in top-10 SQB% than his stretch from 2012 to 2014. In 2014, Big Ben piled up stats without sustained fanfare, posting only five top-10 weeks despite accumulating the fifth-most fantasy points among quarterbacks. Over the past two years, Roethlisberger traded consistent efficiency for a boom-or-bust model, increasing both his top-10 SQB% and his propensity for dud weeks.
While regression back toward his 2012-2014 numbers should make sense, the 2015-2016 version of Roethlisberger might just be who he is now. The emergence of Le’Veon Bell helps sell the boom-bust narrative for Pittsburgh’s passing game. On top of Roethlisberger missing an average of two games per year since 2012, Bell’s excellence affords Big Ben leeway to occasionally take a back seat in the Pittsburgh offense. We saw it happen this season in Week 11 at Cleveland and Week 14 at Buffalo. In the right match-ups, however, Bell can take the offense to soaring heights, and it rubs off on Roethlisberger’s fantasy performance. All told, I don’t see Ben as a big regression candidate. He’s a risky buy in drafts, but the upside he offers is huge if he can manage to stay healthy for a full 16 games.
I intentionally glossed over Brian Hoyer to talk about bigger names, but his SQB% numbers are a gentle reminder not to get our hopes up for 2017 if he lands a starting gig. Compared to his average output between 2012 and 2015, 2016 was a career year for Hoyer. Keep in mind he only played six games, and none of his opponents featured tough defenses. His best performances came against Dallas, Detroit, and Indianapolis — three teams who generated fantasy-friendly game scripts for most opposing quarterbacks they faced. Hoyer failed to throw a touchdown pass in any of his other games (vs. PHI, vs. JAC, at GB). Buyer beware next season.
I’ll close this already-too-long piece out with a look at Jameis Winston. We only have two years of data on the young Buccaneer, but he backslid significantly in top-20 SQB% after his rookie season, while improving his top-10 SQB%.
- 2015: 16 GP, 18.8% Top-10, 81.3% Top-20
- 2016: 16 GP, 25.0% Top-10, 62.5% Top-20
- Diff.: 0 GP, -6.2% Top-10, +18.8% Top-20
As with the other two-year samples we’ve examined, I expect Winston’s top-20 SQB% to fall somewhere in between his 2015 and 2016 numbers. A 65% to 75% range of outcomes seems fair. Call it regression to the average. Conversely, I don’t predict negative regression in his top-10 SQB%. His 2016 mark of 25% seems about right as a baseline. Considering his youth, I anticipate a small amount of improvement.
I was a Winston apologist in the 2016 draft season, but I don’t want to overreact too much to one disappointing season. I had rose-colored shades on in expectation of a 2016 value explosion. Disassociating from my bad call is important for future seasons, but I can’t dismiss Winston’s potential for improvement, even if it’s not on the breakout level I wanted to see last year. Objectivity is tough to attain in these scenarios, and he didn’t deserve the top-12 ranking I projected onto him in 2016, but Wintson merits consideration alongside the wealth of other viable QB2s in the player pool for 2017.
Programming Note: I’ll be back soon to profile SQB%’s positive regression candidates. Stay tuned.
UPDATE: Check out Greg’s piece on positive regression candidates here.
Latest posts by Greg Smith (see all)
- 2QBXP Podcast – The Analytics of Dynasty with Jordan McNamara - January 22, 2019
- Week 20 Game Flowbotics (Conference Championships) - January 18, 2019
- Week 19 Game Flowbotics (Divisional Round) - January 10, 2019