Over the past three weeks, I've introduced True Fantasy Points (TFP) for QBs, WRs, and TEs. Running the reliability analyses for RBs takes twice the work because of having to find regression to the mean (RTM) factors for both rushing and receiving stats, so I'll be introducing TFP for RBs next week. In the meantime, today's installment will include some preliminary diagnostics about TFP's hits and misses for QBs since Week 5, as well as updated stats for QBs through Week 7.
Before we get started, here's a quick reminder of how TFP works for QBs -- in five easy steps:
- Use reliability analysis to derive RTM factors for yards per attempt (YPA), TD rate (TD%), and INT rate (INT%).
- Calculate each QB's True YPA, True TD%, and True INT% by applying RTM factors to their current stats.
- Calculate each QB's TFP by applying his True YPA, True TD%, and True INT% to his current number of pass attempts.
- Calculate a +/- stat for each QB by subtracting his TFP from his actual fantasy points.
- Combine said +/- stats with more subjective factors (e.g., schedule strength, injuries, offensive schemes, etc.) to identify buy-low/sell-high candidates.
tfp Hits and misses: quarterbacks
So far, TFP after Week 4 has correctly identified the future fantasy trend for 20 of 31 total QBs,1 going 10-for-16 on sell-high predictions and 10-for-15 on buy-low predictions. Furthermore, the correlation between QB +/- stats through Week 4 and the change in their points per game from Weeks 1-4 to Weeks 5-7 equalled -0.486. This means that the higher a QB's +/- was after Week 4, the more his scoring averaged decreased in Weeks 5-7 -- which is the kind of pattern we would expect given how RTM works.
Dividing those 31 QBs into quartiles based on their +/- stats three weeks ago, we find the following results:
| +/- Quartile thru Wk 4 | Avg thru Wk 4 | Avg Wks 5-7 | Chg | TFP Correct |
|---|---|---|---|---|
| Top 8 | 21.9 | 18.5 | -3.4 | 7 of 8 |
| Second 8 | 18.5 | 20.7 | +2.2 | 3 of 8 |
| Third 8 | 15.7 | 17.2 | +1.5 | 6 of 8 |
| Bottom 7 | 15.9 | 18.0 | +2.1 | 4 of 7 |
Aside from the "Second 8" quartile, this pattern -- i.e., the "Chg" column steadily increasing as you go down the table -- is also what you would expect to see from RTM. It's worth pointing out, however, that the outlying "Second 8" result is slightly misleading because Blake Bortles' per-game average through Week 4 included a "game played" in which he never actually stepped foot on the field. If we used 17.8 points as his average rather than 8.9 points, then the within-group difference for "Second 8" would decrease from +2.2 to +1.0, and TFP accuracy would be 4-for-8 instead of 3-for-8. To be sure, that's still less accurate than you would expect, but at least we're quantifying the miss in a more reasonable manner.
Moving on, four of the five QBs I featured in my Week 5 column saw their scoring averages move in the direction predicted by TFP:
- Buy-low candidate #1: Contrary to TFP, Nick Foles' average actually decreased from 19.7 to 17.9.
- Buy-low candidate #2: Consistent with TFP, Drew Brees' average increased from 21.3 to 23.8.
- Buy-low candidate #3: Consistent with TFP, Tom Brady's average increased from 13.4 to 27.2.
- Sell-high candidate #1: Consistent with TFP, Andrew Luck's average decreased from 28.3 to 24.1.
- Sell-high candidate #2: Consistent with TFP, Eli Manning's average decreased from 19.9 to 16.7.
Foles was a miss, but only a slight one. That said, TFP failed to hit the broad side of a barn on two non-featured QBs. First, Peyton Manning's scoring average has actually increased by 7.3 points thanks to his TD% increasing from 7.21% in Weeks 1-4 to 10.4% TD% in Weeks 5-7, which means it's now even more out of whack with his True TD% (6.49%) than it was three weeks ago. Second, Tony Romo's averaged 22.9 points in his last three games after averaging 17.7 in his first four games, an increase that's highly in conflict with what his +/- was at the time (+2.8). The reason? He's gone from overperforming his True TD% to now overperforming both his True TD% (6.54% vs. 5.44%) and his True YPA (8.36 vs. 7.82).
tfp through week 7: quarterbacks
Below is a table showing updated TFP and +/- stats for FBG's Top 36 QBs Forward:
| QB | Obs FBG | TFP14 | Rk | TFP | Rk | +/- 14 | Rk | +/- | Rk |
|---|---|---|---|---|---|---|---|---|---|
| Andrew Luck | 185.6 | 167.6 | 1 | 155.3 | 1 | +18.0 | 5 | +30.2 | 1 |
| Philip Rivers | 163.1 | 132.2 | 8 | 134.3 | 8 | +30.9 | 3 | +28.8 | 2 |
| Peyton Manning | 165.4 | 126.3 | 10 | 138.9 | 6 | +39.1 | 1 | +26.5 | 3 |
| Joe Flacco | 143.7 | 133.5 | 5 | 121.7 | 14 | +10.2 | 6 | +22.0 | 4 |
| Aaron Rodgers | 154.7 | 120.3 | 14 | 132.7 | 9 | +34.4 | 2 | +22.0 | 5 |
| Tony Romo | 139.5 | 120.3 | 15 | 124.5 | 13 | +19.2 | 4 | +14.9 | 6 |
| Eli Manning | 129.7 | 120.8 | 13 | 114.8 | 15 | +8.9 | 7 | +14.9 | 7 |
| Jay Cutler | 142.3 | 139.9 | 3 | 137.7 | 7 | +2.4 | 14 | +4.6 | 8 |
| Colt McCoy | 10.4 | 6.5 | 35 | 6.6 | 35 | +3.9 | 10 | +3.8 | 9 |
| Carson Palmer | 63.4 | 59.8 | 29 | 59.6 | 29 | +3.5 | 12 | +3.8 | 10 |
| Alex Smith | 100.4 | 98.9 | 24 | 96.7 | 25 | +1.4 | 15 | +3.7 | 11 |
| Matt Ryan | 148.9 | 148.7 | 2 | 145.7 | 3 | +0.2 | 16 | +3.2 | 12 |
| Charlie Whitehurst | 56.2 | 52.7 | 32 | 53.6 | 32 | +3.5 | 13 | +2.6 | 13 |
| Mike Glennon | 74.3 | 75.0 | 27 | 72.6 | 27 | -0.7 | 18 | +1.7 | 14 |
| Ryan Tannehill | 105.5 | 112.5 | 16 | 105.4 | 19 | -7.0 | 27 | +0.1 | 15 |
| Johnny Manziel | 0.0 | 0.5 | 36 | 0.5 | 36 | -0.5 | 17 | -0.5 | 16 |
| Colin Kaepernick | 125.0 | 121.3 | 12 | 125.5 | 12 | +3.6 | 11 | -0.5 | 17 |
| Russell Wilson | 102.6 | 94.5 | 25 | 103.4 | 21 | +8.1 | 8 | -0.8 | 18 |
| Austin Davis | 107.9 | 109.4 | 18 | 110.9 | 16 | -1.5 | 19 | -3.1 | 19 |
| Ryan Fitzpatrick | 101.5 | 103.3 | 20 | 104.7 | 20 | -1.8 | 20 | -3.2 | 20 |
| Brian Hoyer | 98.4 | 101.5 | 21 | 101.7 | 22 | -3.1 | 21 | -3.3 | 21 |
| Matthew Stafford | 124.1 | 132.6 | 7 | 127.4 | 11 | -8.5 | 30 | -3.3 | 22 |
| Tom Brady | 135.3 | 131.0 | 9 | 139.8 | 5 | +4.3 | 9 | -4.6 | 23 |
| Kyle Orton | 61.5 | 65.4 | 28 | 66.4 | 28 | -3.9 | 23 | -4.9 | 24 |
| Robert Griffin | 15.3 | 21.1 | 33 | 21.3 | 33 | -5.8 | 26 | -6.1 | 25 |
| Jake Locker | 50.2 | 58.4 | 30 | 57.4 | 31 | -8.2 | 29 | -7.2 | 26 |
| Cam Newton | 102.6 | 108.1 | 19 | 109.9 | 17 | -5.5 | 24 | -7.3 | 27 |
| Michael Vick | 2.4 | 10.3 | 34 | 10.5 | 34 | -8.0 | 28 | -8.2 | 28 |
| Andy Dalton | 89.8 | 99.2 | 22 | 98.5 | 24 | -9.4 | 32 | -8.8 | 29 |
| Ben Roethlisberger | 129.9 | 133.3 | 6 | 141.5 | 4 | -3.4 | 22 | -11.6 | 30 |
| Derek Carr | 86.5 | 99.0 | 23 | 100.5 | 23 | -12.6 | 33 | -14.1 | 31 |
| Nick Foles | 114.4 | 123.4 | 11 | 132.4 | 10 | -9.0 | 31 | -18.0 | 32 |
| Geno Smith | 89.3 | 111.9 | 17 | 108.0 | 18 | -22.6 | 36 | -18.7 | 33 |
| Teddy Bridgewater | 39.6 | 58.1 | 31 | 59.1 | 30 | -18.5 | 34 | -19.5 | 34 |
| Drew Brees | 132.8 | 138.5 | 4 | 152.6 | 2 | -5.7 | 25 | -19.8 | 35 |
| Blake Bortles | 68.1 | 88.5 | 26 | 89.9 | 26 | -20.4 | 35 | -21.8 | 36 |
As a reminder of how to read the table, take a look at the row associated with Aaron Rodgers. So far this season, he's scored 154.7 standard fantasy points in the passing game. Applying his 7.50 True YPA, 5.42% True TD%, and 2.12% True INT Rate (INT%) in 2014 to his 211 pass attempts, "TFP 14" says he should have 120.3 fantasy points. "TFP," on the other hand, uses "true" stats based on all of Rodgers' throws in the Packers offense (8.09 True YPA, 6.12% True TD%, and 1.99% True INT%), so it says he should have scored 132.7 points in his 211 attempts. Finally, subtracting Rodgers' "TFP 14" from his "Obs FBG" produces a "+/- 14" of +34.4 points, while subtracting "TFP" instead produces a "+/-" of +22.0. Both of these values are positive, so they both suggest he's overperforming his "true" fantasy scoring ability so far this season.
Now, to put a bow on the previous discussion, we can see a similar story in the table, just told in a different manner. As TFP predicted, Tom Brady has risen from 36th in +/- through Week 4 to 23rd through Week 7, and Eli Manning has dropped from 4th to 7th. Meanwhile, Peyton Manning and Tony Romo have defied TFP by jumping from 14th to 3rd and 13th to 6th, respectively.
buy-low candidate #1: blake bortles
To be honest, outside of repeat underachievers like Brees and Foles, TFP isn't all that helpful right now for identifying great buy-low candidates. That's because, as I wrote previously, TFP has a blind spot for younger players -- read: smaller sample sizes -- and most of the lowest-ranked QBs according to +/- (or +/- 14) are youngsters.
That said, if there's one young QB that I would make a play for based on the totality of TFP and other factors, it's Bortles. First and foremost, his low TFP is being driven by the two more-random QB stats. His current TD% is nearly half the league average, while his current INT% is nearly double the league average. However, because these stats take longer to stabilize, RTM suggests his True TD% is 4.52% -- not his current 2.87% -- and his True INT% is 2.65% -- not his current 5.75%. In other words, we should expect to see a higher TD% and lower INT% from Bortles going forward.
What's more, this likely trend could pay dividends given Jacksonville's increasing willingness to put the offense in Bortles' hands.
SELL-HIGH CANDIDATE #1: joe flacco
Last time around, I was reluctant to include Flacco as a sell-high candidate even though his No. 8 ranking among QBs looked fishy in the context of an 11th-ranked +/-. The reason for my trepidation was another TFP blind spot: The Ravens installed a new offense this season, and Flacco's +/- 14 of -0.8 through Week 4 suggested an entirely different conclusion. That's no longer the case, however, with the table showing he's among the Top 6 overachievers regardless of which +/- we look at.
That said, I'm still leaning towards relying on Flacco's +/- 14 for the time being, so here are the details on why it's so high right now. Primarily, Flacco is overperforming his True TD%: In 245 attempts this season, he should have 12 TDs, not 14. Also factoring into the equation is that True YPA and True INT% say he's thrown for 46 more yards and 1 fewer INT than expected.
Finally, if you're looking for non-TFP reasons to sell high, Flacco's stock is pretty high right now, and he's likely a backup QB who you can deal for a starter at another position. Further to that point, trading Flacco to a team that will be starting him means said team will be up a creek without a paddle come Week 11, when the Ravens are on bye. Ideally, you hope that the team desperate for a QB and the team you play in Week 11 are one and the same. I call this tactic -- i.e., trading a player with a future bye week to the team that you play on said week -- the "screw your neighbor" method of fantasy football sabotage; but that's an article best left for the offseason.
SELL-HIGH CANDIDATE #2: PHILIP RIVERS
In short, Rivers is hitting the overachievement trifecta by throwing for far more yardage and touchdowns per attempt than his "true" ability, while at the same time throwing far fewer interceptions. This season, he currently sits at 8.53 YPA and 7.39 TD%, with an INT% of 1.30%. However, based on over 4,300 previous attempts in San Diego, his True YPA is 7.87, his True TD% is 5.38%, and his True INT% is 2.50%. Even if you think the 3,500 or so attempts in Norv Turner's offense don't tell us much about River's "new normal" in Mike McCoy's offense -- which is a perfectly reasonable thing to think -- his +/- for 2014 only suggests that omitting those attempts makes him even more of an overachiever.
Aside from these TFP reasons, it's also advisable for Rivers owners to sell high because his remaining schedule includes three top-8 defenses against fantasy QBs, as well as San Diego's Week 10 bye.
1 The Week 5 table listed 36 QBs. I'm only using 31 of them for this comparison because four of the five omitted QBs (Robert Griffin III, Shawn Hill, Matt McGloin, and Johnny Manziel) haven't had a pass attempt in the past three weeks. The other, Kyle Orton, was known to be the starter going forward at the time, but had yet to throw a pass for the Bills, so he had a TFP of exactly zero and a +/- of exactly zero; hardly the kind of player we should be relying on to assess TFP accuracy.
