One year ago today, I introduced my True Fantasy Points (TFP) system. (If you haven't read it, stop and click on that link. I'm not going to go into methodoligical minutiae here.) Using TFP is a four-step process:
- Calculate a player's "true" efficiency stats via the application of a basic regression to the mean technique. This adjusts the stats to account for their inherent randomness, essentially separating skill from luck.
- Multiply the player's "true" efficiency stats by his opportunity volume (i.e., passes, runs, etc.) as projected by our very own David Dodds. This results in a player's projection for "true" stat totals.
- Use a scoring system -- in this case, Footballguys standard scoring system -- to convert the player's projection for "true" stat totals into TFP.
- Compare the player's TFP to his Dodds' point projection to gauge how much, if at all, Dodds is over- or under-estimating the player's "true" efficiency.
Just so no one's confused from this point forward, I'll give a preview of what's to come by using Matt Ryan as an example. In 4,530 pass attempts with the Falcons, Ryan has produced a Yards per Attempt (YPA) of 7.23, a touchdown rate (TD%) of 4.46%, and an interception rate (INT%) of 2.36%. These three efficiency stats differ with respect to how inherently random they are, which means that 4,530 attempts has given us a better gauge of his "true" skill in a relatively stable efficiency stat like YPA than his "true" skill in a wildly volatile efficiency stat like INT%. After taking this into account via regression to the mean (Step 1 above), we arrive at Ryan's "true" efficiency stats: 7.25 True YPA, 4.56% True TD%, 2.39% True INT%.
From there, the math is pretty simple. As of July 27th, Dodds projects Ryan to throw 570 passes. If we multiply each of his "true" rates by 570, we get a projected "true" stat line of 4,131-26-14 (Step 2). Converting that into standard Footballguys fantasy points (and adding in Dodds' 73-0 rushing projection), Ryan's projected to score 304.1 TFP (Step 3). Now compare (Step 4). Dodds projects Ryan to score 305.9 points, or 1.8 points higher than his TFP projection. As you'll see, this is one of the smallest differences among quarterbacks this year, which means we can conclude that Dodds' projection properly reflects Ryan's "true" skill as it relates to scoring fantasy points in 2016. If the difference was much higher, say +20.0, we'd conclude Ryan's being highly overestimated due to the unlikelihood that he wildly overachieves his True YPA, True TD%, and/or True INT% -- and vice versa.
In terms of the accuracy of my TFP system, this is only its second year, so a broader assessment will happen when I have more data. That said, reviewing last year's article, it did remarkably well at the margins. Picking a specific cutoff to delineate where exactly "the margins" are located would be disingenuous of me, but if you click on the link and take a look for yourself, you'll see more and more hits than misses as you move towards either end of the TFP table.
Alright, so let's dive into this year's numbers...
True Stats Through 2015
Below are actual passing stats and "true" passing stats for the 28 quarterbacks starting for the same team they played with last season (sorted by True YPA):
|Actual Stats||True Stats|
I'll briefly use Tony Romo's stats as an example of how to interpret all this. With Dallas (since 2007), Romo's thrown for 7.82 YPA, which ranks 5th among quarterbacks in the table. However, when we apply regression to the mean, Romo's True YPA falls to 7.79, and his ranking falls one spot. Similar comparisons can be made for TD% and INT%, both of which also reveal that Romo is who we think he is at this point in his Cowboys career: good for yards and touchdowns, but a tad interception-prone.
Only four quarterbacks rank in the Top 10 of all three true stat categories, but only two rank in the Top 4 across the board, and even ardent fans are probably surprised by one of the two. The obvious one is uber-efficient Aaron Rodgers, who, after having the best all-around true stat profile heading into last year, retains that lofty status entering 2016. The not-so-obvious one is Russell Wilson, whose true stats indicate is a more efficient all-around quarterback than the other two Triple Top-10-ers, Ben Roethlisberger and Tom Brady. I doubt most people put Wilson in the company of Rodgers in terms of passing efficiency. I think that's mostly because the formula Seattle's used to win so many games over the past four years -- from roster construction to game-planning to play-calling -- gives off the impression that Wilson is simply a game manager; more like the guy drafted No. 1 overall instead of Rodgers in 2005, if you will. Hell, even respected former players that are paid to study and talk about football on national television have fallen into this trap.
And just so you don't think that Wilson's high standing is a byproduct of small sample size, here's some stat geekery to assuage that concern. His 1,735 career attempts with the Seahawks means that, for all three categories, we're past the point where luck plays more than a 50/50 role in Wilson's actual passing statistics. Specifically, his YPA reflects 81% skill vs. 19% randomness, his TD% reflects a 62/38 split in favor of skill, and his INT% reflects a 51/49 split. If I were the Seahawks, I too would be talking about putting even more of their offense in Wilson's hands.
Another player you might be surprised to see ranked so highly in all three true stat categories is Tyrod Taylor, perhaps for the same "running quarterback on a run-heavy offense" reason as Wilson. If my stats are to be believed, however, he's an efficient passer, even after heavily regressing his 2015 performance to the mean. He's nowhere near as great as Wilson, mind you, so I'm not advocating Buffalo start passing the ball all over the yard, but Taylor's true stats suggest he's off to a better start as a passer than you'd think.
On the opposite end of the spectrum are Derek Carr and Blake Bortles, two quarterbacks whose "true" stats don't warrant their perceived imminent ascent into the upper tier of the position. Some reading this may say, "But Danny, it's not fair to include their awful rookie years. It's clear their performance took a leap in Year 2." Although I believe it's best to include all observations when we're trying to draw statistical conclusions, fine, I'll play along. If I throw out Bortles' rookie year, then his True YPA increases to 7.36 (which would rank 14th), his True TD% increases to 5.28% (6th), and his true INT% decreases to 2.60% (25th). As for Carr, doing the same thing increases his True YPA to 7.16 (22nd), increases his True TD% to 5.20% (7th), and actually increases his True INT% to 2.42% (10th). While a 2015-only true stat profile definitely reflects better on them, both Bortles (via True INT%) and Carr (via True YPA) still have a glaring hole in their statistical game even when cast in the most favorable light.
But here's the thing about this little statistical exercise. Remember what I said in the introduction about TFP seeming to have higher accuracy at the margins last season? Well, it turns out that the TFP system correctly identified Carr as being massively underrated heading into 2015. And that analysis was based entirely on his rookie season. This is why it's dangerous to aggressively slice and dice your data based on subjective beliefs. There are situations where it's warranted, but they're few and far between. Eliminating half of your sample when that half of the sample just correctly predicted what you're trying to predict isn't one of those situations.
The conclusion to be drawn from the "true" stats for Bortles and -- especially -- Carr heading into 2016 isn't that they're bad; or even average. It's that they're not as good as they're perceived to be relative to other quarterbacks. Before last season, the system didn't think they were as bad as their rookie passing rates indicated, and it was right. After said predicted improvement in their sophomore seasons, the system now thinks Bortles and Carr aren't as good as their two-year passing rates indicate. This is exactly how regression to the mean works. Like a pendulum, stats swing back and forth, but with smaller and smaller swings over time, until at some point they reach an equilibrium. They're getting close, but Carr and Bortles haven't reached that point yet.
True Projections for 2016
So now that we have our true rates for these 28 quarterbacks (Step 1), we can calculate TFP by multiplying these rates by the number of passes David Dodds projects each to throw this season (Step 2), and then applying Footballguys standard scoring system (Step 3). Here's a table with the results of those calculations (sorted by "DIFF," which is the difference between Dodds' points projection and my TFP projection [Step 4]):
|Dodds Stats||True Stats||Points|
After examining the table, I've identified a group of six quarterbacks worth highlighting: two that are underrated per TFP, two that are overrated per TFP, and two that are just right.
Underrated: Russell Wilson
After seeing Wilson's true stats earlier, this one's a no-brainer. But given that he's already ranked so highly in Dodds' projections, I'm including him here to emphasize that the sky's the limit. If Seattle ends up allowing Wilson to throw significantly more than the 484 passes that Dodds projects, he could easily end this season as the overall QB1.
Underrated: Marcus Mariota
The massive difference between Dodds and TFP regarding Mariota is due to discrepancies in all three stat categories. We saw earlier that Mariota's True YPA is 7.52, whereas Dodds projects 7.12; 0.40 YPA adds up over the course of 510 projected passes. We also saw that Mariota has a Top 10 True TD% at 5.02%, whereas Dodds projects 4.52%. Finally, even though True INT% was Mariota's worst true stat, it's also the one that improved the most via regression to the mean because INT% takes the most pass attempts of the three to reliably indicate "true" interception-avoiding skill. In other words, Mariota's INT% is much more likely to go down from 2.70% -- and yet Dodds has it increasing to 3.14%. If Mariota's able to stay healthy and throw as many or more than the 510 projected passes, he should exceed his current QB14 ADP.
Overrated: Andrew Luck
Luck was actually at the top of the DIFF column last season, but I chose not to highlight him in the discussion because I trusted my subjective intuition over objective stats. Fool me once, shame on you; fool me twice, shame on me. And I say that totally irrespective of Luck's nine-game injury absence; focusing on rates, not totals. At the time of last year's article, Luck's actual YPA through 2014 was 7.14, his True YPA was slightly higher at 7.19, and Dodds projected 7.73 YPA. While healthy in 2015, his YPA ended up being 6.42. (!!!) The same phenomenon happened with respect to Luck's historical, "true," projected, and actual TD% and INT% last season.
This year, Dodds' projected YPA (7.20) and INT% (2.68%) are more in line with Luck's True YPA (7.11) and True INT% (2.55%), but his projected TD% (5.35%) remains much higher than Luck's True TD% (4.86%). Compared to the first seven games of 2015, the only meaningful -- and at this point purely theoretical -- improvements to Luck's situation in Indianapolis are a) better pass protection, and b) larger roles for the talented, young duo of Donte Moncrief and Phillip Dorsett. But we have statical evidence to tell us that sack rates are more about the quarterback than the offensive line, and we have the Law of Cause and Effect to tell us that -- Ryan Fitzpatrick excluded -- wide receiver performance depends on quarterback performance, not the other way around. It's not impossible for Luck to throw three more touchdowns than his True TD% suggests, but it's not likely either.
Now, add on top of all this Luck's QB3 ADP. No thanks, I'll pass. You should too.
Overrated: Ryan Tannehill
Just above David Carr and Blake Bortles in the earlier table, you'll find Tannehill, my personal, perennial source of preseason perplexity. Tannehill's true stats were just as bad as that duo's; maybe even worse when you consider how much farther along he is in the statistical skill-vs.-luck continuum. Whereas we're still hundreds of passes away frrom Carr's and Bortles' actual stats reflecting their true ability more than randomness, Tannehill's already 1,852 attempts past the YPA threshold, 1,189 attempts past the TD% threshold, and 567 attempts past the INT% threshold.
And of course, there's the oft-cited (by me at least) research from colleague Chase Stuart showing how rare it is for a below-average quarterback through 32 starts to become a great quarterback thereafter. Note: Even through twice that many starts, Tannehill still hasn't produced above-average passing stats, let alone those on par with franchise quarterbacks.
It's for these reasons that TFP's projection has the 6th-highest divergence from Dodds'. Tannehill's True YPA is 6.96, while Dodds projects him to improve slightly, from 7.18 in 2015 to 7.20 in 2016. Most significantly, the biggest point differerential comes from dissenting touchdown rates. Tannehill's True TD% is 4.22%, but Dodds projects him to leap in a single offseason from a 3.87% career rate to a 4.52% rate in 2016.
To be fair, this is a difference in degree, not in kind. Few, if any, fantasy football players are drafting Tannehill with the expectation of QB1-level stats in 2016; but they are drafting him for value in the hope of it. His ADP is QB21 right now. Dodds ranks him QB15, which feeds into that hope. I'm simply here to temper your expectations: expect the expected (i.e., QB18-QB24) from a player from whom we pretty much know what to expect.
Just Right: Aaron Rodgers
I already discussed Rodgers' entrenched status as the best "true" passer in the league, so I'm just including him here to highlight that the YPA, TD%, and INT% rates implied by Dodds' projections are in near-perfect agreement with that conclusion. All indicators point to a bounce-back season from his low-for-Rodgers-standards QB7 fantasy performance in 2015.
Just Right: Teddy Bridgewater
With a TFP projection almost identical to Dodds' projection, Bridgewater is the 2016 Baby Bear's Porridge Award winner. There are subtle differences in the individual stat rates, but they offset each other. Although Dodds projects a slightly lower YPA than his True YPA (7.14 vs. 7.30) and a slightly higher INT% (2.58% vs. 2.47%), the resulting decrease in projected points is recompensed via a slightly higher True TD% (4.52% vs. 4.23%).
Of course, while it's nice to see such harmony between projection systems, even Bridgewater's ADP is in the same range (QB25-QB30). We're still 6 weeks from the regular season, but if there are no significant changes between now and then, Bridgewater is the fantasy quarterback for which subjective analysis, objective analysis, and public perception agrees the most.