Dynasty, in Practice: Touchdowns Really Do Follow Yards. (Yards Still Don't Follow Back.)

A re-examination of the hypothesis that touchdowns follow yards.

Two weeks ago, I wrote about the tendency for the ratio of yards to touchdowns to stabilize within a relatively well-defined and narrow range. I further noted that this stabilization tended to be the result of touchdowns falling more in line with what would be predicted by the yardage totals, and not yards falling more in line with what would be predicted by the touchdown totals.

In short, touchdowns follow yards, (but yards don’t follow back).

Ordinarily, I’d be opposed to revisiting a topic so soon after I first wrote about it, but I wanted to try something different. I wrote last year about how someone gains expertise in a field, and I used the example of chicken sexing, (which is the art of determining the gender of young chicks, and not what it immediately sounded like).

In order to acquire expertise, prospective chicken sexers needed something called “deliberate practice”. To quote from my previous article:

…the role of deliberate practice in improvement is unquestioned. What separates deliberate practice from mere repetition is the fact that it is intentional, effortful, and involves immediate feedback. Think back to the apprentice chicken sexers- they made a determination, then immediately learned from the veteran chicken sexer whether it was correct or not.

Obviously immediate feedback is a problem for fantasy football, where little comes immediately. Instead, we make a prediction during the offseason and by the time we know whether we were right, our faulty mental operating system has completely forgotten the original prediction, (or, worse, altered it to something that casts ourselves in a more favorable light).

I was just thinking to myself that this concept that touchdowns follow yards, (but yards don’t follow back), is the perfect example to see deliberate practice in action. The article two weeks ago posited a hypothesis:

So if you want to look like some sort of crazy future-predicting robot, look for players who are converting yards into touchdowns at an abnormally high or an abnormally low rate… and predict they will cease to do so going forward. It’s the easiest bet in fantasy football.

While there will be noise in the data, that hypothesis is pretty immediately testable. Indeed, three of the top four players marked for an increase in touchdowns going forward two weeks ago scored multiple touchdowns this last weekend.

But that’s anecdote, and anecdote is no substitute for data. So let’s get some data.

A Big Pile of Numbers

Here is a chart of the 100 players from two weeks ago, minus the four who did not play over the past two weeks, (Lance Dunbar, Percy Harvin, Jamaal Charles, and Karlos Williams). “Rank” is where they ranked in the original regression-predicting chart.

I’ve included each player’s fantasy ppg average, as well as their yards per game and touchdowns per game, for both the first five weeks of the season, (labeled “pre”), and the last two weeks, (labeled “post”).

RkPlayerPre PPGPre YPGPre TDPGPost PPGPost YPGPost TDPG
1 T.Y. Hilton 7.6 76.4 0.0 20.2 112.0 1.5
2 Melvin Gordon 6.3 71.2 0.0 1.9 29.0 0.0
3 Todd Gurley 11.1 111.3 0.0 28.3 163.0 2.0
4 Jarvis Landry 8.5 79.5 0.0 16.7 77.0 1.5
5 Alfred Morris 5.8 58.4 0.0 2.0 19.5 0.0
7 Justin Forsett 11.3 97.0 0.2 10.3 72.5 0.5
8 Jeremy Maclin 10.8 96.2 0.2 4.8 48.0 0.0
9 Giovani Bernard 10.7 94.6 0.2 13.3 73.0 1.0
10 Willie Snead 8.3 72.7 0.2 4.0 40.0 0.0
11 Demaryius Thomas 9.1 83.2 0.2 11.1 111.0 0.0
12 T.J. Yeldon 8.5 72.8 0.2 18.4 124.0 1.0
13 Brandin Cooks 7.1 60.5 0.2 6.1 61.0 0.0
14 Jordan Matthews 7.7 65.0 0.2 2.7 36.5 0.0
15 Thomas Rawls 7.7 65.0 0.2 2.5 24.5 0.0
16 Eddie Lacy 7.7 64.6 0.2 2.0 20.0 0.0
17 Calvin Johnson 7.2 64.4 0.2 18.6 126.0 1.0
18 Michael Crabtree 7.6 63.6 0.2 12.3 63.0 1.0
19 Shane Vereen 7.4 61.8 0.2 3.4 34.0 0.0
20 Matt Forte 14.6 121.8 0.4 16.9 89.0 1.0
21 Duke Johnson 7.1 58.6 0.2 7.3 73.0 0.0
22 Jermaine Kearse 6.7 54.8 0.2 3.2 32.0 0.0
23 Antonio Brown 13.3 104.6 0.4 7.7 77.0 0.0
25 Alfred Blue 6.2 49.6 0.2 0.8 8.0 0.0
26 Kamar Aiken 5.7 48.6 0.2 4.4 14.0 0.5
27 Cecil Shorts 6.9 59.0 0.3 6.3 63.0 0.0
28 Darren Sproles 5.9 47.0 0.2 2.4 23.5 0.0
29 Marquess Wilson 5.9 46.8 0.2 5.4 54.0 0.0
30 Mike Wallace 7.3 58.3 0.3 3.0 29.5 0.0
31 Chris Thompson 5.8 46.4 0.2 3.8 38.0 0.0
32 Darren McFadden 5.8 46.2 0.2 22.2 162.0 1.0
33 LeSean McCoy 9.6 76.3 0.3 12.9 99.0 0.5
34 Chris Johnson 11.5 91.2 0.4 11.4 83.5 0.5
35 Markus Wheaton 6.2 45.6 0.2 1.2 12.0 0.0
36 Torrey Smith 6.1 45.4 0.2 7.8 48.0 0.5
37 Latavius Murray 11.1 86.8 0.4 14.6 86.0 1.0
38 Malcom Floyd 5.5 43.0 0.2 5.1 51.0 0.0
39 Danny Woodhead 10.9 84.6 0.4 14.6 86.0 1.0
40 Emmanuel Sanders 10.8 83.6 0.4 16.9 109.0 1.0
41 Amari Cooper 9.8 77.6 0.4 19.3 133.0 1.0
42 DeAndre Hopkins 15.6 115.6 0.6 15.9 99.0 1.0
43 Steve Smith 12.3 93.3 0.5 13.8 107.5 0.5
44 Isaiah Crowell 9.5 70.6 0.4 2.5 25.0 0.0
45 A.J. Green 13.5 99.0 0.6 3.6 36.0 0.0
46 Charles Sims 9.0 65.6 0.4 4.6 66.0 0.0
47 Julio Jones 15.0 106.3 0.7 12.3 92.5 0.5
48 John Brown 8.7 62.8 0.4 15.1 130.5 0.5
49 Adrian Peterson 15.6 116.0 0.8 8.4 83.5 0.0
50 Vincent Jackson 8.9 61.2 0.4 1.3 13.0 0.0
51 Keenan Allen 12.5 88.8 0.6 12.3 123.0 0.0
52 Rashad Jennings 8.2 58.4 0.4 4.1 51.0 0.0
53 Leonard Hankerson 6.9 48.5 0.3 1.9 18.5 0.0
54 Odell Beckham 12.2 86.2 0.6 7.8 48.0 0.5
55 Allen Hurns 11.8 86.0 0.6 10.2 41.5 1.0
56 Carlos Hyde 12.1 84.6 0.6 5.0 50.0 0.0
57 Julian Edelman 15.0 105.0 0.8 8.3 53.0 0.5
58 Dion Lewis 14.5 104.5 0.8 3.9 39.0 0.0
59 Le'Veon Bell 19.6 136.3 1.0 11.3 112.5 0.0
60 Mark Ingram 12.6 89.5 0.7 19.2 102.0 1.5
61 Doug Baldwin 7.8 53.6 0.4 2.1 21.0 0.0
62 Pierre Garcon 8.1 53.4 0.4 7.2 41.5 0.5
63 Brandon Marshall 14.0 100.0 0.8 10.9 89.0 0.5
64 Kendall Wright 9.4 64.0 0.5 7.0 40.0 0.5
65 Doug Martin 14.4 99.8 0.8 17.1 171.0 0.0
66 Frank Gore 10.2 74.2 0.6 8.5 84.5 0.0
67 Khiry Robinson 6.1 41.2 0.3 9.9 39.0 1.0
68 Theo Riddick 6.8 47.6 0.4 6.3 63.0 0.0
69 Marvin Jones 7.1 47.0 0.4 15.5 95.0 1.0
70 Anquan Boldin 7.0 46.2 0.4 7.1 70.5 0.0
71 Ronnie Hillman 6.9 45.0 0.4 11.5 115.0 0.0
72 Terrance Williams 6.9 44.6 0.4 7.0 70.0 0.0
73 Ameer Abdullah 6.5 44.6 0.4 5.6 56.0 0.0
74 Chris Ivory 16.9 108.7 1.0 18.5 124.5 1.0
75 Rueben Randle 6.7 43.4 0.4 5.6 56.0 0.0
77 Bishop Sankey 8.3 53.0 0.5 1.0 10.0 0.0
78 Darrius Heyward-Bey 7.0 41.8 0.4 0.0 0.0 0.0
79 Travis Benjamin 13.0 82.2 0.8 7.2 82.0 0.0
80 Steve Johnson 8.1 51.3 0.5 5.0 50.0 0.0
81 Allen Robinson 12.8 80.4 0.8 15.2 92.0 1.0
82 Joseph Randle 12.3 75.2 0.8 2.4 24.0 0.0
83 Rishard Matthews 11.5 69.5 0.8 11.0 80.0 0.5
84 Donte Moncrief 9.2 55.6 0.6 11.2 51.5 1.0
85 Ryan Mathews 8.2 54.4 0.6 11.4 83.5 0.5
86 Matt Jones 8.0 52.0 0.6 5.1 51.0 0.0
87 DeAngelo Williams 8.6 50.2 0.6 0.9 8.0 0.0
88 Larry Fitzgerald 16.6 98.0 1.2 6.6 66.0 0.0
89 Randall Cobb 11.3 64.8 0.8 3.8 38.0 0.0
90 DeMarco Murray 10.6 60.8 0.8 12.8 98.0 0.5
91 Devonta Freeman 23.4 133.5 1.7 20.3 143.0 1.0
92 James Jones 13.9 78.8 1.0 9.0 30.0 1.0
93 Ted Ginn 9.7 51.5 0.8 6.0 60.0 0.0
95 Tavon Austin 9.7 48.6 0.8 6.4 64.0 0.0
96 Eric Decker 12.0 60.0 1.0 9.7 76.5 0.5
97 LeGarrette Blount 11.7 56.7 1.0 11.2 50.5 1.0
98 Marcel Reece 7.0 33.8 0.6 1.5 15.0 0.0
99 David Johnson 10.2 45.6 1.0 2.1 20.5 0.0
100 Jeremy Hill 9.4 38.0 1.0 12.9 69.0 1.0

Okay, so that’s just a massive data dump and it’s hard to draw any big conclusions from it. So let’s drill down into the data a bit.

For starters, let’s calculate a few correlations. Correlations coefficients, (sometimes referred to as “R”), are quick-and-dirty statistical tools designed to measure how well one cluster of data would seem to predict another. Correlations range from 1, (which indicates two sets of data that move in perfect lock-step), down to 0, (which indicates two sets of data with no linear relationship), then back down to -1, (which indicates two sets of data that again move in perfect lockstep, but with a negative relationship).

The correlation of yards per game in the first five weeks to yards per game over the next two weeks was 0.479. That’s a pretty decent correlation— the guys who got a ton of yards early tended to get a ton of yards later, too. The correlation between touchdowns per game over the first five weeks to touchdowns per game over the last two, on the other hand, was 0.031.

I’ll repeat it, because that’s kind of a big deal: the correlations between touchdowns over the first five weeks and touchdowns over the last two weeks was 0.031. That’s virtually indistinguishable from zero. Statisticians sometimes say that R^2, (which, as the name implies, is merely the correlation coefficient squared), represents how much of the fluctuation in the second sample is “explained” by the data in the first sample. And by that definition, touchdowns per game over the first five weeks explained just 0.09% of the variation in touchdowns per game over the last two weeks.

That's not a typo. I did not mean to put 9%. That's zero-point-zero-nine percent. Touchdowns per game over the first five weeks, it turns out, did absolutely nothing to predict touchdowns per game over the last two weeks. Zilch. Nada.

Given the relative stability of yards per game and the relative instability of touchdowns per game, we should expect to see another tiny correlation between a player’s yard:TD ratio over the first five weeks and his yard:TD ratio over the last two weeks. And, indeed, that’s exactly what we see, with a correlation of just 0.033.

Let’s look at fantasy production, though. Fantasy points per game over the first five weeks had a correlation with fantasy points per game over the last two weeks of 0.385. That’s pretty substantial. Guys who scored more points early were likely to continue scoring more points.

But let’s break that down into its component parts. The correlation between yards per game early and points per game over the last two weeks was 0.467, while the correlation between touchdowns per game early and points per game over the last two weeks was 0.118. Yards alone predicted who would be successful at a substantially better rate than fantasy points did. Touchdowns alone predicted who would be successful at a substantially worse rate.

Finally, to explicitly test my thesis that touchdowns follow yards, but yards don’t follow back, the correlation of yards over the first five weeks to touchdowns over the last two weeks was 0.305. The correlation of touchdowns over the first five weeks to yards over the last two weeks was 0.161.

Or, using the idea that R^2 represents how much of the second sample is explained by the first, past yards predicted future touchdowns at a rate 3.6 times higher than past touchdowns predicted future yards.

Touchdowns follow yards, but yards don’t follow back.

Enough With the Correlations, Give it to Me Straight

So far we’ve had a big data dump, a bunch of inscrutible decimals, and some statistical jargon. Those tell a pretty compelling story. But perhaps the best way to illustrate the difference is to look at the actual numbers.

Based on yard:TD ratio over the first five weeks, I’ve split the sample up into the top quartile, the bottom quartile, and the players in the middle, (what is known as the “interquartile range”). Since we have 96 players, this effectively means the top 24, the bottom 24, and the guys ranked from 25th to 72nd.

Here’s how each group produced over the first five weeks and the last two weeks:

 Pre PPGPre YPGPre TDPGPre RatioPost PPGPost YPGPost TDPGPost Ratio
Top Quartlie 8.43 74.05 0.18 419.6 8.45 60.38 0.41 147.2
Interquartile Range 9.81 71.18 0.46 155.9 9.02 69.73 0.36 194.8
Bottom Quartile 10.92 62.58 0.79 78.8 7.75 57.40 0.35 164.0

Now, be cautious reading too much into the data. The sample size is large enough to be useful, but still relatively small overall. (The “last two weeks” sample consists of 39 games for the top quartile, 40 games for the bottom quartile, and 81 games for the interquartile range.)

But it’s hard not to be floored at some of those results. Touchdowns per game for all three groups were essentially identical, with the “low-touchdown” group actually having the highest of the three groups and the “high-touchdown” group having the lowest! (Again, one shouldn’t read too terribly much into the numbers. The “low-touchdown” group collectively accounted for 16 touchdowns; had that been just 14, instead, the three groups would have been virtually indistinguishable.)

But look at the fantasy point per game differences. The “high-touchdown” cohort outscored the “low-touchdown” cohort by 2.5 points per game over the first five weeks. Over the last two weeks, the “low-touchdown” cohort has led by 0.7 points per game, a stunning swing of over 3 points per game.

Most incredibly, look at the change in the yard:TD ratio for each group. Over the last two weeks, all three cohorts have settled into the 100-200 yards per touchdown range that we identified as the boundaries over a long timeline. Both the low-touchdown and the high-touchdown cohort settled near the 150 yards per touchdown mark that we identified as the long-term NFL average.

In short… regression is really, really, really real. Really.

Where do we go from here?

Two weeks ago we generated a hypothesis. We were able to test that hypothesis and receive nearly immediate feedback, allowing us to engage in the sort of deliberate practice that is the key to gaining fantasy expertise.

What you choose to do with this knowledge is again up to you. But apropos of nothing, I’m just going to put this chart of the season-to-date yard:TD ratios of the top 100 fantasy players right here. Not for any reason in particular or anything.

PlayerGamesFPtsPPGYardsTouchdownsRatio
Jeremy Maclin 6 58.9 9.82 529 1 529.0
Demaryius Thomas 6 56.7 9.45 527 1 527.0
Willie Snead 7 50.1 7.16 461 1 461.0
Brandin Cooks 7 50.4 7.2 444 1 444.0
Duke Johnson 7 49.9 7.13 439 1 439.0
Jordan Matthews 7 43.8 6.26 398 1 398.0
Shane Vereen 7 43.7 6.24 377 1 377.0
Thomas Rawls 7 43.4 6.2 374 1 374.0
Antonio Brown 7 81.7 11.67 677 2 338.5
Justin Forsett 7 75 10.71 630 2 315.0
Giovani Bernard 6 66.6 11.1 546 2 273.0
Todd Gurley 4 61.7 15.43 497 2 248.5
Travis Kelce 7 60.9 8.7 489 2 244.5
T.J. Yeldon 6 60.8 10.13 488 2 244.0
Matt Forte 6 87.8 14.63 698 3 232.7
Keenan Allen 7 85 12.14 690 3 230.0
Jonathan Stewart 6 57.3 9.55 453 2 226.5
LeSean McCoy 5 54.7 10.94 427 2 213.5
Le'Veon Bell 5 81.4 16.28 634 3 211.3
Adrian Peterson 6 79.1 13.18 631 3 210.3
Chris Johnson 7 80.3 11.47 623 3 207.7
T.Y. Hilton 7 78.6 11.23 606 3 202.0
Isaiah Crowell 7 52.3 7.47 403 2 201.5
Rashad Jennings 7 49.4 7.06 394 2 197.0
Charles Sims 6 49.4 8.23 394 2 197.0
Darren McFadden 6 51.3 8.55 393 2 196.5
Steve Smith 6 76.8 12.8 588 3 196.0
John Brown 7 73.5 10.5 575 3 191.7
Calvin Johnson 7 73.4 10.49 574 3 191.3
Michael Crabtree 6 50.1 8.35 381 2 190.5
Jimmy Graham 7 49.5 7.07 375 2 187.5
Anquan Boldin 7 49.2 7.03 372 2 186.0
Marshawn Lynch 5 50.5 10.1 365 2 182.5
Theo Riddick 7 46.4 6.63 364 2 182.0
Jarvis Landry 7 74.3 10.61 543 3 181.0
James Starks 6 46.1 7.68 361 2 180.5
Frank Gore 7 68 9.71 540 3 180.0
A.J. Green 6 71.1 11.85 531 3 177.0
Emmanuel Sanders 6 70.7 11.78 527 3 175.7
Charles Clay 7 47 6.71 350 2 175.0
Carlos Hyde 7 70.3 10.04 523 3 174.3
Amari Cooper 6 68.1 11.35 521 3 173.7
Latavius Murray 6 70 11.67 520 3 173.3
Jason Witten 6 44.4 7.4 344 2 172.0
Ronnie Hillman 6 46 7.67 340 2 170.0
Nate Washington 5 45.7 9.14 337 2 168.5
Doug Martin 6 89 14.83 670 4 167.5
Ameer Abdullah 7 43.5 6.21 335 2 167.5
Rueben Randle 7 44.9 6.41 329 2 164.5
Ben Watson 7 42.5 6.07 325 2 162.5
Torrey Smith 7 46.3 6.61 323 2 161.5
Vincent Jackson 6 45.9 7.65 319 2 159.5
DeAndre Hopkins 7 109.6 15.66 776 5 155.2
Doug Baldwin 7 43 6.14 310 2 155.0
Lamar Miller 7 84.6 12.09 606 4 151.5
Danny Amendola 7 42.3 6.04 303 2 151.5
Danny Woodhead 7 83.5 11.93 595 4 148.8
Terrance Williams 6 41.3 6.88 293 2 146.5
Greg Olsen 6 61.9 10.32 439 3 146.3
Julio Jones 7 103 14.71 730 5 146.0
Martellus Bennett 6 41.2 6.87 292 2 146.0
Brandon Marshall 6 77.8 12.97 578 4 144.5
Travis Benjamin 7 77.5 11.07 575 4 143.8
Dion Lewis 6 78.9 13.15 569 4 142.3
Mark Ingram 7 96.5 13.79 685 5 137.0
Odell Beckham 7 76.7 10.96 527 4 131.8
Arian Foster 4 55 13.75 390 3 130.0
Rishard Matthews 7 74 10.57 500 4 125.0
Pierre Garcon 7 53 7.57 350 3 116.7
Jordan Reed 5 51 10.2 350 3 116.7
Chris Ivory 5 87.5 17.5 575 5 115.0
Kendall Wright 6 51.6 8.6 336 3 112.0
Marvin Jones 6 51 8.5 330 3 110.0
DeMarco Murray 6 67.9 11.32 439 4 109.8
Ryan Mathews 7 63.9 9.13 439 4 109.8
Ted Ginn 6 50.6 8.43 326 3 108.7
Jamaal Charles 5 80.1 16.02 541 5 108.2
Larry Fitzgerald 7 96.2 13.74 622 6 103.7
Matt Jones 6 45.1 7.52 311 3 103.7
Gary Barnidge 7 81.4 11.63 514 5 102.8
Allen Hurns 7 79.3 11.33 513 5 102.6
Julian Edelman 7 96.7 13.81 607 6 101.2
Joseph Randle 6 64 10.67 400 4 100.0
Allen Robinson 7 94.6 13.51 586 6 97.7
Devonta Freeman 7 153.1 21.87 931 10 93.1
Rob Gronkowski 7 106.6 15.23 646 7 92.3
Randall Cobb 6 60.2 10.03 362 4 90.5
Eric Ebron 5 44.8 8.96 268 3 89.3
DeAngelo Williams 7 44.7 6.39 267 3 89.0
Eric Decker 5 55.3 11.06 333 4 83.3
Tavon Austin 6 54.7 9.12 307 4 76.8
Donte Moncrief 7 68.1 9.73 381 5 76.2
Ladarius Green 6 58.4 9.73 304 4 76.0
James Jones 6 78.4 13.07 424 6 70.7
Khiry Robinson 7 51.8 7.4 278 4 69.5
LeGarrette Blount 6 64.5 10.75 345 5 69.0
Karlos Williams 4 51.2 12.8 272 4 68.0
Tyler Eifert 6 70.2 11.7 342 6 57.0
David Johnson 7 54.9 7.84 269 5 53.8
Jeremy Hill 6 59.9 9.98 259 6 43.2


More articles from Adam Harstad

See all

More articles on: Dynasty

See all

More articles on: Stats

See all

More articles on: Strategy

See all