/cdn.vox-cdn.com/uploads/chorus_image/image/66461786/SOW.0.jpg)
Why can’t we just use total yards?
Not all yards are created equal. I’ll demonstrate the problems with total yards, at the team and player level, with two examples:
- Take two TCU games: Against Texas Tech, TCU gained 535 total yards. Against Texas, TCU gained 435 total yards. In which game did TCU’s offense play better? Were both of these games “good” offensive performances? Looking at only total yards, you would be tempted to think that the Texas Tech game was the better offensive game. But it wasn’t - in fact, TCU was 3% better on a points-per-drive basis against Texas than Texas Tech (2.84 to 2.75)! While the magnitude doesn’t seem shocking, the point sstands - total yards is a confusing and noisy stat that can lead us to prioritize and evaluate the wrong aspects of college football offenses.
- On the other hand, consider Oklahoma State running back Chuba Hubbard’s dominant performance against TCU this past fall: 20 carries for 223 yards and two touchdowns. That’s an impressive performance and has lead many to speculate as to the existence of a fundamental flaw in TCU’s defense. I would quibble with not the fact of Hubbard’s dominance, but the magnitude. Hubbard had two huge runs against TCU - and hear me say this: those runs involved talent on Hubbard’s part, execution by his offensive line, and a great play-call/scheme by his coaches - one of 92 yards and one of 62 yards, both touchdowns. Outside of those two runs, Hubbard had 18 carries for 69 yards, a pedestrian 3.833 yards per carry. 69.05% of Hubbard’s yards that day came on two plays! The primary reason he got so many yards on those two plays was not some threshold of skill he possesses, but rather the fact that he got past a defender 92 and 62 yards away from the end zone! Hubbard’s runs, had they come 18 and 25 yards away from the end zone, would’ve required the same level of skill from Hubbard, and he would’ve ended the day with 112 yards on 20 carries, a still very good 5.6 yards per carry, but nowhere near the dominance of his 11.5 yards per carry listed in the box score. The credit we attribute to Hubbard because of those yards, past a certain point, should diminish! The magnitude of Hubbard’s performance is inflated by looking at total yards out of context.
In this article, I will propose two alternate measures to properly evaluate offensive performance in context to help us think through the problems of comparing teams and players with total yards.
Methods
Before diving in, it’s important to remember the goal here: the goal is not to pretend that certain plays didn’t happen or to try and spin any single outcome to confirm my priors. Instead, the goal is to examine some alternate measures of yardage to help us properly assign credit and blame, ultimately providing us more meaningful insight into what happened in any given game.
Here are three measures of total yards I’ll examine today:
- First, total yards = sum(yards_gained)
Why we use this? Always presented in the box score summary, and tells you how far the offense moved the ball. As a general rule, fine: Most of the time, a 600-yard game is better than a 100-yard game, but as those totals get closer and closer, it becomes less and less clear who wins a game based on total yards differential.
- Log_yards = sum(ln(yards_gained))
Why we use this?
The natural log function is often used in economics to eliminate the leverage, or “influence on the total values”, of outliers, or extreme values. We can do the same thing here to “smooth” the effect of yards, dampening the credit we give to teams as yards gained on any given play gets higher. This can help minimize the outliers and assess real quality - when analyzing teams, we often overestimate the magnitude of a single play. By reducing the effect of outlier plays, we can get a better sense of what happened on the aggregate. If log_yards is substantially lower than total yards, then perhaps a single play is biasing our perception of the relative quality of two teams.
- Value_yards = sum(yards_gained/yards_to_endzone)
Why use this?
Not all yards are created equal. The concept of Expected Points Added and Success Rates, both commonly used in football analytics, both get at this idea and do a pretty good job. In using value_yards, I’m trying to capture the value of 3 yards on first and goal from the 3 versus three yards on third and three from your own 45. Both of those are successes, both of those generate positive EPA, but when considering aggregate performance, both of those three-yard-gains have different values. This boils down to a refinement of Bill Connelley’s success rate - he awards a team a success if it gains 50% of yards to go on first down, 70% of yards to go on second down, and 100% of yards to go on third and fourth down (these numbers are broadly consistent with an EPA-based measure of success, as well). Here, I’ll effectively be assigning partial credit to teams for yardage they gain. A gain of three yards from your own 45 will count for .054 yards (3 yards gained/ 55 yards from end zone), whereas a gain of three from the 7 yard line will count for .428 yards (3/7).
Data
Thanks to collegefootballdata.com and the magic of the cfbscrapR package in R, I gathered the play-by-play data for all FBS versus FBS games in the 2019 regular season. The average team gained 5.75 yards per play last year (5.78 in non-garbage time), 5.10 on rushes, 6.46 on passes (5.08, 6.52 in non-garbage time, respectively).
Some quick descriptives: TCU gained 5.38 yards per play, 5.62 yards per rush, and 5.11 yards per pass attempt (5.318/5.62/4.98 in non-garbage time). LSU, the nation’s top offense, average 7.815 yards per play, leading the five teams who averaged more than 7 on a per play basis (Oklahoma, Alabama, Clemson, Ohio State). Akron brought up the rear with a measly 3.82 yards per play, with Power 5 stalwarts Duke, Rutgers, Northwestern, Vanderbilt, and South Carolina not too far ahead. Fun fact - in non-garbage time, LSU averaged 8.12 yards per play!
Here’s a graph of every FBS team’s non-garbage yards per rush and yards per pass attempt.
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/19775673/ypp.png)
And here are all FBS teams plotted for their value_yards (x axis) and log_yards (y axis):
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/19775675/ly_vy.png)
(note: in this graph, don’t focus on the axis values so much as the relative positions!)
Teams further to the right got more yardage in opponent territory, which correlates to more points and finishing drives better. Teams further up had more consistent chunks of yards gained (although the chunk could be 3 yards (bad, like Northwestern) or 7 yards (good, like Ohio State)). Most teams fall on a positively-sloped line, indicating a positive relationship between value_yards and log_yards.
Results
We finally get to the payoff graph: the change in rankings from total yards to value_yards (x axis) and from total yards to log_yards (y axis):
:no_upscale()/cdn.vox-cdn.com/uploads/chorus_asset/file/19775685/ranks.png)
This is a crazy graph! It doesn’t tell us much in the way of relative comparison; I can’t say anything from this graph about Alabama and Northwestern, for example, other than the fact that Alabama was better than total yards suggest (sustainably explosive) and Northwestern was worse!
A large drop in the rankings from Total Yards to Log Yards implies that most of your offensive success came from chunk yards/big plays, which implies you were a sustainably explosive team. A large rise in the rankings from Total Yards to Log Yards implies that most of your offensive success came from small yardage plays, meaning you were very bad at moving the football and had little explosiveness.
A large drop in the rankings from Total Yards to Value Yards implies that most of your offensive success came in your own territory/away from the opponent’s end zone, which implies that you failed to finish drives. A large rise in the rankings implies most of your offensive success came from capitalizing on field position and finishing drives.
BOTTOM LINE: Big drop in Total Yards to Log Yards means your offense is better than total yards suggests, a big drop in Total Yards to Value Yards means your offense is worse than total yards suggests.
From this graph, then, we see that Tennessee, Mississippi State, Virginia Tech, and Washington had underrated offenses, in that they got much of their yardage in opposing territory. We also see that Iowa State, Navy, Houston, Louisville, and Middle Tenn. were underrated in the sense that they sustainably had explosive plays!
Let’s focus on TCU’s offense.
TCU’s Average Ranks:
TCU Total Yards Ranks
Total Yards | Log_Yards | Value_Yards |
---|---|---|
Total Yards | Log_Yards | Value_Yards |
65th | 40th | 75th |
95th | 72nd | 101st |
29th | 40th | 40th |
Overall: 60th, Rushing: 36th, Passing: 90th
TCU rose 25 spots in Log_Yards, indicating the lack of explosiveness from TCU’s offense, generally. That rise puts TCU in the bottom third of college football in offensive production. Whereas a rank of 65th for total yards suggests TCU’s offense was merely average, the Log_Yards metric better informs us that TCU was mainly gaining short yardage on most plays. TCU fell 10 spots in Value_Yards, which suggests that more of their offensive success came on their own side of the fifty. In fact, 60% of TCU’s total yards came more than fifty yards away from the endzone, while only 56.3% of their plays did. The Texas Tech game is a perfect example of that! While TCU had a large total yards number, 61.3% of their yards gained came more than fifty yards away from the endzone, while 56.8% percent of their plays did! TCU gained a lot of offense on their own side of the field but failed to gain yards in plus territory, indicating that the 500+ total yards statistic did not provide us with an accurate assessment of TCU’s offensive performance in Lubbock.
Astute fans of TCU’s offense will recall the offensive decline of TCU’s last few games of the season. While Max Duggan was improving over the season and learning how to be an FCS quarterback, the inconsistency and injury on the offensive line ruined a lot of plans for what might have been a “hot” finish to the season. You need no more evidence than this graph:
this graph:
— parker (@statsowar) March 7, 2020
the information it contains about the quality of TCU's 2019 offense: https://t.co/ZngFKf1ev1
Playing against a bad team, TCU struggled to move the ball in meaningful situations and finish drives, and those trends continued against Oklahoma and West Virginia, where TCU’s offense scored 17 and 24 points in losing efforts.
Discussion
What’s the moral of the story? The moral of the story is that total yards can be deceptive, especially if you’re trying to predict future performance by diagnosing past performance. College football fans, writers, and analysts can do better than total yards when evaluating teams and box scores. Ideally, you would look at success rates, points per drive, starting field position, points per successful play, and turnovers, called the “Five Factors” of College Football by our friend Bill Connelly. Additionally, there are a variety of easily accessible metrics out there that take per-play and per-drive efficiency along with context to rank and evaluate teams. As fans, let’s be better than total yards.