Caution: The problem with Yards Per Point as a football efficiency metric

The smart football fan is always hoping to find a single number that captures the strength of a team. Much like baseball has on-base percentage for hitters, this metric would rate the skill of an offense or defense. It would separate the contenders from pretenders.

At the Sloan Sports Analytics Conference in 2013, football coach Tim Chou proposed Yards Per Point as this essential metric for football. For offense, take yards gained and divide by points scored. Substitute yards allowed and points allowed for the defense.

On offense, lower values for Yards Per Point mean a more efficient offense. This unit tends scores touchdowns instead of field goals after long drives. On defense, higher values of Yards Per Point corresponds to better defenses. These defense bend but don't break, letting the opposition to sustain long drives before forcing a field goal or turnover.

In his talk, Chou showed that the difference in Yards Per Point (defense minus offense) correlated strongly with winning percentage in college football. Hence, Yards Per Point is a good efficiency metric to judge a football team.

However, there are problems with Yards Per Point.

An efficiency metric should reflect skill

In sports, a useful metric reflects the skill of a player or team. A skill should be repeatable over the course of the season and less subject to the vagaries of luck.

To determine whether a metric reflects a skill, it is common to look at the correlation between its early and late season values. If the correlation is weak, then this metric regresses to the mean. Randomness plays a large role in its value. If the correlation is strong, this metric represents a skill that persists throughout the season.

For example, consider forcing turnovers in football. While many people think this is a defensive skill, early season turnovers have almost no correlation with late season turnovers. There is a lot of randomness in forcing turnovers, and good handicappers have used this for years to predict the outcome of football games.

Yards Per Point tends to be random

For the difference between defensive and offensive Yards Per Point, the correlation between early and late season values is weak. In college football, the values in the first six games of the season explains seven percent of the variance in the values the remainder of the season (correlation coefficient: 0.27).

Yards Per Point tends to regress to the mean because random events like turnovers affect its value. For an extreme example, suppose an offense gets the ball on the opponent's 2-yard line after a turnover. If the offense scores a touchdown, they earn a Yards Per Point of less than 1, an amazing value for an efficient offense. However, the turnover played a huge role in this score, and turnovers are as predictable as flips of a coin.

As an example, consider Kansas State's offense from 2012, the unit led by quarterback Collin Klein. Most traditional statistics did not rate this unit as elite. For example, the Wildcats ranked 28th in the nation for yards gained per play. However, Kansas State's offense led the nation in Yards Per Point (10.3). The 31 turnovers their defense forced (14th best in the nation) most likely played a big role in their efficient offense.

Yards Per Point does not isolate a unit

It's also useful when a metric can isolate a unit on the field, like offense, defense and special teams. However, Yards Per Point does not do this well.

For offensive Yards Per Point, defense and special teams play a crucial role in this metric. First, these units can score points. These scores lower Yards Per Point since no offensive yards are gained on these scoring plays. This assumes that return yards are not included in Yards Per Point, the method Chou used in his talk.

Second, the defense and special teams affect field position. The 2012 Kansas State team is a great example of how special team can enhance an offense's Yards Per Point. They ranked first and second in kickoff and punt returns. Their average of almost 30 yards on kickoff returns makes the offense look good by Yards Per Point.

For Yards Per Point on defense, the offense and special teams also play a role through turnovers, touchdowns allowed and field position.

Yards per play is a better efficiency metric

Given the problems with Yards Per Point, is there a better efficiency metric for football?

You're probably already familiar with S&P+ (Success Rate plus Points Per Play) that Bill Connelly tracks at Football Outsiders. I'm also a big fan of yards per play. It correlates with winning percentage almost as strongly as Yards Per Point. (For the 246 teams in the bowl and championship subdivisions in 2012, yards per play and point explained 67% and 71% of the variance in win percentage respectively). It's surprising that yards per play comes this close, since this quantity does not consider points, the currency of wins and losses.

Moreover, early- and late-season values correlate much more strongly for yards per play than Yards Per Point. While Yards Per Point explains seven percent of the late season variance, yards per play explains almost 40 percent. Yards per play is more of a skill because turnovers do not affect its value.

Finally, yards per play isolates a unit on a football team. Defense and special teams have no effect of offensive yards per play.

But yards per play does not have all the answers. Its value is sensitive to big plays. Suppose a defense forces two straight three-and-outs without allowing a yard but then gives up a 70 yard touchdown. The defense has a poor 10.0 yards per play (compared with the FBS average of 5.45) but has not given up points on two of three drives.

Still, I believe yards per play is a useful efficiency metric in football. At The Power Rank, we adjust this metric for strength of schedule to rank offenses and defenses. To check it out, click here.