MMW: Introducing The Win Probability Metric
Cold, Hard Football Facts for August 16, 2011

By Luis DeLoureiro
Cold, Hard Football Facts Waterboy

On Thursday night, the New England Patriots racked up 476 yards of net offense.  Tom Brady didn’t play a down, but the Pats still managed over 300 yards in passing offense.

But, as is the case for many teams in many games – preseason or otherwise – much of New England’s production came when the game was well in hand.  Ryan Mallett threw for 164 yards in the second half and had almost no impact on the outcome of the game.

The NFL has long used volume statistics to rank offenses and defenses.  

But, lately, we’ve been asking ourselves if there is a way to differentially score an offense’s production based on the game situation – and, as such, the goals of an offense in each unique situation.

We decided to rank offenses based on each drive’s impact on a game.  

Armed with drive information (and our trusty excel spreadsheets), we created a model that estimated the historical probability of victory based on unique game situations:  current score differential, time remaining, down and field position.  

This model allows us to estimate the impact an offense has on a game’s outcome by comparing a team’s probability of victory at the beginning and at the end of each drive.

In some situations, an offense is only successful if points are scored.

For example, in Week 1 of 2010, the Chicago Bears were losing 14-13 to the Detroit Lions when their offense took the field with 3:21 to go.  Because of the score and time remaining – and the fact that the drive started at the Chicago 44 – The Bears started the drive with a 34.7% chance of winning.

The Bears scored on the drive to take a 19-14 lead with just over a minute and a half to play – leaving them with a 76.8% chance of winning.

The drive yielded a swing of 42% in the probability of victory.

On the other hand, if the Bears had been 20 points ahead or 20 points behind, the touchdown would have had a negligible impact on the likely outcome.  In fact, in such a situation, the time elapsed may have a bigger impact than the yardage accumulated or points scored during the drive.

To calculate our metric, we added up the win probability increases (or decreases) over all of each offense’s 2010 drives.  

Until we think of something better, we’ll call this the Win Probability Metric (WPM).

The results are in the table below:




A few notes:

ONE – Despite ranking eighth in total yards, the New England Patriots finished first in WPM.  Because of an underperforming defense, New England’s offense was frequently on the hook to score points in close games.  

And they usually came through.

TWO – The AFC Champion Pittsburgh Steelers finished 28th in WPP.  Why?  Because Pittsburgh’s special teams and defense frequently put the offense in favorable positions.  In other words, the Pittsburgh’s offense is not prolific because it doesn’t need to be.

Consider this example from early in the 2010 season.  Antonio Brown gave the Steelers a very early lead when he returned the opening kick-off 89 yards for a touchdown.  The Titans fumbled the ensuing kick-off, giving the Steelers the ball at the Tennessee 46.  

By the time Dennis Dixon came on the field to take his first snap, the Steelers already had a 66.8% chance of winning.  The metric is intended to reward an offense for doing whatever a situation calls for – e.g., eating up the clock when the team gets an early lead.  But, because Pittsburgh’s offense was put in favorable positions so often, they had less opportunity to move the WP needle.

THREE – The top five teams in WPM all made the playoffs.  Only three of the top five in total offensive yards made the playoffs (The Eagles and Colts made it while the Chargers, Texans and Giants did not). Of course, the Seahawks and Bears were 31st and 32nd, so there's that.

Finally, WPM could lead to the first real quantification of clutch play.  Clutch situations are more often associated with larger swings in probability of victory.  Points are relatively more beneficial while turnovers and three and outs are relatively more costly when the game is on the line.  

We can’t make any promises, but a future metric teasing out and quantifying clutch play may be on the radar………of course, ten years ago, we also thought a healthy middle age would be on our radar….