Should your coach go for it on 4th down? How about a 2 point conversion instead of the extra point?
Where do we see the skill in players, decisions makers and journalists? When does randomness play a bigger role than skill?
Analytics will always fight an uphill battle in football. The sport suffers from playing many fewer games than baseball or basketball, making it more difficult to obtain meaningful statistics.
In addition, important players like offensive linemen do not accrue any statistics. Analytics will never make the impact in football as it has in baseball or basketball.
However, we can make progress, especially on the team level.
This guide summarizes the top 9 articles on football analytics, all of which are freely available on the web. Let it guide you towards what we know about America’s favorite sport.
Is there randomness in big plays?
Big plays matter in football. Duh.
But how much control do teams have over those 80 yard touchdown runs? In particular, does an explosive offense over the first part of the season predict the same type of big plays the remainder of the season? This would suggest skill in the creation of big plays.
The conventional wisdom says yes, explosive plays should be predictive. Why else would Nick Saban and Urban Meyer spend so much time recruiting the nation’s best athletes?
Bill Connelly at SB Nation wanted to see whether the data backed up the conventional wisdom. He started with the idea of successful play, which requires half of the necessary yards on first down, 70% on 2nd down, and all the rest of the yards on 3rd and 4th down.
To measure explosiveness, he only looked at successful plays. This weeds out the bad plays that don’t go anywhere, but still leaves a decent sample size to gauge explosiveness.
For these successful plays, he measured the point value of the play based on the starting and ending yard line. For example, 80 yard touchdown runs are worth more than 20 yard runs.
In his study, Connelly found that success rate is predictive. This means that a team with a high success rate the first half of conference play tends to have a high success rate the remainder of the season.
However, explosiveness does not persist from early to late season. Teams at the top of the nation in explosiveness the first half of the conference season regress to the mean the remainder of the season. This result also holds for defense.
Randomness plays an enormous role in the creation of big plays. Connelly has made a benchmark discovery in football analytics, and it will impact research done in football analytics over the coming years.
To read Connelly’s article on the randomness of big plays, click here.
Why luck plays an increasing role for NFL quarterbacks
Michael Mauboussin, a banker and academic, looked at how skill evolves over time, and how this paradoxically makes luck more important. His ideas impact not only sports but also investing and business.
Mauboussin was inspired by Stephen Jay Gould, the famous biologist who also wondered why no baseball player had hit .400 since Ted Williams in 1941. In exploring the data, he found that batting average has been consistent since that time, which suggests hitters and pitchers have improved at the same rate.
However, Gould found that the variance in batting average has dropped considerably from the 1940s to the first decade of the 2000s. Compared to 60 years before, a higher fraction of hitters cluster near the .260-.270 average. This implies luck is more important in distinguishing top hitters, and it makes a .400 hitter an extreme outlier.
The evolution of skill implies a smaller variance in skill. This means luck becomes more important. This is also true for NFL quarterbacks.
Let’s compare completion percentage for QBs in two periods: the 5 years after the NFL rule changes in 1978 that favored passing and a recent period (2013-2017). I looked at all QBs who had attempted at least 100 passes during a season.
Unlike batting average, completion percentage has increased from 55.7% in the early period to 62.8% in the recent period. Offense has evolved faster than defense in the NFL.
In addition, the standard deviation has dropped from 4.8% in the early period to 4.0% in the later period. The width of the bell curve has become skinnier for the completion percentage of NFL quarterbacks. This 16% decline is almost identical to the drop found by Gould in the spread of batting average.
Interceptions are even more interesting. The pick rate dropped from 4.5% in the early period to 2.4% in the recent period, a stunning decline that has made the forward pass significantly less risky. In addition, the standard deviation has dropped 28%.
With the clustering of NFL quarterbacks closer the mean, luck becomes ever more important in the play of NFL quarterbacks.
To grab the pdf of the Mauboussin manifesto, click here.
When should teams go for it on 4th down?
Brian Burke, the founder of Advanced NFL Stats who now works at ESPN, has done the most complete study on 4th down decisions. The analysis is based on expected points, which measures the strength of field position in terms of down and distance.
For example, consider an offense with a 1st and 10 from their own 27 yard line. Expected points looks at how the next score impacts the game for the offense.
For example, if the offense throws a long bomb for a touchdown, this results in +7 for the offense. Other times, the defense returns an interception for a touchdown, which gives -7 for the offense. In extreme cases, neither team scores on their next 5 possessions before the offense kicks a field goal for +3.
Expected points takes an average of this quantity over all situations in which a team had a 1st and 10 from their own 27 yard line. The calculation gives an expected points of +0.7 for this situation.
To analyze 4th down decisions, Burke calculates the expected points for punting, kicking a field goal and going for it.
The results are startling.
Coaches should go for it much more often than they do. In fact, unless the offense is under the shadow of its own end zone, a coach should always go for it with 2 yards or fewer.
That’s right, Bill Belichick made the correct decision on 4th and 2 on his own 28 against Peyton Manning and the Colts in 2009.
Do you make these 3 mistakes with college football statistics?
For some reason, sacks count as rushing plays in traditional college football statistics. However, a play that ends in a sack began life as a pass play.
To get an accurate assessment of passing and rushing in college football, you must count sacks as pass plays. This adjustment is not necessary for NFL statistics since sacks do not count as rush plays.
In addition, it’s also important to account for pace and strength of schedule in college football. One of my articles discusses these three important points as well as shows how Stanford in 2012 illustrates the importance of these concepts.
Stay up to date on the latest in football analytics
If you enjoyed this guide to football analytics, sign up for my free email newsletter. I’ll keep you updated on all of the latest research findings.
In addition, I offer a curated selection of my best football predictions. This content is not available on the site.
To sign up for the free email newsletter, enter your best email and click on “Sign up now!”
Go for 2, or kick the extra point?
Let’s suppose your team is down 14 points but scores a touchdown. Down 8, should you go for 2 or kick the extra point?
Every NFL coach faced with this decision during the 2015, 2016 seasons has kicked the extra point. However, analytics suggest going for 2.
Kevin Cole, director of analytics at RotoGrinders, provides a simple explanation. The analysis assumes that your team down 14 will score two touchdowns to get back into the game (otherwise, they have no chance to win). It also includes reasonable assumptions about conversion rates simplify the math.
First, consider the case in which your team decides to kick the extra point after both touchdowns. Assuming extra points always convert, the game goes into overtime. Let’s assume your team has a 50% chance to win in overtime.
What if your team goes for 2 after the first touchdown when they are down 8? Let’s assume your team has a 50% chance to make the 2 point conversion. After the next touchdown, your team kicks the extra point to win the game. By choosing to go for 2 down 8, your team have already made up the 50% win probability from kicking two extra points.
However, your team can still win if they miss the 2 point conversion after the first touchdown. After the second touchdown, your team tries for two again. If you convert, they now have a 50% chance to win in overtime. When working out the math, this gives an additional 12.5% win probability.
Analytics favors going for 2 down 8 points. Yes, there’s the chance you lose in regulation by missing both 2 point conversions. NFL coaches are probably thinking overtime when deciding to kick the extra point down 8.
However, the coach can increase his team’s win probability by going for 2 down 8. To check out Cole’s article, click here.
He also references a more in depth study on going for 2 by FiveThirtyEight. To learn why you should go for 2 down 4 but not down 1 late in the game, click here.
What is the NFL’s most efficient play?
This question bothered Robert Mays of Grantland, so he enlisted the help of ESPN’s analytics group. They found ample evidence that the play action pass is the most efficient play.
To show this, they looked at expected points added (EPA) for different plays. EPA is the change in expected points, the same concept used by Burke in his 4th down study, on a given play. This statistic acknowledges that 2 yards on 3rd and 1 is worth more than 2 yards on 1st and 10.
Mays and the ESPN quants found that the play action pass earned the highest EPA of all plays. And it wasn’t even close. Running plays lost expected points on average (-0.04 EPA), while passes averaged +0.04 EPA. The play action gained +0.17 on average, 4 times more than the typical pass.
Deception matters in football. On a play action pass, the offense fakes a run, which freezes the linebackers. This frees up space down the field for a big pass play.
Moreover, the data suggests no relationship between the quality of the run game and play action. For example, Minnesota had a strong rush attack with Adrian Peterson. However, the Vikings were only 21st in play action EPA.
Play action passes are pass plays, and a team needs a good quarterback to make those throws. The top teams in play action efficiency have quarterbacks like Aaron Rodgers and Peyton Manning.
To check the article on football’s corner 3, click here.
Can a defense force turnovers?
Was something wrong with the mighty Steel Curtain? Had star safety Troy Polamalu finally whacked his head into the line of scrimmage one too many times?
The numbers suggest otherwise. Bill Barnwell of Grantland asked whether turnovers forced in the first 5 games can predict turnovers forced the remaining 11 games. Over a huge sample of more than 600 NFL teams, he found no correlation between early and late season forced turnovers.
This lack of correlation implies that turnovers regress to the mean. For a team like the 2011 Steelers, this means they will most likely force an average number of turnovers the remainder of the season.
After Barnwell’s article appeared, the Steelers forced 13 turnovers the remaining 11 games, much closer to the NFL average of 1.6 turnovers per game.
To read Barnwell’s article on turnovers, click here.
The power of simple division in recruiting rankings
Johnny Manziel was a 3 (out of 5) star recruit who went to Texas A&M. He won the Heisman Trophy in 2013 as a freshman.
In contrast, many highly touted high school players flop at the college level. Over a recent 5 year period, only 42 of 316 college players named to All-American teams arrived on campus as 5 star recruits.
However, these stories and numbers are misleading.
Matt Hinton noted that only 158 high school players were 5 star recruits during the 5 year period (2008-2012). With simple division, 26.6% of 5 star recruits ended up as All-Americans. The success rate at becoming All-American goes down for 4 and lower star recruits.
Simple division is a powerful tool in sports analytics. You don’t need the most modern machine learning algorithm to make an impact in the field.
Hinton’s CBS Sports article also looks at how recruiting rankings fare on the team level. To read the article, click here.
Can teams consistently beat the NFL draft?
NFL teams rely on the draft to build their teams. A few stunning drafts can lead a franchise to the Super Bowl. A series of lousy drafts almost certainly gets the GM fired.
Neil Paine at FiveThirtyEight asked whether teams can consistently beat the draft. To measure the quality of a team’s draft, he used Approximate Value, a system that assigns a point value for every player’s performance no matter the position.
Then he looked at the correlation of the value of a team’s draft one year with the next year. A high correlation implies skill in drafting players, as good teams can do this year after year, while the bad teams suck every year.
Instead, he found zero correlation from year to year in the quality of a team’s draft. He got the same result when looking at year to year data for the same GM.
In a last ditch effort to find skill, Paine looked at how a 3 year period of one GM affects the next 3 years of his tenure. This restricts the data set to executives that survive for 6 years, an eternity in the modern NFL. This data shows a weak correlation, which does suggest a tiny amount of skill in picking players.
However, randomness plays a big role in a team’s success at any one NFL draft. To read Paine’s article, click here.