Football Analytics Primer: Part 1

Sabermetrics, if you will, are becoming a bigger part of sports today. Baseball is way ahead of the game, with many analytical statistics. Baseball is the easiest sports to evaluate with sabermetrics, because it’s essentially ‘Pitcher vs. Batter’, and there are limited outcomes. Basketball has also added an analytical side in recent years. And then there’s football. Football is the toughest sport to evaluate with sabermetrics because there are many different variables and outcomes in a given play, and it is the ultimate team sport, with 22 different players each affecting the outcome of any and every play. Although it is difficult for sabermetrics to play a role in football, it’s not impossible at all, and that’s what we’re going to look at in this article.

Efficiency Stats vs. Counting Stats

Let’s start with how to evaluate individual players statistically. First and foremost, let’s look at efficiency metrics vs. counting stats. Counting stats include you’re basic yards, touchdowns, etc. Efficiency stats measure effectiveness or productivity on a per attempt/play/drive basis. Here’s a basic example, comparing Tom Brady’s 2007 season and Peyton Manning’s 2004 season:

Peyton Manning `04 Season Counting Stats: 4,557 yds, 49 TDs, 10 INTs, 13 Sks

Tom Brady `07 Season Counting Stats: 4,806 yds, 50 TDs, 8 INTs, 21 sks

Based off counting stats, you’d assume that Tom Brady’s season is superior. But what counting stats don’t tell us is simply just the amount of passes each quarterback attempted. Everyone would agree that a QB will throw for more yards and more TDs and more INTs with more attempts. Let’s take a look at their efficiency stats:

Peyton Manning `04 Season Efficiency Stats: 9.2 YPA, 13.6 YPC, 9.9 TD%, 2.0 INT%, 2.5 SK%

Tom Brady `07 Season Efficiency Stats: 8.3 YPA, 12.1 YPC, 8.7 TD%, 1.7 INT%, 3.5 SK%

Efficiency stats put the players on equal footing. Brady had 81 more pass attempts than Manning did, so his counting stats get a leg up. Yards per attempt is a simple, yet very effective stat, that just divides total yards by pass attempts. As we can see, on any given attempt, Manning will likely gain more yardage than Brady despite having less total yards. Yards per completion averages how many yards a QB gains per each pass that is completed. Touchdown % gives us a number that tells us what percent of the player’s pass attempts end up as touchdowns. Interception % does the same but for interceptions, as does sack % for sacks. The reason efficiency stats need to be looked at over counting stats is because it takes away any advantage a player might have because of more attempts.

The Hidden Game of Football

The Hidden Game of Football is an influential book on football sabermetrics. It was written in 1988 by Bob Carrol, John Thorn, and Pete Palmer. The book’s approach was simple - to analyze American football based on advanced statistics. It revolutionized the way some fans perceived football and it put innovative football statistics on the map. This book is the Godfather of football analytics, and a must read for anyone interested in the topic.

The Hidden Game of Football explored “success rate”, “win probability”, “expected points”, “point differential”, and more. The one we’re looking at here is success rate. Play success rate is a great stat, because it reveals whether or not a team achieved "success" on any given play. The formula put forth in this book is to view a successful first down play as one that picks up 45% or more of the yardage needed to gain a first down. On second down a team needs to get 60% of the remaining yardage necessary to pick up the first. Finally, on third down, a team needs to get enough yardage to pick up the first down, no matter what the distance is.

DVOA and DYAR

The best website for football analytics is www.footballoutsider.com. Their main statistic is DVOA, which stands for Defense Adjusted Value Over Average. The best part about this metric is that it takes into account every play of the NFL season. It compares what a team/player did in each specific situation to the league average in that same situation.

The basis of DVOA is success rate. It’s best explained, of course, by themselves. This block quote
paragraph should help you to understand more about this stat:

This system finds the difference between a player/team’s success in a situation and the expected success/league average success in that same situation. It compares each different play only to plays in similar situations, so therefore it accurately depicts how good a team truly is compared to the league as a whole. Each play is adjusted by the defense’s average success in stopping that type of play. Defenses are adjusted by the average success of the offense they are facing. A positive DVOA represents an above average offense/player or a below average defense. A negative DVOA represents a below average offense/player or an above average defense.

So that’s DVOA. If you still don’t understand it and want to learn more about it, do so here: http://www.footballoutsiders.com/info/methods#dvoa.

Next up is DYAR, which stands for Defense-Adjusted Yards Above Replacement. DYAR incorporates a popular concept in baseball analytics to football, and that’s the concept of the replacement level player. The replacement level player is independent from whom would actually replace an injured player (like Curtis Painter for Peyton Manning), as it’s the same for every QB, the same for every RB, WR, and TE. It’s basically the league average replacement level player, because it would be impossible to find out how many yards Manning contributes above his actual replacement (Curtis Painter) unless Painter is in that situation.

DYAR is similar to DVOA in many ways. DYAR also uses success value points of a player. Instead of comparing the success points to the league average in that same situation, it compares the player’s success value points to the average replacement level player. Like DVOA, DYAR is defense adjusted, or adjusted for the strength of opponent. The success value points are converted into a yards figure – for example: Tom Brady led all QBs with a 2,137 DYAR. This means Brady contributed 2,137 yards above the average replacement level player. So the main concept in DYAR is that it computes success value points (like DVOA) and compares it to the replacement level player (which is independent of team).

Per Drive Stats

Nothing is more fundamental in football than the drive. Every drive has a result. It can be one of 8: TD, FG, Punt, Turnover, End of Time, loss of downs, Missed FG, Safety. With limited outcomes, it becomes much easier to evaluate a sport statistically. By charting the result of each and every drive, both on offense and on defense, you can start to accurately assess strengths and weaknesses. You discard much detail but the results will indicate how efficiently a team used its chances to score and prevent scores.

It’s become commonplace in basketball to evaluate team’s on a per possession basis. The main reason for that is because the number of possessions for teams can differ so much in basketball. In football, the possessions do not differ nearly as much. Still though, the number of drives can still skew or basic stats in football. For example, the Giants had 201 drives on offense while the Patriots had 158. With a difference of 43 drives, we know that the teams are not on equal footing. By eliminating the number of drives each team had, and instead averaging what happened on each drive, we’ll end up with a better idea of a team’s efficiency. Take a look at this example of the "Drive Chart Stats" for last year’s Super Bowl teams, the Packers and Steelers:

Team		Offense				Defense
Team	TD%		PPD	Punt%	TO%	TD%	PPD	Punt%	TO%
GB	24.0%		2.06	40.6%	7.4%	12.1%	1.24	42.9%	16.5%
PIT	21.1%		1.98	42.3%	5.1%	11.4%	1.25	45.7%	17.1%

These stats are simple. TD% is the percentage of drives that end up in touchdowns. PPD is points per drive. Punt% is the percentage of drives that end up in punts, and TO% is the percentage of drives that end up in turnovers.

We can also calculate yards per drive, average starting field position, and FG%. Additionally, we can look at the net per drive stats – which subtracts the offensive per drive stats by defensive per drive stats. For example, the Packers averaged +0.82 net points per drive and the Steelers averaged +0.73 net points per drive.

Another interesting stat that we can compute with drive stats is “Drive Success Rate”, which was introduced in Pro Football Prospectus 2005. It’s defined as:

Drive Success Rate (or more precisely, series of downs success rate), or DSR, measures the likelihood that a team's offense will get another first down (or a touchdown, which the official NFL statisticians also count as a first down) in a given set of downs. And the equivalent defensive number measures how often a defense will allow another first down.

It can be calculated with the formula: (First Downs + Pass TDs + Rush TDs) / (First Downs + Drives)

The Packers had a net drive success rate of +7.3%, while the Steelers had a net of +4.4%. Drive chart stats, which look at how teams perform on a per drive basis, have a high correlation to wins. When picking NFL games, I’m always taking out the per drive stats and comparing the two teams.

Advanced NFL Stats by Brian Burke

Advanced NFL Stats features a variety of analytical techniques and applications. The main stats include WPA and EPA. WPA stands for Win probability added. This is a win probability model, as used in baseball, which estimates the chances either opponent will win the game. The win probability fluctuates throughout the same, since the probability a team will win the game is always changing based on every play. Using this, they are able to calculate the win probability a certain player or coaching decision adds to the game. For example, Drew Brees added about 0.29 value points in win probability per game.

Another interesting stat they use is EPA, or Expected Points Added. Instead of evaluating plays in how many yards were gained, they use a formula to show how many expected points were gained. Just like WPA, they are able to calculate how many expected points a player or coaching decision adds to the team.

They also complete many research articles dealing with strategy, home field advantage, coaching decisions, the draft, etc. www.advancednflstats.com is a great site for football analytics and statistical analysis.

Recap

I hope you learned a lot from this article. Obviously, this is just the beginning of NFL sabermetrics. There’s many more sites dedicated to the analysis of advanced nfl stats. If this becomes a popular article, I’ll be glad to publish a part 2 on this football analytics primer, which will dig even deeper into NFL sabermetrics. Even traditional sports sites like ESPN seem to catching on with advanced metrics, as they will release their new stat called Total QB Rating tonight. I’ve read a bit about that, and overall, I think it could be a useful stat. It’s certainly superior to the flawed Passer rating. As with any stat or metric, you cannot use it as your sole evaluator. If your basis for the argument that Matt Ryan is better than Joe Flacco is that Ryan has a better total QB rating, it’s clearly a flawed argument. We need to look at all the stats, not just one, to make a sophisticated argument.

Thanks for reading, hope you enjoyed. I may write an article next week of part 2 to this football analytics primer – that will hinge on the success of this article.