website statistics

Everyday during the month of October, I'll be marking the re-launch of AFA by highlighting a feature here at the site. To help re-build the AFA community, please spread the word. Tell your friends and follow AFA on Twitter, Facebook, or via RSS

The very first feature I'll highlight is the Advanced Team Stat Visualization. That's a mouthful, but it's just a really simple way to get a picture of the entire league at a glance. The viz plots each team according to their offensive and defensive Expected Points Added (EPA) per Game. Good offenses are to the right, and good defenses are up on the plot. So teams toward the top-right are the best performing all-around teams.

EPA is a measure of point expectancy. In this application of EPA, it offers a quantum leap over yards or points as a measure of team or squad effectiveness. For example, if a team with a good defense but poor offense is always defending a short field, they'll end up allowing more points than they are truly accountable for. And a good offense with a poor defense might not be getting the credit they deserve due to the long fields they face.

Since I created this a few years back, I've seen a few imitations--versions for college, versions that plot yards, and even versions for baseball and other sports. Although I'd love to claim credit for inventing the X-Y plot, I think the credit belongs with this guy.

The viz is interactive and offers all kinds of drill-down data for individual teams. Hover your cursor over a team logo and you can see their week-to-week EPA performance against each of their opponents. You'll also highlight the other teams in the same division for comparison.

The viz goes back through the '99 season, and you can isolate stretches of selected weeks as well. For example, in 1999 through week 8, the Jaguars defense was literally off-the-chart good. (Yes, kids. The Jags were scary good back in the later days of the old AFC Central.) 


Pro tip - put the season slider to All and see the average team 'character' over the last 15 years. My hometown team is up and left, exactly where I imaged. You can zoom in by double clicking or (better) click-dragging a box around an area.

Right now the Bengals, Lions, Seahawks, Ravens, and Colts are on the Pareto Frontier.  You can look that one up on your own.

Here's the link, and thanks for supporting AFA!




Here is the player statistics page. This page lists the top players for offensive skill positions by year, offering individual player Win Probability Added (WPA), Expected Points Added (EPA), Success Rate (SR) and a load of other stats. This is the only place to see which players really made the biggest impacts on their team's success--in terms of wins and points. Of course, all football stats are not truly individual stats. Until the day when we figure out how to isolate an individual's contribution quantitatively, individual stats will be shorthand for team stats when this player is involved.

You can isolate games from only the regular season or include playoff games. Each stat is sortable, and for the more obscure stats hovering the cursor over the column header with give you a very brief explanation. For more detail you can visit the Glossary page.



Here is Quarterback Air Yards. This is a fun look at how each QB is getting his passing yards. Is he relying on his backs and receivers to gain gobs of yards after catch, or is he airing the ball downfield to get his yards? For example, would you be surprised that the QB with the highest yards per attempt this season is also 33rd in Air Yards per attempt? Would you be less surprised if I told you that Qb was Andy Dalton?

When I first developed this stat back in 2007 I thought it would be some kind of major breakthrough, but it's really just part of a much larger puzzle. Still it's a novel way to look at QB stats, and tells a lot about their scheme, style, and the players around them. When I first published air yards stats, I explained the concept this way: Why do we credit a QB for a 2-yard dump-off pass to a RB who dashes 50 yards for a TD?



Here is the Time Calculator widget. It's purpose is to quickly and accurately tell when a trailing team can get the ball back in an endgame situation. This is very handy when trying to evaluate late-game 4th down situations and onside kick situations. It's more complicated than you think--down, timeouts, and the two minute warning are all critical factors. Besides, doing math under pressure, even relatively simple arithmetic, is not advised.

Like a lot of the features here AFA, this tool was something I created for myself, but thought would be useful to everyone. I first built it in 2012 to be able to crunch the numbers for the Field Goal Choke Hold scenario. In other words, when should a team intentionally allow their opponent to score a touchdown and get the ball back, rather than let them kneel out the clock and kick an easy field goal?



Here is the advanced offensive line stats page. Offensive lines go overlooked statistically because developing meaningful and quantitative offensive line stats is hard. Really hard. My approach was to start with the basics. What's the purpose of an offensive line and how do we measure it? In one way, the offensive line is strictly defensive. Their job is to protect the ball carrier from tacklers, whether the guy with the ball is a QB in the pocket or a RB darting past the line of scrimmage. Put simply, the purpose of an offensive line is to minimize the damage done by opposing front-sevens (defensive linemen and linebackers).

So I add up all the havoc created by opposing front-sevens (except for LBs in pass coverage) and assign that damage to offensive lines. The damage is measured by summing all negative plays in terms of WPA and in terms of EPA. Here's a more detailed explanation.




Here is the Player of the Week pages. Each week the offensive skill players and the defenders with the biggest impacts on their games are highlighted. For example, this week Tony Romo led all offensive players with +0.75 WPA, meaning his plays accounted for a net improvement in his team's chances of winning by 75%. Considering the teams begin with 50%, anything above 50% is fairly heroic. On defense, Gerald McCoy terrorized the Saints all day Sunday, accounting for +0.45 +WPA. 

The page always defaults to the most recent week, but you can see the top players for each previous week as well.



Here is the archive and search tool (over there to the right). Yes, I know a search box is a pretty lame feature, but let me explain. The search is a custom Google search that spans both all the old articles at the pre-2014 site and all the new articles here on the new platform. Just type in a search term, say game theory, and boom--pages of archived articles dealing with the topic. There are over 1,500 articles there brimming with original research and analysis.

Unfortunately there's a catch. The Google crawlers haven't re-indexed the old site. Until they do, the links all point to "www.advanced..." The address for all the old articles should be "" So as a work-around, simply click on the article you want, suffer through the "404 page not found" error, and replace www with archive in your address bar. You can do that with any of the links at the old site. When Google re-indexes things, this work-around won't be needed.


Here is the Advanced Team Stats page. This page lists and ranks the core numbers for each team's offense and defense--Win Probability Added, Expected Points Added, and Success Rate, each broken down by run and pass. Each stat is sortable, so for example NO, SEA, and CLE lead the league in EPA per run attempt. (But what about DAL and DeMarco Murray? 4 lost fumbles can be costly.)

The numbers go back through the '99 season and can be filtered to either include or exclude post-season games.


Here  is the Defender Stats page. Stats for defensive players can be very tricky. Lots of tackles are a good thing for defensive lineman but can be a bad thing for defensive backs. Player EPA and WPA would be nice, but should a defender be penalized for racing to make a touchdown-saving tackle on a play that gained 50 yards? 

Fortunately, AFA stats such as Tackle Factor and +EPA solve a lot of those problems. The numbers go back through the '99 season and can be sorted however you like. 



Here is the Quarterback Season Visualization. This series of charts illustrates each selected QB's production game by game. It provides a quick and rich way of comparing player production levels and trends. I couldn't decide how to best present the data, so I created a number of different ways to skin the cat. I've come to prefer the version that the page defaults to, which is a game-by-game bar chart. 

You can select any season back through '99 and choose which QBs you'd like to compare.


Here is, what else?...the Live Win Probability Graphs. It's a pretty simple concept at heart. I use the WP model to plot how likely each team would win in real time. The chart updates every few seconds. It includes nifty little features like the probability of a first down, the  Expected Points on the current drive. Hovering over each point on the graph will give you additional information about the game state or play. Every now and then the data feed has a bug or two, and the graphs go bonkers, and I always appreciate a heads up on Twitter or email. 

The live graphs were inspired by their baseball equivalents at


Here is the Player Stats by Team pages. I'm often asked 'Why isn't player X on the stat leaderboard page?' The answer is that he's not in the top 40 players at his position in WPA. But you can find every player at his team page. There's a page for every offense and a page for every defense.

And as bonus features today, I'll point you toward the player pages. Each offensive skill player gets his own career page, listing his advanced stats by season. Drill down even further by clicking on the season link, and you'll see a game-by-game breakdown of the player's advanced stats complete with a direct link to each games Win Probability graph.


Here is the Top Plays of the Week post at Deadspin's little brother Regressing. Each week I provide the biggest plays in terms of their effect on each game's outcome, and Ross Benes write up the article. Some of the plays are heroic and others are horrible blown plays, and for some reason Jay Cutler keeps showing up on the list each week.

Here is the Plays of the Week post from this week (week 6).


Here is the offensive and defensive run/pass visualization. It's actually a little-known part of the advanced team stat visualization. Like its parent viz, it plots team performance in terms of Expected Points Added. These plots put passing performance on the horizontal axis and running performance on the vertical axis. The best offenses and defenses will be in the upper right quadrant of the plot. 
To find the run/pass visualizations, go to the main Advanced Stat Viz here, and click on the Run/Pass EPA or Def Run/Pass EPA tabs.

The teams in the extreme bottom right might do better to almost never run the ball.


Here is the Bayesian Draft Analysis Tool. I realize it's not draft season, but this is one of my favorite features. The tool uses Bayesian inference to estimate the probability each available player will be drafted in each slot, and by extension, the probability each player will still be available at each pick number. This is a helpful bit of information for team personnel decision makers targeting certain players.

The tool is currently frozen in time at the end of the second round of this year's draft. That's when I gave up updating it in real time. Next year's version will be even better and include much richer information.


Here is the Running Back Season and Career Visualization. Just like its brother, the QB Viz, it presents RB performance in terms of Expected Points Added by game and by season. You can select any combination of players and seasons to create your own plot. There are a few different ways to view the data, selectable by tab at the top. You can really see the difference between the impact of running vs passing in the modern game by comparing the RB and QB plots.

The WP Calculator was first launched in 2009. Like most tools at the site, it was something I built for myself to help me analyze game situations quickly. Digging around the inner workings of the model frequently resulted in errors and confusion, so I created a widget to streamline everything. I soon saw the value in making it available publically, and it quickly became one of the most visited features on the site. 

Last year the model and tool got major upgrades, adding the new overtime format, relative team strength, and kickoff considerations. The publically available version is based on the original WP model (version 1.x). The new version available to league teams and other clients, adds a number of additional features and offers more precision and accuracy, plus it adds a cool visualization component.

Here's the link: The AFA WP Calculator

MVP Visualization is probably a terrible name for this fun tool. I built after the regular season ended a couple years ago so I could avoid writing an All-AFA Team write up because, as long time readers know, I'd much rather be making models and crunching numbers than writing about making models and crunching numbers. The MVP Viz plots the top players at each position according to their cumulative Win Probability Added and Expected Points Added. The purpose was to identify the players who truly had the biggest impact on their team's outcomes. Players to the upper right have the biggest impacts. The players above the diagonal line have more WPA (game impact) than we'd expect given their EPA (score impact), meaning they tended to make more high-leverage or 'clutch' plays. Players below the diagonal had more of their production during low-leverage situations, commonly known as trash time.

In all honesty this is the one of the things I really look forward to checking out each Monday morning. If you want a real kick in the pants, click on the DEs and check out where J.J. Watt stands relative to the pack. Or click on 'all' years to see which players define the last decade and a half. The only thing I wish I could do is unite the players whose careers straddle multiple teams. For instance, Peyton Manning ranks as the clear #1 QB of the era just as a Colt. Plus, his time as a Bronco ranks 12th in the entire era. Amazing.

The viz goes back through the '99 season. In addition there are alternate tabs that rank each player with a simple bar chart. Have fun.

Here's the link: The Player Stats by Position Visualization, aka the MVP Viz.


This isn't really an AFA feature as much as it is a New York Times feature, but it still runs on the WP model and methodology AFA readers are familiar with. AFA just makes the Bot's brains and the super talented folks at the NYT do everything else. Some people have the wrong idea about the Bot. It's not intended to be the final word on every 4th down, but rather a 1st cut---although a very good 1st cut.

The 4th Down Bot is a boon to me personally. For all of you bored by reading about 4th down this and 4th down that, trust me, I'm ten times as bored as you. It's nice to be able to offload that stuff to the Bot. And as predictable as it might be to you, the hardcore football analytics fan, 4th downs analysis is still new and interesting to the vast majority of general football fans. Plus the fact remains 4th downs are the juiciest, lowest hanging fruit on the tree of football analytics, and perhaps in all of sports analytics. 

I, for one, welcome our new 4th down overlord. Follow it (him?) on Twitter.


This is another one of my personal favorites. If you go to the QB Viz and click on the Career Total WPA tab and Nth Best EPA tab, you'll see a unique way of comparing QB careers. It takes a little more explaining than I can do in a quick Feature of the Day post, but here is the full explanation. In short, the Career Total WPA graph plots the cumulative WPA produced by each selected QB starting from his first season. And the Nth Best EPA graph plots each selected QB's cumulative EPA production, sorted from his best season to his worst. The QBs with the most 'area under the curve' have enjoyed the more productive careers.

If these plots look familiar to baseball sabermetric guys, it's because they are direct imitations of Fangraphs' very cool WAR graphs.


Here is the Advanced Stat Boxscores for every game. The boxscores list the EPA, WPA, SR adjusted YPA, and other interesting numbers for the top players on each team. It even includes some illuminating offensive line stats. The box scores can be found under the WP graphs. For the current season use this link or the scoreboard ribbon above, and for previous seasons please use this link. They go back through every game since the '99 season opener.

Caution: Clicking through all your most memorable games can be addictive and can result in hours of your work day lost.


I've already highlighted the 4th Down Bot as a feature, so I can't leave the 4th Down Calculator out of the mix. This tool allows you to quickly get a full analysis of any 4th down situation you like. The calculator does two independent analyses, one based on the Expected Point model and another based on the Win Probability model. The tool is based on version 1.x of the WP model, but team clients get access to the new WP model which considers timeouts and a number of other improvements.

The calculator doesn't say Go, Punt, or Kick in black and white. Instead it produces a comparison of the total expected value of each alternative along with a break-even probability of success needed for a conversion attempt to be worthwhile. 

Here's the link: The 4th Down Calculator.


Here is the Game Matchup page. By clicking on any of the future games in the scoreboard ribbon above, you'll get a full team comparison of stats, from simple efficiency rates to advanced metrics like defender Success Count and Tackle Factor. You can select any two teams you like for comparison. Each of the tables are sortable by any column.

Here's a look at Sunday night's Packers-Saints matchup.


Here is the Game Matchup page. By clicking on any of the future games in the scoreboard ribbon above, you'll get a full team comparison of stats, from simple efficiency rates to advanced metrics like defender Success Count and Tackle Factor. You can select any two teams you like for comparison. Each of the tables are sortable by any column.

Here's a look at Sunday night's Packers-Saints matchup.


Here is the Franchize-Season Visualization. This interactive chart lets you track team "identity" over the last 15 years or so through the lens of Expected Points Added. Like several of the other team visualizations, the horizontal axis represents offensive performance and the vertical axis represents defensive performance. Some teams have had characteristically great offenses and mediocre defenses, like IND. Some teams have had much better defenses than offenses, like BAL. Some teams' seasons are packed densely together, like the Jets. Others, like JAX, are spread out all over the place.

The selected team's year-to-year performance is plotted below the main chart, so you can see how a team responded to the arrival of a new coach or QB. You can also see where the teams have tended to cluster and track their evolution from season to season. Only a very few teams have a balanced history, with seasons centered around the league average. SEA, the current default team for all things here at AFA and reigning champs, is one of those teams. 

This was created at the suggestion of Chase Stuart, so kudos to him for a fun idea.


Here is the Top Game Finder. This fun tool lets you search for the most interesting games in recent years. There are two criteria: Excitement Index (EI) and Comeback Factor (CBF). You can search and sort according to year, team, and either stat. Links to the game and its advanced box score is provided.

EI was something I created to measure how exciting a game was. It measures the total vertical travel of the game's Win Probability graph. Close, exciting games with large swings in WP create very high EIs. Boring blowouts have very low EIs. 

Big comebacks are equally compelling. They might not have a lot of total WP movement because one team has a large lead for part of the game, so I created this stat to identify how unlikely it was for the eventual winner to have won. In other words, CBF looks at the lowest WP during the game of the winning team. For example, a CBF of 25 means there was about a 1 in 25 chance that the winner would come back to steal win. Bigger CBFs mean bigger comebacks.


Here is the Adjusted Success Rate page. This is one more interesting way to view team performance. It looks at how consistent teams are at moving the chains and stopping opponents from moving the chains. Plus, it adjusts of opponent strength.

Success Rate was a handy rule of thumb outlined in the book Hidden Game of Football. It defined success as whether an offense gained at least 5 yds on 1st down, half the distance remaining on second down, and a conversion on 3rd down. Football Outsiders' "DVOA" is largely based on this rubric, which demonstrates how useful it is in measuring team effectiveness. When I created a full, 4-down Expected Points model, I thought there might be a alternative way to define SR. I defined AFA SR as whether a play improved point expectancy or not. This approach accounts for things like red zone and goal line "field compression" without any additional adjustments.


Here is the AFA Podcast with host Dave Collins. The podcast is in its second full season, and we've got 30 episodes in the vault. If you haven't sampled it, it's primarily an interview show with a weekly guest who has been a significant contributor to football analytics. Past episodes have featured Virgil Carter, David Romer, Pete Palmer, Jeff Sagarin and Wayne Winston, plus other legends in the field.

The vast majority of the episodes are meant to be timeless and not focused on the league events of the day. So if you haven't been listening regularly, it's worth your time to work your way back through the catalog.

There are several ways to catch the podcast: RSS, iTunes, Stitcher, or here on the site (and in the archives). The latest episode can be played right here on the homepage. And the microphone icon in the upper right corner will take you to the library of past episodes at our podcast hosting site.


Here is the Playoff Projections. The projections feature probabilities of division winners, wildcards, seeding, and playoff outcomes. The projections are based on Monte Carlo season simulation that uses the weekly team efficiency rankings as game outcome predictors.

Toward the end of last season, I developed live in-game playoff probabilities, which were one of the most popular features ever on the site.

We usually don't begin these until at least midway through the season, so stay tuned. But as a taste, here's how the AFC wildcard race looked throughout the last Sunday of 2013:


Here is the weekly game probabilities, found at Sports on Earth this season. The game probabilities was the very first thing I did crunching football numbers. A friend and I were debating the merits of the "defense wins championships" adage, so I thought I might be able to make use of some leftover software from grad school, download some stats, and come up with an answer. I was completely unaware of baseball sabermetrics or other approaches to football analysis, so I was operating with a blank slate.

The real value of the game prediction model isn't really the final output of a game prediction. It's the insight that the model provides. To me, predicting games is mostly just a test of the validity of the underlying model. The model represents a way of understand the inner workings of the sport. And if the model is accurate enough, it validates our understanding of what makes teams successful.