Published: 31 December 2014
I promise this is not filler content, an excuse to have a post on an otherwise slow holiday week. This is the sixth year I've done a best of post, and each one does a good job highlighting the progress made by the site. The previous best of
posts can be found in the Required Reading module at the Archive
. If you're a new reader, then these will help you catch up.
2014 was undoubtedly a year of change around here. 2014 brought a name and address change and the site was rebuilt on a completely new platform. It has a new look with far fewer ads, and it has a new more professionally-oriented focus. The Win Probability model was completely revamped from the ground up, and several new tools and features were added throughout the year including a game simulator tool and a Bayesian draft analysis tool. I think any one of those things would have made the year a fruitful one.
Our most popular features continued largely unchanged, including Sterling's rankings posts, Keith's game analysis, and Dave's AFA Podcast. The weekly game probabilities continued but moved to Sports on Earth
, and the 4th Down Bot
continued its real-time analysis at the New York Times.
Here is a selection of some of the best and most interesting research and analysis at AFA for 2014. First I'll touch on some of the best stuff from our regular contributors, and then I'll walk through the year by month highlighting the rest of the content most worthy of your attention.
Keith Goldner continued his weekly game analysis column with looks at some of the more interesting questions in today's game. He looked at whether Seattle should have refrained from scoring a touchdown
in their playoff game against New Orleans last January. Here's Keith's take on how Seattle was upset by St. Louis
earlier this year. And here's Keith's analysis of some of Chip Kelly's frustrating decisions
earlier this season.
In addition to his weekly team efficiency rankings write-up, Sterling Xie does some great analysis. Here's Sterling's look at whether late-season momentum can explain big playoff upsets
. Sterling also looked at how young QBs' interception rates tend to change
(or don't change). Here's Sterling's updated look at run-pass balance in the red zone using EPA and WPA
AFA continued its partnership with the Harvard Sports Analysis Collective this year. Here's one of my favorites from HSAC on whether Wonderlic scores correlate with QB performance
You might think that AFA is a football website with a companion podcast. Pretty soon you might think of it as a podcast with a companion website. Dave Collins has been working hard bringing interesting guests on the podcast. Here is a collection of my favorite guests this year: Ravens rookie guard and published mathematician John Uschel
, professor and draft researcher Cade Massey
, Football Outsiders founder Aaron Schatz
, and ESPN national analyst Mike Sando
Now on to my own contributions. January began with a look at the worst article about football analytics from the previous year. I took a few moments to dismantle it point-by-point
Surprise! The AFC Championship game pit Peyton Manning and Tom Brady against each other. Here's a neat visual way to compare the careers of great players
My multi-part series on momentum in football concluded with a search for series-level momentum
One of the final frontiers of game analysis is estimating the value of timeouts. Here are two posts that reported the first time
I was able to gain real traction on the topic
The KC-IND playoff game was one of the most remarkable post-season games ever. KC blew a 28-point lead in the 3rd quarter. I wondered whether KC might have been better off kneeling out the entire second half
. After writing that post I realized we'd need some new tools to answer outside-the-box kinds of questions like this.
February means it's time for the MIT/Sloan Sports Analytics Conference. I wrote an essay on what I thought about what defines the field of football analytics and where AFA stands
within that field.
I was added at the last-minute as a panelist at Sloan
. I was a fill-in for (no kidding) Julian Edelman, who dropped out a few days ahead of time. I think that fortunate for the conference, because otherwise there would have been no football analyst on the panel about football analytics. Here's a video of the event
For much of the rest of the off-season, I spent my time building new models and tools. In March I took a closer look at how and when coaches use their timeouts
April is lacrosse season in these parts. I took a break from football to build a simple WP model for lax
The new Bayesian Draft Analysis Tool began to take shape
. It became fully functional
in time for the Draft in May.
Most of the work building the site's new platform took place over the summer, but I took some time in July to do some deep thinking about Win Values for the NFL
In August I began gearing up for the new season with a sneak peek at the all-new Win Probability model
The 2014 season arrived in September, and it means plenty of fresh analysis at AFA. To begin the year, I reported on my project to predict the entire season, including the outcome and score of every game
, for ESPN The Magazine
. It was, of course, an impossible task, but it turns out the season went according to script with very few exceptions.
I partnered with Deadspin's
sister site Regressing
to create a feature on the biggest plays of each week
The finer grain of the new WP model allows us to measure very small differences in a team's chances of winning. It turns out teams are leaving little bits of free WP all over the field
The new model also factors in timeouts, which means we have for the first time an effective method for analyzing replay challenges
The 2-pt chart that almost every coach uses was created for college football in the early 70s. It was time for an update
. I also created a new online tool and visualization for teams to use for 2-pt decision support
October marked the re-launch of AFA on a completely new platform
. It's much more than a new look. The new site allows me to do some really neat stuff for team and media clients.
The all new WOPR game simulation engine lets us ask questions outside the realm of what has happened before. For example, when should offenses start running their four-minute drill
? Hint: It's not at four minutes. And what if a team never ran and always passed
In November I looked at whether kick returners would be better off always taking a touchback
New onside kick analysis tools
were added to the client side of the site.
The WOPR helped answer a long-standing football conundrum. Should offenses down two scores late take an immediate field goal to save time for a subsequent TD drive
Here's a good example of using the new WP model to analyze a replay challenge
Atlanta may have blown a game (and ultimately the NFC South championship) with some questionable clock management decisions
In December I set out to make the WOPR Simulator publically available. I had two purposes--to expose the model for public validation
and to allow users to do their own analysis
In collaboration with a couple young developers, we launched a mobile version of the Live WP Graphs. Check out the app 'Fifty' on the App Store
And finally, live playoff probabilities were relaunched
for the final couple weeks of the season. The good news is that the revised version can work throughout next season, so we can get live playoff updates all year long. This season, they'll continue to run to update Super Bowl probabilities in real-time as the playoffs progress.
I say this every year: I always think the preceding year was a lean one with not much new content or many new features. Then I do the annual Best Of
post, and I'm amazed at everything we've accomplished over the year.