When the season kicked off in September, I had promised several new and interesting features. The game graphs are now on NBC Sports, we've started a great podcast, added new site features, and added new contributors. But there was one feature that has been in the works until now.
In partnership with the New York Times, I'm pleased to unveil the bane of every football coach in America--The 4th Down Bot. At its heart, it is a real-time application of the 4th Down Calculator feature here at ANS. It uses both the Expected Points model and the Win Probability model to estimate the best option for every 4th down as a game is in progress.
As always, the models' estimates are league-average baselines, and local factors do matter. The recommendations are starting points of analysis, not the end of the discussion. Still, this should be a fun and entertaining way to spread the word about the capabilities of analytics.
It's been five years since I began applying both the WP and EP models to in-game decisions and strategy, and four since the 4th down heard round the world. But the 4th down issue remains the lowest hanging fruit on the analytics tree. Even I'm tired of analyzing 4th downs, so it's nice to be able to delegate the task to a Twitter feed.
My thanks to visualization and interactive multimedia pro Kevin Quealy for his hard work putting the project together. Well, that's it. I better get back to work. These football stats aren't going to crunch themselv...never mind.
Official link
Subscribe to:
Post Comments (Atom)
Brilliant
This is perfect! I wish I had this for Madden!
This example is an impossible sell, going for it from your own 17, up by 7, with only 4:38 remaining in the game:
http://www.nytimes.com/newsgraphics/2013/11/28/fourth-downs/play.html?gameid=12012013_ATL@BUF&playid=215936
It will be tough for this bot to gain credibility if it makes suggestions that no coach, no matter how aggressive, would follow. Even if the math behind it is 100% accurate.
I wish I had a dime for every time a commenter gave me advice using the word 'credibility'. It's usually in the context of the team rankings or game probabilities. Did it every strike you guys I'm not particularly concerned about credibility? I do the math and let the credibility take care of itself.
This is fantastic, even with the finite sample size in comparison to the ridiculous amount of possibilities. I'd bet it would whole-heartedly agree with Belichick's decision to go for it in his own end against the Colts a few years ago.
That Bills-Falcons example does seem a bit odd, since the EP model doesn't say to go for it in that field position unless you were at more like 4th and 3 or 4, or to go for it on 4th and 7 unless you were basically at midfield. I'm guessing there might be something a little fluky in the numbers for that late in the game; the bot says that the 4th and 7 at 4:38 is converted 57% of the time but the three other 4th and 7's earlier in the game are converted more like 42% of the time.
Also, irony: there is now a robot publishing on the NYTimes website, but I have to pass a captcha to publish a comment here.
I don't see any instances of coaches going for it and the bot saying punt. I haven't looked at them all yet, but is this just another sign that coaches are way too conservative? coincidence?
This bot absolutely needs a nickname, i'm going with "Riverboat RAM".
"I don't see any instances of coaches going for it and the bot saying punt."
Just had a look at my new hero, Riverboat Ron.
Against MIA on 4th-10 he was going for it. Bot says no. He got more aggresive than the bot. I love it.
Karl, Germany