OT- 4th down bot picks best choice live on Twitter, running for rest of NFL games

Submitted by iawolve on December 2nd, 2013 at 10:47 PM

NYTimes is supporting this bot that analyzed 10 years of statistical data and will evaluate the proper 4th down choice during the remaining NFL games for the rest of this year. Just putting a front end on the available tables. No reason any team could not use the technology if it is eventually commericialized. I find it really interesting what big data could bring to in game decisions.


The original article is here



Can also follow it on Twitter for live updates




December 2nd, 2013 at 11:02 PM ^

Well, if I tallied it correctly, the Detroit Lions got it right 63 out of 75 occasions, if the bot is indeed accurate. I would consider an 84% rate of "Good call" on that algorithm rather good, if some other random teams that I've looked at are representative of the NFL as a whole.

Interesting link. Thanks for sharing this. 

Hair Raid Offense

December 3rd, 2013 at 1:40 AM ^

It's pretty easy for a computer to make these calls to punt or to go for it. But in real life the guy who has to make that call is risking his dream job as an NFL coach, his paycheck, livelihood, and reputation. I love coach's that go for it, especially Hoke's 2pt conversion call, but there is certainly a human element that this article cleary misses

Two Hearted Ale

December 3rd, 2013 at 8:15 AM ^

Coaches shouldn't need this. Their job is to know what to do and they should study this to the point it is rote. Good poker players know their odds on every hand. Even a novice poker player should be able to reasonably figure out their odds. It isn't too much to ask that someone on the staff know what to do on fourth down.

While we are on football decisions every punt returner should know whether to field a ball or not based on where it hits. How many balls go into the endzone when they hit the 10 yard line? 5? 3? I'm amazed more punt returners aren't strangled by their coaches when they call for a fair catch on the 5 yard line.

swan flu

December 3rd, 2013 at 8:52 AM ^

The difference is in poker there are 52 cards that never change.  Football is a bit more complicated.  


Either way, these automated decision functions are heuristics at best.  They should never ever considered to be deterministic.


Though ultimately I agree with you.  A good coach should be able to use all the available data (both historical and situational) and consider the constraints and the specific situation in order to make the best call for the given situation.  The best call isn't always going to be the one dictated by these charts.  


December 3rd, 2013 at 8:59 AM ^

I've been making a living at poker for three years and I'm still often not sure of the best play. Basic probabilities are easy to memorize -- like you're about 2-1 against to hit a flush draw with two cards to come -- but what to do with that information isn't obvious. Do you call, hoping to hit and get paid off? Do you raise as a semi-bluff, expecting your opponent to fold often enough to make it profitable? Etc.

Likewise, these football decisions are unique and often difficult. Look at Brian's analysis of Michigan's decision to go for it on fourth down rather than kick a field goal late in the fourth quarter. The numbers swing in unexpected ways based on whether there are 7 minutes, 5 minutes, or 3 minutes left. There are a lot of weird quirks to the numbers. And the specific strengths and weaknesses of each team matter too. These are very difficult decisions that have to be made in a few seconds.