Friday, February 28, 2014

Sloan Sports Analytics Conference, 2014 Edition

Hello from the 2014 Sloan Sports Analytics Conference.

Another year has passed and there's been some major strides in college football analytics. The most visible introduction of analytics to the casual fan is ESPN's Football Power Index, whose predictions were scrolling across the screen during every single bowl game. Behind the scenes there's been good work at Football Study Hall, and I really wish I had more time to do the kind of analysis and data collection they're doing over there.

Looking forward, I think the biggest advances in football understanding are going to come from computer vision and automated understanding of where players are and what they're doing. Current efforts at data collection are manual, spotty, and time-consuming. Computer vision is the up-and-coming technology which has the most potential to provide detailed and reliable information about where the players are, where they were moving, where the ball was, and what a play did or didn't succeed.

The counter-argument is that it's difficult to spot the football, or it's difficult to know where a player was supposed to go on a certain play, but I believe issues such as  those can be overcome. Compared to the money necessary to hire a person, adding more cameras connected to vision software is a more efficient, economic, and scalable solution. Hardware will get faster. Software will get better.

The question after that, obviously, is how to use that data? Right now I believe there are two broad types of questions we're using stats to answer: what happened, and how did it happen? The "what" part can be something like how many (opponent-and-pace-adjusted) points did a team score. The "how" part is whether they got those points on long drives or big plays or because of field position or some combination of the two.

There's a spectrum, and as someone who tends to approach problems from the big picture and drill down, I'm initially more interested in the "what" part since that has immediate predictive value. I realize that other people are more interested in digging into the data and explaining what happened at the line or in a certain type of pass route, but my interest is more in whether or not they should be calling a run play or that type of route in the first place.

Given that it's the college football offseason, there's not too much happening on an ongoing basis right now, and it'll be kind of quiet until August. However we invite you to look at some of the "best of" we've produced, as other posts of interest, including
During the regular season you can expect to find weekly posts showcasing
This is all on top of predictions for each and every game between two FBS teams.

All-in-all it's grown into a pretty complicated system backed by a lot of code we've written over the last few years. If you have any questions or feedback for us, don't hesitate to email us at our tempo-free-gridiron.com addresses (justin@ or eddie@), leave a comment here, or hit us up on Twitter.

Enjoy the conference, and we hope to see you there.

Follow us on Twitter at @TFGridiron and @TFGLiveOdds.

Friday, January 3, 2014

Wednesday, January 1, 2014

Week 19: Wednesday In-Game Win Probabilities

Last updated: Thu Jan 2 08:27:03 2014

UCF52Baylor42Final
Nebraska24Georgia19Final
Iowa14LSU21Final
Stanford20Michigan St.24Final
UNLV14North Texas36Final
Wisconsin24South Carolina34Final