Today we’re talking to Ray, better known as Hawkeye Gamefilm on Twitter, a former NCAA defensive back who got an undergrad degree in computer science and did his graduate work in statistics. Who better to do a deep-dive on football analytics with?
Where did you grow up, go to school?
I grew up in Iowa went to school at a P5 school in the Midwest.
What is the current state of football analytics?
There has always been some analytics used in the football, manual down and distance tendencies etc., but I think its usage is relatively immature at the NFL level. A lot of teams are in the ‘feeling it out’ stages and others barely acknowledge it.
It feels like basketball and baseball are running at the forefront of analytics while football is taking baby steps, do you agree, if so what’s holding broader adoption back?
Yes, I would agree with that sentiment. To me, the two biggest factors in football would be the lack of consistent, high-quality data, and the football culture itself. Many of the sources of analytics and research are outside the ‘circle of trust’.
What needs to happen to trigger a broader adoption of football analytics among coaches?
Software systems need to continue making progress, XOS Digital’s platform has been coming along nicely in recent years and is making strides when it comes to accommodating 3rd party data sources right alongside coaches notes in the practice and game play database.
Where do you see analytics becoming most valuable – for players or on an X’s and O’s level?
I think it can be tremendously valuable for both, but it’s easier to adopt on the player/personnel side.
How can analytics help out with play calling?
The 2nd down anomaly is a great example of something analytics can help with. In the NFL for a long very stretch of time (‘02-’12), coaches chose to run the ball 80% of the time on 2nd down following a failed pass on 1st down. However, when a failed run (0 yard gain) occurred on 1st down, teams only ran the ball around 50% of the time. It’s a topic I talked about extensively with our offensive coordinator during my time working full time with an NFL team in ‘13.
I’ve heard you say that lot of teams look at the plays they practice versus what they call in a game, can you elaborate on that?
Yeah, this was a project I’ve done for two NFL teams now. They recorded the play call from both games and practices in the film tagging system, and then I ran comparisons of the two. It was designed to help steer future game prep and attempt to validate that they were spending time where the staff was expecting to.
So, if a team wants to get started with a more refined analytics process – what are the first steps?
- Hire talented, motivated guys who already have a some technical skill-set and training. I recommend a 3 person analytics group to most teams.
- Invest in training of existing staff and ensure every coach and front office member are at least familiar with what your analytics staff is capable of
Visualization is a big part of understanding analytics, any ideas on that subject?
Visualizations are a product of building a quality data store. Collecting and storing high-quality data allows for impressive visual representations to be built. Tagging film is the main mechanism for coaches and staff members to enter data on players. Specifically, for defensive backs I’ve seen a coach ask for custom columns per play: bucket step on break, feet outside the circle, missed jam, DLBFB(didn’t look back for ball) etc. A basic bar chart on top of that data will show you across the board how guys are performing and where they need to improve. No doubt that kind of thing is do-able without analytics but putting up those charts in a position meeting really lets the player know where they stand.
What is the basic stats/analysis a football team should focus on?
For pure efficiency measure I prefer points per drive, and preferably have it adjusted so that it throws out blowouts(+/- 21pts) and garbage time(kneel downs etc).
It’s limited in that it’s only diagnostic but it’s as clear a picture as I’ve seen. Many people measure yards/play on offense and that’s a useful one as well but I feel it lacks the scoring component. I see yards/play as more of an explosiveness measure.
If you could dream how would your analytics reports look like?
In a perfect world, all data would be recorded after each practice/game and immediately update your database and alerts/reports would be triggered for coaches to read in their email after post-practice meetings finish up. It would contain GPS data for each player in their position group and rep counts for each guy from 1 on 1s, team D etc.
What are the minimum stats teams should chart in a game?
As much as you can and still give coherent timely responses on the head-set. It will vary team to team and system to system.
NFL teams I’ve been around tried to do:
Down/distance ,formation, play call, direction/gap ,scheme, and a root cause if the play fails. In addition to this they had assistent OL coaches IDing fronts vs formation and QB/WR coaches IDing coverages.
You have experience from the NFL and D1 level, how do those teams work with analytics in terms of preparing for an opponent?
In the NFL most teams have their combined staff(personnel, video dept & analytics) produce what they refer to as an ‘Advance Scouting Report’. In my experience in college, we had something similar each week but it was very manually produced and extremely labor intensive. There were some components that had to be manual in the NFL(who signals what on the sidelines etc) but about 80% of our weekly report in the NFL was completely automated. We had the ability to pull scouting grades & pro scouting reports on opponent personnel as well as recent game play calling and statistics. The results/impacts were difficult to measure but I can our process in the NFL(in ‘13) was far ahead of where we were in my NCAA days (‘03-’06)
Predictive analytics is something that corporations are taking a hard look at, do you see any applications for football?
Predictive analytics are really the final goal of an analytics platform. It takes a long time to cull all the necessary data to quickly answer questions with a high degree of accuracy. Detecting financial fraud is a common application of predictive analytics and they have a ton of high quality data to develop heuristics for a ‘normal person’ and will sound alarms when they see activity that wildly deviates from that norm. You can apply the concept to football but it is much more difficult. The only experience I have with anything that in-depth involves personnel criteria and eventual success/failure based on key criterion.
If a team only could focus on a couple of things wrt to analytics (like gameplanning, opponent scouting, in-game adjustments etc), what would you recommend?
Game-planning is a big one. I feel often times teams can fail to fully capitalize on a match-up advantage because they don’t want to become predictable. Things like not going after an overmatched CB when he’s 1 on 1 is a common thing. Another common one is you’ve got an advantage with a TE vs LB/S but fail to target him more than 3-4 times in a game.
Tactically, how do you weigh-in and adjust for personnel changes and how it affects the value of previous data?
This is an extremely difficult thing to account for, especially if a major contributor is missing(QB, RB, key OL spot etc). When an opponent loses their QB, any analytics you have gathered could be totally useless. At that point, I’d say you’re better off moving back toward traditional approaches.
When do you start over with your collected data from games and practices? Every season? After major personnel or scheme changes?
Tough question and I think it varies greatly per level of football. The NFL is relatively homogenous and seems to be circulating a limited pool of ideas and concepts. The HS level is where I’ve seen a lot of innovation and more things I would term as attempting to exploit market efficiencies. The NCAA level is somewhere between and it’s always been fun to watch the evolution of more edgy schemes with the level of athletes the NCAA has. In some cases NFL data is worth studying as far back as 25-30 years IMO (within reason, rule changes make this tough at times). The college game seems to have a shorter shelf life as far as the data goes, I don’t find myself going further back than 10-15 years.
Situational, in-game analysis, seems to be the holy grail of football analytics – what are the main challenges in that area?
- Sample sizes. Single games of football have less than 100 plays (observations) per team. You get 12-16 games worth but in the world of statistics, this is still a very small size
- Getting a system/process in place that works. In the NFL computers are not currently allowed in the booth. This limits you to hand charting plays and spotting things with only the human eye. Teams typically shuffle personnel yearly and continuity rarely forms with this process,
Winning Starts Now. Get your head start on the competition with XO Wizard. We are a playbook management and player learning platform that will make your team win more games. XO Wizard makes it super-simple to create, manage, and share your plays, playbooks, and gameplans, while making it easy to collaborate with your coaching staff and engage your players. Take your play diagramming to the next level with rich drawing tools and video directly in your browser or tablet. Chalk it up anywhere: on the field, on the sideline, or at the office.