Football analytics Q&A with data expert Trey Causey

The New York Times' Fourth Down Bot

Ever since Tuesday’s Cleveland Browns hiring of Paul DePodesta, the Internet has been buzzing with descriptions and stories of football analytics. For many, the evolution of sports analytics from the Moneyball days of the Oakland Athletics, to the Daryl Morey-ball play of the Houston Rockets, and also to the NFL, was seemingly inevitable.

But what really is football analytics? What does it mean? How can it help the perennially bad Cleveland Browns? I’ve wrestled with those questions in my head over the past 48 hours. And while I may be the Internet’s foremost sports analytics cataloguer — a bold proclamation, to be sure — I didn’t feel qualified enough to answer those questions myself.

If I were to curate some type of primer on football analytics for the Cleveland sports audience, the bulk of the post would have been citing individuals like Bill Barnwell, Brian Burke, Chris Brown, Chase Stuart, Bill Connelly, Benjamin Morris, et al. Instead of doing that, I felt the best thing to do was just to ask someone more qualified than me and have them answer some questions. And so I asked Trey Causey.

Causey writes online at The Spread, a website he founded that is dedicated to data science and sports. He once wrote an entire post on a blueprint for an analytical NFL franchise. He’s also one of the current minds behind the New York Times’ 4th Down Bot. I met Trey at last year’s Sloan Sports Analytics Conference and we’ve kept in touch since. He’s one of the top thought-leaders in the online sports analytics community, and he just happens to concentrate on football.

Jacob Rosen: So let’s just start from the basics. To you, what is football analytics? I often cite this Sports Analytics Blog post of various definitions of “analytics” from sports industry leaders. I’m curious for your thoughts on football, specifically, and how that might differ.

I think the general set of analytical skills is sport-agnostic, as it’s more of a way of thinking about how to solve problems with data than about specific subject matter expertise

Trey Causey: Football analytics is the rigorous application of data and the scientific method to football to inform decisions and maximize the chances of winning as many games as possible. This could be via player selection (in the draft, in free agency, on game day), in-game decision-making (fourth downs, two-point conversions, etc.), or in the design and analysis of plays. Football analytics is distinct from football “statistics” or football “data,” which I see as just collections of observations about the game which may or may not be useful to anyone.

I think a lot of people equate “analytics” with “numbers,” but that’s simply not the case. Counting things for the sake of counting things doesn’t help anyone. Similarly, building the most advanced statistical model possible with no use case in mind isn’t helpful, either. Analytics is the intersection of counting / modeling and implementation. I think it’s important that football analysts be familiar with concepts like cognitive biases to recognize when how one interprets a data set might be clouded by human error. Why do coaches go for it less on fourth down than the numbers say they should? This isn’t purely a data question, it’s a question of risk aversion, the game theoretic structure of the labor market for coaches, and the psychological implications of both.

JR: Within the context of implementation and football-specific decision-making, one of the main criticisms from Browns fans has been that someone who has worked in baseball for two decades might not be that informed in football. You’ve obviously dabbled a bit in some different sports, but it seems like you’ve done the bulk of your sports work in football. How do you think an individual can be successful applying analytical skillsets across different sports?

TC: I think the general set of analytical skills is sport-agnostic, as it’s more of a way of thinking about how to solve problems with data than about specific subject matter expertise. Those skills travel across domains — it’s not uncommon in academia for researchers who are experts in a field to collaborate with statisticians who are better at analysis than they are. I’ve had many conversations with people who work in a variety of sports and the “baseball knowledge won’t travel to football” doesn’t seem like an opinion many serious sports analysts would hold. There’s much more in common than not across various sports. Further, cross-pollination of ideas from different substantive areas is how new, good ideas come about.

JR: For many people, when they think of “football analytics” they’re immediately turned off. They’ll complain that the game is too complicated for the use of analytics like we’ve seen in baseball or basketball. How would you respond to that line of thinking?

TC: My day job is working as a data scientist in tech. I’ve worked for one of the biggest tech firms in the world, medium-sized firms, and startups. Every single firm has faced tremendously complicated problems with thousands of variables and up to billions of users. The number of times that these firms said that their problems were too complicated for analytics? Zero. Quite the opposite — the thinking is that the problems are so complicated that you must use analytics in order to succeed. The idea that football is somehow special in this regard is preposterous on its face and wouldn’t be taken seriously in almost any other industry.

One way that football is unique compared to other sports is in the small number of games played each year. This means that small advantages gained via analytics are more difficult to detect. Further, small samples are characterized by high variance, so you have to consistently apply analytical methods and be lucky in order to reap the benefits of analytics. Of course, this is true in every sport. There aren’t any champions that don’t benefit from luck along the way. That doesn’t make them any less successful or less of a champion.

JR: One of your more popular posts from the past year was when you commented on the sorry state on football analytics. Certainly, the public discourse on the topic hasn’t really improved like it has in the baseball and basketball communities. Do you have any more hope for this now with some of these moves? What do you think comes next?

TC: It’s way too early to tell based on these moves. I think the Browns are taking a very interesting approach and certainly seem to be moving in a certain direction that, from the outside looking in, is analytical. However, I don’t think that a few hires by a single team indicate any kind of shift or move me away from the argument that I made in that post. Many teams already have analytics staffs in place — how they use the advice from those staffs is key. If they ignore those staffs, it doesn’t really matter who they hire. If they don’t give the staffs enough time to make a difference, again, it doesn’t matter. Adopting an analytical approach on the one hand and trying to reconcile it with a “win now” mentality on the other is likely to fail.

JR: While the state of analytics within football front offices is still in its infancy, how do you think the public discourse can change for the better? When there are only 16 games, a lot of mainstream narratives will ride on just one game or one play. To a Browns fan or a Browns journalist, what more do you think they could learn or discuss or write about in order to be more involved in the analytics world?

TC: The only way for public discourse to change for the better is for more analytics people to find a voice in the media, to have the media take them seriously, and to have more teams publicly embrace analytics. It doesn’t help that there aren’t a lot of pro-analytics people in mainstream football media. It’s a much broader issue than just in football; thinking about statistics and probabilities is very hard to do and the brain isn’t really equipped to do so without a lot of practice. Of course, there are lots of smart people who write very well about analytical topics, but you have to seek them out. Mike Lopez, Ben Alamar, and Seth Partnow are all good examples of people who are great at writing about analytics without overpromising or dismissing people who aren’t convinced.

Many thanks to Trey for answering my questions. Make sure to follow @treycausey on Twitter.