Takeaways from the OSUSAAC Football Analytics Panel
The Sports Analytics Association at Ohio State put on a wonderful conference today. Among the many highlights was the panel on football analytics, featuring several experts whose work I’ve been following. I’ve condensed my notes from the panel below, separated by each panelist, for anyone interested in football analytics. There is much to be hopeful about.
Josh Hermsmeyer (FiveThirtyEight)
Optimal outcomes in football due to analytics are more exciting for the fan (going for it on 4th down, attempting 2 point conversions, etc) than those seen in baseball.
Defense is incredibly difficult to analyze because of a lack of solid benchmark statistics.
View the industry as a marketplace where you have to stand out. Aim to solve people’s problems and let people know about it. The 3 steps to focus on are (1) learn a skill set, (2) leverage it to find a solution other people want, and then (3) get them to notice it. They all relate to networking.
For there to be a GM hired solely on the basis of analytics, analytics would need to prove itself in a way that nobody is incentivized to show now. If a team is currently winning with analytics, they’re not talking about it. Furthermore, those people probably came from another sport where they saw firsthand the result of information being democratized.
Sarah Bailey (Rams)
Baseball is very discrete in nature, whereas football is continuous with an infinite number of different scenarios. There are so many problems that are difficult to solve now because it’ll take years to get a big enough sample size for a team to be interested in it. That leads to having more freedom and creativity in what to look into.
Part of using analytics in college scouting is taking info from college and trying to find bits you can’t see on film. Specifically, analytics can be used to form slight edges in later rounds of the draft. For example, it’s possible that Combine performance could be more predictive of future performance for later round draft picks.
Zach Feldman (NFL Next Gen Stats)
There’s a fine line to walk between bringing in more tracking data and keeping it simple enough to be understood by the average fan. Next Gen Stats used to just be how fast someone was running mid-play, but now it’s much more advanced. We can have more immediate context for in-game performance, like how successful a WR is at in-breaking routes at halftime as opposed to waiting for hours after a game.
Relatedly, new tech can speed up the process of “padding” a game, and getting the results quickly allows you to attack other areas.
People looking to break into the field have to be curious enough to self-learn a lot of material using real-world data as opposed to clean data from a classroom setting.
Eric Eager (PFF)
Football is becoming less of an “insider” sport. We now have data, so we can validate whether players should be making the pro bowl ten years in a row. Whereas these decisions used to be primarily based on reputation.
The evolution towards maxima is not going to be as fast as in baseball because of slow sample sizes. The problems that are being solved now will be different than those being solved 5 years from now.
Baseball and other sports deal with a lot of symmetric information. There’s not a different scheme you can run in baseball to win unlike in football. As a result, football teams have inherently different value systems for players. For example, people might say that player XYZ is a bad guard, but a team might say they can use him a certain way and that he’s cost effective.
Currently, the biggest possible value add is being able to deal with heterogeneity. If I can (1) weigh what’s stable or not, (2) properly value mean projection and uncertainty, and (3) make bets that way, that adds value.
Doing well in the classroom isn’t enough anymore. The next frontier is doing work publicly.
Teams have a lot on their plate week-to-week, and PFF helps reduce that workload. It’s primarily self-scouting, scouting opponents, etc but there is little thinking through many edge case scenarios. For example, Coach Vrabel of the Titans and their 12th man penalty against Houston. He likely had the free time to think about that in advance.
Lastly, there wasn’t enough time for my question to be answered, but there was little mention of analytics in salary cap management. In a recent ESPN analytics survey, not one person surveyed mentioned cap management as an area of football currently affected by analytics. There are several reasons I can think of, and some were mentioned today, but I see it as a field worth further exploration.