All opinions are my own and do not necessarily reflect those of Novo Nordisk.

A.  They’re both being studied via mobile tech to create their ethograms.

What’s an ethogram?  I had no idea until I saw this PLOSone paper on using inertial sensors such as accelerometers and gyroscopes to measure the movements and behaviors of dogs, so as to create an ethogram, or collection of behaviors and actions, characteristic of labrador retrievers and Belgian Malinois.  For scientists studying the behavior and action patterns of different species, building an ethogram is essential to studies of animal behavior.  Without a standardized, objective catalog of behaviors, it can be easy for the perspectives of the observer to get in the way.  And it can make comparisons of data among different researchers (or coaches, as we’ll discuss in a bit) difficult.

And just as I was mulling over how that study shows the power of technology for behavioral research, the latest issue of Sports Illustrated came in the mail and I read a short, fascinating article by Tim Newcomb about how eight NFL teams have signed up with the company Catapult to integrate small GPS sensors into practice and game uniforms. This data allows a more accurate, granular and comprehensive view of how different receivers, for example, play the game.  Basically, building the receiver ethogram (using the term rather loosely). Sadly, this article is currently only in the print issue and not online that I can find.  But it’s at newsstands now.  You can go pick one up.  I’ll wait.

So let me delve into each of these articles a little more.

The paper on dog behavior starts by describing the issue of measurement of behavior, including the vexing issues of animals that move quickly, or are hard to observe, or move in groups, or all of the above.  In addition, the observers often can only watch one animal at a time and even with recording devices, cataloging behaviors from each animal takes time (that is, running a lot of replays) and often can only be done from a limited number of vantage points.  You know, camera positions.  Sounds a bit like evaluating players on a football field to me.

The researchers, who mainly come from Hungary with one in the United Kingdom, chose a combination accelerometer and gyroscope device to track dog movements as the dogs were taken through several standard behaviors such as sitting, lying down and others.  They studied twelve individuals from the two mentioned breeds, taking them through several behaviors, and cross-referenced the data with video of the training sessions.  The dataset was then taken through a machine learning process, including a training and validation set process, to create a model that would identify specific behaviors from the different recorded data patterns.

At the end of the study they compared how well their model performed relative to the actual behaviors and found better than a 90% accuracy rate when using within-breed comparisons.  Now, I can understand people saying, so what?  Why not just look at the dog and you’ll know whether Fido is sitting or lying down?  But the important distinction is, you can get a 90% success rate without looking at the dog.  And now imagine you’re in the forests of Borneo and you’re trying to figure out just what all those orangutans are doing up in the trees when you can only get a glimpse of them once in a while in the gloom.  This experiment is a proof of concept that these kinds of devices can help record and classify behaviors without any human having to go and observe.  Yes, it’s easy to tell when Fido is sitting; less simple to tell when a spider monkey is doing an aggressive display up in the trees.  You still have to temporarily capture and equip those monkeys first, though.

Which is why the NFL has it a little easier.  In addition to having far more money than any behavioral researcher, or possibly all of them put together, the NFL works with people, to whom you can explain the reasons behind wearing a specific device. They’ll even put it on themselves. In Tim Newcomb’s article he described how the GPS devices being used are allowing teams to get a much better handle on exactly how their players are practicing and performing in games.

These devices give qualitative and quantitative additions to the data teams previously were able to gather on their players. Here are just a few things coaches were able to better understand and act upon with this increased data flow: relative ability of receivers to cut and decelerate, resulting in different calls for stop and comeback routes; asymmetries in movement to left and right which could indicate an injury or difficulty in adapting to a new position; and that standing for greater than 80% of practice leads to a greater chance of injuries later.  There was a lot more in the article, and I’m sure a lot more information yet the teams aren’t sharing.

Both of these reports highlight how mobile tracking devices and tools are extending the range of the things we can quantify and as a result are coming up with novel insights and discoveries.  I’ve written before (here, here, here) about the opportunity that’s coming in combining animal behavior research tools, the quantified self movement, and sports.  In this particular case, I’m particularly struck by the potential gain in efficiency via the granularity of data, and the parallelization of data recording.

As I alluded to above, video watching has been the standby of detailed performance analysis for years.  But few can pay detailed attention to more than a couple of specific players at once.  So, a lot of rewinding and rewatching.  As these monitoring technologies get better, building off work like that by Linda Gerencsér and colleagues, repeated viewing may no longer be as necessary to understand how players performed on the field.  Watching video as an instructional tool will still happen, but culling through the game film again and again to locate the right examples may not be necessary.

It’s also not too hard to envision how this kind of approach will also really change practice and training as coaches are able to much more quickly identify good behaviors and areas for improvement.  It will also take out more of the human biases that sometimes can get in the way of effective coaching.  It’s one thing to think and believe a player is dogging it.  It’s another thing to see data that shows the player is practicing at exactly the same speed and pattern as before.  Or vice versa.

To speculate even further, the cataloging and standardizing of behaviors should trickle down to the college and high school levels, with the benefit of democratizing positional coaching.  Young football players could compare their recorded efforts to those of more experienced and skilled players via their datastreams in addition to video study, providing increased feedback on areas for improvement. In essence players would be pattern matching, trying to make their datastreams match as closely as possible those of established stars.

I don’t know if this approach will reach that far, but I also don’t think it’ll be long before we find out.