Michael Lopez @ NFL | Being forward-thinking & anticipating questions | Data Science Hangout

Transcript#

This transcript was generated automatically and may contain errors.

Well, happy Thursday, everybody. Welcome to the Data Science Hangout. If this is your first time joining us here today, it's so nice to meet you. Thanks for spending your Thursday with us. Actually calling in from the POSIT office. Nobody's here with me today, but I have this nice view of the water. But if we haven't met yet, I'm Rachel Dempsey. I host the Data Science Hangout and lead customer marketing at POSIT. This is our open space to chat about data science leadership, questions you're facing and getting to hear about what's going on in the world of data across many different industries.

And so we're here the same space, same time every Thursday. So if you're watching this recording on YouTube later, you can add it to your calendar with the details below. But together, we're all dedicated to making this a welcoming community for everybody. And so we love hearing from everyone no matter your years of experience, titles, industry or even languages that you work in. So every week I'm joined by a different leader here from our community who joins us to share their experience and answer questions from you all.

So to let you know how to ask questions, first of all, it's totally okay to just listen in if you want to. But there's also three ways you can jump in to ask questions. So you can raise your hand on Zoom, and I'll be here looking out. You can put questions in the Zoom chat. And feel free to just put a little star next to it if you want me to read it, or else I can call on you to ask the question live. And then lastly, we have a Slido link, which Tyler will share here in right when I said that. Thanks, Tyler, where you can ask questions anonymously too.

But with all that, thank you so much for joining us here today. And I'm so excited to be joined by my co-host, Mike Lopez. Mike is Senior Director of Data and Analytics at the NFL National Football League. Mike, I'd love to have you introduce yourself, maybe tell us something you like to do outside of work too, and share a little bit about your role.

Thanks, Rachel. And thanks to everyone at Posit for organizing these. It is an exciting time to be in football data. And we just have a lot of questions that we have to try and use sort of naive or advanced models for, because there is so much about the world of football that we are just starting to learn. My background was, I was a high school teacher, turned PhD student, turned assistant professor at college. And at that point, the influx of football data became apparent. The league started getting next gen stats, which is each player wears a pair of RFID chips in their shoulder pads. And you sort of went from an NFL data history where every game had 160 plays, so 160 rows of data, which everyone was sort of analyzing locally on their laptops, to now a world where we have this tracking data, both RFID based and then now even video based. And it's just hard. There's just a lot that we are learning and trying to figure out.

So my job is to organize a group of really intelligent young data scientists that are helping us and helping the league make better decisions. Awesome. Thanks, Mike. Our preseason starts tonight. And I sort of now realize that from now until the end of March, most of my spare time is going to be keeping an eye on what happens in everything related to the NFL. So my spare time was over the summer. I have three young kids, and so hanging out with them is really how I spend my spare time.

But tonight, that is now done with an eye on the NFL game that's on the television. So watching football is what you do in your spare time. Yeah. It's a nice side gig.

If you share the data, they will analyze it, especially in sports data.

If you're on a data science team and that your job is to build interpretable metrics and to communicate those and to do a good job of validating your models and things like that, that's what the big data role is all about. So we've seen a lot of, one of the neat things is that now people are copying football and that was never a place where football was before. There was the big data derby and horse racing. There was the big data cup, which staff leads put on in hockey. I think there was like now a fantasy football one. And it's pretty humbling to know that largely started with the NFL sharing data and being open about it.

So we're certainly open to more leagues doing it. In terms of advice, the number one thing that we found is that the crowdsourced metrics are always better than anything we've ever come up with ourselves. And the more thoughtful and sort of, I guess, sort of thinking forward about what people will analyze, I think companies will get the most out of it. Like we've had competitions that have immediately turned into metrics that are shared on air . And our 2020 competition was on rushing yards. And within maybe 11 months from the contest ending, you watch the NFC title game, which has maybe 30 million viewers. They're watching a metric that came from our competition. And that's sort of a really exciting aspect of that competition. And so the better questions you ask, the better answers you'll get. But I think overall, we've certainly found that the crowdsourced approach has really helped.

Structured vs unstructured data

Yu, I see you've had your hand raised for a bit here. Do you want to jump in? Sure. Hi, Michael. Thank you for being here. I have one question. So what type of data are you using for the analysis? Are they tabular structure data or unstructured data like images or recording from the game directly? If the data are unstructured, how do you process them before you can use them for analysis and modeling?

Yeah, good question. I would say with each passing year, we're getting to more unstructured data. But I think the league has done a pretty good job. And when you're at a league office, we have a large number of data engineers that help that ensure our SQL tables are clean and easy to read from. I would say most of it's pretty structured. And the other reality is with player tracking data, at least in the NFL, because it's RFID based, it's very accurate. And so it's a tabular format. In fact, if you go to our big datable site, you can download it as a CSV and it's a big data set, but it comes out clean. That's not necessarily the case with maybe some of the newer data sets we'd be getting into.

If you talk about like video tracking, which would be like taking from an image and then trying to look at player location. One of the ways that football is going is that we have RFID tags on all 32 NFL teams and all the players and all the games. However, those players come from a pool of college players. The NFL can't make college players wear those same RFID tags. And so when you want college data, you either have your traditional stats or you've got to use video based data. And so a lot of that video based data on college players is stuff that is a little messier. And so that would be where you would have to extract it from imaging and things like that.

There are companies out there that do that. And it's sort of, I don't know if it's an arms race, but I do know that NFL teams are really interested in those skills because of the way that before you would have to have a dozen scouts go out and watch videos of the teams in the northeast and the southeast and all over, you'd send scouts. And teams still do that. But now if you can get some of that insight by coming up with an algorithm that can take the data from the video and turn it into a nice usable format, that can certainly save you some time and maybe get you some insight you weren't getting otherwise.

Transition from academia to the NFL

Lisa, I see you had a question a bit earlier on Mike's transition from academia to the NFL. Do you want to jump in? Sure. Yeah. I'm Lisa. I was just curious about how you made the transition from academia to working for the NFL and how you think that academic experience was helpful. And then maybe again, this is probably selfish, but like how did you sort of market that academic skill set?

I think the most helpful part about being in academia is that you are evaluated. And I was at a small liberal arts college. You're evaluated on your teaching. So you have to be a good teacher, right? I was a tenure track assistant professor and evals would come in. And when you weren't doing a good job communicating to your students, that's going to ding you or whatever. So I think that's the most helpful skill set is that my job was to take statistical methods and distill them down to students who were learning it.

I've been teaching R for maybe some software for a decade. But when you do that in 2020 or whatever, I couldn't be teaching the same stuff I was teaching in 2010 because software was entirely different. So you also have to sort of keep up to date. And that's kind of a lot of what we do now. Like our job is to communicate statistical results or methods to folks that aren't statistics experts. Now, instead of students, now it's executives. And one of the things that we are constantly working on is what do we show? How much is too much? When do we lose them? How many plots is too many plots? Do they need to know the ins and outs of our models? And whereas when you're teaching, you really need to communicate all the ins and outs. Certainly, we don't do that now.

But I think there's a lot of the experience that was most helpful was teaching and communicating results. We don't do as much writing. I don't have as many research papers maybe as I used to would have. But now, for better or for worse, I spend a lot of time making presentations. And I think a lot of that's the same approach where you're trying to take what you've learned and distill it into a template that others can understand. I started in 2018 when I worked at the league office, maybe doing one PowerPoint deck every couple months. And now, for better or for worse, I have to do that a lot because I have a really talented group of data scientists that are coming up with so many cool things. And I am now helping them curate material to share with executives.

And there's, like I mentioned at the start, there's a lot that we can do. And trying to figure out how to communicate that goes back to that experience in academia where you're writing a journal article or you're writing a summary of your work. So that's where it's been most valuable. And I also think one of the cool aspects about where we are now is that there's just still a lot to learn. And that's no different than being in academia where you're an expert in something and you're trying to build the next tool or you're trying to build the next model or the next paper. And there's, you know, I think it's not that, you know, I like to often refer to our work as we just, we have a laundry list and sometimes our job is to just start at the top. What's the most important thing we can do? And that's not much different when you're in academia and trying to do the same thing.

Communication skills and anticipating questions

Can we stay on this topic for just a bit here? When you were saying, when you were talking about the communication skills that have transferred over, I was curious if you do have certain tips for us for communicating to executives or what you've learned in many of those presentations you've given. Yeah. I mean, we've tried and failed a lot. I mean, there's things that I've pushed for or we felt strongly about that other people were like, yeah. And then there's other things where maybe I didn't feel as passionate about. And then suddenly maybe every, maybe the presentation got shared a lot more widely and I read about it the next day on ESPN.com. So I, it's hard to really know, like I haven't kept a spreadsheet of like what's worked or what hasn't in terms of communicating results.

I think the, we have to have good data visualization. If we don't, we just, we can't do our job. And I think I spend probably, my team would probably complain about how I do this, but like, I probably complain about things that I, if this was a student paper that like, I probably wouldn't, right. Because I have to make sure that the work that is being presented is as clear and concise as it possibly can be. It can't be confusing. It has to have the right annotations. The stuff that has come out in ggplot in the last couple of years and also in GT with like the sort of visualizing and tables has been really helpful in that front because for a while, like we would have to make tables in an old school format, right. Or I'd have to add annotations in PowerPoint because, you know, you couldn't code that. So that's, that's probably the most important thing is, is your, your viz has to be top notch.

I also, and this just comes partly from experience, partly because I played football in college and my dad was a football coach and I grew up watching way too much film, but the more closer you are to the subject level matter, the better you are at your job. It doesn't mean that you need it for all roles. You can be a great coder and sort of learn that skillset along the way. And I mean, football is in some sense a complicated sport, but in another, anybody can learn it if you try to learn it, right. So it's not like it's too far away from, from somebody to learn.

But a lot of the things that we do is we just answer their questions. And I think the sort of second part of communicating that result is sometimes we are answering the questions that we know they're going to ask before they've asked them. And I think, I think the best presentations we give are the ones where they're asking a question and then I just click next and it's on the next slide, right. Or they ask a question and I say, yep, it's in the appendix, click a couple of slides forward and say, here it is, because you have to anticipate the questions they're going to ask. You have to, sometimes we're collecting data before they've asked us to collect data because we can tell, like we hear something in the media or we hear something that they're asking in the replay room or their coaches are asking. And we're like, all right, we're going to go build this because they didn't tell us to go collect data on quarterback slides. But we now know that quarterbacks are sliding more than ever before. We know quarterbacks are the most important position in football. So last year, Andrew Patton basically led a NFL data collection process in quarterback slides because we knew that when we got to the offseason, somebody was going to say, hey, how are our quarterbacks sliding? And it's like, actually we have that because we went out and collected it. So I would say that it's a combination of great data viz and then also being purposeful and sort of forward thinking with the questions that people are going to be wanting to ask.

I think the best presentations we give are the ones where they're asking a question and then I just click next and it's on the next slide, right. Or they ask a question and I say, yep, it's in the appendix, click a couple of slides forward and say, here it is, because you have to anticipate the questions they're going to ask.