Talking Analytics with Alex Stern

Alex is currently working towards his Master's in Data Science at the University of Virginia. Not only has he been working for the UVA football team in an analytical capacity, but he's also presented his own research at the NFL's Big Data Bowl in 2020.

I came across his paper and was eager to reach out to him. Our discussion covers the deficiencies of traditional box scores, judging information in context, predicting individual vs team performance, and how all this work will go from behind the scenes to on the field.

Links mentioned:
ESPN analytics survey
Virgil Carter - Operations Research on Football
Pulling the Goalie: Hockey and Investment Implications
Data, Decisions, and Basketball with Sam Hinkie
Inside Football Podcast with Bill Polian

Timestamps:

1:51 Alex’s time at Big Data bowl and his direction for research

4:56 First exposure to data analytics

6:52 Traditional box scores

9:44 Analytics and fantasy sports

12:19 Judging information in context

15:52 Player vs team modeling

19:38 EPA vs WPA

24:38 Increasing variance as the underdog

30:02 How does this work go from behind the scenes to on the field?

35:44 Unpopular roster management decisions

38:42 Individual performance, success rates, and tracking data

40:48 Adopting analytics, communication

44:05 Spin on 25th hour question

Transcript:

Mike (1:21) I came across, through LinkedIn I believe, your presentation at the NFL Big Data Bowl popped up across my feed, and that got me going down this rabbit hole of reading your paper and unfortunately we must have just missed crossing paths at Virginia, but I'm definitely interested in learning more about your research, your time at UVAm and talking a lot of NFL analytics.

Alex (1:54) Absolutely, yeah, where to start. It's been pretty crazy ride over the last year or so since then, but yeah the NFL Big Data Bowl was a competition where they released a bunch of player tracking data on running plays and kind of held like a research competition where basically college students were in their own section and they basically posed a very open question. Tell us something about this data. What can you tell us about running plays, about running backs, what is success, what is not successful, and write a paper about it. Extremely open-ended. Come to us with something novel. So I really took the time to think about it. There's going to be a lot of people doing this. These are not just undergrads like myself but people with Master’s in PhD programs. So I wanted to think very outside the box. So who are the most overlooked players on the field? Offensive linemen. And so when it comes to running the ball, the kind of counterpoint is “okay, let's say running backs don't actually account for.. their skill level doesn’t actually account for you one hundred percent of their rushing production.” Well then what else would? The offensive line. They create space for the running back to move through and if they don't do their jobs properly the running back is never going to get free, find a hole, and be able to do their jobs properly. So I did a lot of research, you know, football analytics is certainly behind other sports such as baseball, soccer, basketball. So in the soccer space there's a lot of research on space creation because they want to create space for players to move through both with the ball and for those players who don't have the ball who might possibly receive the ball from a pass. So putting together some of those ideas I was able to put together an algorithm that calculates how much space these offensive linemen were creating in front of their running backs. And again, the second point of it was kind of normalizing for opportunity. So if you are, you know, sending a player, a running back up the middle against a stacked box, you’re not expecting to get you know a maximum number of yards. But, when my running back goes up the middle against six guys in the box, he's getting an extra half yard every time compared to the average running back going up against a stacked box up the middle. And those are the kinds of questions I try to answer, and then I compared that to salary data. Are these guys who were good at creating space getting paid as they should? And the answer was no.

Mike (4:36) I want to take a step back to the beginning. I really want to get into the weeds in a minute, but before we get into that. You got into all this by applying these analytical tools to the medical world. 

Alex (4:58) That was actually my first exposure to analytics. I was in high school and I was volunteering at this retirement home and they had some people on staff who had exposure to data analytics, had connections to people who were willing to do some extra analytics on the side for charity, and they said okay let's model the rehospitalization rate of our people who stay in our home because that’s our largest problem. And they came up with a model, and this model predicted who was most likely to be be rehospitalized after coming back to the home from the hospital, and they were able to take extra care of them, keep an extra eye on them, maybe alter something about their daily lives and the rehospitalization rate went down. That was a significant, real-world effect on peoples’ lives. Sports is a lot of fun, but this is something that was legitimately, positively impacting people's lives. I was just astounded by that. So I did my best to research and get into this field. I came up with an internship in high school with another company I was  very fortunate to come into contact with, and that really just sent me down this path of data science, statistics, and really applying analytics to real-world problems. 

Mike (6:15) Now, most people hear data analytics or analytics and sports, and this was certainly the case when I was younger. You know you watch SportsCenter you see you crawl the bottom of the screen, or even today you look at the crawl at the bottom of the screen, you got the basic quarterback statistics, wide receivers, yards, catches, all that stuff. I've got my own opinions, but I want to hear your thoughts on the shortcomings of relying solely on a traditional box score to sort of parse out how games go.

Alex (6:52) Definitely. That’s a hot point of contention these days, and not to say that the traditional scores and metrics are not helpful or important, but there are certainly things that get lost. The biggest point that we as analytic folk, or whatever you want to call it, make is that volume statistics don't always tell the full story, and volume can kind of come from places that don’t positively affect the team. So one of the largest things we’ve tried to do is cut out garbage time. So you do that by creating a win probability model, and if your team has less than 20% chance of winning, you're probably chucking up passes. Interceptions are probably occurring at a higher rate, but you don’t care because you’re trying to win. Similarly if you're up by 80% or more you were probably going to run the clock out. You are not really concerned with scoring more, as eating up the clock and moving down the field little-by-little. So those kinds of things don’t contribute to winning, they’re a result of already winning, or already losing. So we try to factor out garbage time, and as well, when you're on the field, what exactly are you doing that is improving your ability to score or win. So if you’re in first and ten, after a touchback, and you have 75 yards to go, you are probably expected to get maybe two or three points on this drive, right. Now if you have a forty yard pass, and it’s first and ten again, thirty five yards to go, you’re probably expected to get four or five points on this drive, but not all the time. So the expected value of that forty yard pass is massive compared to a run of two yards, and so it’s an assessment of how much does this play improve your ability to score or win. 

Mike (9:01) In a minute we’re going to cover EPA vs WPA, I’m fascinated by that, but if you could change the way box scores were presented in the mainstream media, and I've got some things here that... I think ESPN does Total QBR, I think that’s a bit more mainstream now, but especially with people who play fantasy football. For wide receivers you’re looking at aDOT, WOPR, all these things. What stats do you think, that exist in the analytics community now, are overlooked by the mainstream?

Alex (9:45) It’s funny that you bring up fantasy sports. I really think that that has gone a long way to improving the public’s understanding of more advanced metrics and increasing their intrigue in them. The fantasy football people want to dive in as deep as they can to figure out who's going to help their team. It's not just about how many yards did they rack up. One of the biggest things for us is opportunity vs performance and outcomes. Football is, at the end of the day, a game of luck to some extent, and the things that are really predictive of future performance down the road are opportunity and expectations of performance versus actual performance and outcomes. I throw you the ball 4 times and you catch it 3 of those 4 times. I judge you on your, let’s call it expected completion percentage, 75% vs that one time you dropped a pass. And I think in football we really get caught up on certain highlight plays, whether they're good or bad, but at the end of the day it's a matter of what can we expect out of you on the next play, as an unseen play.

Mike (11:05) I think the biggest mistake,  I probably did this early on in fantasy football years ago, if you're looking to chase last weeks points. If you’re looking to pick up a free agent that scored a lot the week before, yet if you don't look any closer, you might miss that they caught one pass on one target for 90 yards and a touchdown. Which I would argue is the least likely thing to happen again in the future. Instead maybe someone else had eight catches on 12 targets for 60 yards, with maybe less points the week before. That's what you want to chase going forward, like you were saying, the opportunity.

Alex (11:45) Exactly, it’s the Henry Ruggs and the Mike Williams of the world, they’re really throwing us off. Also, at the end of the day, these teams are smart. They’re going to watch the film of the guy bursting out for an 80 yards catch and they're going to do their best to cover him and make sure that he doesn't do that next week.

Mike (12:05) That’s where you get these unicorns like Tyreek Hill who seems to be the best of both worlds.

Alex (12:12) Absolutely. If you’re Tyreek Hill and you can do that on a consistent basis, that is some serious skill and talent.

Mike (12:20) And I wanted to tie in something you mentioned earlier about having everything in context. Judging opportunity in context and all these things. If I was going to tie in a financial world example here. You turn on CNBC and you got the Dow Jones or whatever and they're breaking it down by industry and by stocks and some things are up some things are down. You see Tesla's trading at $460 a share, Apple is at this amount per share. Some people, and this was me at some point too, are basing their evaluation of each company off the dollars per share. Off just that nominal amount, knowing nothing else about Tesla, Apple, Pfizer, or whatever company it is. Some people think, “oh, Ford's trading at $5 a share, and Apple is trading at $500 a share, and Amazon trading at $2,000 a share, whatever it is. That must be the relative valuation of these companies.” Where if Ford is trading at $5 a share, or General Motors, or whatever it is. You could have one share of the company at $5, and it could be a $5 company. Or you can have a trillion shares of the company, and it’d be a 5 trillion dollar company. You were mentioning these things about judging in context. You're after, with offensive line in the running game, is you’re after the full set of information.

Alex (13:49) Yeah, and I think there's a disconnect, we like to chase what we’ve seen and what seems flashy. There’s these numbers that have floated around the NFL to this day where if you run the ball 25 plus times in a game, you’re virtually guaranteed to win. And yeah that’s honestly a true correlation, but running the ball 25 plus times  should not be your goal in order to secure a win. It happens when you are already up in the fourth quarter, you’re trying to drain the clock, and you’re running the ball consecutively. That will kill clock time. It’s not those carries that led you to that winning position, but at the end of the day yes, when you win the game you are probably running the ball more often. We just saw this last week with I believe it was Pete Carroll who said his new mark was 32 or more carries and completions in a game. With not a ton to back that up. Completions and carries are two very different things, and they're going to add totally different types of value to your game.

Mike (14:51) I think most people would be familiar with, if you’ve heard the stat that, teams that take a knee at the end of the game win 100% of times. So why wouldn't you just call a kneel on the final play of the game? You’d be guaranteed to win. Now, from my understanding there is player modeling and team modeling. So, within the player model, this is where all this stuff gets complicated. Within modeling player data, my understanding is you're trying to separate these statistics into what's descriptive of past performance and what's predictive of future performance. So we were talking about, there might be some things that don't repeat in the future, as opposed to things to do. This is what your research was touching upon with the offensive line. So, how do you separate these stats into these two buckets?

Alex (15:52) I think that’s really the question of the day. Football certainly does it’s best to make that as hard of a question to answer as possible. What is predictive of future performance? Football is an incredibly complex game, you have 22 people on the field interacting at any given time. It’s kind of tough to, say you have one guy, is he positively affecting the play in this way. One of the things we try to do is find measures of different actions that are a little bit what we call stickier in the long-term. These things are more predictive from year to year, but still lead you to the result you want. A couple examples would be monitoring pressures created by defensive linemen instead of sacks. Sacks are a result of pressures and so certainly the more pressures you get, the more sacks you are expected to have, but they're also kind of a flukey thing. You as the defensive linemen are not really in control of the sacks. The smart quarterbacks are mobile, they throw the ball away, they avoid sacks themselves. We can get into the whole “are sacks a QB stat?” debate another time, but pressures are certainly more indicative of a defensive lineman’s talent. Similarly with defensive backs, passes defended lead to interceptions but they’re a lot stickier stat. They can be fluky, the ball can tip off someone’s hands, a lot of random stuff happens. But passes defended is a lot more indicative of defensive back talent. 

Mike (17:33) My eyes were opened to this when, on Twitter you can find people going down rabbit holes for anything, and there were some people that would just study quarterbacks at such an in-depth level, and they were tracking what they called interceptable passes, as opposed to just interceptions. Like you were talking about. So something that might have deflected off of three people's hands might not count as one if the receiver had an easy catch. Versus something that wasn't an interception but easily could have been. 

Alex (18:04) Definitely. That's another perfect example, those interceptable passes that the cornerback may have dropped for some flukey reason. Nine times out of 10 if you see that look as a cornerback it's going to be an interception, and vice versa. Like you said, something wacky happens, that's not actually the quarterback’s fault, and then you have the third thing. I'm originally from Minnesota, I was watching the game this past weekend and the first touchdown pass of the game, Mike Glennon throws the ball and it bounces off of the back of the Vikings safety, like skids off his back and flies into the endzone and Shenault catches it. Easy touchdown. But you know, that's not skilled, that's just random football.

Mike (18:49) I wasn't watching the game but, please tell me it wasn't Anthony Harris.

Alex (18:54) No, no, our boy is doing quite well.

Mike (18:57) Thank goodness, and moving on to the team level now. I'm always fascinated reading the differences between modeling a game and expected points added vs win probability added because, from my understanding, for most of the game there's a big overlap. But when you get to these end of game situations, and it's happened multiple times this year. I remember the Falcons earlier this year and one or two other times, sometimes the play of the end of the game that gives you the most expected points added is not the best play in terms of win probability.

Alex (19:38) Yeah, I think you're right. There's definitely a disconnect at the tail end of the win probability. So while you're in this sweet spot of 20 to 80% or 40 to 60% where the game is really up in the air, the things you do, the actions you take, and the points you score matter a lot more for the outcome. But when you're up 21-0 and you let the other team score a garbage-time touchdown, year, there's expected points added in that, and there's some level of skill, but it doesn't affect the outcome of the game. You might’ve put your second stringers in, your guys might have been playing lazier, but when you take into account the win probability, you see which things matter when everyone is at the top of their game. So for example for the UVA team one of the things I do every week on Sunday mornings is create these win probability graphs for the coaching staff. It shows the win probability of both teams throughout the game as well as it flags the top live games the top five plays in terms of win probability added. Those plays are never at the very beginning or the very end of the game because what happens is, in the middle of the game, when these crucial turnovers occur, or these 4th down conversions, if you've seen our defense this year it's a lot of interceptions and sacks. So they've really helped win us games, and that's fun to see on this win probability graph. You see three or four of the top five plays of the game were from our defense.

Mike (21:09) These are some really niche examples, but say you're a team, I think this was the exact situation the Falcons were in earlier this year. They were up a point, with the ball, and less than a minute left on the opposition's goal-line, when they probably could have taken knees to end the game. Instead they scored. No matter how you slice it, unless they went for two, it still would have been a one-possession game, and maybe they did, or whatever it was. In that situation they scored a touchdown, but they went from giving the other team, it might have been the Atlanta-Dallas game, where they gave Dallas virtually no shot to win, and even though it was a slim shot, it increased Dallas's chances of winning when Atlanta scored.

Alex (22:00) Yeah, and that's a fantastic example of what goes into win probability. You have your chances of scoring more points, it's your current score differential plus the expected points of wherever you are on the field, and one of the main things that takes into account is time remaining on the clock. So when you score that quickly, instead of kneeling or whatever happened, I think Todd Gurley meant to kneel down at the 1-yard line and just fell in. It's kind of crazy, that's the game of football, that swung things and the opposing team ran down the field. Obviously it's hard to construct a game with less than 2 minutes. Props to the other team for doing that, but Atlanta certainly gave them the opportunity to.

Mike (22:49) The last point on this topic, this would certainly be like you said, whether sacks are a QB stat, the whole argument on when to go for two or not to go for 2. Or when to go for 4th down or punt it or kick a field goal. I just came across this paper by Virgil Carter that he wrote when he was the quarterback for the Cincinnati Bengals under Bill Walsh, decades and decades ago. And he was looking at expected points added in the 60s. It was based on where you were on the field, and so the decision to go for it on 4th down, he was saying, or to punt it, there was a huge discrepancy over how valuable that decision was based on where you were on the field. And it seems like these are things that were known back then, and he was doing research on this decades ago, and I don’t think some teams have advanced past that point.

Alex (23:52) It’s pretty crazy. It's a very conservative game, nobody wants to lose or give up the big play, but at the end of the day you have to be aggressive, especially when you’re the underdog. You can't keep playing into these situations where you give up the ball and stay safe and obviously they keep the opportunity for turning the ball over at the 50-yard line and your team having a great opportunity to score, but I think it was the, I forget what game it was, but a team punted instead of going for it on 4th Down and because they were on their own 20-yard line, and they punted and the other team got it around the 50. First play of the drive, 50-yard touchdown, scored. And so what it comes down to is increasing variance as the underdog. You want to be as aggressive as possible within reason, to increase your chances of winning. If you play into the other team's hand you play conservatively, you can't possibly expect to win as this giant Underdog. We saw this last year, UVA, if you watched either the ACC Championship game against Clemson or the Orange Bowl game against Florida, we played very aggressively in terms of maximizing the variability on every place. I think at least in the ACC championship against Clemson, we were massive fourscore underdogs, we ran either precept pre-snap motion or something funky on almost every single play, because when you come into a game against Clemson as UVA you have to do something weird and increase the variability of the game if you want a shot of actually winning

Mike (25:35) There's a few things that reminds me of. The one that, I don't know if you read this, the paper, the hockey analytics paper. It makes a logical argument that teams should be pulling goalies with as much as 10, 12 minutes left in the game as opposed to the traditional, maybe if there's under two minutes left, maybe then you could pull the goalie, even if you're down by a ton.

Alex (25:59) Yeah. I'm not very well-versed in hockey but I know what you're talking about. It's a scary strategy. It opens you up to a lot of criticism, and the media can be ruthless against coaching staffs and front offices, and so pull your goalie 6 minutes early and that doesn't go well, and you get scored on an extra two or three times, they're never going to let you hear the end of that. I certainly understand where the conservatism comes from.

Mike (26:24) The other example I think works against Virginia, like last year looking at the basketball side of things. I think it's well-known that Virginia basketball has, on offense the past few years, among the lowest amount of possessions per game on offense because of their play style. I'm not well-versed in basketball. I'm pretty sure you're a lot more interested in basketball so I want to hear your opinion on this. Assuming that Virginia is often the favorite in basketball games, it seems counterintuitive to be playing a style, like you said, with fewer possessions is seemingly increasing the variance in result.

Alex (27:16) So actually, this may or may actually not be a hot take, but I think it's fantastic what Tony Bennett has done, slowing down the game. It's actually the opposite of what we were just talking about with increasing variability as underdogs. When you limit the number of possessions per game and limit the number of opportunities to score, more often than not the better team is going to win. If you're the better team on paper you want to do the opposite. You want to minimize the variance in this game, because at the end of the day, the fewer opportunities that team has to go on a run that they wouldn't be expected to, or stop you from scoring, because you're expected to score, you're the better team. You should be able to score on them, given minimal opportunity. There's obviously extends to both of these things, how aggressive should you be, how slowly should you play in basketball. And Tony Bennett certainly does his best to push the limits of that sometimes, but we love it. That pack line defense certainly knows how to slow down teams, and as long as we can score, I hate to say it, 45 points on offense, we can go out there and win games. It's kind of crazy.

Mike (28:29) That's the magic number. It'd be a weird dichotomy seeing Virginia scoring 45 points per game to win on offense and having very few possessions, and then Duke would be the complete opposite. Scoring 80+ points per game and having the most possessions in a game. So it's kind of the two sides of the coin there, and both teams executed really well on the strategies last year.

Alex (28:56) Yeah, definitely. And you do have a disadvantage when you're UVA and you come into that kind of game, because when you get down early or even in the middle of the game it's tougher to come back from a deficit when you play like we do. But I'm excited to see this year, we definitely got some better shooters than we had in the past, and Hauser is a fantastic addition to the team. So I'm excited to see the uptick in offense this year from last year.

Mike (29:22) Getting back to an area I'm more familiar with. Getting back to football here. I want to get your input, you spent time recently with the coaching staff and all the different areas of the Virginia football team. How do you see coaches, and in your research you were touching upon front-office with offensive line salaries, how do you see this line of work being implemented by in-game decision-makers and even in front-offices. I guess what I'm asking is how does this work go from behind-the-scenes to on-the-field?

Alex (30:04) It's definitely, I think, it might have been in that ESPN survey, but it kind of talked about the visible versus the invisible effect of analytics on teams. And you could see this very visibly in teams like the Ravens, who have certainly adopted analytics more so than the majority of teams in the NFL, and they've become one of the most aggressive fourth-down teams in the NFL, and that's certainly the analytics back up being more aggressive. There's also a lot of invisible things, a lot of stuff is kept secret and I do personally believe a lot of teams are more advanced than we like to think, but you just can't see what they're doing. I think in the next decade or so, a lot of player scouting will include analytics, and it's not to say that analytics are the end-all-be-all, a lot of people hate the idea of these math nerds from Princeton coming in and taking their jobs, and destroying their scouting departments with their laptop. But it's another seat at the table, it's another scout, it's another box you have to check. So when the scouts are watching film and noticing these more intangible aspects of a player's game agree with the predictive analytics and the models that the nerds are pulling up on their computer, those are really the opportunities to find great players. So, even if it's analytics saving time we like to talk about. If you separated players in two tranches and you said the analytics really like these top third guys, we want scouts to check out and review the film. This is where you should spend your time. Then these middle guys, check them out if you can, these bottom guys we don't think it's worth your time to do. At the end of the day when you're spending all this time watching film, reviewing players, it's an incredible value to save time and really only look at these guys that we think have value.

Mike (32:18) And I know from experience that time, especially in football, is your most valuable resource.

Alex (32:25) Absolutely, any way we could help the scouting staff make better use of their time and possibly uncover future talent at a higher rate, I think that’s a fantastic contribution that analytics makes. As well as I think the salary cap is a great place to start. I don’t think anyone is really doing that right now, but at the end of the day it’s a giant optimization problem. With a hard salary cap you're allowed to spend this much money and you need players at these positions, and the variability of skill at each position is different. So where do you want to spend your money?

Mike (33:05) And I think that’s an area where basketball, I was just listening to the interview with Daryl Morey of the 76ers, and Tim and Sam hinkie and other Executives in basketball, and even baseball to have brought a completely data-driven mindset to roster decisions. From what you could see publicly in football I guess what you can see is paying for past performance in football as opposed to other sports

Alex (33:36) Absolutely. That's hitting the nail on the head. If anyone hasn't heard it, that Sam Hinkie interview was fantastic. He's an incredibly interesting person, but yeah, how much can you expect out of this player in the future even if he's done so well? And I think the place where that bites teams is, at least recently, running back contracts. It's been another point of contention between the analytics community and others, like this is Zeke contract, just signed to a massive contract, and now the Cowboys haven't had the best season, and it's about where are you spending your money properly. Yes, you want to make your franchise guy happy, it makes the fans happy, it sells jerseys, tickets to games, etc, but what is really contributing to your winning? And more importantly, what will contribute the most to you winning in the future?

Mike (34:30) And part of it might just be a fanbase thing. Or it could be a sport thing, a fanbase thing, because I think everyone is desensitized, in basketball too with LeBron and these superstars just changing teams every year. I think people just expect that's going to happen, someone will get traded, whether that's a thing inherent to the sport, or the way their contracts are structured, just the way that plays a part in it too. But in football, and I know from being a big Patriots fan and following and reading all about Bill Belichick's career, he was vilified for benching Bernie Kosar when he was the coach in Cleveland, or running Drew Bledsoe out of town for Tom Brady. And Drew Bledsoe was one of the highest paid quarterbacks in the league at that point. Now I think we're going to see a similar huge decision coming up with Carson Wentz and the Eagles, so maybe it's a mix of factors, but in football it seems like if you've got a franchise guy, whether he's worth the money or not, it's a big issue if the fan base revolts against the coach.

Alex (35:44) Certainly, they can be as ruthless as ever, and the only way you can get around that is with results. Tom Brady came in, won games, won Super Bowls, won over the fan base. The funniest example that I've heard recently is, there was an interview with Tyreek Hill, where a week or two ago they were asking him about Patrick Mahomes, who is the franchise quarterback for the next 10 years. They were like what did you think of him when he first came in here and he sat on the bench for a year after Texas Tech. What did you think of Patrick Maomes? He was like, I thought he was horrible, I thought he was awful, I thought he was going to be a terrible quarterback. He was pissed at the front office for drafting him, he was like why did you draft a quarterback? Why did you guys draft this kid? Etcetera etcetera. But then when you produce on the field, you really shut everyone up.

Mike (36:37) Yeah, especially in football. Results speak for themselves. I think that's why Joe Judge is now getting all these Coach of the Year nominations in people's minds, and public opinion has certainly turned on him for the better, the past five weeks when they've been winning. As opposed to the chatter I was hearing over the summer was that he was trying to be Belichick, and a disciplinarian, and all this stuff. So the winning recently has certainly quieted that down.

Alex (37:07) Certainly, and all props to Joe Judge, I mean losing Saquon at the beginning of the season, losing Daniel Jones two weeks ago and still coming out here and winning, and obviously a lot of the notoriety is because they're in the NFC east and it's certainly an easy place to be notable these days, but they have been winning games. The win over the Seahawks was impressive. No Saquon, no Daniel Jones, so all props to Joe Judge for sure. 

Mike (37:35) Something you had mentioned earlier, and other sports like baseball and soccer, from what I've read, one of the big differences between football and baseball is with baseball you've got separate, identifiable events that you could always track. You've got a pitcher throwing to a batter. You could see exactly where the pitch is going and all that stuff, and it's just those two players. And in football there is really no equivalent, and it's very hard in football to attribute anything reliably to individual performance. I'll give an example related to exactly what your paper was about. When the Giants, you were just mentioning, made Nate Solder the highest paid offensive lineman in football a couple of years ago, that decision seemingly backfired, and now everyone is thinking after the fact: how much do you attribute Nate Solder's play to Tom Brady being behind center in New England all those years? 

Alex (38:43) It's an incredibly complex problem. I think the addition of player tracking data in recent years certainly brought about a data revolution. You no longer have people who can calculate success rate, but you really need to find these people who are well-versed in computer science at a deeper level to kind of parse through this incredibly dense stack of player tracking data. You have millions of data points for a given game now. As opposed to a hundred something plays. So it becomes a lot harder, and I think we've barely scratched the surface at this point for what we could do with player tracking data and the insights we can gain from it, so it'll be really interesting over the next decade or so to see what people come up with. 

Mike (39:28) There's a lot of room for improvement in the future. Have you been listening to, I'm not a big fan of Bill Polian, but I'm a big fan of his podcast, since he's one of the few people who have been an NFL GM who have had a regular podcast. Have you heard any of those podcast episodes?

Alex (39:46) No, I haven’t.

Mike (39:47) You’d get a kick out of this. Some of the things were very eye-opening and informative. He talks about roster building, managing the salary cap, all of that stuff. He was going on about is time in Buffalo, when he hired Marv Levy as the coach, and he was saying how we comes in on the first day and is talking to the team and he's like “we're going to win if we run 40 times a game” and all this stuff, and I'm just chuckling. He's going on and on about the importance of establishing the run, too. Which nowadays we know is not all that it's cracked up to be, analytically. But I think that's still how people, who are around front offices now, who were still around back then... I don't think with these teams being run by people who have been around football a long time, maybe that's a contributing factor to why teams have been slow to adopt this new research. 

Alex (40:48) Certainly. Part of that burden falls on us, on the analytics side of things. Communication, I like to think is the most important skill any of us can have as analysts. Partially through the data, and discovering new metrics and all that stuff is fantastic, but at the end of the day it really doesn't matter if you can't properly communicate what you're doing and why it's important to the stakeholders, the people who are currently in the front offices. You need to generate a certain level of buy-in to both your level of skill and the value that your metrics are adding to what they already do. Otherwise there's no reason for them to listen to you, but it's certainly a slow-moving train. You heard it on the broadcast just the other night, we're actually more concerned with rushing attempts than rushing yards because we want to wear down the defense. These ideas that really do sound great when you say them but there's not a ton of evidence to back it up. 

Mike (41:49) Exactly, and I was just going to ask about how important you thought the role of communication was in your role. So that checks that box. One of the last things I want to know before wrapping up here is, what are some lessons specifically that you've taken away from working at UVA? I know I have a huge laundry list of things I've taken away from Coach Mendenhall, but what are some of the takeaways you had?

Alex (42:17) I would say, again the communication thing. Every stakeholder, you want to call them, has their own point of view. Everyone has their own job in this operation, and everyone wants to win. Everyone wants to get better. It's not that these guys on the coaching staff are here to disregard analytics, but you have to show them the value that lies in these things. So, when I talk about wide receivers, and I talked about production versus opportunity, it definitely has improved my communication skills exponentially, and it has really helped me discover what I can really possibly do to help them. What is my role in this? So now I try to create more interactive visualizations for the coaching staff to use when game planning. So instead of me just handing in a spreadsheet of what I think this team is good at and what this team is bad at, and here we can exploit X team because they're awful against passes over the middle between 5 and 15 yards to the tight end. That's where we exploit them. Or where they're going to run a screen to the left on 3rd and medium or 3rd and long. Instead of me telling them that, I felt it's helpful to create interactive visualisations for coaches to use on their own and play around with, and kind of visualize these things for themselves. Because that's how game planning in football works. 

Mike (43:55) Going back to something you said earlier. I'll do a different spin on the question I was going to ask. Instead of the typical 25th hour question I'll get into here is, you were mentioning about analytics and time-saving. If you could save an hour of the coaching staff's time, with your work, what hour would be the most beneficial to save?

Alex (44:25) That is a fantastic question. I think game script, I would say. What do you do on first and ten? What do you run as the first play on your drive? Or when you turn the ball over, what do you do? I think this is something we get very caught up in, and you can go down a million rabbit holes of how you want your drive to go, and what you want it to look like. At the end of the day, analytics has really helped prove that passing on early downs is helpful, and it increases your expected points of your drive. And it increases the probability you'll have a successful drive, whatever you define that to be. So I think that would definitely be the first thing to cut out is: first and ten, touchback, pass the ball.