#141 How Data Science is Transforming the NBA

Hi, I'm Maggie, VP of Curriculum at datacamp and I know firsthand how overwhelming it can be to keep up with the rapid changes in data and AI. But staying ahead doesn't have to be hard. That's why our curriculum team has curated the best learning content to help you and your team stay ahead, covering everything from data and AI literacy for busy executives to cutting edge generative AI for developers, and everything in between. Whether you're upskilling yourself or your team, we make learning practical, engaging and flexible. It just works. Power your data and AI transformation journey on datacamp.com.

I think there is an important, more broadly applicable lesson about the importance of having that knowledge of the subject matter, just so you're aware of what the data you're working with actually says, actually represents. It's an abstraction of reality. What is it abstracting away? What is it capturing? What isn't it capturing? I think that having a decent level of understanding of that really gives you the first level of sniff test of almost QA ing your own work.

Welcome to Data Framed. This is Richie. With the NBA Finals ongoing, it's an exciting time for basketball and a perfect opportunity to talk about basketball analytics. How I personally find that I enjoy sport more when I understand how the players and teams are thinking about their decisions in the game. And data analysis is a great way to make sense of those choices. Joining me today is Seth Partnow, the director of North American sports for Statspot and an NBA analyst for sports news site the Athletic. Seth was previously the director of basketball research for the Milwaukee Books and he's also the author of the basketball analytics book the Mid Range Theory, which provides a colourful description of how analytics has changed the game of basketball. In short, Seth has a real insider insight into the world of basketball analytics and I'm keen to pick his brains on the subject.

Hi, Seth, great to have you on the show.

Thanks for having me.

Yeah, I'd love to just kick off with what basketball teams care about. So can you start by telling me what sort of problems the teams try to solve with data?

First of all, when we're talking about sports organizations, the first sort of split is between the sports side and the business side and for certainly at the top level of sports, not just in the U.S. but across the world, like the people who are doing the analysis, the recruitments, those things are very separate from the people who are worrying about ticket prices and, you know, projecting concessions and all those fun data challenges that that sort of revolve around revenue generation and drawing eyeballs and Stuff like that. So that's, that's the first one. And I am wholly on the sport side. I was with the bucks side. There were people who did similar jobs on the business side, but never the twain shall meet. I mean, they were friends of mine, but we didn't work together other than some pool technical resources. But broadly speaking, it's very different jobs.

Now.

Within the sport itself, there's three areas really where data can have an impact. The first one is sports science, which I know there's some differences in what that means across the pond from here, but when we say sports science in the US we're really talking about injury prevention, return to play, load management, kind of training and medical stuff. I did not work much on that side, so I was wholly more on the on court side part, which was on one hand, player recruitment, personnel deciding who to draft, who to sign, who to trade, who to no longer resign, things like that. And also kind of game analysis, whether self scout or discovering tendencies, strengths, weaknesses of an upcoming opponent. So in broad terms, those are three buckets and I think those are broadly speaking similar across sports.

All right, it's kind of amazing because you think, okay, well it's just about winning games, but actually there's all sorts of like data evolving. These kind of supporting folks of like, oh, how do you can like optimize how many like tickets you sell or how many cans of beer you sell ringside or whatever. Today we'll focus on the sports side of things. So can you tell me, are there any teams that you think make particularly good use of data?

That's always a little tough to say from the outside. It's more anecdotal from talking to people inside organizations, it's always hard to say, well, that was a data decision they made. It's much more about how integrated in the process is the use of data. Is it, there's two poles and most teams fall somewhere in between, but they tend to one side or another. On one hand it's using data as like a encyclopedia resource where, all right, we've done all the real work, let's consult the data to see what they have to say about it. And on the other hand, it's something that is integrated at every stage of the information gathering, dissemination, analysis, decision process. And I think early in the days of sports analytics, many more teams were on the, well, let's go ask the nerd after we've largely made the decision side. And it's, this isn't just basketball, this is every sport. Now that data analytics has become a thing that gets noted. And ESPN does rankings every year across sports of which teams are the most data forward and it's not great PR to be last in those. So there are teams that have those groups just to say they have them, not necessarily making great use of them or really getting anything out of it because it's this siloed off thing. And there are other teams again that have it fully in internalized in the process of everything they do. So it's almost hard to separate it out from the broader decision making process.

And do you have a sense of whether the teams mostly do their analytics in house or are they using consultancies or data platforms, things like that?

For the most part at the NBA level it's mostly in house. Now there are some use of consultants for certain problems of certain degree of complexity. I think a lot of a fair amount of kind of systems development work is maybe outsourced. This is a difference between baseball and I think other American sports is baseball is so integrated with use of data that there's almost two data groups within a baseball organization. There's one that's the analysis and the other one that's systems and platform and stuff like that. But that's having a 15 year head start. And certainly with respect to basketball, a much larger player pool to deal with. A baseball team will have the major league team and then their minor league subsidiaries and looking at drafting players at volume from all over the world. Whereas an NBA team has 15 guys and maybe a G league team which has maybe four or five guys they care about. And the universe of players who conceivably matter is considerably smaller. So just the headcount is just naturally smaller to deal with that more contained bits of analysis you're doing.

That's really interesting that it seems like recruiting players then is a bit more of a. I don't know whether is it an easier problem or is it just like you've got a smaller pool? I'm not sure whether it makes it easier or harder.

I don't know if it's easier or harder, but it is. Certainly it's the acceptance of using data in that effort is an easier road because the evaluation of players has always included some level of statistical. Now, well, we have better stats now, so let's use the better stats and more advanced methods to get those better stats. And all right, we're still doing basically the same thing we were doing before. We're just, you know, using a little bit of technology, a little bit of statistical technique to be a Little sharper at it. That's a difference in degree, but not in kind. Whereas certainly with the advent of player tracking data, the application of that to on floor in game strategy, that's a little bit newer, a little bit different, probably a little bit more directly threatening to the existing kind of coaching structure. So while that's got certainly has gained inroads there, that was a tougher nut to crack initially than the personnel side.

All right, I'd like to talk a little bit about your book. So do you want to tell us about what Mid Range Theory is about?

Sure. So the Mid Range Theory is in some ways, I'll use the analogy of a band in that the first studio album a band records, they have the entirety of the material they've ever developed over the course of their lives and they can pick the best stuff and throw it in there. And on some level, that's a little bit what the Mid Range Theory is and was is a lot of topics that I've written about, researched, thought about, talked about pretty extensively over the time that I've been doing basketball analysis and sharpened it up and put it together and put it out in the world.

I have to say it is a great read. Like really amazing sort of details and folklore about basketball. But one of the things that was a recurring theme was that it's really hard to figure out the attribution of which player contributed what. Can you tell me why this is a problem in basketball?

I think it is of the team sports where credit attribution is fairly straightforward. They're one on one confrontation games. So baseball, cricket, sports like that, a bowler or pitcher throws the ball, a batter or batsman hits ball, things happen. It's not too terribly difficult. Obviously how the fielders interact and stuff like that is a little more complicated. But largely speaking, either the pitcher or the bowler is able to fool the batter or not. And either the batter is able to execute hitting the ball well or not. That's pretty straightforward. In terms of the one person's success is to some degree is the other person's failure. In a sport like basketball or soccer or American football or hockey or rugby or Australian Rules Football or any of these other more dynamic sports, the interaction between players often has a great deal of impact on what the outcome is. I can certainly look and see, okay, I know what percentage of shots a player made. We certainly have the data now to know that the quality, the expectation of those shots going and the difficulty of those shots can vary widely from player to player. And figuring out why a player is getting easier or harder shots is quite difficult. Maybe a player is very good at freeing themselves. Maybe they have teammates who draw a bunch of attention or very good passers or any number of other things. So how do you divide the credit for, okay, this player made a lot of shots. They got very open shots. Is that them doing something well? Is that teammate doing something well? How would that change in another context where they were being asked to do more or less? It becomes very complex on the offensive side very quickly. And then on the defensive side you're trying to measure what didn't happen and who gets credit for what didn't happen on the defensive team. Or if there's no credit at all to be awarded to the defensive team, the offensive team just failed at executing what is a difficult challenge. The an uncontested three pointer in the NBA is made across the league about 38 and a half percent of the time. Player misses an open jumper. Did the defense do anything or is that just one of the just under two thirds of shots that guys just missed? And do you get any, do you get any credit for standing near the guy when he does that? Probably not. So those problems just that stock problem is really at root of trying to figure out who's good and who's not.

It does seem like there's maybe a bit of subjectivity in terms of like how you decide like who gets what sort of credit. But it seems like because basketball analytics is getting fairly mature, there are a lot of sort of standard statistics that people track to try and measure these things. So can you just give us an overview of like one of those things like you mentioned, like the number of shots someone takes. But I guess there are a lot more of these things.

So I think that when we're talking about individual players, a place to start is the intersection of scoring volume and efficiency. I said earlier that you can't always tell how difficult the shots a player is getting based on just the accuracy of those shots. There is a pretty strong correlation in the higher the frequency of shots you take. There's a strong tendency for the average difficulty and certainly the marginal difficulty of the additional shot to be more difficult. If you think about it, if you only shot wide open layups under the basket, you could probably. You might get to take two or three of those over the course of a 48 minute games. Now say you have to take 25 shots and progressively they get more and more. If we have to fill that quota, they get more and more difficult because the defense is doing more and More to try to stop you. So on one axis is its usage is the stat and it's just basically the play ending efforts, whether it's taking a shot, getting fouled, attempting a shot or turning the ball over, that a player takes per a given number of scoring chances for their team while they're on the court. And the other level is efficiency, which is how many points are you getting off of those attempts. And the intersection between those two stats tells you a decent amount about a player. Now having said all that, it of course is highly contextual because the player who might be the best player on one team at a certain level might be the third best player on a better team. And now the mix of shots they're getting for the volume, what will tend to happen there is their volume will decrease and their efficiency will increase. Because to get back to our, the first thing we talked about like they're getting easier shots because they're being relied upon less. But still the first pass is this player is taking 25% of the team's attempts and doing so at a rate of efficiency that's league average, above, below, what have you. That's a good first pass at what a player is doing strictly from a scoring standpoint. And that's not the entirety of a player's contribution, but that's the biggest chunk of it.

So do you have any examples of like players who were like maybe good on volume but not good on efficiency or like the other way around?

Sure. So I mean one of the current big examples of someone who is in the technical term is a chucker, someone who takes a lot of shots. And over the latter half of his career, Russell Westbrook has been a player who has had very high usage and very middling efficiency. And some of that high usage has been the result of the talent around him. And some of it has just been that's the player he is. On the other hand, there's any number of players who of more specialists. Not to get too deep in the basketball weeds, but you may have heard the expression of a 3 and D player, that's a mid sized player who's a good defender, but they don't necessarily have a lot of skill relative to the rest of the NBA with the ball in their hands. So their offensive role is mainly to stand on the perimeter and wait for the more talented players to create opportunities, pass them the ball and they shoot open three pointers. The better of those players will tend to be low volume, high efficiency players because mainly the shots they're taking are either layups or open team created three pointers, which are. Those are two of the higher expectancy types of shots in the game. So that. And then there's a sort of related player type which is a certain kind of big man who is basically like a garbage man around the basket who only shoots if he's right at the rim and he can dunk the ball. Again, you have a pretty low usage. You probably are maybe using 15% of a team's chances, but you're doing it so at a very high rate of efficiency because dunking the ball is. Those tend to go in quite a lot.

Okay. Yeah. So there's this big trade off between just like do you take tons and tons of shots even though they're not great chances, or do you just wait for the actual chance? So bit of tactics there. So just in terms of like how do you go about tracking plays? The new metrics being developed are the different things that are being tracked in terms of player performance compared to historically. Yeah.

So there's been three levels of data prior to about and this, I'm speaking specifically the NBA level. I think that at other levels of play, the data that's available is definitely sparser, less reliable. So I'm going to. Speaking specifically at the NBA level of play, up until around 1996, there was box score data, which is, you would know there's been iterations of it. But the, the, the version that was around for the last couple years of that era was how much a guy played, how many shots a guy took and made, how many free throws he took and made some rebounds, assists, steals, blocks, turnovers, fouls, you know, all those things. And that's. You can tell some things about that that does a reasonable job of saying who has contributed what on the floor. Then starting in 96 and then probably a couple years later, the data quality became sufficient to really rely on the NBA started doing detailed play by play. And that meant not only a little bit more information about not just how many shots did the guy take, but, but where he took them on the floor. If the shot was assisted, who assisted it, if the shot was blocked, who blocked it, if he was fouled, who fouled him. But additionally, since it's also tracking substitutions, the key change there was you could know which 10 players were on the floor for any one game event. And then that allowed you to start doing some correlation, doing some analysis. Okay, well, when this player's on the floor, what happens when this player's off the floor, what happens when these combinations of players. And so that unlocks another level of analysis. And then starting 10 years ago or so, starting for the 2013-14 season, across the league was implemented. It's known as player tracking data. It's essentially a number of sophisticated cameras placed in the rafters of every arena that allow for players to be identified 25 times per second and XY location and the ball. And this was another kind of exponential growth in the kind of things that you could know because when we got to play by play, you knew kind of some events that were happening. Now we know a little bit more about why you can start to see the, okay, you know the dish that's produced when a player, okay, player X made 17 foot jump shot from this location on the floor, assisted by player Y. Okay, we knew that. Now with the player tracking data, you're allowed to say, all right, the player who shot the ball, he cut from this spot on the floor and got a, received a pick from that this player. And the pass was delivered at this time and he caught it, didn't take a dribble, defender was however far apart, he was closing at whatever rate of speed he was or wasn't closing and then took the shot. So that's an exponential increase in the amount of data that you have. And it allows stuff like measuring. We had a colloquial understanding of maximum basketball is get good shots, have a pretty good idea of what those are. But now with this tracking data, you can actually much more directly measure the quality of shot. I mean if you're from the uk, so you're probably more familiar with soccer. It's a similar concept to XG from this spot in these circumstances, how often is this attempt successful? And so for the last 10 years we've had that, and not just that end of the chain, but also all the events that happened beforehand, whether it's whether it's a pick, whether it's a drive, how many dribbles, how many passes, how long a player has possessed the ball before all these things happen. All of that is a, just a vast increase in understanding the context of how the possession ending thing happened.

Okay, so that's kind of incredible. So growth in the use of analytics and just from a few simple metrics up to like really sophisticated like camera data.

I'll add one more by the way, starting next year. The tracking data currently is just like a dot at center mass of a player. Starting next year it's going to be data at the same frame rate which is going to be pose data which I Forget off the top of my head how many data points on each player it is, but it's some number in the low double digits. I don't remember if it's a dozen or two dozen somewhere in that area. But so you can track arms, hands, feet, things like that. Now what's going to come out of that data remains to be seen, but it's another exponential increase in the amount of raw information that's out there.

So I guess that leads to a natural question around like how do you go about harnessing that? Because once you have these sort of dizzying amounts of data, it's very easy to get lost. I was struggling. If you have any thoughts on how this is going to be used or how it's to make them useful, you.

Can just think about it. In the tools that you had, the number of data points when it was the box score, there might be 80, 100 data points for a game. When you had play by play data, we're talking a couple hundred data points a game, maybe depending on how you want to count them, maybe as many as like 1500 data points a game with the tracking data with it when it's just center mass. Now we're talking about 900,000 data points per game and the pose data that's coming in next year. Now all of a sudden we're talking probably getting close to 8 figure up into the tens of millions of data points for games. So just the technology that you would use to be able to address that amount, that size of data has obviously changed. You know, you could still do all your analysis in Excel when you're working, even at the play by play level. Once you get to the tracking data level, that's just not possible. Now we're talking about database solutions now just to pull the events like the basketball events. The CEO of Second Spectrum, which is the company that's had the rights to the provider of the NBA's tracking data for a number of seasons, describes it as pulling basketball words from the data to do that. Now we're talking about applying some pretty involved algorithmic and machine learning techniques to identify, okay, this pattern of moving dots is a pick and roll. And this pattern of moving dots has these features, these characteristics of a pick and roll, whether it's the players involved or some categorization of the style of what they're doing and the outcomes.

Okay, let's file it under fancy deep learning and AI. And it's probably going to turn crazy big data into something meaningful on a sort of basketball level. I'd like to talk a little bit about how basketball is changing. And when I think of what makes a good basketball player, the obvious thing is they have to be really tall. But you make a case in your book that actually height's been becoming less important to basketball since about 1980. The era of the big man has ended at that point. I'm sorry, why do you think people. That's been the case.

So height is becoming less important only in relative terms. If you want to, you could be reductive and say like the two traits in basketball players are skill and size. Obviously there's a lot more than that. Each of those things has more. But because of rule changes, because of style of play, because of realization of the value of certain areas on the floor relative to others, the perceived and probably actual returns to skill have increased relative to size.

Okay, so maybe height is as important as it was. Skills become relatively more important to compensate there.

And there's some demographic aspect of it too in that like if you need a certain level of skill, certainly once you get to NBA size, every kind of additional inch in height is a drastic reduction in the sheer number of people, not just players who fit that bill. So naturally, like there are far more six footers in the world than seven footers. If the key thing that you're indexing on is a certain skill based ability, you're much more likely to find a six footer that can do it than a seven footer. Now, height is still important enough in basketball that in most cases you don't actually want the six footer no matter how good they are at whatever the skill is. So you do tend to move up higher. But again, the relative impact of a guy who is six' six with certain skills and a guy who's seven feet without those skills has moved more towards the smaller player with the skill. Again based on rule changes and the competitive environment and dominant strategies and all those things.

So there's some sort of optimization process and it's, it's shifted slightly from to skill. So the other sort of notable change seems to be that it's been like the rise of the three pointer shot. That's a lot more three pointer shots being taken than two pointers. And beyond the obvious three points is more than two, why do you think this is kind of better?

So I think that a key inflection point from this was the development of the, was the implementation of the play by play data. There is sort of a bit of conventional wisdom that's like, all right, you're not A great three point shooter take a take, take one dribble and get an easier shot. Now it stands to reason, and it's true that taking a shot 19ft away from the basket is going to go in more than taking a shot from 24ft away from the basket. And for the entirety of the time that, that we've had the three point shot in the NBA, two point shots have been made at a much higher percentage than three point shots. When you factor in the extra point, maybe not at the higher level of efficiency, but that is a plausible story to tell, that even accounting for the extra point, you will make the 19 foot or more more often at a higher rate, more often than the 24 footer, that even the extra point does not overcome that increase in efficiency, that increase in accuracy. Now empirically what we found is that's not true. Once we had detailed location on where shots were taken from, you see that shots near the basket go in a whole lot. It declines very quickly until you get out to about five feet. And then once you get out to about nine feet from the basket, it basically stays more like pretty level, declining slowly until you get about 27ft away from the basket. So you might make the 19 footer 2 percentage point higher rate than the 24 footer. The extra point of the three pointer swamps that advantage. So really the empirical recognition that relative to shots just a little bit closer, the three pointer is better, that's one thing. And then you realize also, oh, so if I'm not doing anything, the better place for me to stand isn't 19ft away from the basket, it's 24ft away from the basket. And then that has some knock on effects in terms of, well, my defender has to stand further away from the basket too. So the person with the ball who's more skilled and is trying to get to the basket because shots right at the rim are the most efficient, I've basically created more space on the court for them. Also, it's that kind of changing of the geometry, the, the topography, if you will, of the court that was unleashed by the kind of the empirical understanding. And there were people who even in the absence of data, intuited this by observation. But with the data it's demonstrable that it's no, there are very few players for whom the shot from that one step further in is actually worth doing than the shot from further out. But until we had that data, it was an argument that was plausible.

Either way, that's cool that you actually need the data in order to make the decision about that like a pretty major strategic component of what you're doing there. And I do find it amazing that there is such a drop off in terms of the shot success by distance. You'd imagine it'd be fairly linear, but you're saying it actually just drops off very quickly once you get a few feet away from the basket.

Essentially, once you're shooting jump shots under NBA conditions, the accuracy tends to be in the low 40s to mid-30s when you're closer to the basket and you can more for a variety of ways you can shoot the ball. Like whether it's use either hand or little flip shots, use the glass, what have you, to avoid the defender, that's where. Or just take the ball and dunk it in the basket, which is the easiest shot. Again, height is good because it's easier to dunk if you're tall. Again being slightly reductive. But I think that's the mechanism we're talking about is when you're close enough that you can. You have that variety at your disposal. That's it just, it's easy, easier than if you. Well, I'm going to shoot a jump shot now because I'm far enough away from the basket. That's the only way to really do it.

And are there any other ways you think that analytics has changed the game of basketball?

I mean, I think in many ways it has just, I think related to that. There are certain types of ways to attack a defense that have been empirically shown to be more or less efficient than others. Now whether or not that's strictly an analytic like this is where it gets tricky because is that an analytical thing? Is it a rules based thing? There are certain ways of defending because the way the rules change used to be able to defend certain ways that you can't do anymore. And that I think fairly obviously will make certain strategies more or less viable than they used to be. You can probably more swiftly determine the efficacy of new strategies using statistical analysis than you would maybe by long term trial and error. So I think that's certainly one way where the speed of discovery has probably increased because of being able to. If I'm watching film to try to figure, if I'm watching a game tape to try to figure this out. I can watch about maybe a game an hour, probably a little less if I'm giving it real attention. If I'm using statistical analysis, I can watch every game from last night in five minutes at a lower level of detail. But to find out like if I'VE defined these events, these strategies correctly in the data. It's a press of a button and I can know a lot very quickly. I can look at several seasons of data of several, several thousand games worth of trials in a few minutes and versus something that would be essentially impossible for, for a human to do in more analog fashion. So I think that's. That has certainly the speed of kind of exploration I think has definitely increased because of this.

I'm sure like if you happen to watch videos of games, then yeah, there's something new to the number of videos you can watch at any given time. So I've met them automatically. It seems that like a huge improvement. I'm wondering are there any big challenges left in baseball, basketball, analytics? What's still to be worked on?

The whole analysis of defense is still very difficult. Again, you're almost inherently looking at counterfactuals. Defenses is as much about changing the decisions an offense makes as it is about the outcome of what they. Because as I said earlier, like sometimes guys just miss shots and even over the course of a season the natural variance is going to in many cases overwhelmed whatever defensive skill might have been involved there. But if we can look at the choices that an offense makes, the offense of the process by which they follow and see those things change, then we can maybe start to really directly measure what a defense is doing instead of just doing so by imputation of kind of what happens to the offense when they're there. But that's really difficult. Like any counterfactual analysis of what would have happened had the player with the ball done X instead of Y. Like the level of complexity there becomes very quick, very fast. And so defense is. All of which is a very long winded way of saying defense is hard, especially crediting individuals the defensive levels of achievement their team is seeing.

Okay, we've come back to that attribution problem again. Is trying to work out who did bot and why.

I mean that's, I mean that's on some level that's the whole game because okay, we had success. Why if we want to have more success, what should we change? If we change that one thing, what are we losing in other areas like this is the difficulty of it. Because it's not just. You're not just like adding a widget and it's purely additive. Like you're removing a piece, you're doing a different strategy, you're playing different players. So you're not just the benefits of the new piece, but what you're losing from the old piece and it's multifactorial. It's, it's in all these different areas this player does interesting things and we're nowhere close to really understanding what all those are. In some areas we have a close enough understanding that we can make some reasonable predictions about it. But I don't think that's anywhere close to saying the whole thing is solved.

So I'd like to talk a little bit about careers in sports analytics. So maybe and tell us how you.

Got into basketball athletics by blind chance and luck. I think certainly I'm of the early enough generation of people who work in the field that I stumbled into it from doing other things and there are much more defined pathways to getting in now. For better and worse. I think largely for worse if I'm honest. But I did a number of different things. I graduated from undergrad right as the first.com boom was crashing, the first.com bubble was bursting. I worked for an e commerce startup which failed after I'd been there for a little bit and then went back to school because I didn't know what else to do. So I became a lawyer. And while I was in law school I had learned that I had a. So my relative to people who do these kind of jobs, my level of formal statistical training is pretty low. But I played poker professionally for a number of years. I had a talent for estimating distributions, essentially which I discovered. And gambling for a living is a graduate level, perhaps even PhD level crash course in applied statistics, if not formal statistics. If you're going to be successful, you just have to develop that statistical intuition. So I did that for several years and then I moved into sort of educational consulting. While I was doing that, I started like writing about basketball on the side. I played small college basketball and so I've always been involved in the game. I played a lot growing up. I've always been interested in the statistical side of sports. And so I just started my own website and started writing. And this was around the time when the NBA was releasing the first tranche of the tracking data. And because I think the combination of my statistical intuition and my kind of basketball as basketball experience and knowledge, I don't think I'm being arrogant by saying I was doing more with that new data than most people in the public sphere were. And from that I got both better opportunities to write and do analysis for larger outlets, but also started to get contacted by teams. And I was old enough at the time that I was not going to take an entry level job. But the team that eventually hired me. The Milwaukee Bucks created a role for me that was commensurate with someone the middle of their career rather than at the start of their career. That's how I got there. I don't know how useful that is necessarily to other people other than saying that the parts that have served me best in that career are the practical and subject matter training that I've gotten from, whether it's basketball, whether it's learning hard lessons about incentives in other fields have served me well in ways that that pure training in statistical programming or formal statistics and machine learning and stuff might would not have.

So I do actually love that you've got this winding path to getting into data. And it seems that it's very common that people discover midlife that, oh, actually I really do enjoy working with data and end up in the field. So for any of our listeners who want to jump in, do you have any advice on what skills they need?

So there's several different answers to that question. The first one is a little bit of introspection about what they mean when they say they want to work in basketball or sports analytics. There's two different questions there. Do you want to work in sports and sports analytics, or do you want to work for a team? On the sports side, one of those is a much larger target to aim at than the other. And also is the costs of doing so and of succeeding is lower because working in, whether it's the business side of a sports team or as a data. Working for a data provider as I do now, is much closer to a stable workday. Work, life balance, plan your vacations, do all the normal career stuff. Whereas it's almost like working for a team is you're almost working chef hours. When I worked for the Bucks, I went to 41 home games a year. That's 41 nights that basically like my workday did not end at. Not that my workday ever ended at 5 anyway, but my workday ended at 11. Maybe later, if something happened in the game that needed an immediate response. And really. And considering there was 41 home games and it might even happen on a road game, even if I was at home watching on tv, I'm still at work. And really, when you're working for a sports team, you're all kind of at work, even when you're not at work, because things come up at any given time. This isn't in any way trying to dissuade people from chasing that if that's what they want to do. It's just understanding that's the choice that you're making and being upfront about. Are there rewards to that in terms of being in a competitive high performance environment? Is that your thing enough for it to be worth it? So that's the first thing said. That has nothing to do with technical skills or anything like that. That's just setting appropriate career goals. Aside from that I think is very important because a lot of what is very common for people to land one of these jobs and then start to do the job and it's like this kind of actually sucks from a life perspective and that's no good for anybody. It's no good for the team that hired you, it's no good for you. So trying to, if that's not going to be your thing, don't do it. And that's probably good career advice regardless of anything. But I think in this field it's worth pointing out now from a more job skill standpoint, sports analytics are becoming mature enough that a lot of the roles that one would see in a technologically based role are starting to see. I mean teams are now hiring for data engineers and they're hiring for analysts, hiring for front end and back end developers. They're hiring for all of these different roles that touch data. And so finding your lane in that is from a standpoint of identifying what job you're going to chase is the next most important thing to do. What is your competitive advantage on the hundreds of people that apply for any of these jobs with the team? And if you want to be a data engineer, like if that's your skill set, then like focus on that. If it's more on the analysis side then honing not just your statistical techniques but your communication ability, whether it's written, whether it's verbal, whether it's doing it writing a bulleted deck, whether it's a data visualization stuff like that's a communication job as much as it is a technical job. And the closer you are to that, the more the subject matter knowledge of the sport becomes important on a certain level. You're setting up a data pipeline. Data is data. It certainly will help to know what to expect from the structure of data you're getting from a given sport. But it's not as necessary if you're in a standpoint where you're going to be presenting to coaches and front office executives your findings. You better have the terminology right, otherwise there's no quicker way to lose buy in. So that in parallel with kind of the technical skills is important. And I think even though it's probably more so in sports than other industries. I think there is an important, more broadly applicable lesson about the importance of having that knowledge of the subject matter. Just so you're aware of what the data you're working with actually says, actually represents. It's an abstraction of reality. What is it abstracting away? What is it capturing? What isn't it capturing? I think that having a decent level of understanding of that really gives you the first level of sniff test of almost QA ing your own work. You get a result that doesn't make sense. I did something wrong. My data is bad. Let me go examine that again before I look what I found. And then you reversed a sign somewhere in your code or the data was faulty in some ways that is skewing your the analysis badly to the point. I mean, you're in a garbage in, garbage out scenario then. And being able to catch that yourself can save you a lot of pain. So I guess if I was going to reduce what I just said to two maxims, it would be like, pick your lane and learn what? Learn about the thing you're talking about. Learn about the thing you want to talk about so you can talk about it intelligently.

Nice. I mean, it does seem like there's a pretty sort of broad set of skills you need. You need the technical skills, communication skills. You need all the sports knowledge. And it sounds like chick, if you want to work for a team, you got to really love the game. It's more of a lifestyle than a job.

I think that's a great way of putting it. And I will say teams, like for a long time, teams were not very, very good at. When they put job descriptions out, they weren't very good at describing what they were actually looking for because it was like a full stack developer with eight languages and this and that and this and that. We're going to pay you $40,000. That's not really the case anymore. I think teams have become mature enough that they can start to know what we actually need is an analyst. What we actually need is an engineer. What we actually need is a backend developer. And putting job requirements that increasingly, though not perfectly, reflect the tech stack they're actually working with. So finding a job description that actually matches skills you have is or can work towards is a little bit more of a guide than it might have been even five years ago.

That's very promising. So we should actually talk a little bit about what's going on in the NBA right now. So have you been involved in any analysis for the playoffs this year, this.

Is the best, worst time of year because it's. There's a ton of games happening and they're all interesting and meaningful. So yes, I've been. We're at the point we're recording on the day after the first game of the conference finals. So we're finally at the point where we're down to no more than one game a day, which is when work, life, balance can return or when some balance can return to the force a little bit. But yeah, no, I mean it's watching in detail and analyzing across the first two rounds of the playoffs. Now that I no longer work for a team and I'm not responsible for, for day to day content produced off of basketball, it's come full circle into like almost a, a semi professional hobby to analyze what's going on now.

And have there been any kind of interesting data moments, any upsets or things like that or remarkable stats coming out after the games?

So it's a. An interesting kind of under undertone, under background bit of NBA analysis is the style of play, the strategy of play, the value of certain players has always diverged between regular season and playoffs. To some degree it's very different preparing to play one team seven times across a couple weeks as it is preparing to play against a different opponent every other night and sometimes even consecutive nights. And like the level of detail you can get into about what up team a player is or is not good at will or won't do just allows you to do different things. And second of all, just like when you cut off to go from regular season to playoff, you're basically eliminating half the teams. So it's maybe it's not one to one, it's not just the top half of players, but your average opponent is much better in the playoffs than they are in the regular season. So whether your skill set translates as well against better opposition as it does against worse. So there's always been this divergence between regular season and play, I think over the last several years. And why that could be is a whole other several hour podcast. But that divergence has only increased and this year it almost feels like we're at the at a peak of regular season and playoff divergence. Though maybe not because this was a year we're in a transition period in the NBA where the previous dynasties have largely run their course and the newer ones haven't quite risen up. So this was the most compressed that the league was top to bottom of relevant teams in terms of how good they were, which is in many ways, I think fits well with these playoffs being less predictable. Maybe they were predictable in the fact that, like, teams were closer, so things were much closer to coin flip than they normally are. And when things are closer to coin flip, weird stuff happens and you shouldn't be surprised by it. So either the regular season was not very predictive of what would happen in the playoffs, if you look at the rank ordering of teams, or it was very predictable that weird things would happen if you looked at, like, the narrowness of the distribution between teams.

And I guess that's the most exciting time when all the teams are pretty close and it does become more unexpected. So it sounds like some great games coming up.

What's the data show?

I'm like, which.

It's unexpected. It's exciting, but it all. No, it's exciting, but it also has a tendency to make you feel stupid because it's like, I have. I. I've made all these predictions and they're all wrong except for the ones that are right, but there are still a lot of them are wrong. So do I know anything or is it just weird in terms of what is the da? I mean, I don't know.

That's a.

Like, the playoffs are also difficult. Billy Beane was quoted in Moneyball, which is obviously a fairly seminal text in the development of sports analytics. He's quoted as saying, my shit doesn't work in the playoffs. And there's something to that because, again, it's A, inherently the sample sizes are smaller, B, they're biased sample sizes because you're only playing against one opponent instead of the league as a whole, and C, it's just that, like, the style of play is weird and more intense and things that sort of can just happen and you brush off over the course of the season, all that'll even itself out. Losing an extra game here or there means is the difference between winning and losing a series, advancing or not winning a championship and not. So what do we make of those? Statistically, that's always tough. Like, you can maybe, well say this, that, well, this team made a historically high rate of shots in this game, and that's why they won. That's neat. That's interesting. What does that say about next game? The expression is it's a make or miss league. You made shots tonight, you might not make them tomorrow night, and we have no idea why. You just have to live with that and hang on for the ride. That's not a very satisfying data answer, but learning to say I don't know is pretty powerful. And when we're like, you have game one of the Eastern Conference finals tonight. On paper, the Boston Celtics are much superior to the Miami Heat. Miami Heat have been a better team over the playoffs than the Celtics have. Which one of those is going to hold Sway? I don't know. And even figuring out, once we actually see what happened, trying to unpack why it happened will be its own challenge.

All right, sounds like it could be anyone's final length, so really exciting stuff. Okay, so with that, thank you very much. Seth. It's been great having you on the show.