We’ve had very few NBA games over the past week, but there’s still a lot to discuss due to the Sloan analytics conference and a new feature I’ll be doing: answering random stats-based questions at the end of the article instead of a mini-deep dive into stats. I’ll be taking the time with my stats studies instead, and most of those deserve the full attention of an article.
There’s also a lot going on with the league in general, from how dominant the Rockets are when healthy to the new-look Cavaliers, along with monster games from Anthony Davis and Damian Lillard, to the madness with the Mavericks work place. And with that, let’s look back at the last (half) week in basketball.
The 2018 MIT Sloan analytics conference: Day 1
Over the weekend, a large number of people with opinions on font styles and graph margins descended onto the sleepy town of Boston, Massachusetts for an annual conference on numbers in sports. It’s an expansive conference with a large number of papers and discussion panels that you can sink your teeth into ranging a number of sports. I’d recommend our own recap here for the discussions, but I’ll quickly share my own thoughts on some of the subjects.
For the rule changes, I know we’ve all discussed ad nauseum about intentional fouls and playoff seeding and the like, but it was nice to see they addressed the free-throw slow-down. I’ve considered this for a while, and it makes me feel saner others are considering it: changing free-throws so that you take one to make two points (and perhaps one shot for three points.) This would substantially decrease the length of games, and I don’t think there’s a reasonable counter to this. Sure, players would be taking fewer free-throws, and in a sense that increases randomness — but that’s not bad, and free-throw percentage is pretty stable anyway. But the league isn’t receptive to the change; they don’t see its value. I’m not sure what they’re missing. Is it useful to have those little breaks in the game? Would we break the fundamental rhythm to basketball? I’m not sure, but it’s an experiment worth undertaking in the G-league at least.
The other topic that interested me was Bhostgusters, a player tracking application that would let coaches sketch out a play on a pad and get near-instantaneous feedback on how effective it would be based on a deep-learning algorithm using “ghost” defenders. (It’s a play on Ghostbusters, I assume, as they wanted to tip-toe around the copyright issues; or they’re crazy.) I wanted to highlight this because it may be the future of not just the league but the entire world. This kind of output from software may overtake our lives, and it’ll be fascinating to see how that plays out in the NBA, where people pout and throw fits over “advanced” stats that are basically just simple percentages. I’m just warning you all — be prepared.
That one famous guy we can’t talk about (no, not Steve Guttenberg)
For reasons unknown to me and the vast majority of the media, we cannot discuss a certain panel that was led by some super famous guy. It’s quite bizarre that we can’t talk about the contents of the panel; that has no precedent, and it’s not like they were discussing nuclear arms or alien contact (I hope.) Instead I’ll point to a very funny video embedded below that may or may not have anything to do with the person in question. The Game of Zones videos are popular, and I can’t believe we’ve gotten to a place in society where you can parody a fantasy series with Barack Obama, Sam Hinkie, and use prominent NBA guys like Vlade Divac and Charles Barkley as the butts of the joke. And can we get the reactions from everyone parodied in the video? This is nuts.
The 2018 MIT Sloan analytics conference: Day 2
You can see a recap of day 2 here. The basketball talk was more general that day and wasn’t tied to specific research topics. But one panel did catch my eye: “inventing basketball” with Daryl Morey, Steve Nash and Shane Battier, and moderator by Jack McCallum. Obviously, the panel implied the catalyst was the 2005 Phoenix “seven-seconds-or-less” Suns. But other people, like Don Nelson, should get more credit. He was experimenting with lineups for years, and his 2007 team, the Golden State squad that upset the No. 1 seed Mavericks, used Stephen Jackson as the nominal power forward and played long stretches with Al Harrington, a 6-foot-9 outside shooter, as the center. He was also the engineer of those early 2000’s Mavericks teams, funnily enough, that sometimes used Dirk Nowitzki as the center. Don Nelson has been marginalized during this skill ball/small ball revolution; he deserves more credit for at least being ahead of his time.
Finally, I want to comment on the “process” in Philadelphia, where martyr Sam Hinkie made an appearance. Some people complained that the problem with the process, or tanking in general, is that you are swinging for the fences hoping for a superstar while ignoring the role players. But that’s demonstrably false. Some of those lottery picks can turn into valuable role players, like Dario Saric, and tanking hard can increase the value of your second-round picks, leading to someone like backup Richaun Holmes — in fact, I can’t believe people forgot how obsessed with second-round picks Hinkie was; it was a big part of their team-building. And by eschewing veterans who would have no future with the team anyway, they could take more chances on undrafted guys, like Robert Covington and TJ McConnell. The process did highlight role players; it was about a lot more than shooting for the next superstar — at least that’s what I’d have said if I had been (shockingly) invited to the panel.
Jimmy Butler suffered a meniscus injury during the first quarter of the Wolves’ loss to the Rockets last Friday. Reports state he’ll be back in 4-to-6 weeks after the surgery on Sunday. Losing a player of his caliber is awful, and we should all hope it does not impact his play when he comes back. But the injury is more concerning the more you study it. Butler was leading the league in minutes per game, and he asked to sit out the All-Star game due to fatigue — and was injured immediately after the break ended. What did the staff miss? Also, based on the timeline, the surgery is probably the “quicker” fix where part of the meniscus is shaved off, instead of sewing the tear together for a more complete heal. The latter would have meant he’d be out for the rest of the season, which is likely why they chose the other option. But is his future worth that?
The last great tankathon
Incrementalism is the NBA’s mode of change. Tanking has been one of the larger issues with the league’s quality, and we have indeed gotten a change but it’s not a drastic one: the three teams with the worst record will receive the same lottery odds, instead of the worst record being gifted 25 percent odds, second-to-worst 19.9 percent, and third 15.6 percent. The change, and the fact that the draft seems strong right now, is supposedly causing an epic tank race to the bottom. While perception is everything, I don’t think the lottery change will in reality have that much of an effect.
I actually proposed a similar idea years ago to curtail some of the tanking. Not like it’s a totally original idea, but mine was giving the same lottery odds to the bottom seven teams, and improving odds a bit for the rest of the teams — and the latter was included as well. My thoughts were, if you’re in the bottom seven, you’re not in competition for the playoffs so there’s definitely no incentive to win, and we don’t need that added motivation to lose. Also, if you’re not in that bottom seven, you are more likely distracted by the thought of making the playoffs for the vast majority of the season.
The problem is that the bottom three qualifier is too soft a change. For example, since 2005 the average difference in wins between the bottom three teams has been 4.6 wins, and the average difference between the fourth and third worst records has been a mere 1.8 wins. This means you have to push far into the season before you lock into your final lottery slots, and by that point the season is just about done anyway and many teams have thrown in the towel. Additionally, the average difference between the worst and fourth-worst records has been eight wins since 2005, and you can boost that to 15 wins when you look at the eighth-worst record. You can lock the worst teams into their lottery slots a lot sooner this way. The shifting of the odds away from the top three to the rest of the lottery does help too, but there’s still that incentive to jump into the bottom three.
We’re going to hear a lot about tanking for the rest of the season. Mark Cuban was recently fined $600,000 for telling his team the plan was to lose games. The Kings are sitting their veterans. Normally, teams tank, and players don’t, but this season is getting weird. The Bulls were so awful in the clutch against the 76ers that the most plausible explanation might be tanking. And it’s only February. This might become the most brutal tanking season ever, even surpassing the race for LeBron James in 2003. Changing the lottery odds so the bottom three all have the same record won’t fix everything or even close to nearly enough — we’ll need more.
Drummond leading in DBPM
As a new tweak in my week-in-review series, I’ll be tackling some “ask-me-anything” style questions about NBA analytics at the end of the articles. This is definitely not new, and I’m not the only one, but people ask a ton of questions out in the aether that never get answered and I don’t want a lot of these advanced stats to seem inaccessible.
This question was directed later back at me, and it’s one I’m primed for answering as much as almost anyone else out there: it comes down to a non-defensive stat, assists. In order to fit the data better, BPM used a couple interaction stats: square root(AST%*TRB%) and AST%*USG%, which I’ve termed the ball-boarder and the playmaker measures, respectively. Basically, the players who did well in a long-term plus-minus model were ones with versatile stats, and many great defensive players rebounded and passed well. And those stats aren’t only relevant for offensive BPM: a little over half of the coefficient for the playmaker measure is used for defense, and two-thirds for the ball-boarder measure. That’s just a correlation — it doesn’t mean every player who does both well is an elite defender. But it does greatly improve the prediction power of the stat, which I’ve found first-hand when building my own metrics. But one can argue it’s overfitting to a significant degree, and you can see that with some of the results.
Andre Drummond, thus, benefits this season due to his increase in assists. He takes a decent amount of shots, so the AST%*USG% piece is working for him. And naturally he’s a monstrous rebounder, so he’s crushing it in ball-boards. He’s one of the leaders in the latter, and it’s most certainly helping his stats. If you substitute his 2017 assist rate into this season, his DBPM would drop by roughly 1.7 (without considering team effects or others.) In fact, he’s not the only for seeing benefits on DBPM for assists. You can see the leaderboard below where every player is a high assist player, or at least for their position, until you get to number ten at Clint Capela. That strong pattern dissipates a bit if you include players with at least 500 minutes — many defensive specialists don’t play heavy minutes anyway — but it’s something to consider when citing this stat. BPM is definitely better than PER and most metrics out there, but it too has weaknesses — you just need to understand them.
Table: top qualifying players for DBPM, 2018
Shooting variance in individuals
If I wanted to answer this question thoroughly, I would need a lot of space and time to do so. Luckily, you can see a recent article on variance in general here; it goes over current players using a variety of single game score measures. Near the end, the article notes that Devin Booker has been one of the most inconsistent players, and the Suns have actually won more games because of it since he’s played so well against good teams. In fact, consistency is indeed a property of teams, and it’s important for translating point differential into wins. Basically, if you have a huge amount of variation, you trend toward the mean — i.e. good teams win closer to 41 games than expected, and bad teams 41 as well. Thus, inconsistent players will help bad teams and hurt better teams — but the adjustments are generally pretty small.
Kyrie Irving versus Damian Lillard, career
I believe this question was posed because they have freakishly similar career stat-lines. See for yourself; it’s bizarre. Kyrie Irving has a slight edge in usage rate, but Damian Lillard has an even smaller edge in efficiency. They have similar assist and turnover rates. Lillard takes more 3-pointers and free-throws, but Irving is a better shooter inside the arc. Irving has the edge in steals, but Lillard the edge in blocks. Irving has more offensive boards, and Lillard defensive boards. They even have virtually the same free-throw percentage. And both guys were productive right out of the gate as rookies.
Overall, by most measures Lillard has had the more valuable career so far in the regular season because of his durability. He’s played more minutes even though he debuted a year later. Of course, Irving has more production in the post-season, but that’s largely because he gotten to play with LeBron James. One could cite Kyrie’s clutch performances, but Lillard is no slouch there either, and he’s had big playoff moments as well. I suppose one would say the new Celtic will have a better career in the end because of his age, but players with a lot of injuries usually don’t age well — I’d call that a wash too.
When the box-score is that close, you gotta turn to overall player impact outside of the box-score — and that’s still tricky. Both are notoriously shaky defenders. Irving probably has a better reputation at this point, but I think that’s only because of the caliber of teams he’s played with; although if you add in the playoff reputation and how he supposedly cranks up the defensive intensity, or how he’s changed in Boston, and this should be an easy win for Irving, the guy who’s been on more winning teams. But this isn’t easy: hustle stats aren’t in his favor and multi-season plus-minus models almost always see Lillard as the more valuable player. Remember, plus-minus is just showing which players correlate with a team’s ability to outscore its opponents; it adjusts for who’s on the court. And this may be surprising: Lillard has had a better RPM, which factors in plus-minus and almost every countable stat possible, in every season the stat has been available (since 2014.)
Table: Hustle stats per 36 minutes
|Loose balls recovered
|Off. Fouls drawn
|Contested 2PT shots
|Contested 3PT shots
Overall, this is actually a tough one. Plus/minus stats actually saw Irving as a negative for most of his years in the league, and although that’s surely misleading — there are some issues going on with him and Matthew Dellavedova’s plus/minus that’s screwing up the models — it’s not a great sign. For what it’s worth, it appears fans give a slight edge to Lillard, even if the media is more in love with Irving’s game. Many people cite this reason, and I do agree with it to an extent: I’d rather have Lillard run the offense and the team. Throw in his four-year degree and his belief that the world is round (and, well, his durability), and I’d take his career by a slight edge.