Self-Made Men

Plus an interview with a basketball soothsayer

A quick house keeping reminder before we get into the meat of the newsletter.

The F5 makes money by sharing the code and data used to generate the visualizations in this newsletter. I paused subscriptions when I abandoned the newsletter to work for the New York Knicks back in 2021, but I’m un-pausing subscriptions this FRIDAY, NOVEMBER 1, 2024.

Paid subscribers can still expect to receive the code and data like always. But this year I’m adding additional content for paid subscribers that should be of interest to all readers, regardless if you care about code. More on that Friday.

If you’re a paid subscriber from back then and no longer want to be, I’d suggest canceling your subscription before you incur a $5 charge. You don’t need to do anything if you want to remain a subscriber.

---

I joked at the start of the season that Boston’s head coach Joe Mazzulla would be single handedly responsible for pushing the league average three-point rate above 40% for the first time. Well, as of this writing, the league average is 41.9% (that drops to 41.3% if you remove Boston).

That’s not only the highest in league history, it’s the highest through the first 67 games¹ played in any season.

Last year, Boston led the league in the proportion of shot attempts coming from three at 47%. So far this season, there’s seven teams (Boston, Minnesota, Chicago, Orlando, Brooklyn, Golden State, and Charlotte) that are taking more than 47% of their attempts from three.

There’s more threes than ever. Stop me if you’ve heard that before.

But there’s something different about this season that I think is worth exploring further. To see what’s unique about this season we need to look at what kinds of threes we’re seeing more of.

The chart² below shows the shot attempt rates from the five main parts of the court as defined on pbpstats.com: At the Rim³, in the Short Midrange⁴, in the Long Midrange⁵, from Above the Break, and in the Corners.

What stands out is the bump in Above the Break Threes, which have largely come at the expense of Long Midrange shots. Shot rates elsewhere have have more or less remained the same.

If you’ve come across a hexagon shot chart on social media you already know that The Game Has Changed and likely have strong feelings about the so-called death of the midrange. But from Seth Partnow’s work and others we know that’s at best an incomplete story.

To briefly summarize Partnow’s Midrange Theory, the midrange is alive and well for superstar players. The main reason shot attempts from the midrange have decreased over time is because two-point catch-and-shoot jump shots by role players have been replaced by three-point catch-and-shoot jump shots by role players. The isolation midrangers made famous by Michael Jordan and Kobe Bryant can be found by watching any game featuring DeMar DeRozan, Kevin Durant, Brandon Ingram, and countless star players.

What we’re seeing now with the jump in Above the Break Threes is what happens when there’s no more catch-and-shoot twos to convert into catch-and-shoot threes. Teams that are hungry for even more three-pointers have replaced pull-up twos with pull-up threes at a rate we’ve never seen before.

This is evident when you look closely at “self-created” shots.

A shot is considered self-created if it’s unassisted. But an assist is only recorded when there’s a made basket. So we don’t actually know much about self-created attempts if we’re just looking at publicly available data. But we do know something about self-created makes.

So far this season, about one in every five made threes has been unassisted. That’s an almost five percentage point jump year-over-year.

In the past, superstars were willing to settle for pull-up twos in the midrange because they were usually the least bad shot available. At the same time, opposing defenses that were focused on protecting the rim and determined not to foul were willing to concede pull-up twos in the midrange because they too were the lesser of all evils from their perspective.

But as players have gotten increasingly comfortable pulling up from beyond the arc it’s become more common for players to eschew isolation midrangers for isolation threes. These new isolation threes come almost exclusively from Above the Break, thus explaining the increase in the shot attempt rate from that zone seen in our earlier chart.

To Partnow’s credit, he saw the contours of superstars extending their self-created game to the three-point line back in 2019 when he published his original treatise on the subject.

A similar pattern is only recently starting to emerge as a larger but still select group of players have added the pull-up 3 to their arsenal.

From 2003-04 to 2014-15, league total unassisted made 3s rose from 1,750 to 3,010, an increase of just under one make, total, per game over that 11-year span. In 2018-19, there were 4,935, almost 1.6 more makes per game than just five years ago. The increase is illustrated by the fact that James Harden has set a single-season record in most unassisted 3s made each of the past three seasons, jumping from 196 in 2017-18 to 317 in 2018-19. Those 317 makes last season equal the TOTAL of the top two single-player season marks prior to 2016-17 (Steph Curry in 2013-14 and 2015-16).

I’ll finish by hedging a little and say some of this could be early season noise and things will settle back to their normal levels later into this season. Two of the league’s most prolific midrange scorers in Kawhi Leonard and Joel Embiid have yet to play in a game this season. Those two by themselves could tip the league averages back towards more isolation midrangers and fewer isolation threes. Also, early season offenses tend to be sloppy and players jacking up threes without running any offense is not that unexpected.

That said, the NBA is a copycat league and the two teams that met in the Finals last year were 1st (Dallas) and 5th (Boston) in percentage of threes that were unassisted. Additionally, the players that currently lead the league in unassisted made threes are Jayson Tatum (26 years old), Jalen Green (22), and LaMelo Ball (23). That fact that all three of those players are on the younger sider of the aging curve suggests to me that these self-made threes are a product of a new evolution in superstar shot taking and not something that will come out in the wash once more games have been played.

---

A Refreshing Interview with Sports Mediocre

At the start of every season The Association for Professional Basketball Research (APBR) message board hosts a wins projection contest where entrants submit their estimates for each team’s regular season win totals and the person whose projections are most accurate is crowned the winner at the end of the season.

Last year’s winner was the user sports_mediocre (SM) on Twitter. If you’ve seen SM’s handle before it’s probably from their twitter thread on why Luka Doncic is the most overrated player, which caused some mild controversy in NBA analytics spaces.

I reached out to SM to ask them about defending their win projections title, being known as the internet’s biggest Doncic hater, and what it’s like to post anonymously as a former NBA staffer.

This interview has been edited for length and clarity

---

F5: Let's talk about where your projections diverge from the consensus or more specifically, Vegas. The first thing that caught my eye is that you are considerably lower on the New York Knicks and Milwaukee Bucks this season relative to their Vegas win totals. Do you have a sense for what's driving those differences in your projections?

Vegas Win Totals via DraftKings

SM: For the Bucks, it's a combination of steep aging (Khris Middleton, Brook Lopez, and Damian Lillard), fewer star minutes (Giannis Antetokounmpo had his highest minutes since 2018 last season), and the fact that last year's team was really a bit lucky and only played like a 47 win team. So my projection is really just a slight regression from last season.

For the Knicks, Isaiah Hartenstein is a huge loss. I actually bet on the Knicks two years ago in part because I thought if he got big minutes after leaving the Clippers, he'd make a huge impact. Unfortunately, I was off by a year.

I'm not a big believer in Mikal Bridges. I think new pieces require time to get comfortable with each other.

But most importantly, and I don't think most people realize this, last year's Knicks team was ranked 14th-16th in six of the eight Offensive/Defensive Four Factors.

The exceptions were 6th in Opponent Free Throw Rate and 1st in Offensive Rebounding Rate (ORB%). That ORB% ranking was a product of Hartenstein (gone), Julius Randle (gone), Mitchell Robinson (injured), and Josh Hart. I don't see Karl Anthony-Towns picking up all that slack, and what he brings in efficiency doesn't make up for the lost possessions. So I think they'll be slightly worse than last year, but still good.

Meanwhile, you’re more optimistic about the Oklahoma City Thunder and the Brooklyn Nets than Vegas is. What gives?

For OKC, I really struggled to come to a number on how good they'd be. Before Hartenstein’s injury, I had them winning 64 games because Hartenstein sort of perfectly fills the only void in their team, rebounding. Honestly, I could see them winning 70. And if they ended up closer to the Vegas line despite good health, there would have to be serious regression from young players in roles they've already thrived in, which is just hard to imagine.

The Nets have a lot of pretty bad players, but they don't have many awful players. It's possible they turn up the tanking effort, but their current roster looks more like a -6 Net Rating team than a -9 Net Rating team to me. Honestly, I'm not very confident in my take on the Nets.

Just for fun, I want to push back on the OKC love a little bit. I think everyone basically has them as the best regular season team in the West. But, are you concerned at all that once Hartenstein comes back from injury that they’ll have to go through an identity shift? Right now they get to play five-out with Chet Holmgren at center. Moving him to the 4 seems like it’ll improve their defense at the expense of their offense. But their defense is already good! So I guess what I’m asking is are we sure OKC will be "this" good?

Yeah, fair question. I don't think what makes OKC great is shot selection, which is what might change with Hartenstein at the 5. Last year they were 12th in percentage of shots coming from 0-3 feet and 18th in percentage of shots coming from three. Comparatively, the Knicks, with Hartenstein, took more shots at the rim and from three. I think Hartenstein fits really well even if offensively he's more like Steven Adams on the 2016 Thunder — just setting screens and getting boards while letting shot makers do their thing.

I think the Hartenstein rebounding impact improves them significantly on both ends, and yeah Caruso is going to Caruso. Side note, I didn't realize that Caruso shot 41% from 3 last year until I was working on projections. He really is the perfect fit for OKC.

What's your process for coming up with your win projections? What makes your process different from others that leverage all-in-one metrics like EPM, DARKO, LEBRON and others?

I think at a high level they're all similar: I use player per 100 possessions impact values and minutes estimates to make a bottom up projection of each team's Net Rating and win totals.

But I stray from other processes a little by using a team's previous season's Net Rating as a baseline, as I believe it captures a lot of the player interaction effects that no Adjusted Plus Minus model that I know of has effectively calculated for individuals.

From there, I use player movement between teams, injury minutes adjustments, and aging curves to move the Net Ratings. I also tend to overweight the back half of the previous season

Are you using your own minutes estimate or someone else’s?

I use my own estimates, but they're the most artistic and least scientific part of the process for me. I round heavily, using average minutes per game from previous healthy seasons. Then games played is a very finger in the air guess based off a previous numbers of seasons average.

And for context on why I don't put as much effort into this estimate 1) I think it's the hardest to predict and 2) the difference in 2200 or 2400 minutes for a +2 impact player equates to roughly 0.2 wins, so I think of ballpark estimates as useful enough

That’s interesting because perhaps naïvely I always assumed that the difference between someone who nails win projections vs. someone who doesn’t comes down to who gets closest to properly estimating minutes and therefore injuries, which is sort of the great unknown. But it sounds like what you’re saying is that there’s more value to be had in properly assigning weights to what matters (like a team's Net Rating from the preseason).

I believe that if you look at the difference in accuracy between a model that's 95th percentile in player value accuracy compared to one that's 50th percentile, and then you do the same 95th vs. 50th percentile comparison but for minutes played, the greater difference will be on the player value side.

To give an example of why, I'll use Victor Wembanyama. I've chatted with Kostya Medvedovsky about why he's so high in DARKO. I have him at roughly a +1 value, meanwhile DARKO has him at +5. Two models that have him at +5 playing 2000 vs. 2500 minutes (which is a massive difference) will differ in win totals by less than 2. But if we both had him playing 2250 minutes at +1 vs. +5, those models would differ by more than 5 wins.

Basically the differences in value measurements are more impactful in win projections AND there's more signal to work with in the first place. When I look at why my projections last season performed better than every other one in the APBR competition, it was because I considered players like Fred VanVleet, Dillon Brooks, SGA, and Franz Wagner better than other models (I was by far the highest on HOU, OKC, and ORL), not because I was more accurate on their minutes

You’re probably best known in some circles as the analytics person who is most skeptical of Luka Doncic's impact on winning. I don’t believe you think he's "bad", but I think it's fair to say you think he’s overrated. I'm curious what you think some of the more popular all-in-on NBA metrics are missing when it comes to Doncic? In other words, if he doesn’t drive winning in the way most people think he does, why does EPM, DARKO, LEBRON, etc.. all seem to say otherwise?

Let me start by saying positive things because I think they get missed once I get into the negatives. You're right, I don't think he's bad. Nor do I think he's average. I think he's good, but not great. I'll also say I think his penchant for impressive highlight shots and passes that make him fun to watch beyond any winning impact.

Okay. Got that out.

I think the primary reason those models overvalue Luka is their overweighting box scores. For EPM and LEBRON, they're using single season metrics, and single seasons have so much noise in +/- data, that box scores, being smoother, should generally be weighed more heavily.

DARKO, being multi-season, has generally been lower on Luka than single season all-in-ones. In the thread that started this whole thing, I did point to RAPTOR, which conveniently split the box score and +/- components of the all-in-one, to make the point that Luka, more than any other player, is benefitting from box scores over +/- .

Basically year after year he has great box scores and year after year his +/- numbers are slightly above average. Over the long run, I trust those +/- metrics more than his box score metrics. So long-term RAPM is more important to me than box scores explains most of it.

The other piece is applying the general to the specific. These all-in-ones do a great job of minimizing GENERAL errors in predicting play-by-play data. But for specific players, predictable errors highlight the flaws in general applications of a model.

Five seasons into his career, there was plenty of data showing Luka's otherworldly box scores were not translating to on-court impact. The errors in the models you mentioned would be higher for Luka's possessions, and they'd be higher in the same direction each season.

I think these models provide the best guidelines we have on player value, but I don't think they're randomly wrong. I think they're specifically wrong, and that difference is often most visible on box score outliers like Luka or Wemby.

But you know what, just so my first point isn't forgotten, I'll restate it at the end: I don't think Luka is bad. Nor do I think he's average. I think he's good, but not great. And I'll even add one more positive: post overrated thread, his impact has been much greater.

This reminds me of a conversation I heard between two NBA analytics guys on whether box score informed all-in-ones are as valuable these days because teams are more or less "choosing" who gets stats. In other words, are Luka’s box score stats entirely representative of his true talent or are they inflated, or at least somewhat misleading, because of his role on the team where every rebound, pass, and shot attempt flows through him.

That's definitely a part of what I'm getting at. Like when Russell Westbrook became the primary defensive rebounder for OKC in his triple double run. Those weren't incremental rebounds, they were strategic. With Luka, it's the same.

For his position, his raw rebounding numbers are off the charts. But — and it's surprising this rarely gets discussed — his team is better at rebounding⁶ with him off the court than with him on. And to your point as well, rebounding is just one example. Assists and scoring get inflated when one person has the ball the whole possession.

But I'll end this with some praise: Luka's efficiency, which was not particularly impressive to start his career, has been solid the last two seasons. So at least some of that high volume is now creating real positive impacts, even if he's still taking way too many unforced tough twos

Can we talk about yourself? All I know about you is that you’re an ex-analytics staffer for an NBA team. As a former analytics staffer for an NBA team myself I can understand the logic to remain anonymous. It’s easier not to burn bridges with any future potential employers when they don’t know it was you that was tweeting "this team is ass" or whatever. Can you say anything about why you left your job with a team and what you do for work now?

Fair question, and yeah, you're spot on with the reasoning for anonymity. I'll say I'm still in the analytics space and I loved my NBA job, but it was the right place, wrong time for me.

I'll also add: I absolutely love basketball stats. Maybe to an unhealthy, obsessive level. And honestly, this twitter account started as just an outlet for me to share the random stats and nuggets that I was stumbling into in my free time anyway.

There's a part of me that wants to work in the NBA again, but I get a lot of gratification from "publishing" my findings and work here. This account has given me the ability to interact with so many people in the NBA analytics space whom I revere, and even this interview is a testament to how engaging this account has been. Speaking of which, thanks for the interview. Great questions and better follow ups

Last thing: what's one thing you can't live without during the NBA season? Something you look at every day.

I look at Tari Eason's box score on ESPN and his basketball-reference page an unhealthy amount. Last year was really disappointing since he only played 22 games, but I'm back on Tari watch this season

1. You should be mildly skeptical of any early season analysis that tries to compare averages from a season in progress to completed seasons. With so few games played you should expect to see more noise and outliers compared to a full season’s worth of data. That’s why were just looking at the first 67 games played in each season to get a more fair apples-to-apples comparison.

2. I did my best to recreate the shot zones from pbpstats.com using raw play-by-play data in order to look at the first 67 games of each season, but it’s possible my numbers differ slightly than the aggregated totals on pbpstats.com. Here’s what the chart looks like when we use the aggregated full season data found on pbpstats.com. It tells a similar story. More threes from Above the Break and fewer long 2s.

3. Restricted Area - Shots within the restricted area (less than 4 feet).

4. Short Midrange - Shots from 4 feet to 14 feet

5. Long Midrange - Shots from 14 feet to the three-point line

6. based on his six-year career on/off ORB% + DRB% numbers via pbpstats.com