The Promise and Peril of Second-Tier Shot Creation
An RAPM-assisted investigation of good-but-not-great creators
In NBA draft circles and wider NBA discourse, there exists one pervasive idea at the heart of player evaluation: that on-ball shot creation is the league’s most important ability. This idea is often employed to justify the star status of players like Julius Randle, or going further back, Monta Ellis — players who were often suboptimal fits on playoff-level teams but retained the allure of guys who can go get you a bucket.
Now, I don’t even disagree with this idea that creation is the NBA's premier ability. At the top, that is. Elite creation is certainly the separator between superstars and everyone else. The last 20 years of NBA basketball have seen four players crowned Finals MVP multiple times, and it’s no accident that all are historically great creators: LeBron, Durant, Kawhi, Kobe.
And yet, my contention is this: below the tier of the league’s elite creators, this ability to self-create is often overemphasized. Now, I maintain that good-but-not-great creation is still valuable; yet if it comes without complementary abilities, namely defense and off-ball play on offense, its impact on winning is often more muted than one might expect. “There’s only one ball” may be a platitude. It’s also a reality. From a leaguewide perspective, there’s only 30 balls. And if you, like me, define the bare minimum level of success as making the playoffs, then we’re left with 16 balls on playoff teams. So, unless you’re good enough as a creator to be the primary equity holder of one of those 16 balls, you had better find ways to contribute off-ball and on defense.
Because definitions are essential in a discussion such as this, the term creator must be defined precisely. Here I’m talking about ball-in-hand shot creation, both for oneself and teammates, with an emphasis on half-court possessions, as those situations require advantage creation against a set defense. Specifically, I’ll be looking at the play-type data for isolations, pick and roll ball handler possessions, and post ups, as these are the most self-created play types. It’s not a perfect heuristic, but it will suffice. Furthermore, the term elite creators will be defined as the top tier of primary on-ball creators, a group of around 15 players good enough to serve as above-average number one options. If a player isn’t in the group, they probably fit better on a playoff-level team in a secondary or tertiary offensive role, at which point their off-ball and defensive abilities become increasingly important.
The Second-Tier Creators are a group of high-frequency on-ball creators who sit at an unfortunate intersection: not quite good enough to be that elite, top-15 creator who can rightfully command the status of number one option on a good NBA offense, but also lacking in the complementary abilities like off-ball gravity/efficiency and defensive impact that often define a playoff-level number-two or number-three option. It may be slightly contentious, but today I’d like to dive deeper into the conundrum posed by this archetype of player.
To bolster the data of this exercise, I’ll be looking at the past three seasons of regular season data. A sample of three seasons is necessary because a key part of the quantitative analysis will be based on RAPM. RAPM is great for this application because it’s unswayed by counting stats like points and assists. It solely indexes on how a given player impacts the scoreboard while on the court. If you’re unfamiliar with RAPM, Thinking Basketball has a great primer on it. The only drawback of RAPM is that it needs a ton of data for the noise to be (somewhat) tamed – like a few years' worth of data.
The quantitative analysis will be limited to the regular season due to the smallness and schedule strength variability of playoff data. For instance, Bulls players like Lavine, DeRozan, etc. have only played one playoff series over this three-year span. And that series also came against the elite defense of the Bucks. As much as I’d love to extend the data into the postseason, there’s too many situations like this that make it unfeasible.
In an attempt to quantify the value of a player’s creation, I’ve developed a rather simple metric called Creation Value, which operates on a per-100-possessions basis. This metric aims to estimate the value of a player’s creation by combining self-creation volume with self-creation efficiency. As mentioned above, I’ll be examining the play-type data for isolations, PnR ball handler possessions, and post ups for this metric, which I’ll refer to as “creation” play types.
At a high level, this metric has two terms, or components: a scoring term and a playmaking term. The scoring term estimates the value of self-created scoring usage. It takes the player’s scoring efficiency in creation relative to league average and multiplies that by their number of creation plays per 100. Think of this as scoring efficiency times scoring volume in creation play types.
The playmaking term is a bit rougher of an estimate, but it attempts to measure the value of self-created playmaking usage. It takes the player’s potential assists per 100 multiplied by the proportion of potential assists from creation play types multiplied by the estimated value of a potential assist. I estimate the proportion of a player’s potential assists that arise from self-created play types by equating it to the player’s proportion of usage coming from creation play types. So, if a player’s usage is 50% self-created, we’ll assume 50% of their potential assists to be self-created. This isn’t perfect, and will surely miss high or low on some players, but it’s about the best that can be done when working solely with the publicly-available stats on nba.com.1 If someone wants to hook me up with access to Second Spectrum, I won’t say no.
Here’s the formula in its full glory (ppp = points per play):
Creation Value per 100 = (ppp in creation – 0.97) * (creation plays / 100) + (creation frequency) * (pot ast / 100) * (0.16)
The 0.97 value results from the fact that over the past three seasons, the average half-court play has returned 0.97 points, per Cleaning the Glass. I toyed with using 0.92 ppp, which is around the average efficiency for creation play types, but it hardly changed the results that will be explored later, so I decided to stick with the average efficiency for all half-court plays. Ultimately, I only wanted to reward players for taking on high self-creation usage if it came on efficiency that boosted the team’s overall half-court efficiency relative to league average, and the 0.97 value was better aligned with this ideal.
The 0.16 value arises from the estimate that a potential assist adds, on average, 0.16 points of value to a shot, per studies from NBA Game Charting: the value of a 'Potential Assist' from 82games.com and Data reveals the value of an assist in basketball (theconversation.com). The precise value of an assist or a potential assist is a discussion that could keep philosophers debating for years (538’s RAPTOR methodology has an interesting section on this topic), but for our purposes, apportioning 0.16 points of value to the potential assister will do. Also, a quick note on turnovers: they are included in the “ppp in creation” value, since nba.com considers a turnover a scoring possession in their play type data.
In the context of this article, Creation Value serves the purpose of quantitatively separating those top 15 or so “elite creators” from the second tier of good-but-not-great creators.
Here’s the top 15 in Creation Value per 100 over the last three regular seasons2:
I’m not a huge fan of the “smell test” for judging the validity of a metric, but for those that are, this seems to pass the smell test. Moreover, the idea here is to look at the overall shape of the Creation Value curve, as we will do shortly. Some players may be too low, other may be too high, but overall those errors on specific players will roughly cancel and the shape of the curve is therefore largely unaffected.
It’s important to make the distinction that this is a measure of value, not ability. This is an intentional decision. Again, the NBA is a league with only 30 balls to go around. A player may have plenty of on-ball skill, but if he’s not quite good enough to demand the rock from even more skilled teammates, where’s the value in that? There’s a reason Cam Thomas, for all his wizardry with the ball, struggles to get NBA minutes on any half-decent team. Even players with as much creation ability as Devin Booker often find themselves in situations where much of their scoring value is realized off-ball. This ability to retain value by shifting off-ball next to other creators is often key to fitting into the offensive environment on a high-level team.
Here’s the Creation Value “utility curve” for the top 100 players in Creation Value over the past three regular seasons:
This curve supports the idea that the elite creators are providing enormous value relative to everyone else. The top 15 all deliver over 2.1 Creation Value per 100, and there’s a significant drop off after that – by the time the curve reaches the 25th most valuable creator, the Creation Value per 100 has dropped to around 1. The curve starts to level off here, around rank number 25. The 30th most valuable creator is only providing slightly more Creation Value than the 70th. The entire range from about 30 to the very end of the curve at 100 could roughly be grouped into a single tier of value. This lends credence to the idea that second-tier creation is not highly scarce, and consequently not highly valuable.
Based on this, one might question the impact of these good-but-not-quite-great creators, especially if they’re not supplementing that on-ball creation with efficient off-ball play and positive defense. As mentioned earlier, I call this group of players the Second-Tier Creators. To begin an investigation of the impact of this group quantitatively, I’ve first defined the group with five statistical cutoffs:
Greater than 1500 true shot attempts in creation play types over the past three regular seasons.
Less than 2 Creation Value per 100. This cutoff removes those top-15 elite creators.
Greater than 30% total frequency for creation play types, i.e. isolation frequency + PnR ball handler frequency + post up frequency > 30%. This criterion identifies players who have the creator archetype, as 30% total frequency is roughly the 70th percentile for creation frequency. So for the average 10-man rotation, we’re selecting the three or so players with the highest creation frequency.
Less than 62 TS% on non-creation play types. 62 TS% lies between the 70th and 75th percentiles for efficiency on non-creation plays. So with this cutoff, we’re identifying players who don’t have high-level efficiency off-ball.
Less than +1 Box Defensive RAPTOR. A value of +1 is between the 70th and 75th percentiles in D-RAPTOR. The purpose here is to remove players with significantly positive defensive impact. 3
For our purposes, I think the cutoff of +1 Box D-RAPTOR worked quite well, in that it only filtered out seven names that would have otherwise made the final group: Jrue Holiday, Paul George, Andrew Wiggins, Marcus Smart, Derrick White, Fred VanVleet, and Dillon Brooks. All strong defenders. The remaining 39 names in the final group are almost all a clear tier or two below as defenders (with only a couple possible exceptions, like Dejounte Murray and Kyle Lowry), and to that end this filter was successful.
I’m not including players who were rookies at any point in the past three years in this grouping, as rookies are often spoon-fed minutes and on-ball reps for developmental purposes. Moreover, rookies who are afforded numerous possessions to experiment with creation play types are often on unserious, tanky teams. For these reasons, I think excluding rookies from this exercise will improve the quality of the data.
Once these filters are applied over the past three regular seasons, 39 names emerge to form the Second-Tier Creators. Some names will undoubtedly be contentious, but again, the objective here is to look at the overall characteristics and impact of the group. Adding or subtracting a few names will make little difference to the average plus-minus impact, salary, etc. of this fairly large set of players. The idea here is to key in on the average impact/salary for this archetype as a whole. So for my audience of three valued subscribers, I plead that we focus primarily on the group-level results and only secondarily on specific names.
With that out of the way, here’s the group of Second-Tier Creators4:
The contrast between the LA-RAPM ranks and 2022-23 salary ranks are telling (rookie scale contracts are shaded blue). Of the 35 players in this group not on rookie-scale contracts, 30 (86%) are “overpaid” relative to their LA-RAPM rank5. Furthermore, the average LA-RAPM of the group is only 0.20. There are some issues with translating this number into a single-season ranking, but by comparing it with single-season estimates of RAPM like EPM, LEBRON, BPM, and RAPTOR, this slightly positive impact would rank roughly around 125th in a typical season. However, the group’s $20.5 million average 2022-23 salary for non-rookie-scale members would rank 65th in the league. To add to this, the 35 Second-Tier Creators on non-rookie contracts were ranked, on average, 237 spots higher in salary than in LA-RAPM, compared to a 115-spot difference for all other members of the 263 players with 1000+ minutes in 2022-23. There does seem to be a disconnect here between the market value of these Second-Tier Creators and their plus-minus impact.6
The disparity between ESPN’s NBArank and the LA-RAPM numbers is similar.7 Of the 20 players in the Second-Tier Creators who received a ranking in 2022 NBArank, 17 (85%) were “overrated” relative to their LA-RAPM rank. 12 of these 17 were over 100 spots lower in LA-RAPM than NBArank. Also, the 20 Second-Tier Creators in the top 100 of NBArank were “overrated” by 182 spots relative to LA-RAPM, on average, compared to just 113 spots for the other 80 members of the NBArank top 100.
To examine some concrete instances of this archetype in action, let’s take a look at Brandon Ingram. He’s been in the All-Star conversation for the past few years in New Orleans, and I’d say his offensive value is toward the high end of Second-Tier Creators. And yet, there are issues in his game that lead to him being far from a top-15 primary offensive option and also suboptimal as an off-ball player. His offensive game revolves around his ability to self-create, especially in isolation and as a PnR ball handler. Unfortunately, he’s somewhat inefficient in creation play types (52 TS% over the past three seasons) and lacks the rim pressure to consistently collapse defenses and pry open the juiciest looks for teammates.
This play against the Mavs demonstrates some of the limitations of leaning on Ingram as a primary creator. The lob to Hayes is there – if he probes one dribble further into the paint, forcing Powell to commit and leaving Hayes all alone with no low man rotation. But, Ingram doesn’t have quite that elite level of craft in the PnR, and the opportunity vanishes.
Just one possession earlier, the other side of this dilemma presents itself: Ingram’s lack of impact off the ball. In this play, he idles in the midrange on the weak side – not spotting up for three, not cutting, not engaging in anything productive. This is a pattern for Ingram – over the past three seasons, he’s been allergic to cutting, and only moderately engaged as an off-ball shooter from 3. Only around 2% of his total offense has come from cuts over the past three seasons, or about 0.5 plays per game. He’s also a good shooter, but oddly reluctant to really lean into this skill as a catch-and-shoot player. Over the past three seasons, he’s taken slightly over three catch-and-shoot 3’s per 36 – an underwhelming amount for a perimeter player, especially given that Ingram has converted these at a high percentage.
Ingram also doesn’t have the defensive impact that could make his fit as a number two or number three more palatable. He’s not exactly a lockdown defender on-ball, and he’s not moving the needle off-ball – decent with his rotations, but lacking in activity with low steal and block rates and only slight secondary rim protection.
These issues are compounded by the fact that Ingram is paid an amount commensurate with superstar production. He’s owed around $35 million for each of the next two seasons, and is eligible for an extension this offseason at a significantly richer number – about $50 million annually over three years starting in 2025. This is not an atypical salary commitment for players in this archetype. And yet in a salary-capped league, building a contending-level roster with Brandon Ingram taking up around a third of the cap is an immensely difficult undertaking. At that salary number, either he’s your number one option, and you’re probably running into a hard ceiling of something like 45 wins and a first-round exit; or, he’s a suboptimal fit next to a better creator who’s making at least as much in salary, and the small remaining room under the cap and tax necessitates bargain hunting for three likely below-average starters and minimum contracts to fill the bench. It’s difficult to see this construction leading much further than the former scenario. These complications lead to the Second-Tier creator being much more of a floor raiser than a ceiling raiser.
While I won’t pretend that this analysis is conclusive, I believe it presents sufficient evidence to invite a reconsideration of the value of the Second-Tier Creator archetype. This class of player often raises as many dilemmas as solutions in the construction of rosters and lineups. If, by some cosmic mistake, I were a GM tasked with building a roster around one of these players, these dilemmas would keep me up at night. But luckily for me, and you, and everyone else, I am only building spreadsheets.
As an aside, I did some hand tracking on five players of varying creation play type frequencies to check the accuracy of this estimate for the self-created proportion of potential assists. For each player, I sampled 30 assists from this past season and compared the proportion of assists out of creation play types against their creation frequency. Here are the results:
Trae Young: 18/30 (60%) in creation play types, vs 60.1% estimated
Ingram 14/30 (46.7%) in creation play types, vs 45.7% estimated
FVV 14/30 (46.7%) in creation play types, vs 47.7% estimated
Luka 21/30 (70%) in creation play types, vs 69.6% estimated
Derrick White 10/30 (33.3%) in creation play types, vs 26.8% estimated
Checks out. It appears this estimate doesn’t miss catastrophically on any individual players, and also doesn’t have a systemic upward or downward bias that would affect the quality of the results as a whole.
Apologies to SGA for, um, truncating his name. Be warned, I might truncate a few more names in the next table.
I know D-RAPTOR gets a bad rap for how highly it rates Jokic’s defense, but his case seems to be quite anomalous (if you’re into the aforementioned “smell test,” just check the top 20 in D-RAPTOR from this past season). Also, D-RAPTOR satisfies my requirement that it’s based solely on box-score and tracking data, with no plus-minus component. This is important, because I will be evaluating the impact of this group using plus-minus data (RAPM, specifically), so it would be nonsensical to also define the group using any plus-minus stats. The entire idea of this exercise is to define this group of Second-Tier Creators using strictly box/tracking stats, and then to quantify the impact of that group using plus-minus stats.
Despite my pleas that we not focus too much on specific names in this group, I know a handful of inclusions here may be surprising. Please keep in mind that this category is based on averages over the past three regular seasons. It’s not intended to describe what these players are right now, or what they may be going forward. But for thoroughness, a few words on a few names likely to be the most confounding:
Ja Morant: Again, again, this exercise is about average performance over the past three years. And it may be forgotten, but Ja was hurt/ineffective for much of the 2020-21 season after an early-season ankle injury. Those games comprise over a third of Ja’s sample. Furthermore, for all his jaw-dropping talent, Ja has been surprisingly inefficient as a scorer in creation play types, at only 51 TS% over these three seasons. Subjectively, I would still have Ja as a top-15 creator going forward. Statistically, he wasn’t quite there over this sample.
Dejounte Murray: Murray lowkey isn’t the defender he’s often made out to be. His on-ball defense is actually kinda mid, and his steal rate has dropped for his All-Defense days. There’s really no statistical argument that the Dejounte Murray of the last three seasons is the All-Defense-level defender that would exclude him from a Second-Tier Creator nomination.
Khris Middleton: Again, again, again, this is all based on the last three regular seasons. Unfortunately, Middleton was recovering from an injury and uncharacteristically mediocre for a third of those seasons (this past season). If we’re just talking about 2021 Middleton, he’s obviously good enough as an efficient number two option to be excluded from this group.
Bradley Beal: I think everyone realizes by now that Bradley Beal as a number one option is a recipe for 40-ish wins. If he could just shoot the ball with the precision he did in his earlier years (40% from 3 his first five years vs 34% from 3 the last three years), I would trust Beal more as a number two on a high-level team. But even with the weird cold shooting of the past three seasons, Beal just barely made the cutoff for inefficiency on non-self-created shots at 61.5 TS% (62 TS% was the cutoff).
Jamal Murray: Do I need to repeat that this is about averages over the past three regular seasons? No? Good. Current Murray is way too good for this group, as his off-ball efficiency is elite. Past-Three-Seasons Murray was mostly injured/recovering from injury, and consequently not nearly as efficient as one would expect him to be going forward (59 TS% on non-self-created looks this past regular season, significantly down from his 65 TS% mark in his last healthy regular season of 2020-21).
Brandon Ingram: Discussed at-length in the article.
Pascal Siakam: Subjectively, I probably wouldn’t have Siakam in this category. I think the 2019 title run showed he can ramp up the off-ball scoring and defensive impact when he’s in a secondary role. But, the stats are what they are, and he falls into this group. And again, the specific names aren’t really the point of the exercise. It’s about the averaged impact/salary of players with these averaged statistical characteristics.
Julius Randle: Ask any Knicks fan: Who would you rather run the offense through, Randle or Brunson? And if he’s not on-ball, Randle’s 59 TS% on non-self-created looks (about 42nd percentile for qualifying players) over the past three years leaves much to be desired. I know he’s All-NBA. To quote Seth Partnow, “yay points.”
De’Aaron Fox: I know I don’t have to tell you about this being an average over the past three years. Fox was way too good as a creator this past season (59 TS% in creation play types), and even off-ball around Sabonis, to be in this group. But unfortunately, the prior two seasons he had meh efficiency as a creator (52 TS%) with irresponsible irreverence towards defense. For that, he lands here as a Second-Tier Creator. Going forward, I don’t expect he’ll be in this category.
I’m only citing 2022-23 salary for the market value because salary is a lagging indicator of market value, due to the multi-year contracts of most of these players. Since we’re looking at a three-year period, the third-year salary should be the best estimate of market value. Moreover, I’m not including the rookie-scale players in this conversation about market value, because rookie-scale contracts are unrelated to market value since they’re determined solely by draft position.
There are some quirks in comparing 3-year LA-RAPM with a single-year salary rank. Since a lot of fringe players cycle through the league over three years, and LA-RAPM regresses these low-minute players towards a value of 0, the middle third or so of LA-RAPM is crowded with a number of bottom 5% NBA players (apologies to Stanley Umude). Due to this (along with the inherent noisiness of the stat, which also results in a slight “overrating” of the players at the top end of the statistic, and a slight “underrating” of the players at the bottom end of the statistic), you would expect legitimate rotation-level NBA players to be somewhat under-ranked by 3-year LA-RAPM, even in a perfectly efficient market. Of the 263 players to record 1000 minutes in the 2022-23 season, 64% were “overpaid” by this measure of LA-RAPM rank vs salary rank. Still, 86% is significantly higher than 64%. So, there is still some evidence here that Second-Tier Creators are “overpaid” relative to the rest of the league when indexing by LA-RAPM.
NBArank gets a lot of flak, and not undeservedly, but it is a great measure of the perceived value of the league’s top players. It polls a group of “over 200 reporters, editors, producers and analysts,” so it draws from a wide range of members of the Mediadel. I’m using the 2022 NBArank results, since (a) they’re the most recent, and (b) they incorporate performance of the first two years of our 2020-23 period and project for the performance of the third year. Therefore, 2022 NBArank seems like the best way to capture the public’s perception of the performance of these players over the entirety of the three-year period.
This concept of the "Second-Tier Shot Creator” is fantastic, and something I've thought about as a Knicks fan whose FO is presumably looking to pursue a "second star" to pair with Jalen Brunson. In the past 3 seasons the Knicks greatest weakness has been their awful half court offense, which stems from the lack of quality shot creators on the team (it is not surprising that the 3 main players who took on this role made your list in Barrett, Burks, and Randle). Even in this most recent season where the Knicks somehow finished with the 3rd best offense rating in the league, this was more a reflection of a roster that immaculately synergized elite offense rebounding and played each game with the mindset of “if we chuck enough shots at the hoop something is gonna go in.” Or… “your team might get 70 decent shots, but we’re going to take 100 bad ones”
Either way, while this reverse glass-cannon style offense is a fun footnote in a league that spends most of its time copying each other’s strategies, I think the Knicks FO will be looking to add a bit more of a traditional scoring threat for this coming season. Which leads back into the point of your article - many Knicks fans are very leery of sending our younger assets away for a player that turns out to be one of these “Second-Tier Shot Creators,” particularly guys in your list like Brandon Ingram and Pascal Siakam who’ve been circling in trade conversations and with their 24ppg all-star pedigree may seem like a satisfying move for the FO.
Shelling out max or near max deals to these types of players seems to be the biggest roster-constructing landmine in the NBA - throwing money at a Tobias Harris or Bradley Beal can actively cripple championship aspirations or even simply a shot at 2nd round legitimacy. I wonder if instead, with FOs catching on we will see more teams “camping out” like the 2019 Raptors, 2022 76ers, or this past years Suns, choosing to bypass the dangerous land of unproven high usage players in the off-season for disgruntled Hall of Famers at the trade deadline.