APBRmetrics

Probably could have that potential value or more or far more. Organization inertia and opposition might limit. Would be willing to try. Don't think you get that estimated average level of value or "enough", fire me. The value depends on the use and the experience with what recommendations were tried. I wouldn't ask for $3.5-$10 million, upfront. But worth the offered salary? Say / do one or a few things right/ useful / implemented and the rest is team gravy.


The cited paper was simple but a conversation starter. Right or not on scale, it appears that marginal additional spending on players has less or far less value than average spending or apparent value of front office spending on analysts.


Ask experienced team analysts if they think their work was worth 1+ win.

I will: Experienced team analysts, do you think your work was worth 1+ win per season?

Not that I expect they or their bosses will answer, but that is a real question that has real answers even if not revealed by them here.


Do you doubt this level of value for team analysts Mike? On average or at max? Do others?


How many millions would it be worth to say no, via analytic work, to restaging LaVine-DeRozan? Use 800+ lineups vs. 200 or less? Picking a better top 6 lineups and increasing concentration significantly? Winning 45-48% of lineups used instead of 36-42% or for the minutes? Squeezing one or more wins out of greater selectivity? Playing Bol Bol less than 1000 minutes over last 2 years? Nobody giving Tsheibwe a real chance?

Not drafting Scoot Henderson? Not drafting Wiseman? Drafting Edey over Risacher, Castle or Saluan? Trae Young over Doncic? 57 teams passing on Jackson-Davis? The value someone saw in Jalen Williams or Jokic? Starting with a scout or analyst or a combo of under-paid, underappreciated team employees? The value of taking someone other than Cody Williams or Bouknight or Hood-Schifino and so many others? Not taking Terrance Ferguson or Killian Hayes?

Signing Micic? Paying Jeff Green $8 mil / yr at this stage of career? Not extending Beal or Kuzma at those prices? Not trading for Fox? Hayward? Ben Simmons? Not maxing Paul George at 34? The Markkanen contract and not taking any alternative before then? The Ayton deal? The Vassell deal. The Jalen Green extension. The rejected offer to the rights to Frank Kaminsky? Seeing the value in someone like Kornet before others do, the last time or when it comes up again? Not paying the price paid for D Murray twice? Not locking up Brunson for Mavs? The price paid for Mikal Bridges? Still like the Nembhard deal? Waiving Bitadze? Giving Gabe Vincent his deal? Caleb Martin? The J Murray extension? The Ingram extension? The Bruce Brown deal? The Duncan Robinson and Bertans deals? Finding and not finding Pritchard and Hauser? Getting Horford back?

Getting Steve Kerr to try one lineup in a key game? Not getting Mazulla to change one lineup choice in 2023 playoffs?


Taking 30 years to get to this level of 3pt attempts? Taking 30 years to cut down of mid-range this much?

Selling the Sonics for less than $300 million?? Selling the Warriors?

On and on opportunities to add value or avoid losing value by using analytics well and making recommendations. Millions upon many millions upon tens and many tens of millions.

I was on the "right side" imo on at least 3 dozen of these decisions (not all of the listed but most of them) at the time, almost all in print.

Crow, thanks for the heads up to this recently published article (https://journals.sagepub.com/doi/full/1 ... 5251328264) purporting to show non-trivial value of NBA analytics hires. I enjoyed it and have some thoughts...

First, I would be remiss not to note the unemphasized result, that NBA coaching experience is very small in effect and statistically insignifcant. As one who has long and lonelily espoused the view that NBA Coaches Don't Matter* ™ I found this latest piece of accumulating evidence to be heartwarming. (Also, that regression numero dos suggests that a year of coaching experience is approximately one twentieth the value of a stat guy is coincidentally and arbitrarily delightful, at least for those few who happily recall from the first years of the MIT Sports Analytics Conference, Mike Zarren's sternly-delivered but well-intended "20 Things" homily to the NBA analytics hopefuls. But I digress...)

On to the interpretation of the "Analysts" coefficient...

I'm not buying that an NBA Analyst on average (in most years between 2009 and 2022) brought approximately 1.2 wins...yet.

Were I to have been a journal referee on Wang, et al (2025), I would have asked that other variables be taken into account so as to better interpret the impact of analytics. Of particular relevance for the era in question are game pace and three point attempt rate.

Across these years, the pace of NBA play increased from 91.7 to 98.2 (per 100 possessions) and the 3PAr increased from 0.224 to 0.399. And for both the statistics, the coefficient in variation was cut in half, indicating a strong homogenization in the style of play.

In my view, the increase in 3PAr in this period is less a triumph of Analytics (I mean how embarrassing is it to claim that broadening the recognition that 3 is a bigger number than 2 is perhaps the primary achievement of a movement?) than a continuation of a way overly cautious trend of a backward-looking and intellectually insecure NBA establishment. If one wishes to assign credit for this to Analytics, OK, but it would still be interesting and important to know the independent effect of this one-off episode.

In a similar vein there is the question of of how to attribute the effects of the quickening game pace (and its possible interaction with the Analysts variable).

To my recollection, optimizing the pace of play has never been a hot topic of conversation in the analytics world (at least not in the early years of this forum) which to me was always sort of surprising.

This is not to say that this didn't become part of the special sauce of individual team analytics departments and, correspondingly, there might be some evidence of this in the data. But my sense is that, as a historical matter, this trend was largely and simply motivated by the perception of success of the Seven Seconds or Less Suns (a strategy that pretty clearly did not originate from a coherently thought out, i.e. analytics based, framework).

The conjecture is that the NBA establishment in its usual copycat fashion, decided to have the ball up pushed up the court faster than before, thereby reaping the "unforeseen" benefits of a longer effective shot clock. Is that Analytics? Anyway, so here too it would be interesting to see what the interaction of this one-off episode might have with the Analysts variable, where there is a strong argument for the relationship being correlation and not causation.

Sure, go further, look at more variables.

And I'd look at several time periods. Not sure what cut off dates off top of head, but early, middle and late analytics. Early "Analytics" might have done better against less frequent match-up.

Or maybe there is a critical mass for an analytic shop size. Look at measured impacts by 3 sizes.

Correlation vs causation is always a topic. Some might have more to say on how to address that.

As I said, the article is a modest early piece to a topic only lightly studied.

Other things that could be correlated could be:

size of analytics shop to net value obtained in the markets (player impact / salary and relative to prior year, and draftees value compared to average fir pick number)

size of analytics shop to lineup behavior & results (number & degree of concentration at the top)

Beyond size of analytics shop, could be some consideration of status of head of analytics within organization (by where / when mentioned on organization lists, pay level, GM public mentions or whatever can be found). Any head of analytics paid equal or above top assistant coach? Average assistant coach? Top or average ticket salesperson or middle or high level business manager?

Other hypothetical, won't happen, measure would be money spent on outside consultants for data and / or analysis.

Some positions might be totally or mainly "coaching analytics" or draft analytics. Doubt you could comprehensive get job descriptions or actual practice but having such a dedicated named position would be a heightened marker for work in those areas and would deserve close study.

Or course, impact depends on use, proper use. With use / use rate as unknowns, unless former analysts ever speak up in a way that gives general guidance.

Those with the tracking data could characterize the first pass in the half court into x bins and query the play outcomes. Are there first passes that on average lead to better results? The second or third?

Those with the pose data could characterize 10-30 rebound related moves and sequences. What moves are most correlated with getting the rebound? What specific moves get the most fouls called? What are the reward / risk ratios? What are the best counter moves? Is "hit first" a good strategy? How good?

With step back and sidestep 3s more common, how should standard closeout behavior change based on big data?

On average, playing how far off their man produces the best defensive rating for the matchup and the opponent overall? In general and by floor location and shot clock part? How close does player behavior match the optimal measures?

What does pose data say about optimal pumpfake offense and defense overall and by floor location and shot clock part?

How many teams have done serious research in these areas?

What is the distribution for exactly where / when / how posed do pick n roll ballhandlers commit to a course of action? Defenders? Reverse courses?

What off-ball, don't get the ball movements show some improvement or diminishment to team results? By floor location and shot clock part?

Who has the best pose data program and are teams actively and specifically looking for opportunity to hire away the most important people? What specific apritudes / skills are needed in new staff given the task?

Is awareness of pose data insights leading to more fakes or is awareness too low to make faking common or necessary?

Could try to model games, perhaps at factor level. Simulation?

yes. simulation. for 3 decades now.

I will: Experienced team analysts, do you think your work was worth 1+ win per season?

10-20 wins or more in a season. you recommend the right player, and that player is signed, stays healthy, and plays major minutes, makes all the difference in the world.

Not that I expect they or their bosses will answer, but that is a real question that has real answers even if not revealed by them here.

likely you would want specifics and you clearly know that can't be discussed publicly.

Can you be a "good" analyst if the analysis is not used

could a bench player be better than a starter but the starter play much more because he has a big contract?

I'm not buying that an NBA Analyst on average (in most years between 2009 and 2022) brought approximately 1.2 wins...yet.

** sniff **

optimizing the pace of play has never been a hot topic of conversation in the analytics world

because in various seasons there have been teams that have lead the league in offensive and/or defensive efficiency with the league's fastest game pace, and also the league's slowest game pace.

In my view, the increase in 3PAr in this period is less a triumph of Analytics... than a continuation of a way overly cautious trend of a backward-looking and intellectually insecure NBA establishment.

harsh.

generally leaguewide as 3pters increase FTAs decrease. this season saw the most 3s attempted and the fewest FTs attempted in many decades. drawing less fouls means not getting opponents into foul trouble and sending them to the bench early. more 3s is not always the answer.

a better question would be whether to emphasis (and pay for) defense more than offense, or emphasis offensive rebounding versus getting back on defense.

How many teams have done serious research in these areas?

your questions are a tad esoteric.

better ones would be does the team win more if a certain player shot 2% better on 2s, or committed 1 turnover less per game, or blocked shots at a 10% better rate.

or where exactly should a certain player look to improve - should a specific player look to force more turnovers, or grab more offensive rebounds, or try to draw more fouls. is it better for a player to try to attempt more 3s or draw more fouls on in close shots.

more tangible things that a coach can emphasize and that can then act on if he doesn't get the results he hopes for.

Thanks for the replies where given.

Questions could be better targeted with a greater awareness of the team management environment or questions direct from answer seekers. I tried some freelance brainstorming fwiw.



I suggested some study of team level factors as a guide for study (the basic strength / weakness matchups). Simple but some value over not checking the factor matchups. (Any conversations with Dean Oliver about his presumably team level factor model / Roboscout?)

I mentioned simulation, knowing you do that. I don't recall if I did so actively hoping to draw a response or just recognize the value of such work.

I have a vague memory of your simulation (based on some long ago exposure) as player based, though I didn't know exactly what was under the hood of that seemingly small free program. I also had exposure to a simulation from someone who sought out your advice and built a version that is XO hoops.

I imagine that lineup performance can be adjusted for the specific factor level synergies in that lineup without knowing exactly how deep that adjustment goes, how many interactions. I'll assume that your simulation makes synergy adjustments and that it has gotten more accurate over those 3 decades.

I don't expect you to reveal any specifics on your model but leave the door open for whatever you feel like saying, as it relates to the topic of analytic value. I also wonder if you have any comments / guesses on how common production and use of game simulations are with other teams, to the extent that gossip or shop talk occurs.

Any evidence of Ben Falk using the XOHoops model with NBA teams or any other team using it? Any comments on its behavior? If you know another used a simulation, did that change anything? Model their models and coaches?

If you did feel like brief comments on the model and its use, I'd wonder the extent to which you use it to recommend when lineups are used against which opponent lineups. Do you run every desired lineup against every matchup? Machine learn or manually create new lineups for specific opponent lineups or game situations? Do you program in a Coach's specific play call design or just go random or go based on past behavior? Do you test many play call designs? Model opponent coach play call design and in game adjustments based on results? Adjust player level factors based on specific player matchups or even team specific defensive coverages?

Any interaction with video game designers? Any value from such conversations? Any use of video based big data on motion and pose? Consideration of AI modeling of defense (and perhaps offense)?

Any further refinements to perhaps be made (to model ball movement to specifc locations, beyond just who passes and receives and possibly drive ball movement based on expected points by location...or other dimensions) or locked down at this point? Stand alone simulation project, fairly streamlined, or bolt on everything the organization knows and believes?

Anything on basketball simulations in academic research that has been useful to you? Links?

Can't hurt to think a little and ask.

I'd take comments in private and strictly confidential if you prefer that way.

Simulate drafts or free agency?

Opinions on lineup testing design and concentration?

Any other topic you are inclined to comment on, while you are here?

Directing player (and staff) attention on player improvement, with concern with / awareness of existing key lineup synergies and potential synergies?

Awareness of synergies make their way into contract analysis?

Any comments on RAPM factors, Strong / weak RAPM comparison, other versions of RAPM inquiry?

Any comments on ESPN Net Points?

Do players learn within a game and is player learning a part of a grand model?

How strong are the efforts to design usage distribution in your model and by coaches in real life?

Any comments or structural design activity related to traditinal positions or lineups that break with tradition? 2 PG lineups, 2 true bigs vs use of stretch big?


What other topics, that are not too esoteric, would you consider worth pursuing?

If some of my suggestions seem "a tad esoteric", I assume that means they have not been done by you or your team.


If anyone else is interested in some of these comments, you can comment on them.

Any conversations with Dean Oliver about his presumably team level factor model / Roboscout?

haven't spoken with deano in some time. last i heard he was with the wizards, but now it seems he is with espn.

I mentioned simulation, knowing you do that. I don't recall if I did so actively hoping to draw a response or just recognize the value of such work.

it has value.

I have a vague memory of your simulation (based on some long ago exposure) as player based

yes it is based solely on player statistics. simulations have to calibrate to team values (FG%s, off/def efficiency, etc) but it does not use team data.

though I didn't know exactly what was under the hood of that seemingly small free program

plays basketball player touch by player touch.

I imagine that lineup performance can be adjusted

yes i do lineup analyses all the time. what combinations of players play best together, with limitations based on player positions played.

and that it has gotten more accurate over those 3 decades.

absolutely. can simulate a season and get the overall league pts/poss scored/allowed well within 1/2 percent.

as it relates to the topic of analytic value.

it has alot of value, especially for what-if scenarios.

if you have any comments / guesses on how common production and use of game simulations are with other teams

unfortunately i cannot speak to that.

Any evidence of Ben Falk using the XOHoops model with NBA teams or any other team using it?

i believe he runs his own nba analytics website now, but the name escapes me.

Any comments on its behavior?

have not used it.

I'd wonder the extent to which you use it to recommend when lineups are used against which opponent lineups

do this all the time. can simulate any lineup against any team, give a range of winning percentages for all 5 man lineups.

Do you run every desired lineup against every matchup?

no, too many permutations.

Machine learn or manually create new lineups for specific opponent lineups or game situations?

both.

Do you program in a Coach's specific play call design or just go random or go based on past behavior? Do you test many play call designs? Model opponent coach play call design and in game adjustments based on results?

the simulation does not "run" plays. it simulates touch-by-touch play based on player touches/min, what they do per touch, and each player's individual attributes/statistics, on both offense and defense (def reb, steals, blocks, DFG% allowed). what happens player touch to player touch is determined by player statistics and random number generators.

Any further refinements to perhaps be made

software is never done. improvements are always being made (i'm sure this is the case with all software). the more new data that becomes available, the more you try to see if it helps.

Anything on basketball simulations in academic research that has been useful to you?

nothing i have seen. if you have post it here and i'll sure look at it.

Simulate drafts or free agency?

FA all the time.

Any other topic you are inclined to comment on, while you are here?

again it's great for what-if scenarios. is a team better/worse with another player (via trade/FA), if a player shoots better or worse, rebounds better/worse, commits fewer turnovers, etc.

Awareness of synergies make their way into contract analysis?

always predicting future player performance based on their past performance and the past performance of statistically similar players.

Any comments on RAPM factors, Strong / weak RAPM comparison, other versions of RAPM inquiry?

too noisy in a one year sampling.

Any comments on ESPN Net Points?

oh this is what deano has been up to. no i have not read up on this. but if you have a link for how he rates players for 2024-25 post it and i will take a look.

How strong are the efforts to design usage distribution in your model

player touches/min are a core factor in the simulation. how many touches they average and what they do per touch on offense.

Any comments or structural design activity related to traditinal positions or lineups that break with tradition? 2 PG lineups, 2 true bigs vs use of stretch big?

in the simulation you can simulate any players at any positions. but i wouldn't recommend playing players out of position if you want statistically accurate results. you simulate embiid at PG and he will be guarded by PGs, so not really realistic.

What other topics, that are not too esoteric, would you consider worth pursuing?

the simulation can answer alot of questions - why a team is not winning or is winning, what factors contribute the most, what factors have to be improved, etc.

but it does not look at specific playcalls, just factors that statistics by the numbers affect.

If some of my suggestions seem "a tad esoteric", I assume that means they have not been done by you or your team.

moreso that the simulation looks at a more basic level through the numbers. how much better or worse is a team if a player shoots better, rebounds better, commits less turnovers, plays better defense, i.e. is it better to lower an opponent's FG% or force more turnovers, which would be more effective, and by how much.

Thanks again for the replies.


Net Points ratings

https://espnanalytics.com/nba-net-pts



Wading into the academic literature:

First notable talked about lineup networks, core lineups, connectors and peripheral and rate of re-use. Fine structure but not much beyond that.

This one looks on "shooting style" using player tracking data:

https://arxiv.org/html/2403.13821v1

It will take time to digest. Starts with player clusters for shooting style and offensive role. Some consideration of impact of pairs. Finds:"Top 10 lineups show that in addition to one Big Man, there is typically one ball-handler and one shooter with the other two being all-rounders."

At individual level, the biggest effects are reported. from ballhandlers who mainly shoot 3s but may have all-around profiles. The least from those with a higher concentration on close and mid-range shots.

"lineups with high predicted values tend to consist mainly of about two Isolation Attackers, one or two Spot-up Shooters and Off-screen Shooters combined."

"Conversely, Table 15 shows that lineups with lower predicted values tend to consist mainly of one or two Stretch Big and Post-up Bigs combined and one Primary Ball-Handler."

Table 16 and 17 shows wide spread of estimated impacts of certain combinations.

"previous research indicated the effectiveness of having one strong scorer, our findings suggest that having two strong scorers can further enhance offensive performance. We also found that Spot-up Shooter tend to produce positive effects as well."

Another paywalled offered:

"...support for a predicted positive relationship between shared team experience and team performance that declines as shared experience grows, eventually turning negative..."

Positive relationship with experience expected, eventually turning negative was not. Growth of bad habits? Opponent analysis figuring out weaknesses? Something to think about.

https://www.researchgate.net/publicatio ... k_analysis

Only a preview and I probably can't get in.
Anybody a member of researchgate and have ability to reach article link?

A Sloan paper said:

"Most underperforming lineups have less than two Stretch Forwards and less
than one Ball Dominant Scorer. It is very effective to have at least one Ball Dominant Scorer on the
court and at least two Stretch Forwards.
What we see from our analysis is that most lineups have one high usage player, which means
someone must have the responsibilities of handling the ball and getting the lineup into their sets.
The high usage player will either fall into the category of Floor General, High Usage Guard or Ball
Dominant Scorer. If that player can play as close to a Ball Dominant Scorer as possible (high
efficiency, high assist rate), there is a greater chance of that lineup being effective. The model also
suggests that spacing the court with shooters, combined with a ball dominant player leads to more effective lineups."

"The [best] lineup consists of 1.25 Ball Dominant
Scorers, 2.25 Versatile Role Players 1 Traditional Center and 0.5 Stretch Forwards. The need to
have your high usage player be efficient is evident in our highest predicted lineup. If you switch out
the 1 Traditional Center and 1.5 of the Versatile Role Players to have 0.5 High Usages Guards, 1.25
Ball Dominant Scorers, 2 Stretch Forwards, 0.75 Versatile Role Players, 0.25 Three Point Shooting
Guards and 0.25 Skilled Forwards (lineup 2), the net rating prediction is extremely close to the
original, and still very successful."

Evan Miya has a paid lineup tool for college.

Zach Kram reports:
NBA Five-Man Lineups, 2017-23 by minutes
100+
67% w
33% l
+3.99

250+
80% w
20% l
+5.75

500+
90% w
10% l
+5.95

https://www.theringer.com/2023/04/13/nb ... ineup-data


At least a half a dozen other potentially promising articles are behind pay walls. They usually disappoint. High technique, often low notable findings or reported.

https://medium.com/@xulianrenzoku/model ... c695e95e1d Bigs and dominant ballhandlers seem to make the most difference. Top and bottom 10 lineup mixes given. Iso / pnr ballhandlers appear essential. Most common lineups given.

"Wayne Winston in his book Mathletics indicates that when it comes to a lineup’s +/- rating, the actual performance of the lineup over 48 minutes is normally distributed with a mean equal to the lineups +/- rating and a standard deviation of \dfrac{12}{\sqrt{Games played}} points. "

Bayesian adjustment:

https://412sportsanalytics.wordpress.co ... -approach/


Brian Skinner and network approach:
https://journals.plos.org/plosone/artic ... ne.0136393
touches and location, skills ised in a lineup / network


https://link.springer.com/article/10.10 ... 22-04653-z

high technique, probably worth reading but not replicable for most.

A download link:
https://www.google.com/url?sa=t&source= ... UjCh6OpH7-

Spectral analysis of group contributions at all levels of lineups. I like the approach.


Will want to re-read, think, synthesize takeaways.

I look an academic research every few years. It is remarkable how they are almost never discussed by those I have contact with. I usually do a post like this and move on. I should backtrack to other such posts and seek reminders or new insights. I would hope that presenters and attendees at academic research conferences are having a more robust dialogue about them. Wonder how much team contact they get.

Net Points ratings https://espnanalytics.com/nba-net-pts

have you actually looked at this at all? look at the 2024-25 data. it has an offensive and defensive rating for each player.

hard to take this methodology the least bit serious when it shows nikola jokic with a higher defensive rating (67.15) than both draymond green (55.64) and anthony davis (55.63), and karl-anthony towns with a higher defensive rating (54.1) than both isaiah stewart (51.31) and chet holmgren (47.8)...

yet you can go to the publicly available defensive FG% allowed data at stats.nba.com and see that from <10' of the basket these FG%s allowed:

chet holmgren 43.8% (-15.9 differential)
isaiah stewart 48.1% (-11.5%)
anthony davis 50.5% (-9.7)
draymond green 50.5 (-8.5)
nikola jokic 61.0% (+0.5%)
karl-anthony towns 61.5% (+2.0%)

these differences in shot defense FG% allowed data between jokic/KAT and the others are HUGE - i.e. are far higher/worse - compared to these other players.

jokic and KAT were clearly two of the worst big men shot defenders anywhere near the basket this season.

but Net Ratings says their defense is better than these other players? i don't think so.

so it's clear their methodology in no way properly evaluates individual player defense, and as such is of little use in player evaluation.

I've looked at it modestly so far. I've suggested that they do far more data reporting- player by factor and team for season. Not provided yet.

Jokic is an interesting case. Rebounding is likely playing an important role in that rating. Factor data would show what is happening with his defensive rating.

Okeedokee...

bchaikin wrote: ↑Mon Apr 21, 2025 2:28 am
I will: Experienced team analysts, do you think your work was worth 1+ win per season?

10-20 wins or more in a season. you recommend the right player, and that player is signed, stays healthy, and plays major minutes, makes all the difference in the world.

My first read was that this sentence implied that just acquiring "the right player" (as in a single player) could gain the acquiring team 10 or 20 wins. Is this the point that was intended to be made, and if so, can an example be given of one acquisition having had such an effect, or, failing that, perhaps one could be proposed?

bchaikin wrote: ↑Mon Apr 21, 2025 2:28 am
I'm not buying that an NBA Analyst on average (in most years between 2009 and 2022) brought approximately 1.2 wins...yet.

** sniff **

Nothing to sniff at.

bchaikin wrote: ↑Mon Apr 21, 2025 2:28 am
optimizing the pace of play has never been a hot topic of conversation in the analytics world

because in various seasons there have been teams that have lead the league in offensive and/or defensive efficiency with the league's fastest game pace, and also the league's slowest game pace.

Missing the point. No matter what one's team makeup or coach's predilictions, there are two facts that are true, the first as simple as 3 being a bigger number than two. Wasting time slow walking the ball up the court costs you valuable seconds on the shot clock, and the shot clock is never defeated. Second, again, no matter what one's team makeup or coach's predilictions, there is a theory of maximizing points per possession in the half court. This was essentially never a topic of conversation, at least in these here parts (and when so, in very limited capacity).

bchaikin wrote: ↑Mon Apr 21, 2025 2:28 am
In my view, the increase in 3PAr in this period is less a triumph of Analytics... than a continuation of a way overly cautious trend of a backward-looking and intellectually insecure NBA establishment.

harsh.

generally leaguewide as 3pters increase FTAs decrease. this season saw the most 3s attempted and the fewest FTs attempted in many decades. drawing less fouls means not getting opponents into foul trouble and sending them to the bench early. more 3s is not always the answer.

I wouldn't describe this remark as harsh (a much more severe indictment could easily have been offered for such professional malpractice). Whatever the rhetorical preference, it was more than fair.

As for your remark that more 3s is not always the answer, the correct reply is that, on average, over its entire history in the NBA (with the possible exception of the last few year, but that would imply some checking) more 3s has definitely been the answer. The fact of the matter is that on the margin, trading off 3s for crap twos, even accounting for foregone trips to the foul line (but then again, benefiting from fewer turnovers at the same time) has always been the correct "average" bet. And my sense is that this remains true on the margin for most if not all teams (what I'm too lazy to check).

bchaikin wrote: ↑Tue Apr 22, 2025 2:30 am Net Points ratings https://espnanalytics.com/nba-net-pts

have you actually looked at this at all? look at the 2024-25 data. it has an offensive and defensive rating for each player.

hard to take this methodology the least bit serious when it shows nikola jokic with a higher defensive rating (67.15) than both draymond green (55.64) and anthony davis (55.63), and karl-anthony towns with a higher defensive rating (54.1) than both isaiah stewart (51.31) and chet holmgren (47.8)...

yet you can go to the publicly available defensive FG% allowed data at stats.nba.com and see that from <10' of the basket these FG%s allowed:

chet holmgren 43.8% (-15.9 differential)
isaiah stewart 48.1% (-11.5%)
anthony davis 50.5% (-9.7)
draymond green 50.5 (-8.5)
nikola jokic 61.0% (+0.5%)
karl-anthony towns 61.5% (+2.0%)

these differences in shot defense FG% allowed data between jokic/KAT and the others are HUGE - i.e. are far higher/worse - compared to these other players.

jokic and KAT were clearly two of the worst big men shot defenders anywhere near the basket this season.

but Net Ratings says their defense is better than these other players? i don't think so.

so it's clear their methodology in no way properly evaluates individual player defense, and as such is of little use in player evaluation.

Abstracting from the fact that defense within 10 feet of the basket isn't the be all and end all of a defensive rating, you realize there's a box to tick putting the stat on a per 100 possession basis (and then you can set your minutes played threshold)?

https://espnanalytics.com/nba-net-pts100

Doing this, you can have your defensive priors happily confirmed. Time to apologize to Dean...

P.S. I came up with different FG% allowed within <10' from the basket for the first two players listed that I checked, Holmgren and Jokic, with the differential between the two being 10.9%age points (not the 16.4%age points you got). Not saying my math is correct, but perhaps it's worth a check if interested.

Not sure which public database has the best position assignment system but rule bases systems will only be approximations.

Video-based should be better.

Wonder how much time teams spend on evaluating players by position. Positions are somewhat of an abstraction overall with offensive movement and defensive switching but at possession level it is much of reality.

New lineup detail suggested in lragiuewide lineup analysis thread, execution of first example run in Thunder thread.

A group is working on predictive lifetime RAPM for college prospects. Some draft models are probably similar in intent and possibly on design to some degree. Efforts will of course should receive post implementation performance checking.