APBRmetrics

What hasn't been done in public basketball analysis (or team analysis)?

Could be several or lots of things.

What are your ideas to do anew or pull out of obscurity?



Here's one fairly simple thing that came to mind just now:

Star matchup +/-.

Could be done for single games or most importantly for a playoff series.

Using play by play data find, by query or manual tabulation, what team +/- was when 2 opposing stars were on court together, just one or none.

Could be general direct matchups, absolutely only plays with direct matchups or just simply on the court leadership matchup without much direct guarding.

Could be last season, this season or historically prominent matchups. Star matchup data might show who won / when in a more detailed way than conventional narratives, highlight reels or boxscore analysis. Was it really decided mano v. mano or not so much? It might bring more focus to #2s, benches and the combination.

I'll probably try this later.


To make it even more detailed could go to 2 star × 2 star matrixes of on & off or 3 ×3.


(Could also show full team vs team matrix for all players and combinations but moving from a linear time progression to a summary collapsing stints where matchups or say 4×4 same or 4 of 5 × 4 of 5 reoccur, but that takes it to a different though also new / useful focus.)

I know that 15 team lineups per game is pretty much the minimum but that 30 will occur; but I don't know how many unique (or close to unique as mentioned above) on each lineup matchups occur on average. That shown be known and considered. Does ANYONE know? My quick guess is that it might be known by some quant gamblers and books or could be found fairly quickly is asked and interested. Maybe some teams. Probably not many. Maybe a RAPM builder if they any time perusing and thinking about the stint data outside model building and data cleaning tasks.

My quick guess on this would be 40-60. Highly factured. It could go to near 100. What is the right target? Depends on team and data analysis. Or just slop it out there without knowing and giving clear data based guidance. Do football or other sport analysts know and use? With any enhancements or tricks?

If you really knew mini-game lineup match data better than your opponent, maybe, maybe, you might play a better informed, intelligent version of the dink game. But better than an informed, intelligent version of the lineup concentration game? Now that would be a Sloan Conference topic worth hearing insiders actually discuss... on slight chance they would and on chance that they have evidence based findings.

The right visualization(s) could show that a game in 40-100 mini-games. While that is not unknown, it maybe not have been seen fully or fully absorbed and used.


So something uncommon or new to do and another something uncommon or new to know and then do stuff with.


Average stint length per game or per season can be found fairly easily. It is quite short. Average lineup v. lineup stint length? Even shorter. Might be worth knowing, acting on. If you add time to what you are finding and using or to some degree substitute this for something of lesser importance.



So that leads to another question... What should get less time & attention in basketball analysis or basketball operations? I don't know if there is anything that deserves significantly less time or none at all but IF analytic time is limited, choices are being made explicitly or not.

Is fancier or super fancy data pipeline engineering worth its budget compared to alternative actions?

Visualization production and principal meeting time paying off?

Rigorous statistical confidence measures worth doing?

What other possibilities for review?


Always more to consider. Fwiw.


This post reminds me somewhat of the first analytics post at the original bulletin board.


Anyone with deep exposure to 2 or more sports care to rank basketball analysis on relative sophistication, thoroughness, collaboration or effectiveness or anything else?

on the public side theres probably stuff you can do with covariate RAPM id think, Someone I know tried some stuff I might try that stuff too.

Obviously on the WNBA side pretty much nothing has been done, but applying the stuff from the NBA to the WNBA shouldnt be difficult aside from obvious 40 vs 80 game sample issues for RAPM type stuff.

What types of covariate RAPM have you identified as potential targets?

Another modest idea would be producing splits of team offensive efficiency by top 2-3, then X to 6 or 8 and then the rest. Which groups are you more or less efficient with? The individual data is there but an intermediate group could be worth seeing for adjusting usage splits or play calls, lineups or other.

Thunder, Mavs, Nuggets and Timberwolves played a decent amount against each other in regular season and playoffs.

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

Best and worst matchups overall and by select, potentially determining factors.

At surface level, the most standout factors are:

Thunder: stronger on opp TO, very weak on DR.

Mavs: clearly weakest defg% (for season), lowest foul rate.

Nuggets: Great OR, very weak own FT rate, very low opp TO.

TWolves: weak own TO, strong FT rate, strong opp TO rate, strong DR rate.

Would seem by overall data that Thunder could really hurt TWolves on TWolves turnovers. Nuggets could exploit Thunder on Nuggets offensive glass (but with Hartenstein does it change?). Mavs might shrink Nuggets ft rate to even worse levels. Nuggets struggle to produce turnovers, except maybe against TWolves. TWolves clean up defensive boards except maybe against Nuggets.

How useful is overall data in matchups? I don't know. Find out how to weight overall with specific matchups.

May be limits to "knowing with confidence" but see what the data shows.

In the few of these factor matchups actually tested in playoffs, Nuggets struggled to produce turnovers even against TWolves. TWolves cleaned up defensive boards fine against Nuggets.

So past isn't always future but something to consider.

Thought a few minutes ago about player 3pt fg% differential (offense - defense). Never seen it. Will check into it.


Team differential is easy enough to find but never mentioned either.

J Battle, +20

Dort +5
Caruso -5
C Wallace -3
A Mitchell +12
J Williams +6
SGA -2.5
Wiggins +6...

Frequency matters but to the extent you are trading back n forth, the differential is significant.

Net points by factor coming to ESPN on January.

Not as interested in game level data as weekly, monthly or trend but I guess they could be compiled if not provided.

Conceivably, with AI and / or a dedicated researcher, a list of every stat measurement and ratio every published or contemplated could be compiled. And reviewed for gaps in current practice and imprecise division of focus. And ideas for new ones.

Something that hasn't been done in public basketball analytics to my knowledge and maybe not at all would be looking comprehensively at year to year changes in player metrics or all discrete stats for all players under a coach and addressing changes of coaches and minutes weighting the net changes (and probably age adjusting too) to give an overall player refinement / decline experience / possible coaching impact or at least "responsibility" score.

I'd most want an overall evaluation; but if you did the work, you could break it down by position / role, age, drafted and not under that coach...



Edit: RAPM for coaches works with all the data for 1 year or multiple years? I don't recall. With boxscore priors or not?

It is on the topic not quite the same as this statistical approach.

Not expecting answers, much less comprehensive, but how many team "analysts" have been let go / not renewed in large part because some or many of their analytic findings or opinions based on those findings greatly upset a player, coach, middle manager, executive or owner? Upset with or without a dispute of "the facts"?

I can immediately think of a few possible cases and will think on it more.

An analysis could compile major team flaws and then consider if it was reasonable that "analytics" offered or should have offered 1) understanding and 2) reasonably likely improvement with change in team behavior. Who's "fault" is it? "Responsibility" is on Head Coach and Executive above all on all cases but how much was avoidable?

How many major flaws can be guessed at as starting with "bad" or "not good" analytics?

Is every major or mid-level decision reviewed after the fact with "blame" divided out? Transparently or just in the back of the executive's mind?

Analyze the analysis, the analysts and everybody else for everything they did or did not do, including using the presented analytics, at all or well or not.


Can you be a "good" analyst if the analysis is not used or well for good results?

Will this ever be a Sloan Conference panel or a major article or a podcast topic from those with inside experience?

Do teams, on average, allow / encourage / protect presenters of "contrary" information and / or opinion? Shouldn't they? (within bounds of acceptable preparation or without limit of such judgment?)

One thing little discussed, if at all is the value of analytic staff. This paper is an attempt.


https://phys.org/news/2025-03-basketbal ... y-nba.html


"They found that for every four-fifths of one data analyst, a team gains one additional win in a season. Interestingly, a team can also gain one additional win by increasing its roster salary by $9.6 million."

Full paper

https://journals.sagepub.com/doi/full/1 ... 5251328264

List of suggestions for a new draft site includes some perhaps "new ideas" or things maybe not done often by teams or fully.

https://x.com/bballstrategy/status/1905024610304917837

Could use international translation model beyond league to position / role and perhaps age and size.

Crow wrote: ↑Tue Mar 25, 2025 11:03 pm ... for every four-fifths of one data analyst, a team gains one additional win in a season. Interestingly, a team can also gain one additional win by increasing its roster salary by $9.6 million."

So you could work just 4 days a week and command 6-8 mill and it would be win-win?
Awesome.