APBRmetrics

I'm sorry crow, but what you call analysis I call the banal and meaningless. Citing trivia regarding small samples of lineup data and absently posing but never answering questions is not the path to enlightenment.

You again, in that last comment, fail to understand that in the aggregate of expert opinions there is something approaching the truth.

There is an enormous irony in seeing the use of many metrics as a good thing, an ensemble, or aggregate you might call it, and then claiming that the aggregate opinion of people whose livelihoods rely on knowing basketball, people whose ability to analyse the game continues to improve just as ours does, is not useful because it is an aggregate.

And I have such a model, it is very edifying, it's a strong predictor of team performance, and is fascinating to use to decipher which front offices make unusual decisions relative to the aggregate.

I'm not here to share my models. I am here to discuss methods to attempt to peel back the layers of the onion. I recall now why this place has become such a crypt, and so devoid of useful analysis.

If your general conclusion is that my analysis of lineups and beyond is banal and meaningless, you are imo mistaken.

Right back at ya chap.

You mention an "idea" and say "maybe it is a good idea?" but you actually have it and have proven it? That is not easy to understand or do anything with.

If the place is a crypt it is mainly because very few people make any effort to share analysis or new work.

You mentioned something that could have more discussion and visibility and use, but only you stand in the way in saying more, however much more you want.


Looking at small samples of lineup data and calling them way too small and way too diffuse is important analysis not done often enough. Squeezing out ideas for improvement in lineup management is hard work because of how the so called experts manage it but some gains can be found by looking and working with what is available.

To the right person what I have said is clear. It is fine that to you it is not.

Picking at the English of my post is definitely not a good idea. Perhaps you have never come across a conversational style of writing, within which conclusions are known to the author but not the reader, and so the author chooses to write in a way which invites thought as opposed to dogmatic following.

Your reply, in fact, is an example of not inviting dogmatic following.

As has become clear, I am not for you, and you are not for me.

Perhaps reflect on why people choose not to share their work here any more.

I have laid out the clear and concise idea for a metric which I know adds value and insight, anybody curious, capable and bright will be able to implement the idea, if they want to. I include you in that.

I did not pick on your English. I honestly mentioned that I felt mislead by how much you had done with the idea. If I am mistaken, I am mistaken but it was a fair / reasonable reading of what you said imo.


Use "teammate-marginal player per100 box score stats to predict the teammate-marginal percentage of game time a player is allocated by a coach."

I will think on that. But I am not sure of its value for optimization. Value for prediction, expectation ok. It assumes coaches will use boxscore and not advanced metrics or non boxscore tracking data. Probably true and probably not optimal. How much they actually use boxscore vs. impressions / feelings / other things is unknown.

"Teammate-marginal player" is a phrase I don't know that I understand how you mean it. Help me understand better if you would. It is pretty key that this be clear and imo it is not.

I appreciate both of yall!

I think I agree with a lot of what vzero says, the first post here to me was kind of like weird approach interesting answer, a Non RAPM target for a prior I think does make sense to experiment with and I thought the idea of using expert opinion was interesting altho I wasn’t sure if it’s how I would do it solely because it depends on who you consider experts (didn’t see exactly who he was talking about so not tryna roast anyone lol just didn’t see)

I think RAPM itself and deritivies that base itself from it for AIO metrics can be interesting on an individual level to parse out some cool insights, even if other individual results may not be reality or the magnitude of some suprising results may be a bit overdoing it at times, I think it can be interesting that way on a case by case basis, and also it’s convenient

I think both have a place

"Lineup analysis does have some value to me, but I have not found a good use for it in AIO metric construction, which is the backbone (though perhaps this beast has two backs, 'a' backbone might be more appropriate) of regular season prediction."

If it doesn't have a current good / helpful use in AIO but has value, then it should be pursued by other means.

"Perhaps reflect on why people choose not to share their work here any more."

It can because people horde their work for private gain, have been hired by others, prefer other discussion formats / sites, don't like someone, don't really want critique or the work involved in extended discussion or don't find a compelling variety and quantity of response here. It is guess unless someone or many spell it out.

I have given more response than anyone and lots of good response. Don't have to agree with it all but it was given to my best understanding with the intent of being helpful to discussions.

Others will think and do and not say as they want.

"absently posing but never answering questions is not the path to enlightenment"

That quite a statement.

I ask questions / make suggestions, more than I can follow all of them in my time used, but I follow a number of them as far as I can in that time and to my ability without spending much more time. Plenty for others to take a share or take the lead on. I make my contribution and within its time I am satisfied it has value for those who read and consider it.

I don't fail to understand that in the aggregate of expert opinions there is something approaching the truth. I just question how close they are to it compare to how close they could be with even more disciplined use of analytics.


I didnt question that aggregate information and knowledge isn't present / important / valuable, just not sufficient or "inarguable".

"the aggregate opinion of people whose livelihoods rely on knowing basketball, people whose ability to analyse the game continues to improve just as ours does"

and needs to continue to do so.

I never said it "is not useful". It could be more and better. Especially with regard to lineup management. It is nowhere close to optimal on lineup management to my thinking. Which is what I express, take it or leave it or do better if you can.

In my years investigating these things, I've found essentially zero value in looking at point differential at the individual lineup level... The sources of noise and other variance swamp any signal. It takes multiple years of data, thousands of minutes, to start to get that noise under control... And at that point the long time period adds another source of noise that removes any value.

There can be some value in player pair or trio data, but you really have to control for player and opponents to have any chance at it being meaningful. J.E. did this at one point using a RAPM approach, and very few player combinations had enough signal to trust there were synergies that could be identified purely from point differential.

This is where coaches can recognize good and bad patterns of play and synergy or the reverse through their own pattern recognition and judgement far, far faster (multiple orders of magnitude faster) than it can be identified statistically/mathematically.

"And I have such a model, it is very edifying, it's a strong predictor of team performance."

You are most interested in predicting team performance. I am most interested in improving it.

We have different priorities so we may not overlap that much or that well in conversation.

But others can read and use what is made available according to their own interests and perspectives.

"I've found essentially zero value in looking at point differential at the individual lineup level... The sources of noise and other variance swamp any signal."

Don't disagree for the too few samples tested under 200 minutes or 400 or 600 minutes.

But you either use sublineup data and what lineup data exists as hard as you can for clues knowing the inadequacy, ... or you don't.

If you don't, you are calling on instincts, hunches. Lots of in the moment guessing. Lots of chasing the last play or hopes for the next play over what is in the overall data, such as it is.

More disciplined, concentrated lineup testing / use has been my core message.
Guessing on improvement using the data is my form of guessing under the circumstances relative to the existing coaching guesses. I could use instincts and hunches more, but I sometimes do a bit. Still it starts with the actual performance facts we have. Moreso than many team / coaching examples show.


I have frequently followed my lineup analysis / critique with fully specified lineup rotations for most (or all)minutes.
It is usually hard to track the relative performance of the recommended over the actual because almost all lineups are too lightly used to judge well. But I offer alternatives more promising than what was done. In general design and down to specifics.

I specifically disagreed with a Celtics lineup choice in playoffs 2 years ago in advance of I think game 3 and documented here how the results varied from that moment and how the difference decided the finals, whereas my recommendation appeared by actual performance data to have been enough to reverse the outcome, if actual results tracked past results. Not "banal or meaningless". A micro-decision affecting a title, with a title worth $50, 100 or 250 million or more.

The opportunity to offer different and / or better analysis than I or anyone else offered (and occasionally some do) is open in each and every thread. Either you do it or you don't.