Celtics vs Heat ECF Redux Discussion Thread

kieckeredinthehead

Member
SoSH Member
Jun 26, 2006
8,635
Here’s something sort of related but worth thinking about in the context of postseason game planning. Who shot a higher 3 pt % in last years’ finals? Boston. Who took more free throws? Boston. Who forced more turnovers? GSW. Who had more FGA? GSW, they took 5 additional shots per game. They shot six more 3’s per game. And the average score was GSW +4. If shooting variance is much larger game to game and series to series than we think it is, then maybe defense, especially focusing on turnovers, is a much “safer” strategy in small, 7 game (or fewer) samples.
 

slamminsammya

Member
SoSH Member
Jul 31, 2006
9,193
San Francisco
I think I’m doing both while trying to couch it in a discussion of strategy. Here’s the thing. If you treat a series as having its own true 3 pt shooting percentage (drawn as a random variable from the season average), and each game within it having its own true shooting percentage nested within that, the probability of the Celtics’ shooting performance goes from 0.7% to 11%. For Miami, it goes from 0.3% to 8%. Again, there’s nothing I’m changing about the team’s shooting abilities. I’m just treating the outcome of each 3 pt shot as being dependent on that particular night’s “truth” AND the series’ level “truth” rather than assuming every single shot is independent of every other one. That feels like a much, much better model to describe what actually happened. Saying the Celtics had a 1 in 10 shooting performance and the Heat had about a 1 in 13 shooting performance better captures it, doesn’t it?

Emotionally, that makes me feel better. Celtics had some bad luck that was made worse by Brogdon’s injury, but it wasn’t some insane inexplicable completely impossible shooting performance. Shit happens. I absolutely see the same thing a lot of people say about the impact of confidence, but I’m going to leave that as currently unmeasurable.

Strategically, this approach has very different implications for both a single playoff series and for team building. (Actually I wonder if the Bucks/Bud were working from the same flawed assumption - figuring if they drop, the opponents’ shooting has to regress.) The Celtics doubled down on shooting threes this season. They went out and got Brogdon and Gallinari. They pushed Hauser into the rotation. It was the major offensive identity. They went from 10th to 9th to 2nd on 3PA in the regular season. But if you end up having a much higher probability of a dreadful series from 3, and the opposing team is game planning for exactly that, maybe it makes sense to diversify the offense.
This sounds like straightforward over fitting. Is there any a prior reason why shooting percentage should be a function of a series? Why not say every individual shot has its own true likelihood of going in?
 

kazuneko

Member
SoSH Member
Nov 10, 2006
2,818
Honolulu HI
The Heat aren’t the first #8 seed to play for the NBA title. The first this century, but last century isn’t that far in the past. (Allow me to introduce you to the ‘99 Knicks)
Okay, the second 8th seed ever. The point stands though, comparing Brad Steven’s ECF loss to Miami in 2020 to Mazzulla’s team’s loss this year is ridiculous. This team had far higher expectations than that year’s team and Miami victory was far less surprising.
 

sezwho

Member
SoSH Member
Jul 20, 2005
1,952
Isle of Plum
I think I’m doing both while trying to couch it in a discussion of strategy. Here’s the thing. If you treat a series as having its own true 3 pt shooting percentage (drawn as a random variable from the season average), and each game within it having its own true shooting percentage nested within that, the probability of the Celtics’ shooting performance goes from 0.7% to 11%. For Miami, it goes from 0.3% to 8%. Again, there’s nothing I’m changing about the team’s shooting abilities. I’m just treating the outcome of each 3 pt shot as being dependent on that particular night’s “truth” AND the series’ level “truth” rather than assuming every single shot is independent of every other one. That feels like a much, much better model to describe what actually happened. Saying the Celtics had a 1 in 10 shooting performance and the Heat had about a 1 in 13 shooting performance better captures it, doesn’t it?

Emotionally, that makes me feel better. Celtics had some bad luck that was made worse by Brogdon’s injury, but it wasn’t some insane inexplicable completely impossible shooting performance. Shit happens. I absolutely see the same thing a lot of people say about the impact of confidence, but I’m going to leave that as currently unmeasurable.

Strategically, this approach has very different implications for both a single playoff series and for team building. (Actually I wonder if the Bucks/Bud were working from the same flawed assumption - figuring if they drop, the opponents’ shooting has to regress.) The Celtics doubled down on shooting threes this season. They went out and got Brogdon and Gallinari. They pushed Hauser into the rotation. It was the major offensive identity. They went from 10th to 9th to 2nd on 3PA in the regular season. But if you end up having a much higher probability of a dreadful series from 3, and the opposing team is game planning for exactly that, maybe it makes sense to diversify the offense.
Want to say I’m enjoying this conversation enormously and the concept of presenting nesting or grouped samples within a larger sample is compelling here. The cause of the grouping (bad flight or matchup or great night sleep or whatever) may be impossible to control for (Asimovs Foundation notwithstanding:) but treating these outcomes that are much more likely to present than if they were individual random walks seems important.
 

Imbricus

Member
SoSH Member
Jan 26, 2017
4,813
Again, the actual reason for the non-independence doesn’t matter (although as folks have speculated above, it would probably be really useful if you COULD measure it - then you might be able to say “tonight we move away from the 3 point offense.”)
I'm with kieckered here, and I think this is a difficult-to-appreciate point. Also, I think he's dead on regarding the danger inherent in getting the larger point wrong (the "larger point" being whether 3-point shots are random events, or "nested" within games or series).

If you're coaching a team that shot 20% from three in the first half of the last game of a playoff series, and the team shot 39% during the regular season from three, but is shooting 25% during this series, your advice to the team will be different, depending on how you understand this issue. If you believe outcomes are largely random around a 39% percentage, you are more likely to say, "Keep shooting, we're getting good looks, the threes will fall!" If you believe the outcomes are nested around 25% for this particular series (or game), you are more likely to say, "Look, let's try to run plays X and Y, and get some easier looks." The second coach isn't necessarily saying "don't take any more threes," but simply may lift the bar, demanding that players look for some other shots, and only take really good threes.

I think successful three-point shooting is highly determined and only appears random because it is easy to tip the shot into an adverse outcome. So small things, that wouldn't tip a more robust event, if you will, into an adverse outcome, might do so for three-point shooting. An analogy might be if I were to say, "Cut a less-than-two-inch-wide slice off this beefsteak tomato." Easily done: even if you're half-blind, with a dull knife, doesn't matter much. But if I say, "Cut a slice between 3/16th and 4/16th of an inch thick," all of a sudden, the sharpness of the knife matters, your ability to see well, whether you drank two cups of coffee an hour ago, whether you get nervous under pressure, etc. Ancillary factors become potentially primary factors.

On some level, if a coach sees a team with a "nested" three-point shoot percentage of 25%, even if he thinks the team should be shooting 39% based on its longer-term regular season sample, it doesn't matter what's causing the poor shooting ... unless he's absolutely sure that he understands it, and I would wager, that coaches don't. We have this simplified view of understanding what's important in terms of "open shot," "wide open shot," etc., but I think that's just a (maybe small) piece of what contributes to that three-point shot going in. Why did Boston shoot so poorly from three and Miami so well? Did Boston feel the pressure more? Dented confidence? The way they were getting open? Nagging injuries? Overthinking? Or one of three dozen other things?

Anyway, I think that this "coin flip" view of three-point shooting is kind of silly, and dangerous, especially when coaches use that to justify "just keep shooting three-pointers; they'll fall, because of the long-term percentages."
 

slamminsammya

Member
SoSH Member
Jul 31, 2006
9,193
San Francisco
I'm with kieckered here, and I think this is a difficult-to-appreciate point. Also, I think he's dead on regarding the danger inherent in getting the larger point wrong (the "larger point" being whether 3-point shots are random events, or "nested" within games or series).

If you're coaching a team that shot 20% from three in the first half of the last game of a playoff series, and the team shot 39% during the regular season from three, but is shooting 25% during this series, your advice to the team will be different, depending on how you understand this issue. If you believe outcomes are largely random around a 39% percentage, you are more likely to say, "Keep shooting, we're getting good looks, the threes will fall!" If you believe the outcomes are nested around 25% for this particular series (or game), you are more likely to say, "Look, let's try to run plays X and Y, and get some easier looks." The second coach isn't necessarily saying "don't take any more threes," but simply may lift the bar, demanding that players look for some other shots, and only take really good threes.

I think successful three-point shooting is highly determined and only appears random because it is easy to tip the shot into an adverse outcome. So small things, that wouldn't tip a more robust event, if you will, into an adverse outcome, might do so for three-point shooting. An analogy might be if I were to say, "Cut a less-than-two-inch-wide slice off this beefsteak tomato." Easily done: even if you're half-blind, with a dull knife, doesn't matter much. But if I say, "Cut a slice between 3/16th and 4/16th of an inch thick," all of a sudden, the sharpness of the knife matters, your ability to see well, whether you drank two cups of coffee an hour ago, whether you get nervous under pressure, etc. Ancillary factors become potentially primary factors.

On some level, if a coach sees a team with a "nested" three-point shoot percentage of 25%, even if he thinks the team should be shooting 39% based on its longer-term regular season sample, it doesn't matter what's causing the poor shooting ... unless he's absolutely sure that he understands it, and I would wager, that coaches don't. We have this simplified view of understanding what's important in terms of "open shot," "wide open shot," etc., but I think that's just a (maybe small) piece of what contributes to that three-point shot going in. Why did Boston shoot so poorly from three and Miami so well? Did Boston feel the pressure more? Dented confidence? The way they were getting open? Nagging injuries? Overthinking? Or one of three dozen other things?

Anyway, I think that this "coin flip" view of three-point shooting is kind of silly, and dangerous, especially when coaches use that to justify "just keep shooting three-pointers; they'll fall, because of the long-term percentages."
"silly, and dangerous" - the logic you are using here can literally be applied to a coin flip. At a certain level of analysis, a coin flip is deterministic, you just don't have enough information to predict correctly.

On some level, if a coach sees a team with a "nested" three-point shoot percentage of 25%, even if he thinks the team should be shooting 39% based on its longer-term regular season sample, it doesn't matter what's causing the poor shooting ... unless he's absolutely sure that he understands it, and I would wager, that coaches don't
Can't this logic - that the coach does not have insight into what might be causing what he is seeing - be applied also to say that the coach cannot distinguish between a stretch of bad luck and a stretch of truly playing worse? I don't think basketball coaches are generally versed in hypothesis testing.
 

nolasoxfan

Member
SoSH Member
Jun 11, 2004
6,930
Displaced
Here’s something sort of related but worth thinking about in the context of postseason game planning. Who shot a higher 3 pt % in last years’ finals? Boston. Who took more free throws? Boston. Who forced more turnovers? GSW. Who had more FGA? GSW, they took 5 additional shots per game. They shot six more 3’s per game. And the average score was GSW +4. If shooting variance is much larger game to game and series to series than we think it is, then maybe defense, especially focusing on turnovers, is a much “safer” strategy in small, 7 game (or fewer) samples.
Thank you for the past couple of pages. Like others, I am enjoying the hell out of this discussion.

edit: removed first sentence, which was vague.
 
Last edited:

slamminsammya

Member
SoSH Member
Jul 31, 2006
9,193
San Francisco
The Heat made the finals this year and in 2020 riding insanely hot 3 point shooting. The dynasty of the 2010s was the Warriors, who, ya know. The other team in the finals right now jacks up a very large number of 3s. Are we sure defense and turnovers is a "safer" strategy?
 

chilidawg

Member
SoSH Member
Jan 22, 2015
5,934
Cultural hub of the universe
The Heat made the finals this year and in 2020 riding insanely hot 3 point shooting. The dynasty of the 2010s was the Warriors, who, ya know. The other team in the finals right now jacks up a very large number of 3s. Are we sure defense and turnovers is a "safer" strategy?
I would hope we're all aware enough to know we shouldn't be sure of any of this.
 

DennyDoyle'sBoil

Found no thrill on Blueberry Hill
SoSH Member
Sep 9, 2008
42,298
AZ
I think I’m doing both while trying to couch it in a discussion of strategy. Here’s the thing. If you treat a series as having its own true 3 pt shooting percentage (drawn as a random variable from the season average), and each game within it having its own true shooting percentage nested within that, the probability of the Celtics’ shooting performance goes from 0.7% to 11%. For Miami, it goes from 0.3% to 8%. Again, there’s nothing I’m changing about the team’s shooting abilities. I’m just treating the outcome of each 3 pt shot as being dependent on that particular night’s “truth” AND the series’ level “truth” rather than assuming every single shot is independent of every other one. That feels like a much, much better model to describe what actually happened. Saying the Celtics had a 1 in 10 shooting performance and the Heat had about a 1 in 13 shooting performance better captures it, doesn’t it?

Emotionally, that makes me feel better. Celtics had some bad luck that was made worse by Brogdon’s injury, but it wasn’t some insane inexplicable completely impossible shooting performance. Shit happens. I absolutely see the same thing a lot of people say about the impact of confidence, but I’m going to leave that as currently unmeasurable.

Strategically, this approach has very different implications for both a single playoff series and for team building. (Actually I wonder if the Bucks/Bud were working from the same flawed assumption - figuring if they drop, the opponents’ shooting has to regress.) The Celtics doubled down on shooting threes this season. They went out and got Brogdon and Gallinari. They pushed Hauser into the rotation. It was the major offensive identity. They went from 10th to 9th to 2nd on 3PA in the regular season. But if you end up having a much higher probability of a dreadful series from 3, and the opposing team is game planning for exactly that, maybe it makes sense to diversify the offense.
"silly, and dangerous" - the logic you are using here can literally be applied to a coin flip. At a certain level of analysis, a coin flip is deterministic, you just don't have enough information to predict correctly.



Can't this logic - that the coach does not have insight into what might be causing what he is seeing - be applied also to say that the coach cannot distinguish between a stretch of bad luck and a stretch of truly playing worse? I don't think basketball coaches are generally versed in hypothesis testing.
Great discussion. But yeah — generally I am a bit skeptical of any coach to apply all this in real time as new information comes flooding in. Their best 3 point defender tweaks an ankle or gets fouls, etc.
 

kieckeredinthehead

Member
SoSH Member
Jun 26, 2006
8,635
This sounds like straightforward over fitting. Is there any a prior reason why shooting percentage should be a function of a series? Why not say every individual shot has its own true likelihood of going in?
I’m not fitting anything. I’m taking the Celtics’ shooting percentage in the regular season and simulating a seven game series. In one version, they shoot 267 threes and each one is independent of each other.

In the other, I generate a new “true” shooting percentage randomly from their season for the series. Then in each game, I randomly pick a new “true” shooting percentage from the series. Then each 3 within the game is independent of the other. I do that 10,000 times. Then I see how often they shoot as poorly as they did. And the simulation where shot percentage is nested in game and series makes what we just saw about 1,000 times more likely than if every three is a perfectly independent event. We go from a model with one parameter that says what just happened is a 1-in-50,000 event to a model with three parameters that says it’s 1-in-10. From a likelihood perspective, the new model is much, much, much, much, much better.

This is not a weird thing I made up. This is the only appropriate way to treat these data. In any experimental science if you didn’t do it, you’d end up in one of those Statistical Review papers where they figure out what percentage of published research is doing stats wrong. If you were in banking, you’d cause the 2008 financial crisis. If you were a politician, well…

edit: in any science that uses stats, you have to start with the premise that observations are grouped and work back from there to convince people they’re not. Of course three point shooting is going to be more similar in game to game and series to series. Different teams, different stadia, different refs, different health, different strategies, different fatigue, different confidence, etc. etc. Somebody very smartly pointed out that we do this without ever thinking about it in every other sport: some days your closer doesn’t have it; some days you pull your goalie.

Double edit: I’m also not necessarily saying this has implications for in game decision making. I think it’s very difficult for a coach to know if the three isn’t falling and will continue not to (Tatum Game 6 Philly). I am saying if teams don’t recognize this then they should start, and it would have pretty important implications for how much to emphasize a high variance strategy (3 pt shooting) vs low (defense).
 
Last edited:

slamminsammya

Member
SoSH Member
Jul 31, 2006
9,193
San Francisco
I’m not fitting anything. I’m taking the Celtics’ shooting percentage in the regular season and simulating a seven game series. In one version, they shoot 267 threes and each one is independent of each other.

In the other, I generate a new “true” shooting percentage randomly from their season for the series. Then in each game, I randomly pick a new “true” shooting percentage from the series. Then each 3 within the game is independent of the other. I do that 10,000 times. Then I see how often they shoot as poorly as they did. And the simulation where shot percentage is nested in game and series makes what we just saw about 1,000 times more likely than if every three is a perfectly independent event. We go from a model with one parameter that says what just happened is a 1-in-50,000 event to a model with three parameters that says it’s 1-in-10. From a likelihood perspective, the new model is much, much, much, much, much better.

This is not a weird thing I made up. This is the only appropriate way to treat these data. In any experimental science if you didn’t do it, you’d end up in one of those Statistical Review papers where they figure out what percentage of published research is doing stats wrong. If you were in banking, you’d cause the 2008 financial crisis. If you were a politician, well…

edit: in any science that uses stats, you have to start with the premise that observations are grouped and work back from there to convince people they’re not. Of course three point shooting is going to be more similar in game to game and series to series. Different teams, different stadia, different refs, different health, different strategies, different fatigue, different confidence, etc. etc. Somebody very smartly pointed out that we do this without ever thinking about it in every other sport: some days your closer doesn’t have it; some days you pull your goalie.

Double edit: I’m also not necessarily saying this has implications for in game decision making. I think it’s very difficult for a coach to know if the three isn’t falling and will continue not to (Tatum Game 6 Philly). I am saying if teams don’t recognize this then they should start, and it would have pretty important implications for how much to emphasize a high variance strategy (3 pt shooting) vs low (defense).
Continue this logic and you will model every shot as having "true" shooting percentage as either 0 or 1. Do you see a problem?
 

kieckeredinthehead

Member
SoSH Member
Jun 26, 2006
8,635
Continue this logic and you will model every shot as having "true" shooting percentage as either 0 or 1. Do you see a problem?
That‘s exactly how you treat it - every shot either goes in or it doesn’t. How else could this work? We’re not playing quantum basketball where each 3 has a .37 probability of having gone in

Happy to keep answering questions from people actually interested in understanding how this works but I don’t really have the energy to argue with people who don’t get this and don’t want to. Would also love to see pushback (i.e. evidence) that teams already do this. None of the published literature I’ve seen on basketball stats does.
 

slamminsammya

Member
SoSH Member
Jul 31, 2006
9,193
San Francisco
That‘s exactly how you treat it - every shot either goes in or it doesn’t. How else could this work? We’re not playing quantum basketball where each 3 has a .37 probability of having gone in

Happy to keep answering questions from people actually interested in understanding how this works but I don’t really have the energy to argue with people who don’t get this and don’t want to. Would also love to see pushback (i.e. evidence) that teams already do this. None of the published literature I’ve seen on basketball stats does.
OK, I will stop addressing this if your feeling is I "dont get this and don't want to".
 

PedroKsBambino

Well-Known Member
Lifetime Member
SoSH Member
Apr 17, 2003
31,213
I think strategically one takeaway is that you'd adjust "keep shooting, we'll move towards our season-average % as the sample goes up" to sometimes instead say "we should accept that, for purposes of this game/series, there are one or more variables pushing us towards a new default expectation % and adjust offense accordingly"

That intuitively makes sense to me as a fan. What I would guess is challenging (in addition, as others have noted, to not being able to always correctly identify the micro-variables) is that there's a risk of abandoning your mathemetically-derived overall strategy (threes and layups have highest expected points) too quickly on the assumption you're now in a game/series with a different nesting.

But, knowing that the assumption of "a larger sample will always move towards season average" may not apply in a particular game/series is a powerful improvement to the thinking, if true
 

slamminsammya

Member
SoSH Member
Jul 31, 2006
9,193
San Francisco
I think strategically one takeaway is that you'd adjust "keep shooting, we'll move towards our season-average % as the sample goes up" to sometimes instead say "we should accept that, for purposes of this game/series, there are one or more variables pushing us towards a new default expectation % and adjust offense accordingly"

That intuitively makes sense to me as a fan. What I would guess is challenging (in addition, as others have noted, to not being able to always correctly identify the micro-variables) is that there's a risk of abandoning your mathemetically-derived overall strategy (threes and layups have highest expected points) too quickly on the assumption you're now in a game/series with a different nesting.

But, knowing that the assumption of "a larger sample will always move towards season average" may not apply in a particular game/series is a powerful improvement to the thinking, if true
Something something Bayes versus frequentist
 

kieckeredinthehead

Member
SoSH Member
Jun 26, 2006
8,635
OK, I will stop addressing this if your feeling is I "dont get this and don't want to".
If there’s a question in your last post other than if I “see a problem”, just ask it. There’s no problem. And let me pull rank for a second - that paper you keep posting on different sources of variation is hosted on the website of a stats professor who I studied with in grad school.

I think strategically one takeaway is that you'd adjust "keep shooting, we'll move towards our season-average % as the sample goes up" to sometimes instead say "we should accept that, for purposes of this game/series, there are one or more variables pushing us towards a new default expectation % and adjust offense accordingly"

That intuitively makes sense to me as a fan. What I would guess is challenging (in addition, as others have noted, to not being able to always correctly identify the micro-variables) is that there's a risk of abandoning your mathemetically-derived overall strategy (threes and layups have highest expected points) too quickly on the assumption you're now in a game/series with a different nesting.

But, knowing that the assumption of "a larger sample will always move towards season average" may not apply in a particular game/series is a powerful improvement to the thinking, if true
The problem with random effects is they’re unidentifiable. If we could identify why certain series have higher or lower shooting percentages, that would be great - in theory a lot of that is the coaches’ and analysts’ jobs. But sometimes there’s no (as yet) measurable reason (the stuff slamminsammy keeps harping on… at some point a perfect coin has to come up heads or tails). BUT one place it might have changed things is looking at the Heat-Bucks series. If your worldview is that every shot is independent, then the Heat got lucky. They shot unsustainably well. There was nothing to learn from that shooting streak, and predictably their shooting came back to earth against the Knicks. Great. If on the other hand shooting is a function of series, then maybe it’s worth considering whether they might shoot that well again, and what should the game plan be to make sure it doesn’t happen.
 

PedroKsBambino

Well-Known Member
Lifetime Member
SoSH Member
Apr 17, 2003
31,213
If there’s a question in your last post other than if I “see a problem”, just ask it. There’s no problem. And let me pull rank for a second - that paper you keep posting on different sources of variation is hosted on the website of a stats professor who I studied with in grad school.



The problem with random effects is they’re unidentifiable. If we could identify why certain series have higher or lower shooting percentages, that would be great - in theory a lot of that is the coaches’ and analysts’ jobs. But sometimes there’s no (as yet) measurable reason (the stuff slamminsammy keeps harping on… at some point a perfect coin has to come up heads or tails). BUT one place it might have changed things is looking at the Heat-Bucks series. If your worldview is that every shot is independent, then the Heat got lucky. They shot unsustainably well. There was nothing to learn from that shooting streak, and predictably their shooting came back to earth against the Knicks. Great. If on the other hand shooting is a function of series, then maybe it’s worth considering whether they might shoot that well again, and what should the game plan be to make sure it doesn’t happen.
Right---my belief is that while luck was a factor it was actually more than that. Likely, at both ends---Heat higher % and Celts lower % are not random. So I'm not asking you to define why (I get that's hard) I'm just confirming my belief that, essentially, your hypothesis is a potentially valid statistical rationale for my subjective assessment! I am no stats professor, though I have worked with stats in a variety of ways (generally by seeking to translate and align my thinking to that of smarter stats people like yourself!)
 

slamminsammya

Member
SoSH Member
Jul 31, 2006
9,193
San Francisco
If there’s a question in your last post other than if I “see a problem”, just ask it. There’s no problem. And let me pull rank for a second - that paper you keep posting on different sources of variation is hosted on the website of a stats professor who I studied with in grad school.



The problem with random effects is they’re unidentifiable. If we could identify why certain series have higher or lower shooting percentages, that would be great - in theory a lot of that is the coaches’ and analysts’ jobs. But sometimes there’s no (as yet) measurable reason (the stuff slamminsammy keeps harping on… at some point a perfect coin has to come up heads or tails). BUT one place it might have changed things is looking at the Heat-Bucks series. If your worldview is that every shot is independent, then the Heat got lucky. They shot unsustainably well. There was nothing to learn from that shooting streak, and predictably their shooting came back to earth against the Knicks. Great. If on the other hand shooting is a function of series, then maybe it’s worth considering whether they might shoot that well again, and what should the game plan be to make sure it doesn’t happen.
Thanks for pulling rank, shall we trade cv's? :)

As you probably heard in grad school, all models are wrong, some are useful. Obviously shots are not literally random. Settled.

The original conversation is about when / how a coach or observer would update their prior estimate of their teams shooting ability. Great, shots aren't literally random, but from a decision making standpoint the only practical way to view it is as a random process.

You seem to be arguing for more aggressive updating based on who you are playing in a playoffs series. You obviously cannot use the post hoc observed shooting statistics of the series as you go because, ya know, we can't see the future.

I totally agree the "true shooting" distribution will be a moving target for lots of obvious reasons and over different time scales. Players get better at shooting. Guys have torn forearm tendons. Confidence, rhythm, etc.

If you want to guess at these effects with aggressive updating, go for it. The question is how strongly to weigh the prior and also how to derive the prior. If you use no prior you'll obviously get sone absurd results in the beginning of a series. As you probably know the posterior distribution for a Bernoulli variable is the beta distribution, and the likelihood of very good and very bad true shooting percentages will be way too high, i hope you agree.

The standard way to incorporate your prior model in this setting is to introduce pseudo trials. In our case maybe you take the last K shots for each player, where the size of K determines the strength of the prior. Or you could take the percentage over the last K shots and then scale it down if you want faster updating. Lots of ways to treat this, and you can do fitting to whatever parameters you choose.

I've done this sort of thing before, and the resulting posterior distribution for the "best" updating strategies does not budge as much as you think.

My point about you doing over fitting was precisely that if you drop a prior and treat the series as its own thing your model will not be as predictive as if you used a strong prior.

Hope this clarifies, let me know if I'm misconstruing entirely what you are saying. I just find the "it's not random" take to be true but also, like, totally irrelevant to the conversation.
 

Imbricus

Member
SoSH Member
Jan 26, 2017
4,813
"silly, and dangerous" - the logic you are using here can literally be applied to a coin flip. At a certain level of analysis, a coin flip is deterministic, you just don't have enough information to predict correctly.
"Silly" was a bad choice of words on my part; the more I think about this, the more I start seeing both sides to some degree. Also, yeah, a coin flip is deterministic, but the variables are fewer, and practically immeasurable, so there are differences, but anyway, that's not a rabbit hole worth going down. And I think I get the point about overfitting; it's interesting.

So I think we're all divided partly on how random we perceive three-point shooting to be (around some natural "true" percentage for a given shooter that will be different for a Marcus Smart and a Steph Curry). I tend to think it's less random. That seems to be a key dividing point. If you think Smart's natural true percentage is 35%, and he's shooting 20% over the last 10 games, you're more likely to think that when he takes his next three (assuming conditions are "average"), he'll have a 35% chance of making that shot. But if you think his shooting percentage dropping to 20% is less luck and more something else (and here, I'm not talking something big and obvious, like he's getting double-teamed all the time, because we all can see that as an effect), then you'll have a different, more pessimistic view of how he'll fare on the next shot.

The part that's curious to me is how do you set a "baseline expectation," and under what conditions should that be modified, even if you believe in a largely random element to three-point shooting? E.g., the Celtics shot 37.4% from three during the regular season. So if I'm looking at that, as a coach during the Miami series, maybe I'm thinking, "That's a good percentage, keep taking threes, it'll all revert to the mean of 37.4%!" But the playoffs are a different beast: (1) The pressure is more intense (2) Rotations shorten, so fatigue can start to be a factor (Horford?) (3) Guys are more apt to play through pain, injuries )(Brogdon, maybe Jaylen a bit?) (4) You're up against the same team for up to 7 consecutives games, which never happens in the regular season (5) The coaching becomes smarter. Etc., etc.

Which is to say: even if you believe in a largely random element to three-point shooting success, that needs to bounce around some center point. What if that natural "true shooting" for the Celtics, vs. the Heat, in a playoff series like the one that we just saw, is 25%? So when we shoot 41%, we're actually shooting lights out, and if we shoot 23%, we're not that far from where we'd be expected to be. In that case, if I'm a coach looking for a reversion to 37.4%, I'm kind of deluding myself. Anyway, a few thoughts; I find this fascinating but it's starting to become awfully distracting!
 

DeJesus Built My Hotrod

Well-Known Member
Lifetime Member
SoSH Member
Dec 24, 2002
48,214
"Silly" was a bad choice of words on my part; the more I think about this, the more I start seeing both sides to some degree. Also, yeah, a coin flip is deterministic, but the variables are fewer, and practically immeasurable, so there are differences, but anyway, that's not a rabbit hole worth going down. And I think I get the point about overfitting; it's interesting.

So I think we're all divided partly on how random we perceive three-point shooting to be (around some natural "true" percentage for a given shooter that will be different for a Marcus Smart and a Steph Curry). I tend to think it's less random. That seems to be a key dividing point. If you think Smart's natural true percentage is 35%, and he's shooting 20% over the last 10 games, you're more likely to think that when he takes his next three (assuming conditions are "average"), he'll have a 35% chance of making that shot. But if you think his shooting percentage dropping to 20% is less luck and more something else (and here, I'm not talking something big and obvious, like he's getting double-teamed all the time, because we all can see that as an effect), then you'll have a different, more pessimistic view of how he'll fare on the next shot.

The part that's curious to me is how do you set a "baseline expectation," and under what conditions should that be modified, even if you believe in a largely random element to three-point shooting? E.g., the Celtics shot 37.4% from three during the regular season. So if I'm looking at that, as a coach during the Miami series, maybe I'm thinking, "That's a good percentage, keep taking threes, it'll all revert to the mean of 37.4%!" But the playoffs are a different beast: (1) The pressure is more intense (2) Rotations shorten, so fatigue can start to be a factor (Horford?) (3) Guys are more apt to play through pain, injuries )(Brogdon, maybe Jaylen a bit?) (4) You're up against the same team for up to 7 consecutives games, which never happens in the regular season (5) The coaching becomes smarter. Etc., etc.

Which is to say: even if you believe in a largely random element to three-point shooting success, that needs to bounce around some center point. What if that natural "true shooting" for the Celtics, vs. the Heat, in a playoff series like the one that we just saw, is 25%? So when we shoot 41%, we're actually shooting lights out, and if we shoot 23%, we're not that far from where we'd be expected to be. In that case, if I'm a coach looking for a reversion to 37.4%, I'm kind of deluding myself. Anyway, a few thoughts; I find this fascinating but it's starting to become awfully distracting!
Great post.

I suspect the factors you highlight as potentially limiting to playoff three point shooting percentage apply for two point shooting as well. That - and the level of playoff rim protection (the Cs faced some decent opponents in this department the last few rounds)- probably informs the strategy of continuing to take open looks from deep.

I also agree that this entire discussion is fascinating.
 

lovegtm

Member
SoSH Member
Apr 30, 2013
12,001
The Celtics' strategic options once 3s weren't falling and Tatum was hurt were very limited:

- force drives to the paint against a loaded up D
- keep shooting somewhat marginal 3s and hope for regression

Jaylen and DWhite actually tried option 1. Jaylen turned it over every time. White got some buckets, but it wasn't sustainable because he wasn't....hitting 3s, so Miami could load up more.

If you don't have a healthy star to tilt the floor, you need to hope 3s going down. Hilariously, that is the EXACT situation Miami found itself in: Jimmy played like utter dogshit after going up 3-0, but eventually, in game 7, all the 3s started falling for Miami, and they won.

It sucks to admit we invest so much mental energy in something this random, but it is what it is.
 

dhellers

Member
SoSH Member
Oct 31, 2005
4,209
Silver Spring, Maryland
Pre-2008, finance incorrectly treated loan failures as independent events. In 2016, election prognosticators incorrectly treated state level outcomes as independent events. I’m getting the sense that some teams (the Houston Hardens, the Slingin’ Mazullas) are incorrectly treating single three point shots as independent events rather than as nested within game and within series. There is literally no way that you get the Heat and Celtics 3 pt shooting this series through shot to shot random chance. p<0.0001. I don’t know WHY the Heat shot like Curry and the Celtics shot like Giannis, but that’s why they call them random effects.

Edit: to use the metaphor above, I bet a lot of teams are treating shot outcomes as being from the same coin, but it seems pretty obvious at this point that each game uses a differently weighted coin selected from a bag that changes series to series.

double edit: Just want to emphasize I’m using the 08 financial collapse and 16 election as examples of very smart people with a lot more to lose getting this wrong before anybody blithely insists that teams’ analysts are much more sophisticated than we could ever imagine.
Nicely done.
A serial correlation rv model seems possible, one with a decay term of unknown length ,(1 period,? 1 game, ...,). Orva panel model with a period specific draw ( or a game specific draw). In either case, ones guess as to the duration of the impact(a period? a game) would effect how quickly you gave up on the 3s.

Oof...didn't see 3 pages of comments. The question seems to be: would a sophisticated model with robust fit metrics, but loosely based on observables, be actionable? Say, some neural network/random forest non parametric melange? Perhaps it could flag "dial back the 3s for a while" states of nature.
 
Last edited:

wade boggs chicken dinner

Member
SoSH Member
Mar 26, 2005
30,506
The question seems to be: would a sophisticated model with robust fit metrics, but loosely based on observables, be actionable? Say, some neural network/random forest non parametric melange? Perhaps it could flag "dial back the 3s for a while" states of nature.
I think people are trying to find out if there are biometric indicators that might predict better or worse performance.
 

slamminsammya

Member
SoSH Member
Jul 31, 2006
9,193
San Francisco
Nicely done.
A serial correlation rv model seems possible, one with a decay term of unknown length ,(1 period,? 1 game, ...,). Orva panel model with a period specific draw ( or a game specific draw). In either case, ones guess as to the duration of the impact(a period? a game) would effect how quickly you gave up on the 3s.

Oof...didn't see 3 pages of comments. The question seems to be: would a sophisticated model with robust fit metrics, but loosely based on observables, be actionable? Say, some neural network/random forest non parametric melange? Perhaps it could flag "dial back the 3s for a while" states of nature.
To be actionable you'd also need to do the same thing for the offensive alternatives right? As lovegtm mentioned, if you arent shooting 3s especially when they are open the defense eventually adjusts to make the rim even harder to attack. Sadly in today's league ya just gotta hit the 3s when you get them.
 

OurF'ingCity

Member
SoSH Member
Apr 22, 2016
8,469
New York City
The Celtics' strategic options once 3s weren't falling and Tatum was hurt were very limited:

- force drives to the paint against a loaded up D
- keep shooting somewhat marginal 3s and hope for regression

Jaylen and DWhite actually tried option 1. Jaylen turned it over every time. White got some buckets, but it wasn't sustainable because he wasn't....hitting 3s, so Miami could load up more.

If you don't have a healthy star to tilt the floor, you need to hope 3s going down. Hilariously, that is the EXACT situation Miami found itself in: Jimmy played like utter dogshit after going up 3-0, but eventually, in game 7, all the 3s started falling for Miami, and they won.

It sucks to admit we invest so much mental energy in something this random, but it is what it is.
This is why it's so hard to win 4 straight in an NBA series, particularly for a team that relies heavily on 3-point shooting. There is going to be some night where those shots just aren't falling. So I don't have a problem with Game 7 so much because they just got unlucky in a lot of ways - the problem was in Games 1, 2, and 3 (particularly Game 2 where the Heat shot 34.6% from three and the Cs still couldn't win).

I'm just echoing what many others have said on here already but particularly in the playoffs I don't think the "shoot tons of threes and eventually some will go in" strategy is optimal unless you have absolutely lights out shooters (like the Curry/Klay Warriors at their peak). You're basically guaranteeing a loss or two just due to shot variance which then means there is much less room for error in the remaining games.
 

Auger34

used to be tbb
SoSH Member
Apr 23, 2010
9,303
This is why it's so hard to win 4 straight in an NBA series, particularly for a team that relies heavily on 3-point shooting. There is going to be some night where those shots just aren't falling. So I don't have a problem with Game 7 so much because they just got unlucky in a lot of ways - the problem was in Games 1, 2, and 3 (particularly Game 2 where the Heat shot 34.6% from three and the Cs still couldn't win).

I'm just echoing what many others have said on here already but particularly in the playoffs I don't think the "shoot tons of threes and eventually some will go in" strategy is optimal unless you have absolutely lights out shooters (like the Curry/Klay Warriors at their peak). You're basically guaranteeing a loss or two just due to shot variance which then means there is much less room for error in the remaining games.
I think almost every team has to shoot tons of threes to win, you just have to have a counter if they aren't going in or a guy you can trust to go get an easy bucket.
Denver couldn't buy a 3 last night but when things got tight, especially in the 4th quarter, they would throw the ball to Jokic and he was able to back Bam down to get an easy turnaround jumper or find an open shooter.
To me, that is the key. If you rely too much on jump shooting, and there's no play or person you can count on to get a two or a high percentage shot, things can snowball incredibly fast and get out of hand
 

slamminsammya

Member
SoSH Member
Jul 31, 2006
9,193
San Francisco
I think almost every team has to shoot tons of threes to win, you just have to have a counter if they aren't going in or a guy you can trust to go get an easy bucket.
Denver couldn't buy a 3 last night but when things got tight, especially in the 4th quarter, they would throw the ball to Jokic and he was able to back Bam down to get an easy turnaround jumper or find an open shooter.
To me, that is the key. If you rely too much on jump shooting, and there's no play or person you can count on to get a two or a high percentage shot, things can snowball incredibly fast and get out of hand
The Celtics do have counters. Jaylen and Jayson are two of the best guys attacking the rim in the league, and Brogdon too. But what do you do when they aren't able to get there without turning the ball over, or when the defense collapses?
 

dhellers

Member
SoSH Member
Oct 31, 2005
4,209
Silver Spring, Maryland
To be actionable you'd also need to do the same thing for the offensive alternatives right? As lovegtm mentioned, if you arent shooting 3s especially when they are open the defense eventually adjusts to make the rim even harder to attack. Sadly in today's league ya just gotta hit the 3s when you get them.
Yeah, a proper model would have to suggest switch to x for a while even if your opponent is likely to figure out that is what you are doing

And having better developed alternatives even if they are worse under normal ( not even good) 3 point shooting states of nature could be useful.

Slamming noted "biometric indicators " and although it goes against my 1990s training, modern big data models use everything they can get their hands on. Won't yield explanations but can (maybe?) yield good prediction
 
Last edited:

slamminsammya

Member
SoSH Member
Jul 31, 2006
9,193
San Francisco
This is why it's so hard to win 4 straight in an NBA series, particularly for a team that relies heavily on 3-point shooting. There is going to be some night where those shots just aren't falling. So I don't have a problem with Game 7 so much because they just got unlucky in a lot of ways - the problem was in Games 1, 2, and 3 (particularly Game 2 where the Heat shot 34.6% from three and the Cs still couldn't win).

I'm just echoing what many others have said on here already but particularly in the playoffs I don't think the "shoot tons of threes and eventually some will go in" strategy is optimal unless you have absolutely lights out shooters (like the Curry/Klay Warriors at their peak). You're basically guaranteeing a loss or two just due to shot variance which then means there is much less room for error in the remaining games.
I think the Lebron Cavs were the last team to win a championship that didn't shoot a lot of threes. I guess the Raptors were 10th that year but 5th in the playoffs in 3pa per game. They actually took more than the Warriors.
 

ehaz

Member
SoSH Member
Sep 30, 2007
4,948
I think the Lebron Cavs were the last team to win a championship that didn't shoot a lot of threes. I guess the Raptors were 10th that year but 5th in the playoffs in 3pa per game. They actually took more than the Warriors.
Bubble Lakers.
 

ugmo33

Member
SoSH Member
May 6, 2016
165
The Celtics' strategic options once 3s weren't falling and Tatum was hurt were very limited:

- force drives to the paint against a loaded up D
- keep shooting somewhat marginal 3s and hope for regression

Jaylen and DWhite actually tried option 1. Jaylen turned it over every time. White got some buckets, but it wasn't sustainable because he wasn't....hitting 3s, so Miami could load up more.

If you don't have a healthy star to tilt the floor, you need to hope 3s going down. Hilariously, that is the EXACT situation Miami found itself in: Jimmy played like utter dogshit after going up 3-0, but eventually, in game 7, all the 3s started falling for Miami, and they won.

It sucks to admit we invest so much mental energy in something this random, but it is what it is.
Shouldn't there also be an option 3.) ball movement and player movement, especially utilizing the threat of your stretch 5, to open up the paint and get good looks at the rim/midrange?
 

lovegtm

Member
SoSH Member
Apr 30, 2013
12,001
Shouldn't there also be an option 3.) ball movement and player movement, especially utilizing the threat of your stretch 5, to open up the paint and get good looks at the rim/midrange?
That's the problem though: the stretch 5 has been missing everything since his "elite shooter" game against Philly, and "utilizing the threat" of a stretch 5 implies that he's going to take and make open 3s if the defense gives them to him.........

......at which point you're back to the make-or-miss stuff from 3.

There's no free lunch here. You can do all the cute cutting and ball-movement you want, but if you don't have an elite scoring threat OR the ability to light it up outside, NBA defenses will stifle you.

The Celtics have the elite scoring threat, but he got hurt. They're supposed to have a secondary one too, but he played like dogshit (also possibly hurt).
 

lovegtm

Member
SoSH Member
Apr 30, 2013
12,001
This is why it's so hard to win 4 straight in an NBA series, particularly for a team that relies heavily on 3-point shooting. There is going to be some night where those shots just aren't falling. So I don't have a problem with Game 7 so much because they just got unlucky in a lot of ways - the problem was in Games 1, 2, and 3 (particularly Game 2 where the Heat shot 34.6% from three and the Cs still couldn't win).

I'm just echoing what many others have said on here already but particularly in the playoffs I don't think the "shoot tons of threes and eventually some will go in" strategy is optimal unless you have absolutely lights out shooters (like the Curry/Klay Warriors at their peak). You're basically guaranteeing a loss or two just due to shot variance which then means there is much less room for error in the remaining games.
This is a good point, and would require either a leap forward from Tatum and Brown, or roster recomposition.
 

kazuneko

Member
SoSH Member
Nov 10, 2006
2,818
Honolulu HI
This is why it's so hard to win 4 straight in an NBA series, particularly for a team that relies heavily on 3-point shooting. There is going to be some night where those shots just aren't falling. So I don't have a problem with Game 7 so much because they just got unlucky in a lot of ways - the problem was in Games 1, 2, and 3 (particularly Game 2 where the Heat shot 34.6% from three and the Cs still couldn't win).
I'm just echoing what many others have said on here already but particularly in the playoffs I don't think the "shoot tons of threes and eventually some will go in" strategy is optimal unless you have absolutely lights out shooters (like the Curry/Klay Warriors at their peak). You're basically guaranteeing a loss or two just due to shot variance which then means there is much less room for error in the remaining games.
Even if you do have great 3 point shooters it’s going to lead to more performance variability than other approaches, which is why the NBA playoffs have become less like predictable in recent years. That said, if three point shooting is going to be the focal point of your offense, having a number 1 option like Curry and a number 2 like Thompson is definitely preferable than having Tatum and Brown - who both shot below league average from 3 this season. That's why this needs to be a team whose primary focus is defense - where they have shown a consistent ability to be elite. That said, if they truly want to play offense like Golden State it might be wise to consider swapping Brown for someone who is actually an elite 3pt shooter, which seems beyond what we should ever expect of JB.

This is a good point, and would require either a leap forward from Tatum and Brown, or roster recomposition.
If the Cs want to be built around elite 3 point shooting they need roster recomposition, and the expendable star is clearly Brown. Is there any way to swing a Bane for Brown trade? Until at least next offseason (when Bane might sign his own max contract) their salaries would appear unmatchable. You could do Adams, Bane and Laravia this offseason, but unless Adam’s knee injury is worse than reported or Memphis thinks Brown can provide the type of veteran leadership that might reign in Ja, it's not at all clear why Memphis would be interested. Another name that would make sense is Anfernee Simons, especially if Portland is really considering including the 3 pick as well. Simons is nearly 39% from 3pt range for his career on high volume - and shot near 90% from the free throw line this past year. In other words, while there may be concerns about his defense, his shooting - unlike Brown- is elite. Of course the primary target of a Brown to Portland trade would be the #3 pick - but I think Simons has been undervalued in that discussion. He isn't just salary fodder, as offensively he is the type of shooter the Cs need if their goal is elite shooting. One other name to consider is Lamello Ball. Charlotte is apparently considering bypassing Henderson with the number 2 pick because they don't see Ball and Scoot as compatible. Might they consider trading Ball for Brown instead? Ball has shown more promise than Brown as a shooter and could be the playmaker the Cs are lacking. Ball for Brown doesn’t work for salary matching, but Brown/Smart for Ball/Hayward/M. Williams does (actually Williams isn’t necessary to match but would be awesome to add).
 
Last edited: