Top 25 peaks of the 2001-25: #13-#14 Spots

[Djoker]

It's been interesting reading the last couple of pages of discussion related to the reliability and validity of the different metrics.

However, one thing I'll say is that people generally grossly underestimate how wide the error bars are when using these metrics. Heck, even something like lifetime RAPM where we've got huge 1000+ game samples for most legends; it's still quite noisy. Case in point, I calculated 95% confidence intervals (+/- 2 SD) for the top names in the latest 29-year Lifetime RAPM (1997-2025) from Engelmann.



Pretty wide ranges for 29-year Lifetime RAPM no?

For single season RAPM, the difference between the best player in the league and the average player is statistically insignificant. That's how wide the confidence intervals are. For something like 5-year RAPM, it's much better but it's still noticeably higher than the Lifetime RAPM. For example it might be similar to the confidence intervals for Jordan and Stockton in the Lifetime RAPM above who have 4 and 7 seasons worth of data, respectively. At that point, the top guys are looking at very very wide ranges. For example, maybe #1 to #50.

That's why I've never been a huge proponent of RAPM. It's a nice metric to look at and it is definitely a data point but it's not the end-all be-all yet some people use it like a holy grail stat.

[70sFan]

It's been interesting reading the last couple of pages of discussion related to the reliability and validity of the different metrics.

However, one thing I'll say is that people generally grossly underestimate how wide the error bars are when using these metrics. Heck, even something like lifetime RAPM where we've got huge 1000+ game samples for most legends; it's still quite noisy. Case in point, I calculated 95% confidence intervals (+/- 2 SD) for the top names in the latest 29-year Lifetime RAPM (1997-2025) from Engelmann.



Pretty wide ranges for 29-year Lifetime RAPM no?

For single season RAPM, the difference between the best player in the league and the average player is statistically insignificant. That's how wide the confidence intervals are. For something like 5-year RAPM, it's much better but it's still noticeably higher than the Lifetime RAPM. For example it might be similar to the confidence intervals for Jordan and Stockton in the Lifetime RAPM above who have 4 and 7 seasons worth of data, respectively. At that point, the top guys are looking at very very wide ranges. For example, maybe #1 to #50.

That's why I've never been a huge proponent of RAPM. It's a nice metric to look at and it is definitely a data point but it's not the end-all be-all yet some people use it like a holy grail stat.

I'd add even more to that point - we should always remember that, whatever analytical tools we use to evaluate players, we always have massive error bars in our results. In the end, basketball isn't a hard sport, but it's very complex with a lot of variables that are more or less connected to each other. To unpack all of that, it's extremely hard and we just don't have the access to all data necessary.

That doesn't mean that we shouldn't try - in the end, I am running the project that tries to create an accurate ranking after all - but I think that we shouldn't push our conclusions too hard. It's very safe to say that someone like LeBron James is better than your average all-star. It's far, far harder to differentiate between the top 10 guys here. It's not impossible and we all do that, but we have to remember that we can be wrong on these things.

I think people should realise how extreme praise a player gets from the community if he makes the list, even at the very end. Pushing someone like Embiid, Harden, Manu or anyone else lower than you'd like doesn't mean you are a hater or fanboy - it just means that we reached different conclusions and we can all be wrong. It's not very likely and I think the final product is quite close to the reality, but it's not set in stone.

[DraymondGold]

It's been interesting reading the last couple of pages of discussion related to the reliability and validity of the different metrics.

However, one thing I'll say is that people generally grossly underestimate how wide the error bars are when using these metrics. Heck, even something like lifetime RAPM where we've got huge 1000+ game samples for most legends; it's still quite noisy. Case in point, I calculated 95% confidence intervals (+/- 2 SD) for the top names in the latest 29-year Lifetime RAPM (1997-2025) from Engelmann.



Pretty wide ranges for 29-year Lifetime RAPM no?

For single season RAPM, the difference between the best player in the league and the average player is statistically insignificant. That's how wide the confidence intervals are. For something like 5-year RAPM, it's much better but it's still noticeably higher than the Lifetime RAPM. For example it might be similar to the confidence intervals for Jordan and Stockton in the Lifetime RAPM above who have 4 and 7 seasons worth of data, respectively. At that point, the top guys are looking at very very wide ranges. For example, maybe #1 to #50.

That's why I've never been a huge proponent of RAPM. It's a nice metric to look at and it is definitely a data point but it's not the end-all be-all yet some people use it like a holy grail stat.

Do you have any sources / estimates of the error in 5-year RAPM being higher than lifetime RAPM? Like many have said in this thread (and indeed like many have had to repeat across multiple threads), normal RAPM assumes the player is constant. So as our sample size increases and the statistical standard error decreases, the assumption that the player is the same across the sample gets more and more faulty. SO by the time we're dealing with a 26-year-long RAPM, we're dealing with a case where that assumption becomes really quite undermining in terms of validity compared to something like 5-year RAPM (or even 11-year RAPM in Squared's case).

The faultiness of this assumption should be measurable somewhere. For example, if someone looked at RAPM across sample sizes (from 1-year to 26-year) and applied some sort of prediction / retrodiction tests to measure the stat's accuracy in some sense (e.g. here https://www.nbastuffer.com/analytics101/darko-daily-plus-minus/ or here https://fansided.com/2019/01/08/nylon-calculus-best-advanced-stat/ ; as an aside, is accuracy the right word? ), then I expect the most accurate RAPM would *not* be 26-year lifetime RAPM, as the assumption that a player's constant over the sample would be particularly undermining by this point. Just based on what NBA statisticians tend to prefer, I would expect 3-5 year RAPM to be the most accurate... although I've never found a study explicitly testing this.

I'm not sure how I would expect the error to behave. Part of me would expect the variability of the player over a lifetime sample to inflate the error more than the larger sample size decreases the error. But it also could go the other way. It might also depend on the RAPM version -- e.g. if there's priors, or box inputs, or anything like that. Since different RAPMs scale differently, the thing might be to look at the ranking ranges based on the +/- 2 SD uncertainty, like you did here.

Sounds like you've seen 5-year RAPM have a wider ranking range, so just curious to look at how the ranges compare.

[DraymondGold]

Does reliability not mean reliability any more?

The entire point of the xRAPM variants is that they smack the more pure apm/rapms in reliability.

Yeah seems like a good moment to bust out the old archery diagram:



By validity: APM > RAPM > XRAPM > Box
By reliability: Box > XRAPM > RAPM > APM

I miss long-sample classic, un-regularized, un-Xploited APM studies.

Cool post! I do wonder if the rankings might change as we get to different sample sizes, like I think lessthanjake said somewhere. APM/RAPM might get more reliable for certain sample sizes (traditionally ~3-5 years), and I'm genuinely not sure if the reliability improvement there is enough to change the ranking order. Even Box stats can have some noise in season-long samples, so we'd expect Box stats to also get more reliable.

By the time we get to lifetime 26-year lengths, I'd expect APM/RAPM to become somewhat less valid at estimating a specific version of the player, as the assumption that the player remains constant becomes obviously wrong. Although maybe it would still be valid if explicitly interpreting the values as the player's average, rather than as evidence for the player's peak quality like it's been argued here? Lifetime Box stats pretty obviously are computing a player's average box quality, but it's less obvious to me how the regression process works for a player who's changing drastically throughout the sample... it would presumably approximately tend toward the average player, but it's not clear to me if a given regression/regularization would go towards exactly the player's average based on the specifics of the methodology. Depending on the distribution of pre-prime, prime, peak, and post-prime samples, might different aspects of different players' careers be penalized in the regression process?
Edit: In which case, if one player's e.g. shorter pre-prime sample tends to be penalized more while another player's short peak sample tends to be penalized more, wouldn't those cases be slightly different from unweighted averages? Maybe some sort of weighted average is the way to think of it, where the weightings might differ for different players, based on whichever points the method thinks are most likely to be abnormal (which presumably relates to how much of a pre-prime, prime, peak, and post-prime sample we have for each player).

Regardless, just wanted to point out that I think (?) pure APM can be found on this site, under the confusing labeled column 'Adjusted Plus Minus' (and not APM, which is their xAPM / xRAPM stat): https://www.intraocular.net/apps/best-basketball-player .

I'm not sure if I've ever seen long-sample/multi-year pure APM (without the R ) that you mention, but would be interested in seeing the source if you have something that comes to mind.

[lessthanjake]

It's been interesting reading the last couple of pages of discussion related to the reliability and validity of the different metrics.

However, one thing I'll say is that people generally grossly underestimate how wide the error bars are when using these metrics. Heck, even something like lifetime RAPM where we've got huge 1000+ game samples for most legends; it's still quite noisy. Case in point, I calculated 95% confidence intervals (+/- 2 SD) for the top names in the latest 29-year Lifetime RAPM (1997-2025) from Engelmann.



Pretty wide ranges for 29-year Lifetime RAPM no?

For single season RAPM, the difference between the best player in the league and the average player is statistically insignificant. That's how wide the confidence intervals are. For something like 5-year RAPM, it's much better but it's still noticeably higher than the Lifetime RAPM. For example it might be similar to the confidence intervals for Jordan and Stockton in the Lifetime RAPM above who have 4 and 7 seasons worth of data, respectively. At that point, the top guys are looking at very very wide ranges. For example, maybe #1 to #50.

That's why I've never been a huge proponent of RAPM. It's a nice metric to look at and it is definitely a data point but it's not the end-all be-all yet some people use it like a holy grail stat.

Yep, this is a very good point to raise. Even the large-sample RAPM that is much more reliable than small-sample RAPM still has significant amounts of noise. It’s not that small-sample RAPM is pretty noisy and large-sample RAPM isn’t noisy, but rather that small-sample RAPM is *incredibly* noisy and large-sample RAPM is just pretty noisy.

I will say with RAPM that this sort of thing is why I like to see players that have a lot of great multi-year RAPM spans. To use a player that is relevant to this thread as an example, I quite like that Ginobili has 11 straight five-year spans that are top 10 in RAPM. As you point out, one 5-year span definitely still has a good bit of noise, but the chances of randomly repeatedly being near the top across so many different spans seems quite unlikely to me. You’d need variance to constantly be going in your favor. Intuitively, I think that’s actually even more convincing than someone being high up in lifetime RAPM.

This sort of thing is also why I generally do like to triangulate RAPM with box data and hybrid data. Basically, even if I tend to prefer large-sample RAPM over box data, we can have more confidence in our conclusion about a player whose RAPM and box data *both* look great. The same goes for just looking at lots of data sources in general. Every individual measure is noisy and flawed in serious ways, so I think looking at the whole picture is a good way to avoid indexing one’s view on whatever noise or flaws happen to exist in a particular measure.

I will note that, while I’ve long been a proponent of at least considering all available data, the paragraph just above admittedly does run a bit contrary to what I’ve been saying about Steve Nash in this thread and the last one—where I’ve largely argued for believing his RAPM data over his box data. If we triangulated the great impact data with the not-as-great box data, we’d be left with a conclusion somewhere in the middle—which would probably not be as positive a conclusion about Nash as I have. As I believe I mentioned in earlier discussions in the last thread, I’d normally take the somewhere-in-the-middle point of view in this sort of situation (though that in-the-middle spot would probably still be tilted more towards what the large-sample RAPM says). But with Nash I also just have a huge amount of eye test stuff (he’s probably one of the two or three players in NBA history that I’ve watched the most). And since my eye test is super high on Nash, that results in my assessment tilting much more towards the great RAPM data compared to less-good box data than I normally would with the typical player that I haven’t watched as much of (or that I don’t feel as strongly about the eye test on). The box data on Nash does still give me slight pause, but not nearly as much as it would without the extensive eye test I have on him.

[Doctor MJ]

Does reliability not mean reliability any more?

The entire point of the xRAPM variants is that they smack the more pure apm/rapms in reliability.

Yeah seems like a good moment to bust out the old archery diagram:



By validity: APM > RAPM > XRAPM > Box
By reliability: Box > XRAPM > RAPM > APM

I miss long-sample classic, un-regularized, un-Xploited APM studies.

Cool post! I do wonder if the rankings might change as we get to different sample sizes, like I think lessthanjake said somewhere. APM/RAPM might get more reliable for certain sample sizes (traditionally ~3-5 years), and I'm genuinely not sure if the reliability improvement there is enough to change the ranking order. Even Box stats can have some noise in season-long samples, so we'd expect Box stats to also get more reliable.

By the time we get to lifetime 26-year lengths, I'd expect APM/RAPM to become somewhat less valid at estimating a specific version of the player, as the assumption that the player remains constant becomes obviously wrong. Although maybe it would still be valid if explicitly interpreting the values as the player's average, rather than as evidence for the player's peak quality like it's been argued here? Lifetime Box stats pretty obviously are computing a player's average box quality, but it's less obvious to me how the regression process works for a player who's changing drastically throughout the sample... it would presumably approximately tend toward the average player, but it's not clear to me if a given regression/regularization would go towards exactly the player's average based on the specifics of the methodology. Depending on the distribution of pre-prime, prime, peak, and post-prime samples, might different aspects of different players' careers be penalized in the regression process?
Edit: In which case, if one player's e.g. shorter pre-prime sample tends to be penalized more while another player's short peak sample tends to be penalized more, wouldn't those cases be slightly different from unweighted averages? Maybe some sort of weighted average is the way to think of it, where the weightings might differ for different players, based on whichever points the method thinks are most likely to be abnormal (which presumably relates to how much of a pre-prime, prime, peak, and post-prime sample we have for each player).

Regardless, just wanted to point out that I think (?) pure APM can be found on this site, under the confusing labeled column 'Adjusted Plus Minus' (and not APM, which is their xAPM / xRAPM stat): https://www.intraocular.net/apps/best-basketball-player .

I'm not sure if I've ever seen long-sample/multi-year pure APM (without the R ) that you mention, but would be interested in seeing the source if you have something that comes to mind.

Interesting. I wish they were more consistent with their terminology. I don't mind them using APM as a broader category name and using Pure APM the true vanilla APM as they imply in their writing, but their main page doesn't use "Pure" for what seems like it must be the Pure.

Going back through time, classic things I look for:

Steve Nash in '04-05 should be higher than '05-06 unless '04-05 is being used as a prior, and this is what we find.

'04-05 leaderboard (minutes indicated RS + PS):

1. Ginobili 7.9
2. Duncan 6.8
3. Nash 5.9

'05-06 leaderboard:

1. Wade 6.8
2. Ginobili 5.8
3. Artest 5.2
4. Kobe 5.0
5. Nash 4.9

We should also see the '05-06 Pistons data look extremely problematic due to multicollinearity issues - which cry out techniques beyond pure APM:

Sheed 4.5
Prince 3.5
Ben 3.3
Billups 3.1
Hamilton 0.9

Honestly, that's not as problematic as I remember, but still pretty problematic.

Thanks for the source, will explore more later!

[Djoker]

I'd add even more to that point - we should always remember that, whatever analytical tools we use to evaluate players, we always have massive error bars in our results. In the end, basketball isn't a hard sport, but it's very complex with a lot of variables that are more or less connected to each other. To unpack all of that, it's extremely hard and we just don't have the access to all data necessary.

That doesn't mean that we shouldn't try - in the end, I am running the project that tries to create an accurate ranking after all - but I think that we shouldn't push our conclusions too hard. It's very safe to say that someone like LeBron James is better than your average all-star. It's far, far harder to differentiate between the top 10 guys here. It's not impossible and we all do that, but we have to remember that we can be wrong on these things.

I think people should realise how extreme praise a player gets from the community if he makes the list, even at the very end. Pushing someone like Embiid, Harden, Manu or anyone else lower than you'd like doesn't mean you are a hater or fanboy - it just means that we reached different conclusions and we can all be wrong. It's not very likely and I think the final product is quite close to the reality, but it's not set in stone.

Couldn't agree more. We can be wrong and in fact it's likely that we often are wrong.

Do you have any sources / estimates of the error in 5-year RAPM being higher than lifetime RAPM? Like many have said in this thread (and indeed like many have had to repeat across multiple threads), normal RAPM assumes the player is constant. So as our sample size increases and the statistical standard error decreases, the assumption that the player is the same across the sample gets more and more faulty. SO by the time we're dealing with a 26-year-long RAPM, we're dealing with a case where that assumption becomes really quite undermining in terms of validity compared to something like 5-year RAPM (or even 11-year RAPM in Squared's case).

The faultiness of this assumption should be measurable somewhere. For example, if someone looked at RAPM across sample sizes (from 1-year to 26-year) and applied some sort of prediction / retrodiction tests to measure the stat's accuracy in some sense (e.g. here https://www.nbastuffer.com/analytics101/darko-daily-plus-minus/ or here https://fansided.com/2019/01/08/nylon-calculus-best-advanced-stat/ ; as an aside, is accuracy the right word? ), then I expect the most accurate RAPM would *not* be 26-year lifetime RAPM, as the assumption that a player's constant over the sample would be particularly undermining by this point. Just based on what NBA statisticians tend to prefer, I would expect 3-5 year RAPM to be the most accurate... although I've never found a study explicitly testing this.

I'm not sure how I would expect the error to behave. Part of me would expect the variability of the player over a lifetime sample to inflate the error more than the larger sample size decreases the error. But it also could go the other way. It might also depend on the RAPM version -- e.g. if there's priors, or box inputs, or anything like that. Since different RAPMs scale differently, the thing might be to look at the ranking ranges based on the +/- 2 SD uncertainty, like you did here.

Sounds like you've seen 5-year RAPM have a wider ranking range, so just curious to look at how the ranges compare.

As far as I know, Engelmann never published SE for his 5-year RAPM studies but just logically speaking, the standard deviation for 5-year RAPM would be smaller than 1-year RAPM but larger than lifetime RAPM.

As the words have been used so far in the thread including by DoctorMJ in those pretty bulls-eye graphics, validity = accuracy and reliability = precision. Accuracy and precision are words typically used by scientists and statisticians.

As we increase sample size, we improve the precision/reliability (reduce the standard deviations) but hurt the accuracy/validity because in the context of peak/prime discussion, lifetime RAPM can be irrelevant. Like you said, the sweet spot may be 5 years but as you also said, a study has never been done (as far as I know?) to try to figure out what the sweet spot is. Many players have solid 10-year primes so maybe that would be useful to see. I think 10-year RAPM would be pretty cool off the top of my head.

And even then it's tough because some players have 3 prime years with an injury year in the middle. How in the world do you deal with someone like Kawhi... 2017 then a blank then 2019 then 2020 then basically caputs!

Yep, this is a very good point to raise. Even the large-sample RAPM that is much more reliable than small-sample RAPM still has significant amounts of noise. It’s not that small-sample RAPM is pretty noisy and large-sample RAPM isn’t noisy, but rather that small-sample RAPM is *incredibly* noisy and large-sample RAPM is just pretty noisy.

I will say with RAPM that this sort of thing is why I like to see players that have a lot of great multi-year RAPM spans. To use a player that is relevant to this thread as an example, I quite like that Ginobili has 11 straight five-year spans that are top 10 in RAPM. As you point out, one 5-year span definitely still has a good bit of noise, but the chances of randomly repeatedly being near the top across so many different spans seems quite unlikely to me. You’d need variance to constantly be going in your favor. Intuitively, I think that’s actually even more convincing than someone being high up in lifetime RAPM.

This sort of thing is also why I generally do like to triangulate RAPM with box data and hybrid data. Basically, even if I tend to prefer large-sample RAPM over box data, we can have more confidence in our conclusion about a player whose RAPM and box data *both* look great. The same goes for just looking at lots of data sources in general. Every individual measure is noisy and flawed in serious ways, so I think looking at the whole picture is a good way to avoid indexing one’s view on whatever noise or flaws happen to exist in a particular measure.

I will note that, while I’ve long been a proponent of at least considering all available data, the paragraph just above admittedly does run a bit contrary to what I’ve been saying about Steve Nash in this thread and the last one—where I’ve largely argued for believing his RAPM data over his box data. If we triangulated the great impact data with the not-as-great box data, we’d be left with a conclusion somewhere in the middle—which would probably not be as positive a conclusion about Nash as I have. As I believe I mentioned in earlier discussions in the last thread, I’d normally take the somewhere-in-the-middle point of view in this sort of situation (though that in-the-middle spot would probably still be tilted more towards what the large-sample RAPM says). But with Nash I also just have a huge amount of eye test stuff (he’s probably one of the two or three players in NBA history that I’ve watched the most). And since my eye test is super high on Nash, that results in my assessment tilting much more towards the great RAPM data compared to less-good box data than I normally would with the typical player that I haven’t watched as much of (or that I don’t feel as strongly about the eye test on). The box data on Nash does still give me slight pause, but not nearly as much as it would without the extensive eye test I have on him.

Yep completely agree about the noise in general. But if a guy is doing well across multiple 5-year stretches, that's a big deal. So many consistently high finishes are less likely to be noise. Fantastic point to bring up.

[eminence]

By validity: APM > RAPM > XRAPM > Box
By reliability: Box > XRAPM > RAPM > APM

I miss long-sample classic, un-regularized, un-Xploited APM studies.

I'd add raw plus/minus and on/off on the outside of APM even, as the most valid stats, but suffering from extreme volatility.

Reasonable stuff to bring up. I wouldn't say those stats are more valid than APM as I don't think there's anything inherently invalidating about applying vanilla regression, but I wouldn't object to seeing them as equally valid to APM with considerably worse reliability.

I'd fall the other way, reasonably less reliability, but more valid than an apm (to be fair plus/minus and on/off are measuring something simpler than apm and variants).

[Cavsfansince84]

I want to make a further point which builds upon the idea of combining different data into a more reliable and likely more accurate aggregate. Whether its different 5 year rapm samples or combining +/- with on/off and box score is that each of us combines all of these inputs into a ballot. So if we look at each voter as their own type of metric we are then combining 15 or so of these into a vote. So it's not a big deal at all to greatly disagree with other voters in numerous ways. That's the strength of the system is we are combining 15+ different points of view into one single result. That's what is hopefully giving it some degree of validity after we've discussed things and given rationales for how we vote.

[Doctor MJ]

I'd add raw plus/minus and on/off on the outside of APM even, as the most valid stats, but suffering from extreme volatility.

Reasonable stuff to bring up. I wouldn't say those stats are more valid than APM as I don't think there's anything inherently invalidating about applying vanilla regression, but I wouldn't object to seeing them as equally valid to APM with considerably worse reliability.

I'd fall the other way, reasonably less reliability, but more valid than an apm (to be fair plus/minus and on/off are measuring something simpler than apm and variants).

Hmm, well, I've been using the term "validity" without specifying precisely what that means.

What I mean by it is that we're looking to have a holistic assessment of player impact without systematic bias, and thus with sufficient sample (unrealistic to reach to be clear) we should be able to expect optimized accuracy. In theory +/-, On-Off & APM should all get you there, whereas RAPM injects a bias with its prior, and box production data is always biased by role.

[Doctor MJ]

I want to make a further point which builds upon the idea of combining different data into a more reliable and likely more accurate aggregate. Whether its different 5 year rapm samples or combining +/- with on/off and box score is that each of us combines all of these inputs into a ballot. So if we look at each voter as their own type of metric we are then combining 15 or so of these into a vote. So it's not a big deal at all to greatly disagree with other voters in numerous ways. That's the strength of the system is we are combining 15+ different points of view into one single result. That's what is hopefully giving it some degree of validity after we've discussed things and given rationales for how we vote.

So just to make sure we're on the same page, I'm using "validity" and "accuracy" as synonyms (while "reliability" and "precision" are also synonyms).

When we combine different types of data, I see that as - hopefully - improving reliability at the cost of validity. Done well, it gives you a more effective all-in-one tool in practice, but it is also injecting some form of bias into the algorithm.

[Doctor MJ]

Yep completely agree about the noise in general. But if a guy is doing well across multiple 5-year stretches, that's a big deal. So many consistently high finishes are less likely to be noise. Fantastic point to bring up.

Lots of great stuff in your post, but wanted to hammer home this one:

Adjacent but non-overlapping spans of RAPM are a big deal for giving me confidence in the conclusions.

I'd rather have two adjacent 3-year studies than one 6-year, for example.

[Cavsfansince84]

So just to make sure we're on the same page, I'm using "validity" and "accuracy" as synonyms (while "reliability" and "precision" are also synonyms).

When we combine different types of data, I see that as - hopefully - improving reliability at the cost of validity. Done well, it gives you a more effective all-in-one tool in practice, but it is also injecting some form of bias into the algorithm.

I think no matter what words we use to describe various metrics its up to us as individuals to put it all together and take into account various blind spots/biases and stuff like that. So my point is that we aren't computers. We can see things that go beyond what numbers tell us, combine all of that and then distill that into a vote. I think the brevity of the reasoning for how people vote doesn't really come close to all the different input that goes into that vote. It's like a very small amount of it.

[Top10alltime]

They're good enough for top 15 all-time.

First round exit from 2024 is a top 15 postseason all-time?

That's not what I'm saying, I'm saying the postseason is good enough to push him in the top 15 peaks all-time. There are guys like Steph, Harden, Jokic, Shai, KG, and 2007 Dirk is not even good in the playoffs, so we tax Embiid (who, absolutely clears in the RS over all of these guys with the exception of 2016 Steph and Jokic, and is close in the playoffs to almost all of them), is just proof that realGM is biased and has an irrational hate for Embiid.

It is context to the situation, so it takes away less from them, than you would usually take away.

I don't agree with that. If you consistently happen to be injured in the playoffs, then it's what you give your team. It doesn't matter if it's caused by injury or your weaknesses if you can't overcome that.

Nice. Let's put 2021 Embiid in then. He performed nicely in the playoffs, and availability isn't a problem here to an extent. In the series (let's compare Embiid, RealGM's enemy, to Giannis, RealGM's homeboy, against the same opponent):

2021 Embiid vs Hawks (IA/75)
29.1 pts, 12.2 rbs, 3.7 asts on +2.2 rTS opp adj

Led +4.5 rORtg, and -7.6 rDRtg vs Hawks (opp adjusted, on-court)

2021 Giannis vs Hawks (IA/75)
28.6 pts, 10.8 rbs, 5.9 asts on +4.8 rTS opp adj

Led +3.0 rORtg, and -8.1 rDRtg vs Hawks (opp adjusted, on-court)

Not bad, Embiid! What about average playoffs performance comparison?

2021 Embiid (IA/75): 30.7 pts, 12.2 rbs, 3.7 ast on +6.3 rTS opp adj
Led +10.3 rORtg, and -7.5 rDRtg (opp-adjusted, on-court)

2021 Giannis (IA/75): 29.1 pts, 12.3 rbs, 5.0 ast on +3.0 rTS opp adj
Led -0.4 rORtg, and -11.6 rDRtg (opp-adjusted, on-court)

This is a playoffs dropper, BTW. Now, let's look at Embiid vs Wizards compared to "T10 finals performance ever" Giannis vs Suns

2021 Embiid (IA/75): 35.1 pts, 9.9 rbs, 3.7 asts on +17.5 opp adj rTS
Led +25.7 rORtg, and -7.3 rDRtg (opp-adjusted, on-court)

2021 Giannis (IA/75): 32.6 pts, 12.2 rbs, 4.6 asts on +9.0 opp adj rTS
Led +4.2 rORtg, and -8.3 rDRtg (opp-adjusted, on-court)

Sources: nbarapm.com



So 70sFan, if you still consider Embiid a playoffs dropper, and aren't reconsidering your thoughts, you better provide GOAT lvl reasoning here. He had all-time RS as well (high high end elite at worst). Embiid, at the absolutely lowest, top 15 here.

He had an ATG RS in 2024, as well.

He had an ATG 39 games in 2024, which isn't even a half of the full RS. I don't rate this season even as all-nba worthy, availability is more important than anything you can do.

Fine. Consider 2021 then. High end elite RS, and the description of his skillset written down here:

Embiid should already be here. It's a crime everyone underrated him, even though he had injuries to add context to the playoffs.....

Embiid should probably have gone #7, in place of Giannis (throw him outside the top 20, he's not versatile at all and his skillset neutralized against elite personnel).

Embiid is an elite scorer everywhere, his timing while shooting the jump shots is more valuable to an offense than extremely volatile tough shot-making, he is an elite shot-creator and can be useful in offensive systems (Like with Harden or Maxey). Unlike most big men, he doesn't have a weakness in face-up game. He is an ATG mid-range scorer, and a great post scorer (great mismatch hunter too). Elite footwork and strength makes him a force there. He is an amazing foul-drawer and punishes defenses for it too. ATG scorer at least.

As a playmaker, he has very good processing actually, and is at least a great playmaker because he has great rim pressure, as a roll man he is providing advantages for his team. As a screener, he's a great one. His passing arsenal is very good, bounce passing, pocket passing, overhead passing, PnR or off of screen, and a great transition playmaker when moving. Amazing decision making. Great playmaker at least.

Now, on the defensive end, he is elite, or maybe fringe ATG at defense. Very big with good mobility, an elite rim protector and as an interior defender overall. Pretty switchable, a good on-ball defender and help defender. Good perimeter defense, and ATG drop defense. Anchors top 5 defenses while on the floor too.


Embiid should clearly have been here by now, but people hate him for some reason. They put guys like Shai (bruh) and Giannis (terrible takes) over him.

Embiid is clearly the pick, and if there's another round without Embiid, I don't know what to say about realGM. Must be a group of anti-Embiid people.

So, here's that Embiid case. I don't think you could debunk anything about that.

Christ bless you all anyways, God loves you

And that weak postseason is better than anything from Drob too (who is in the top 25 peaks ever, easily)

This is not a thread for David Robinson, but even then I don't agree with that. Embiid missed 2 out of 11 games in that run and in the 7 games against Boston he had only one game with a positive +/-. Of course, his boxscore production took a massive hit, as always, and he ended the series with a horrible game 7.

If you think that's better than anything Robinson did, then don't put Robinson inside your top 25. I am not going to vote for Embiid anytime soon, we have plenty of great options left.

Woah, I wasn't talking about 2023. 2021 and 2024 playoffs is better than anything Robinson did in the playoffs. I will still put Robinson in top 25, due to his RS (just like Embiid, but worse), and decent enough playoffs to put him in.

Anyways, 70sFan, thank you for your time reading this, Christ bless you and your life here, and live nicely! Bye!

[Top10alltime]

This is ridiculous and a whole lie. This entire list is unjustifiable and there are guys like Harden and Dwight who are bigger playoff dropper than Embiid, including the context Embiid injured makes it worse!

there is literally no star in nba history who is a bigger playoff dropper than embiid. and i say that as someone who likes embiid. for complete careers (or almost complete), my resiliency spreadsheet has karl malone basically in a class of his own. the number from the calculation isn't really scaled to anything so let's just call it -100 as a baseline for karl malone. harden and giannis basically tie for next to last at -67. and then you have a group of guys like KD, nash, steph, robinson, and stockton in the -50 to -60 range. these are the fairly large playoff fallers. embiid after i think 2023 was at like -180! so he's in a class of his own away from karl malone, who is already way away from everyone else. now is some of it because every time he shows up to the playoffs he has a broken leg or a broken eye ball? sure. but the actual declines are enormous. like out of 416 playoff runs i looked at, 2022 embiid and 2023 embiid were 414th and 415th in resiliency.

Good. If you guys care about playoffs performance that much, look at 2021. You obviously don't like Embiid. And I wouldn't take this spreadsheet of resiliency seriously, unless I know what you used to determine this.

[Top10alltime]

I’ve not thought about it all that much and this isn’t an exhaustive list, but I think I’d probably have all of the following above Embiid (in no particular order):

Harden
Chris Paul
Ginobili
Nash
Durant
Anthony Davis
Draymond
Luka
Jimmy Butler
Tatum
Dwight Howard

After that, it gets a bit murkier and maybe I’d actually vote for Embiid. But there’d only be a couple spots left at that point, and guys like Westbrook, McGrady, and Kidd might go above him too. So I’m honestly not even sure I’ll be voting for Embiid at all. Which feels crazy to say, given how good he is when he plays and is healthy, but he’s just a wreck by the playoffs like every prime year, and I have a strong aversion to voting for years like that in a greatest-peaks project. I guess I should probably find my way to voting for 2021 Embiid, since his playoffs wasn’t actually bad that year, but even then he missed like 30% of the regular season and then his team got upset by a completely mediocre team in the playoffs (though to be fair, a lot of that was because of what happened with him off the court—they did solidly outscore the Hawks with him on).

This is ridiculous and a whole lie. This entire list is unjustifiable and there are guys like Harden and Dwight who are bigger playoff dropper than Embiid, including the context Embiid injured makes it worse!

You guys are clearly biased, and cannot analyze Embiid whenever we talk about him. Same with your homeboy Greek.

But when you guys are objective, it's good conversation

As I said, maybe 2021 is a potential year too. It’s his only prime year that doesn’t have a super glaring problem. But even then he missed a lot of regular season games and lost in the second round to a weak team. So it’s not exactly a banner year that I’d be voting for soon.

Once again, another instance we see the double standards when it comes to homeboy Greek, and enemy Embiid (for realGM, ofc). Let me give you an enlightenment...

Nice. Let's put 2021 Embiid in then. He performed nicely in the playoffs, and availability isn't a problem here to an extent. In the series (let's compare Embiid, RealGM's enemy, to Giannis, RealGM's homeboy, against the same opponent):

2021 Embiid vs Hawks (IA/75)
29.1 pts, 12.2 rbs, 3.7 asts on +2.2 rTS opp adj

Led +4.5 rORtg, and -7.6 rDRtg vs Hawks (opp adjusted, on-court)

2021 Giannis vs Hawks (IA/75)
28.6 pts, 10.8 rbs, 5.9 asts on +4.8 rTS opp adj

Led +3.0 rORtg, and -8.1 rDRtg vs Hawks (opp adjusted, on-court)

Not bad, Embiid! What about average playoffs performance comparison?

2021 Embiid (IA/75): 30.7 pts, 12.2 rbs, 3.7 ast on +6.3 rTS opp adj
Led +10.3 rORtg, and -7.5 rDRtg (opp-adjusted, on-court)

2021 Giannis (IA/75): 29.1 pts, 12.3 rbs, 5.0 ast on +3.0 rTS opp adj
Led -0.4 rORtg, and -11.6 rDRtg (opp-adjusted, on-court)

This is a playoffs dropper, BTW. Now, let's look at Embiid vs Wizards compared to "T10 finals performance ever" Giannis vs Suns

2021 Embiid (IA/75): 35.1 pts, 9.9 rbs, 3.7 asts on +17.5 opp adj rTS
Led +25.7 rORtg, and -7.3 rDRtg (opp-adjusted, on-court)

2021 Giannis (IA/75): 32.6 pts, 12.2 rbs, 4.6 asts on +9.0 opp adj rTS
Led +4.2 rORtg, and -8.3 rDRtg (opp-adjusted, on-court)

Sources: nbarapm.com



So, if you still consider Embiid a playoffs dropper, and aren't reconsidering your thoughts, you better provide GOAT lvl reasoning here. He had all-time RS as well (high high end elite at worst). Embiid, at the absolutely lowest, top 15 here.

Same Hawks team comparison, same year in the playoffs comparison, best series comparison, what's the outcome? 2021 Embiid, an "average playoffs performer", "choker", and "failure" beats out Giannis in 2021 whose apparently a "playoffs riser", "GOAT" and "clutch time player". Coincidence? Perhaps not, but I'll give you guys more. 2020 AD and 2025 Shai, and even 2011 Dirk, sound good? Aight, let's check it out

2021 Embiid (IA/75): 30.7 pts, 12.2 rbs, 3.7 ast on +6.3 rTS opp adj
Led +10.3 rORtg, and -7.5 rDRtg (opp-adjusted, on-court)

2020 AD (IA/75): 27.9 pts, 9.8 rbs, 3.5 ast on +10.1 rTS opp adj
Led +6.6 rORtg, and -6.8 rDRtg (opp-adjusted, on-court)

2025 Shai (IA/75): 29.3 pts, 5.2 rbs, 6.4 asts on +0.5 rTS opp adj
Led +1.3 rORtg and -12.0 rDRtg (opp-adjusted, on-court) (unlike 76ers, this team relied on defense (carried by Chet, JDub, Dort, Hart, Caruso) of other players, and not Shai)

2011 Dirk (IA/75): 28.4 pts, 8.4 rbs, 2.6 asts on +8.4 rTS
Led +7.8 rORtg and -6.8 rDRtg (opp-adjusted, on-court)

Yes, this is what a playoffs failure looks like. Please, I urge you guys to reconsider, and rewatch Embiid.

[lessthanjake]

Yep completely agree about the noise in general. But if a guy is doing well across multiple 5-year stretches, that's a big deal. So many consistently high finishes are less likely to be noise. Fantastic point to bring up.

Lots of great stuff in your post, but wanted to hammer home this one:

Adjacent but non-overlapping spans of RAPM are a big deal for giving me confidence in the conclusions.

I'd rather have two adjacent 3-year studies than one 6-year, for example.

So I guess this is getting a bit away from discussing peaks, but this reminds me of something I tallied up somewhat recently, so I figure I’ll just set it out here. Specifically, I looked at what players have 5-year RAPM (as per the NBArapm website) that was in the league’s top 10 in multiple different non-overlapping 5-year spans.

The list I got was the following:

Players with 2+ non-overlapping top 10 finishes in five-year RAPM

Shaquille O’Neal
Tim Duncan
Kevin Garnett
Dirk Nowitzki
LeBron James
Steph Curry
Nikola Jokic
Rasheed Wallace
Manu Ginobili
Steve Nash
Chris Paul
James Harden
Kevin Durant
Draymond Green
Kawhi Leonard
Paul George
Joel Embiid
Vince Carter

A caveat on a few of these guys: Several of these guys have one of the non-overlapping spans be a span where they did not actually play 5 years (i.e. perhaps they’re top 10 in five-year RAPM after their third year in the league). These include James Harden (one of his spans is 2008-2012, but he didn’t come into the league until 2010), Joel Embiid (one of his spans is 2016-2020, but he came into the league in 2017), and Kevin Durant (one of his spans includes the 2020 year he didn’t play). Curry also has one of his spans include the 2020 year he barely played in. Aside from Harden, all of these guys have four years in the data though. There are also other guys who missed a good number of games in the relevant timeframes (for instance, one of Paul George’s timeframes is 2021-2025, and obviously he missed a lot of games in that span).

We could narrow it down even further, to only look at players with three non-overlapping top 10 finishes in five-year RAPM:

Players with 3+ non-overlapping top 10 finishes in five-year RAPM

LeBron James
Kevin Garnett
Dirk Nowitzki
Tim Duncan
Chris Paul
Manu Ginobili

Tim Duncan has the same caveat as above, where one of his spans is a period he did not play the full 5 years in (one of his spans is 1997-2001, but he joined the league in 1998).

Of course, having three non-overlapping five-year spans requires a lot of longevity, so someone who is really great might not have three such spans because they just weren’t great for *that* long. It’s anlso essentially impossible for older players like Shaq to be on that list, since much of his prime was before the play-by-play era, and guys right in the middle of their careers right now (like Jokic) obviously aren’t going to have 3+ spans yet because they haven’t even played enough years. I’ll also note that LeBron is the only one with 4 such non-overlapping spans.

_____________

A lot of this is unsurprising. Obviously guys like LeBron, Garnett, and Duncan were really good over long time periods. And many of these guys have already been voted in. Some things that are particularly interesting to me:

1. Ginobili obviously looks incredible here. The company he is in for having 3+ spans is really elite. Chris Paul is the other guy not yet voted in here who is on that list. But I think Ginobili would probably be the biggest surprise in the 3+ list for most people.

2. The most surprising entries in the 2+ list have to be Rasheed Wallace and Vince Carter. Vince Carter is a very interesting one, since he was a really good player in his early years (being 7th in RAPM in a couple overlapping spans in his early years) and then managed to barely crack the top 10 in five-year RAPM in a couple much later spans that didn’t overlap with the early-years span (being 9th in both 2006-2010 and 2010-2014). Even those two later spans only barely overlapped. So yeah, oddly enough, Vince Carter almost had three non-overlapping spans finishing top 10 in five-year RAPM! A big surprise to me!

I believe Vince Carter and Draymond Green are the only ones who do not have any overlapping top 10 finishes in between their non-overlapping five-year spans. So like, while Carter had top 10 finishes in 2000-2004, 2006-2010, and 2010-2014, none of the five-year spans in between those were in the top 10. A similar thing is true of Draymond, who is top 10 in 2015-2019 and 2021-2025 and not top 10 in any of the timeframes in between those. For Draymond, that’s probably largely because 2020 needed to be washed out of the system in order for him to be top 10 again. For Vince Carter, I’m not entirely sure what it is. The data makes it look like 2005 threw him off. His on-off on the Raptors was awful that year so that was probably it.

3. Rasheed Wallace is another interesting case. I think he’s actually quite underrated and was one of the best players of his generation, and I’ll probably be considering him near the very end of the votes for this era in this project.

[iggymcfrack]

This is ridiculous and a whole lie. This entire list is unjustifiable and there are guys like Harden and Dwight who are bigger playoff dropper than Embiid, including the context Embiid injured makes it worse!

there is literally no star in nba history who is a bigger playoff dropper than embiid. and i say that as someone who likes embiid. for complete careers (or almost complete), my resiliency spreadsheet has karl malone basically in a class of his own. the number from the calculation isn't really scaled to anything so let's just call it -100 as a baseline for karl malone. harden and giannis basically tie for next to last at -67. and then you have a group of guys like KD, nash, steph, robinson, and stockton in the -50 to -60 range. these are the fairly large playoff fallers. embiid after i think 2023 was at like -180! so he's in a class of his own away from karl malone, who is already way away from everyone else. now is some of it because every time he shows up to the playoffs he has a broken leg or a broken eye ball? sure. but the actual declines are enormous. like out of 416 playoff runs i looked at, 2022 embiid and 2023 embiid were 414th and 415th in resiliency.

Good. If you guys care about playoffs performance that much, look at 2021. You obviously don't like Embiid. And I wouldn't take this spreadsheet of resiliency seriously, unless I know what you used to determine this.

Are you talking about the 2021 season where Embiid played 51 regular season games, played 95 minutes TOTAL in the series the Sixers won, and then blew a 2-1 lead to get upset in Round 2 by the 41-31 Hawks who would ultimately get bounced by the Giannis-less Bucks? And you're talking about the playoff run where on a rate basis, Embiid's box stats still didn't measure up to anyone in contention's peak except for Draymond or Nash?

In what world does that season compare to CP3, Manu, Draymond, AD, Westbrook, Harden, Luka, or Butler?

[lessthanjake]

Yep completely agree about the noise in general. But if a guy is doing well across multiple 5-year stretches, that's a big deal. So many consistently high finishes are less likely to be noise. Fantastic point to bring up.

Lots of great stuff in your post, but wanted to hammer home this one:

Adjacent but non-overlapping spans of RAPM are a big deal for giving me confidence in the conclusions.

I'd rather have two adjacent 3-year studies than one 6-year, for example.

So I guess this is getting a bit away from discussing peaks, but this reminds me of something I tallied up somewhat recently, so I figure I’ll just set it out here. Specifically, I looked at what players have 5-year RAPM (as per the NBArapm website) that was in the league’s top 10 in multiple different non-overlapping 5-year spans.

The list I got was the following:

Players with 2+ non-overlapping top 10 finishes in five-year RAPM

Shaquille O’Neal
Tim Duncan
Kevin Garnett
Dirk Nowitzki
LeBron James
Steph Curry
Nikola Jokic
Rasheed Wallace
Manu Ginobili
Steve Nash
Chris Paul
James Harden
Kevin Durant
Draymond Green
Kawhi Leonard
Paul George
Joel Embiid
Vince Carter

A caveat on a few of these guys: Several of these guys have one of the non-overlapping spans be a span where they did not actually play 5 years (i.e. perhaps they’re top 10 in five-year RAPM after their third year in the league). These include James Harden (one of his spans is 2008-2012, but he didn’t come into the league until 2010), Joel Embiid (one of his spans is 2016-2020, but he came into the league in 2017), and Kevin Durant (one of his spans includes the 2020 year he didn’t play). Curry also has one of his spans include the 2020 year he barely played in. Aside from Harden, all of these guys have four years in the data though. There are also other guys who missed a good number of games in the relevant timeframes (for instance, one of Paul George’s timeframes is 2021-2025, and obviously he missed a lot of games in that span).

We could narrow it down even further, to only look at players with three non-overlapping top 10 finishes in five-year RAPM:

Players with 3+ non-overlapping top 10 finishes in five-year RAPM

LeBron James
Kevin Garnett
Dirk Nowitzki
Tim Duncan
Chris Paul
Manu Ginobili

Tim Duncan has the same caveat as above, where one of his spans is a period he did not play the full 5 years in (one of his spans is 1997-2001, but he joined the league in 1998).

Of course, having three non-overlapping five-year spans requires a lot of longevity, so someone who is really great might not have three such spans because they just weren’t great for *that* long. It’s anlso essentially impossible for older players like Shaq to be on that list, since much of his prime was before the play-by-play era, and guys right in the middle of their careers right now (like Jokic) obviously aren’t going to have 3+ spans yet because they haven’t even played enough years. I’ll also note that LeBron is the only one with 4 such non-overlapping spans.

_____________

A lot of this is unsurprising. Obviously guys like LeBron, Garnett, and Duncan were really good over long time periods. And many of these guys have already been voted in. Some things that are particularly interesting to me:

1. Ginobili obviously looks incredible here. The company he is in for having 3+ spans is really elite. Chris Paul is the other guy not yet voted in here who is on that list. But I think Ginobili would probably be the biggest surprise in the 3+ list for most people.

2. The most surprising entries in the 2+ list have to be Rasheed Wallace and Vince Carter. Vince Carter is a very interesting one, since he was a really good player in his early years (being 7th in RAPM in a couple overlapping spans in his early years) and then managed to barely crack the top 10 in five-year RAPM in a couple much later spans that didn’t overlap with the early-years span (being 9th in both 2006-2010 and 2010-2014). Even those two later spans only barely overlapped. So yeah, oddly enough, Vince Carter almost had three non-overlapping spans finishing top 10 in five-year RAPM! A big surprise to me!

I believe Vince Carter and Draymond Green are the only ones who do not have any overlapping top 10 finishes in between their non-overlapping five-year spans. So like, while Carter had top 10 finishes in 2000-2004, 2006-2010, and 2010-2014, none of the five-year spans in between those were in the top 10. A similar thing is true of Draymond, who is top 10 in 2015-2019 and 2021-2025 and not top 10 in any of the timeframes in between those. For Draymond, that’s probably largely because 2020 needed to be washed out of the system in order for him to be top 10 again. For Vince Carter, I’m not entirely sure what it is. The data makes it look like 2005 threw him off. His on-off on the Raptors was awful that year so that was probably it.

3. Rasheed Wallace is another interesting case. I think he’s actually quite underrated and was one of the best players of his generation, and I’ll probably be considering him near the very end of the votes for this era in this project.

Just to add further to this, since I realize I’d previously dug up the following as well:

Here’s the same thing as above, except looking at players with 2+ non-overlapping spans in which they were top 3 in RAPM (rather than merely top 10 in RAPM). This is again using the NBArapm website’s RAPM:

Players with 2+ non-overlapping top 3 finishes in five-year RAPM

LeBron James
Manu Ginobili
Tim Duncan
Kevin Garnett
Stephen Curry
Chris Paul

Interestingly, this list is virtually identical to the list of players that had 3+ non-overlapping top 10 finishes. Which is probably not a surprise, since virtually every player would only get that third non-overlapping top 10 finish in years that weren’t really their prime, so they’d likely need to be *easily* making the top 10 in their prime to still finish in the top 10 when largely outside it.

But again, I think Manu really sticks out here as the most interesting one. The names are basically Manu + Chris Paul + the four best play-by-play era players that aren’t either still in the middle of their career (Jokic) or had a lot of their prime before the play-by-play era (Shaq). Definitely reflects well on Chris Paul too, but Manu seems like the most interesting one.

[Top10alltime]

there is literally no star in nba history who is a bigger playoff dropper than embiid. and i say that as someone who likes embiid. for complete careers (or almost complete), my resiliency spreadsheet has karl malone basically in a class of his own. the number from the calculation isn't really scaled to anything so let's just call it -100 as a baseline for karl malone. harden and giannis basically tie for next to last at -67. and then you have a group of guys like KD, nash, steph, robinson, and stockton in the -50 to -60 range. these are the fairly large playoff fallers. embiid after i think 2023 was at like -180! so he's in a class of his own away from karl malone, who is already way away from everyone else. now is some of it because every time he shows up to the playoffs he has a broken leg or a broken eye ball? sure. but the actual declines are enormous. like out of 416 playoff runs i looked at, 2022 embiid and 2023 embiid were 414th and 415th in resiliency.

Good. If you guys care about playoffs performance that much, look at 2021. You obviously don't like Embiid. And I wouldn't take this spreadsheet of resiliency seriously, unless I know what you used to determine this.

Are you talking about the 2021 season where Embiid played 51 regular season games, played 95 minutes TOTAL in the series the Sixers won, and then blew a 2-1 lead to get upset in Round 2 by the 41-31 Hawks who would ultimately get bounced by the Giannis-less Bucks? And you're talking about the playoff run where on a rate basis, Embiid's box stats still didn't measure up to anyone in contention's peak except for Draymond or Nash?

In what world does that season compare to CP3, Manu, Draymond, AD, Westbrook, Harden, Luka, or Butler?

If you think it doesn't compare, you didn't watch Embiid and completely have an irrational hate for him. Unsurprising , since this is RealGM.