New Impact Metric: MAMBA

[lessthanjake]

They are actually. Though I think they prefer I leave it ambiguous.

Creator made a comment some might find interesting on this a couple months ago:


They expound on this in the original explanation as well (though they had a longer section on lebron specifically they cut out)

Similar with what we see with LEBRON and RAPTOR where 15-17 Steph vs Lebron generally depends on to the degree available box-scores are weighed.

What you’re quoting there effectively just amounts to saying that, while Steph did better than LeBron in this metric, LeBron did better than Steph specifically in the RAPM component of the metric from 2015-2017.

And your statement effectively amounts to saying "Lebron looks better by results, steph looks better by a box-score", as we've seen with every one of these so far. it improving the data for 400 players doesn't change that. Box-scores are still sophisticated eye-tests at the end of the day. Validity of that for specific comparisons has been legislated plenty and will be legislated more. but the impact component is the only part relevant for titles like "impact king".

If it improves the data overall, then it *likely* improves the data as it relates to comparing any two players (though there is of course a chance it won’t). This is in part because an individual player’s measured impact is highly dependent on how the model evaluates all the other players that are on the court with and without them, and so a model that improves the data “for 400 players” will be likely to improve the data as it relates to any individual player too, unless the prior is really off for that player.

In any event, logically, the argument you’re making leads to a conclusion that we should use pure RAPM with no prior, since your argument is that pure RAPM amounts to “results” and any prior isn’t really “impact.”

But the problem there is that we have pure RAPM with no prior on the BasketballDatabase website, and it too tells us that prime Steph had better impact than LeBron. Here is who is ahead in three-year and five-year pure RAPM for each time period that is from 2014 onwards:

Pure RAPM: Three-Year

2014-2016: Curry
2015-2017: Curry
2016-2018: Curry
2017-2019: Curry
2018-2020: Curry
2019-2021: LeBron
2020-2022: Curry
2021-2023: Curry
2022-2024: Curry

Pure RAPM: Five-Year

2014-2018: Curry
2015-2019: Curry
2016-2020: Curry
2017-2021: Curry
2018-2022: Curry
2019-2023: Curry
2020-2024: LeBron

In other words, Steph was clearly ahead of LeBron in pure RAPM in the last decade. Of course, even leaving aside all-in-ones, most RAPM measures aren’t actually pure RAPM. Instead, they usually have some sort of prior, often using basic things like minutes per game and whatnot. Those priors seem to often improve LeBron’s standing relative to Steph, compared to what we see in pure RAPM. But applying your logic, these priors that tend to help LeBron aren’t actually “impact” and it’s the pure RAPM we should look at instead. The problem is that the pure RAPM says Steph is more impactful.

This leaves you with a bit of a conundrum. The impact metrics with the most sophisticated priors (such as this MAMBA metric) have prime Steph consistently ahead of LeBron. Meanwhile, the impact metrics without any priors also have prime Steph consistently ahead of LeBron. You are therefore left arguing in favor of the impact metrics with less sophisticated priors (because they tend to show more even results between the two players). Not exactly a good rhetorical position to be in. And that’s perhaps why you often actually just reject the usefulness of RAPM entirely, in favor of other forms of analysis that you think can get you to the conclusion you want.

Fwiw, creator made an argument for lebron much like yours for Moses. Emphasis mine.

A case example. Awhile ago, I saw a pretty bad Article on BBall index.com. Now, I do really enjoy the site and like what it stands for, and to be clear, this WAS NOT WRITTEN BY TIM (also known as Cranjis Mcbasketball). Tims a smart guy and he’s pretty chill to talk to so he wouldnt write something like this, but the gist of the article was basically one of the other writers clickbaiting off of the olympics doing a “Lebrons not top 10 and I’ll tell you why with FACTS and STATS” and it just being a guy pulling out the LEBRON metric…

But it actually is relevant to this, because Lebron represents probably the clearest example (That I know of) of a high profile player that represents a bias. While I don’t want to go on a 10 page tangent defending Lebrons honor from LEBRON on a spreadsheet in Capslock, what I’ll say is that, especially on the defensive end, for pretty much his entire post Miami career (at the very least),any available “Box Score” component for an all in one of Lebron’s data severely undershoots him defensively. The 2 exceptions, 2018 and 2022, are the only years where his actual adjusted defensive impact data wasn’t good (according to RAPM). This is the case for LEBRON, DPM, and Mine (I’ll release the overall numbers, I can give the priors to anyone who asks but this is a first draft still so need to do some tuning) etc. . On a deeper level though, despite his great box scores, what you end up seeing fairly consistently is the more you weight box scores, the less impressive his All-in-One data can be. This doesn’t mean “Hey maybe his impact data overrates him” because that’s really not how it works if it’s this consistent for long periods of time for a high production player, it means Lebron is better than his box score production indicates. To be clear, Lebron’s career age adjusted impact data is by far the greatest in history, and if you only get playoff RAPM (there are caveats to doing it that way beyond the scope of this post), he’s basically a lone dot at the top even without adjusting for age, and that’s with him being in LeCoast mode in the Regular Season since 2014. All in one data ironically shrouds the case here, but for his Career Lebron is pretty much the Undisputed king in the realm of impact data (Although obviously now he’s no longer undisputed #1 there). I’[b]m sure there are other examples (I feel KG would be another guy?)

I don’t know that this argument is really right. Again, let’s go to pure RAPM from BasketballDatabase for LeBron’s post-Miami career. LeBron’s league ranking in three-year DRAPM for those years looks like this:

2015-2017: 51st
2016-2018: 532nd (!!!)
2017-2019: 267th (!!!)
2018-2020: 53rd
2019-2021: 5th
2020-2022: 30th
2021-2023: 97th
2022-2024: 204th

LeBron’s league ranking for five-year DRAPM for those years looks like this:

2015-2019: 153rd
2016-2020: 49th
2017-2021: 38th
2018-2022: 96th
2019-2023: 26th
2020-2024: 29th

His post-Miami pure DRAPM doesn’t look very good overall (and indeed is outright bad in the first several years). I don’t think all-in-ones are undershooting this. I think what all-in-ones might undershoot is DRAPM that uses some priors. But making this type of argument about that doesn’t really make much sense, since it basically amounts to saying a sophisticated prior must be wrong because it differs from a less sophisticated prior.

[lessthanjake]

Volume alone means only so much. And defense is a thing.

Also, the 2019 Bucks won 7 more games than the Rockets, and Giannis WAS a 28/12/5/6 guy on 64.4% TS himself, 2nd in the DPOY race and all that.

"Robbed" is a big word, which is violently inappropriate. One can make an argument that Harden should have won the MVP that year, but "robbed" is very much not an accurate description.

60 wins is nothing special historically. Im old enough to remember when there were about 4 or 5 60 win teams in a season. Those suns/mavs/spurs teams of the early 2000's regularly reached those win totals

What harden did on the other hand was historic

As for defense, for what it's worth harden averaged 2.1 steals a game that season

Ah, that thing which barely correlates with team defensive success!

I don’t feel strongly at all about the actual underlying discussion (i.e. Harden vs. Giannis in 2019), but I do want to reiterate that this assertion you sometimes make about steals “barely correlat[ing] with team defensive success” is deeply flawed.

We have plenty of data telling us that an individual’s steals are very impactful: https://forums.realgm.com/boards/viewtopic.php?f=6&t=2387860&p=113832610&hilit=Steals#p113832610

Spoiler:

I think it’s worth noting that, yes, steals are very impactful. This is pretty obvious, but I saw a claim here otherwise.

538 did an article on this a while back: https://fivethirtyeight.com/features/the-hidden-value-of-the-nba-steal/. They also did some follow up stuff on it, which you can find links to here: https://fivethirtyeight.com/features/steals-are-predictive-but-are-they-that-important/. There’s a lot to this analysis, but basically they regressed impact on box stats and found steals to be the stat that’s most predictive of impact, and then there’s some analysis of why that might be (including that it’s not obviously replaceable like a lot of other box stats are).

I would say that BPM actually is indicative of something similar, with it basically being based on box stats regressed against RAPM, and having steals with a very high coefficient, including as part of DBPM specifically (so it’s really not just about the offensive benefit of steals). So this is yet more indication that steals are highly impactful defensively.

…

This is intuitively obvious, since steals genuinely end the opponent’s possession, and, at the individual level, are something that are probably uniquely (though not entirely) attributable to the player that does it, compared to other stats. People sometimes counterargue that with steals comes gambling and gambling leads to easy baskets for the opponent, but that’s already priced into the above data that finds that steals correlate quite well with impact.

One thing I want to add to this is that I think the flip side of the gambling thing is that players who go for disruptive plays a lot will have a deterring effect on opponents. If you know that a guy who is fantastic at getting in the passing lanes is lurking nearby, you are going to be less likely to try to make a difficult pass that might lead to an easy bucket. If you choose not to make that pass, the disruptive defender who deters it gets nothing in the stat sheet, but the result of that decision will likely end up being a less efficient shot attempt on that possession. That is impactful. In a sense, this is like the effect of an elite cornerback in American football—their value is less about the interceptions and more about the fact that they shut down whole areas of the field because the offense is afraid that they’ll intercept it. Indeed, Kenny Smith invoked this analogy to Jordan—calling him “the Deion Sanders of basketball” because teams didn’t want to even run plays on his side of the floor. This is also conceptually similar to deterrence by a rim protector—the fact that people are afraid of the blocks alters behavior in a way that actually probably has more impact than the blocks themselves that end up in the stat sheet. Intuitively, I’d think that this deterrence is at its largest with guys that steal balls in the passing lanes—which is something Jordan was incredible at. But it can also come into play with forcing turnovers by ball-handlers. If a player you’re guarding is afraid that you’ll pick his pocket if he drives, he’s more likely to settle for a contested jump shot. Similarly, if a player in the post is afraid that they’ll be stripped when the help comes, they’re more likely to settle for a worse attempt, before the help gets there. Again, when they make that decision, it doesn’t show up in the scoresheet for the guy who deterred the player, but it will have defensive impact.

Since you know this, you carefully word this assertion to talk about barely correlating with “team defensive success.”

But, as an initial matter, that’s not right. There is actually a correlation between a team’s steal rate and their defensive rating:

Spoiler:

There’s also some analysis that has been run at a team level that back up the value of steals on defensive rating: https://m.numberfire.com/nba/news/3425/the-correlation-between-team-stats-and-offensive-and-defensive-efficiency-part-1-steals

And, in any event, it doesn’t actually logically follow that a correlation (or lack thereof) between team stats and team results must correspond to a correlation (or lack thereof) between individual stats and team results. See here for discussion about this: https://forums.realgm.com/boards/viewtopic.php?p=113844175#p113844175.

Spoiler:

I’d also note that team-level data is not the same as individual-level data for these purposes. The point is to assess how much impact an individual player’s steals have, so obviously data on the impact of an individual’s steals on team results is most relevant! I’ve not seen any of the info you’re referring to, and you don’t link to it here, while I linked to something that says otherwise. However, team steals very well may not correlate as well with impact as individual steals do, for a lot of team-level reasons. Team-level steals stats will depend a lot on how the team gets attacked. To give one example, steals are more common in transition, so if a team gets stuck defending in transition a lot (maybe they’re bad at getting back, just bad at offense, or perhaps they tend to force the game to a high pace), they’ll likely rack up steals but give up lots of points. Steals are also relatively common in pick and roll actions, so if a team is easy to just effectively attack with PnR, we’d expect teams to attack them that way more, in which case they’d likely give up more points but have more steals. Relatedly, at a team level, higher steals can also be caused by running certain types of lineups where the plus is more steals but there’s inherently a minus that comes along with it—for instance, a team that goes for small ball will likely get more steals but give back that impact in other ways (giving up mismatches, bad rebounding, etc.). It’s also the case that we wouldn’t even necessarily expect the really-high-impact steals guys to have teams with tons of steals overall, because the highest-impact steals guys will deter risky passes from being made in the first place. The value there will actually come from worse opponent FG%’s, fewer FTs, etc., as a result of high-value actions being deterred, rather than their team getting more steals. This can mean that the teams with the most steals aren’t actually always the teams with the most impactful guys in this regard—instead, they actually may include teams that opposing teams take tons of risky actions against because they’re not all that deterred. The result there might be relatively more steals than if those actions *were* deterred, but worse defense overall, since the high-value actions would be succeeding a lot (hence why opposing teams are trying them more). Looking at the impact of steals at an individual level will do a better job of picking up on this stuff, and avoiding being confused by team-level factors. It’s also just obviously more directly relevant to the question at hand.

The point you were responding to is about individual steals. We know that there is a correlation between an individual’s steals and team results. It does not make a lot of sense to make an argument to the contrary by talking about correlation between teams’ steals and team results (even if there wasn’t also evidence to the contrary on even that claim), since there’s a lot of reasons that a bad defense might get more steals (for instance, as noted in the spoilered portion above, small-ball will tend to lead to bad defense, while nevertheless generating more steals; steals are more common in transition but obviously transition offense is highly efficient, so a team that is bad at getting back will tend to give up more points but have more steals; etc.)

[AEnigma]

What you’re quoting there effectively just amounts to saying that, while Steph did better than LeBron in this metric, LeBron did better than Steph specifically in the RAPM component of the metric from 2015-2017.

And your statement effectively amounts to saying "Lebron looks better by results, steph looks better by a box-score", as we've seen with every one of these so far. it improving the data for 400 players doesn't change that. Box-scores are still sophisticated eye-tests at the end of the day. Validity of that for specific comparisons has been legislated plenty and will be legislated more. but the impact component is the only part relevant for titles like "impact king".

If it improves the data overall, then it *likely* improves the data as it relates to comparing any two players (though there is of course a chance it won’t). This is in part because an individual player’s measured impact is highly dependent on how the model evaluates all the other players that are on the court with and without them, and so a model that improves the data “for 400 players” will be likely to improve the data as it relates to any individual player too, unless the prior is really off for that player.

In any event, logically, the argument you’re making leads to a conclusion that we should use pure RAPM with no prior, since your argument is that pure RAPM amounts to “results” and any prior isn’t really “impact.”

But the problem there is that we have pure RAPM with no prior on the BasketballDatabase website, and it too tells us that prime Steph had better impact than LeBron. Here is who is ahead in three-year and five-year pure RAPM for each time period that is from 2014 onwards:

Pure RAPM: Three-Year

2014-2016: Curry
2015-2017: Curry
2016-2018: Curry
2017-2019: Curry
2018-2020: Curry
2019-2021: LeBron
2020-2022: Curry
2021-2023: Curry
2022-2024: Curry

Pure RAPM: Five-Year

2014-2018: Curry
2015-2019: Curry
2016-2020: Curry
2017-2021: Curry
2018-2022: Curry
2019-2023: Curry
2020-2024: LeBron

In other words, Steph was clearly ahead of LeBron in pure RAPM in the last decade… The problem is that the pure RAPM says Steph is more impactful.

… Do you just assume no one fact checks anything?

On that site you linked, and starting in 2014, Lebron is ahead of Steph in 2014-15 RAPM (omitted), 2015-16 RAPM (omitted), 2014-16 RAPM (directly contrary to what you claimed), 2016-17 RAPM (omitted), 2015-17 RAPM (directly contrary to what you claimed), 2014-17 RAPM (omitted), 2019-20 RAPM (omitted), 2016-20 RAPM (directly contrary to what you claimed), 2020-21 RAPM (omitted), 2018-21 RAPM (omitted), 2017-21 RAPM (directly contrary to what you claimed), 2020-22 RAPM (directly contrary to what you claimed), 2020-23 RAPM (omitted), 2023-24 RAPM (omitted), and 2021-24 RAPM (omitted), in addition to the 2019-21 and 2020-24 RAPM you did bother to acknowledge and accurately represent.

Pathological.

[OhayoKD]

60 wins is nothing special historically. Im old enough to remember when there were about 4 or 5 60 win teams in a season. Those suns/mavs/spurs teams of the early 2000's regularly reached those win totals

What harden did on the other hand was historic

As for defense, for what it's worth harden averaged 2.1 steals a game that season

Ah, that thing which barely correlates with team defensive success!

I don’t feel strongly at all about the actual underlying discussion (i.e. Harden vs. Giannis in 2019), but I do want to reiterate that this assertion you sometimes make about steals “barely correlat[ing] with team defensive success” is deeply flawed.

Uhuh.

We have plenty of data telling us that an individual’s steals are very impactful: https://forums.realgm.com/boards/viewtopic.php?f=6&t=2387860&p=113832610&hilit=Steals#p113832610
(including that it’s not obviously replaceable like a lot of other box stats are).

Your source describes a regression using WOWY of an unknown sample of unknown players presumably including, and hopefully not limited to, 2014 Ricky Rubio and Kevin Love suggesting steals are better predictors than other common box-score stats if you simply look at 1 of each. What it doesn't do is measure whether steals are very impactful":

“Because we’re particularly interested in how each stat compares with points scored, I’ve set the predictive value of a single marginal point as our unit of measure (that is, the predictive value of one point equals one, and something five times more predictive than a point is five, etc.).”


The actual result is that the regression coefficient for a steal has 9 times the magnitude of the regression coefficient for a point scored in a regression predicting the final margin of victory/defeat. It’s disappointing that this wasn’t in the main body of the article.

https://statmodeling.stat.columbia.edu/2014/03/26/steal-really-worth-9-points/#:~:text=Even%20if%20taken%20at%20face%20value%2C%20the%20results,are%20also%20good%20at%20other%20aspects%20of%20defense.


He then released a follow-up saying:

One of the more surprising side-findings in my analysis was that steals don’t seem to predict much about defense at all

A shame he considered an actual measure of "very impactful" too much of a side-finding to share the results but others have picked up the slack:
https://statpadders.com/which-team-stats-correlate-most-strongly-with-winning/

What we put together isn’t a complete list by any means (we won’t go into more advanced metrics here for example), but it does provide a good look at how certain team stats have impacted winning percentage in the NBA. To quickly summarize, we plotted every team’s per game production in a given statistic (x-axis) against their winning percentage from that same season (y-axis). Every team from every year since, and including, 1997 was plotted, which leaves us with a sample of roughly 700 individual seasons.

Steals fall at the bottom in this group of stats, and it makes sense. We would likely see a very similar plot for blocks. Why? There are just so few steals and blocks in any given game. To put this perspective, look at that data point all the way to the right. The ’98 Celtics averaged 12 steals per game, the most any team has averaged in the last twenty-four seasons. Andre Drummond has averaged more than 13 rebounds per game by himself every year since 2014. The overwhelming majority of possessions end in a shot (or a dead ball turnover), and that drowns out the impact of steals (and blocks) over the full course of a game. That doesn’t mean they aren’t important when they occur – they can often be crucial for jump starting the fast break and easy scoring opportunities – but the volume is just too low for a massive overall impact.

https://statpadders.com/do-team-steals-really-matter/

Let’s start with the positive effect. Does racking up a number of steals tend to result in wins?

The answer is pretty clear – no, at least not with any consistency. The data doesn’t leave much room for debate. There’s simply no correlation between placing near the top of the league in steals and doing the same in wins.

https://statpadders.com/which-team-stats-correlate-most-strongly-with-winning/

Here we see two tiny R squared values, which act as a measure of correlation. To put it simply, the bigger the R squared value, the better the data fits the spotted line you see above. The value falls on a scale of 0 to 1, and we see two numbers very close to 0 here, suggesting little to no correlation.

Just like last time when I put in your link I got this:


Luckily I was able to access the article by googling the title and clicking a google link:

However, when expanded to 10 years of data, the trend becomes less apparent. There is still a positive correlation; however, it shrinks significantly. In the past 10 years, a steal would, on average, lead to an extra 0.27 points per 100 possessions...

...

So as our R² drops from 0.098 in the first two years to 0.005 in the total ten, an inference can be made that while it still shows a positive correlation, there in fact might be no correlation.

What happened between the 2-year data and the 10-year data in regards to offenses, was the opposite with defenses. While the data after two years showed correlation between steals and good defenses, it wasn’t significant enough and led me to believe there was no correlation. However, once the 10-year data was graphed, it shows that there might be a correlation after all

...

Our R² value for defenses is still not great[0.06], but it does show more consistency than our offensive data. On average, in the past 10 years, a steal has equated roughly to a savings of 1.02 points per 100 possessions

...

The jury is still out on the value of steals, and the data is far from definitive. We have some years like last season, where a top-three steals team in the 76ers ranked bottom-five in adjusted defensive rating. However, some years we have the exact opposite.

Going by your source, yeah, there's a correlation but it's a weak one with neither offense or defense reaching an rs squared of even .1

FWIW there was a tracking as opposed to impact approach to this too I found where a team of scouts tried to track every defensive action over hundreds of games:
https://www.sportsinfosolutions.com/2021/07/07/rethinking-defensive-impact/

At SIS, we’re lucky to have a staff of basketball experts capable of consuming and evaluating the hundreds of thousands of possessions prospects put on film each draft season.

Among others, we have at our disposal an overall defensive impact statistic called Defensive Winning Impact (DWIMP). Instead of approximating what we don’t see, we’re empowered to watch everything and log every defensive contribution a prospect makes (or fails to make) for an entire season. With such a powerful metric at our disposal, we are enabled to investigate questions like the validity of stock percentage as a defensive proxy.

Our findings? Neither stock percentage nor its component parts are particularly useful stand-ins for defensive impact

Meanwhile, the limitations of what blocks and steals don’t capture far outweigh what they do. Typically, blocks and steals provide results-based information on the conclusions of 2-3% of a prospect’s defensive possessions, while entirely missing the rest of the approximately 80% of possessions in which a prospect engages in at least one meaningful defensive action, according to our research.

Regardless, it seems what you find "intuitively obvious" is not actually particularly obvious. As of now
no study has been provided or even alluded to suggesting a strong correlation between winning and steals (highest r-value came from first article at 0.11), and the "massively impactful" citation is not coming from a measure of actual impact or correlation but a measure of how individual steals compare to individual other weakly correlated box-stats.

And, in any event, it doesn’t actually logically follow that a correlation (or lack thereof) between team stats and team results must correspond to a correlation (or lack thereof) between individual stats and team results. See here for discussion about this:

[/quote][/quote]
Individual stats is not what is being discussed. What is being discussed is what steals say about an individual's impact. If the delta between steals for teams says little about a team's defensive (or overall) success, then that puts an immediate cap on what we can generally derive regarding the value of players who contribute to that delta.

Never mind individual cases like the shooting guard who gambled a bunch and was blown by a bunch racking them up, was being given teammate steals (I'm guessing those are rather replaceable actually!), and happens to have the biggest single-season outlier in terms of home-cooked steals for the era.

All due respect to Kenny.

[lessthanjake]

And your statement effectively amounts to saying "Lebron looks better by results, steph looks better by a box-score", as we've seen with every one of these so far. it improving the data for 400 players doesn't change that. Box-scores are still sophisticated eye-tests at the end of the day. Validity of that for specific comparisons has been legislated plenty and will be legislated more. but the impact component is the only part relevant for titles like "impact king".

If it improves the data overall, then it *likely* improves the data as it relates to comparing any two players (though there is of course a chance it won’t). This is in part because an individual player’s measured impact is highly dependent on how the model evaluates all the other players that are on the court with and without them, and so a model that improves the data “for 400 players” will be likely to improve the data as it relates to any individual player too, unless the prior is really off for that player.

In any event, logically, the argument you’re making leads to a conclusion that we should use pure RAPM with no prior, since your argument is that pure RAPM amounts to “results” and any prior isn’t really “impact.”

But the problem there is that we have pure RAPM with no prior on the BasketballDatabase website, and it too tells us that prime Steph had better impact than LeBron. Here is who is ahead in three-year and five-year pure RAPM for each time period that is from 2014 onwards:

Pure RAPM: Three-Year

2014-2016: Curry
2015-2017: Curry
2016-2018: Curry
2017-2019: Curry
2018-2020: Curry
2019-2021: LeBron
2020-2022: Curry
2021-2023: Curry
2022-2024: Curry

Pure RAPM: Five-Year

2014-2018: Curry
2015-2019: Curry
2016-2020: Curry
2017-2021: Curry
2018-2022: Curry
2019-2023: Curry
2020-2024: LeBron

In other words, Steph was clearly ahead of LeBron in pure RAPM in the last decade… The problem is that the pure RAPM says Steph is more impactful.

… Do you just assume no one fact checks anything?

On that site you linked, and starting in 2014, Lebron is ahead of Steph in 2014-15 RAPM (omitted), 2015-16 RAPM (omitted), 2014-16 RAPM (directly contrary to what you claimed), 2016-17 RAPM (omitted), 2015-17 RAPM (directly contrary to what you claimed), 2014-17 RAPM (omitted), 2019-20 RAPM (omitted), 2016-20 RAPM (directly contrary to what you claimed), 2020-21 RAPM (omitted), 2018-21 RAPM (omitted), 2017-21 RAPM (directly contrary to what you claimed), 2020-22 RAPM (directly contrary to what you claimed), 2020-23 RAPM (omitted), 2023-24 RAPM (omitted), and 2021-24 RAPM (omitted), in addition to the 2019-21 and 2020-24 RAPM you did bother to acknowledge and accurately represent.

Pathological.

What are you talking about? My post clearly referred to “the BasketballDatabase website,” which can be found here: https://thebasketballdatabase.com/index.html. I used this because the website specifically says the following, which makes very clear it uses raw RAPM without any prior (other websites are less completely clear-cut about this or definitely do have a prior): “The values presented on this site are "vanilla" RAPM - the result of a pure ridge regression without any priors to tell the regression what the results "should" look like.”

I referred to three-year and five-year RAPM on that website. You claim I was wrong about 2014-2016, 2015-2017, 2016-2020, 2017-2021, and 2020-2022. This is all false.

Let’s take those one-by-one:

1. For 2014-2016, BasketballDatabase RAPM has Curry ahead of LeBron: https://thebasketballdatabase.com/2015-16RegularSeasonPlayerThreeYearRAPM.html

2. For 2015-2017, BasketballDatabase RAPM has Curry ahead of LeBron: https://thebasketballdatabase.com/2016-17RegularSeasonPlayerThreeYearRAPM.html

3. For 2016-2020, BasketballDatabase RAPM has Curry ahead of LeBron: https://thebasketballdatabase.com/2019-20RegularSeasonPlayerFiveYearRAPM.html

4. For 2017-2021, BasketballDatabase RAPM has Curry ahead of LeBron: https://thebasketballdatabase.com/2020-21RegularSeasonPlayerFiveYearRAPM.html

5. For 2020-2022, BasketballDatabase RAPM has Curry ahead of LeBron: https://thebasketballdatabase.com/2021-22RegularSeasonPlayerThreeYearRAPM.html

Meanwhile, you refer to some other timeframes that I did not refer to and that do not even exist on this website (it has one-year, three-year, and five-year RAPM).

You obviously seem to be confused, which is fine but made a bit galling given how rude your post was (i.e. calling me “Pathological”).

[AEnigma]

… So rather than use the site you link in your original post, the one called “NBARAPM.com”, which explicitly lists “Pure RAPM” and is probably where most people would have first been exposed to the metric at subject in this thread (considering that it has been on the page for a few months now), you instead decided to reference a completely different site, with completely different values, which coincidentally all happen to give you the exact RAPM outputs you want. What an interesting approach.

[lessthanjake]

.

I don’t think this discussion matters enough to respond to every little thing and I don’t want to derail my own thread. So I’ll just say that I don’t think anything you responded with about steals really makes a good point. You do not dispute that steals look very valuable at the individual level (most important here) and at the team level. There’s evidence that both equate to a lot of impact, which you cannot dispute. You talk a lot about the r-squared of different stats, but it seems pretty unsurprising that the r-squared of an event that is directly implicated on few possessions would be lower than things that are directly implicated on way more possessions (such as FG%). That doesn’t mean that it isn’t really impactful. It would obviously be essentially impossible for team steals to correlate as much with team results as FG%, so that’s just an unrealistic bar. The fact that something that happens so rarely even has what you refer to as a “weak correlation” with overall results is indicative of it being significantly impactful when it happens (hence why the estimated impact of steals is high in these models). And that’s one reason why your point is a flawed one. Furthermore, again, I provided information showing that steals are highly impactful at an individual level—which is what is most important in a discussion about an individual player’s steals stats.

You also don’t meaningfully address the fact that the effect of an individual’s steals on team results is not conceptually the same as the effect of a team’s steals on team results, since there are many team-level factors that would generally lead teams as a whole to get more steals while giving away that impact in other ways. I gave many examples, including small-ball lineups. To take that example, with small-ball lineups, we’d expect a team to get more steals. But we’d also expect the team to give away a lot of defensive impact via bad rebounding, lower rim protection, etc. In that case, the result would likely be more steals but worse defense. However, that doesn’t mean that an individual player on that team getting steals isn’t really impactful.

[tsherkin]

Ok let's play, in that case since steals don't matter for guards then blocks and rebounds for a 7 footer in Giannis shouldn't matter either . Deandre Jordan led the league in rebounds and blocks before was he a great defender... it was either Deandre Jordan or Andre Drummond going tit for tat with being the blocks and rebound champ. Hell didnt brook Lopez have several seasons averaging more blocks per game than Giannis. If Im 7 foot tall and remotely athletic in this era if you play enough minutes you can accidentally run into 9 rebounds a game with all of the bricked 3 pointers being launched

But that's wrong as well, and is also somewhat irrelevant to why Giannis was regarded as well as he was on D. It's true that you can be Hassan Whiteside and block a bunch of shots and have only so much utility on defense, but you're also less likely to be violently out of position as you are when you're gambling for steals. But even if you treat them equally, it's a non-sequitur to the idea that steals don't make any positive commentary on Harden's defensive value versus Giannis'.

And Lopez is ALSO a DPOY-level defender... And doesn't really game for just blocks. Also, Giannis is a < 1.5 bpg guy, so this whole exercise has largely been a waste of time...

[lessthanjake]

… So rather than use the site you link in your original post, the one called “NBARAPM.com”, which explicitly lists “Pure RAPM” and is probably where most people would have first been exposed to the metric at subject in this thread (considering that it has been on the page for a few months now), you instead decided to reference a completely different site, with completely different values, which coincidentally all happen to give you the exact RAPM outputs you want. What an interesting approach.

Okay, so a few things here:

1. Your post was not suggesting I should’ve used a different source, but rather that I had misrepresented the source that I had claimed to use. For instance, you said “Do you just assume no one fact checks anything?” This clearly suggests I provided factually inaccurate information. You were 100% wrong about that, and so now you’re falling back to saying I should’ve used a different source.

2. Your reasoning for why I should’ve used a different source is that I linked a different website in my OP. But I did not link that website to talk about RAPM at all, but instead linked it simply to note that it seems to have the original version of MAMBA.

3. As I just explained, I used BasketballDatabase’s RAPM because it was a discussion where I was specifically talking about RAPM with no priors and that website *very explicitly* says that its RAPM has no priors. The nbarapm website uses the label “pure rapm,” but it’s not actually clear whether that really means no prior or just means that it’s not a sophisticated all-in-one like the other measures it includes. To put a point on this, in a context alongside LEBRON and RAPTOR and DARKO, I can imagine someone calling something like Cheema’s RAPM “pure rapm” even though it does actually have a basic prior. With the BasketballDatabase RAPM, we know for sure that there’s no prior at all, so obviously that’s the one I used when making a specific point about what RAPM with no prior tells us.

4. I believe the old NBAshotcharts website also did not use a prior for its RAPM (though I don’t recall if it was completely explicit about that like the BasketballDatabase website is—I think maybe it was, but can’t remember and the website doesn’t exist anymore). And, as I’ve posted about before (example here: https://forums.realgm.com/boards/viewtopic.php?p=114668707#p114668707), Curry was ahead of LeBron in every 5-year RAPM that NBAshotcharts had starting at the 2013-14 season. That completely matches the BasketballDatabase data, which has Curry ahead of LeBron in every five-year timeframe starting at 2013-2014 except the one that post-dates the NBAshotcharts website (i.e. 2020-2024). So I believe we have two examples of RAPM with no prior that are consistent with each other and both show prime Curry consistently ahead of LeBron.

5. It is of course *possible* that nbarapm’s “pure RAPM” also has no prior and just has different results than the BasketballDatabase (and NBAshotcharts) RAPM. If that’s the case, then it’d likely have to do with differences caused by methodological decisions that do not relate to whether there’s a prior (perhaps something like what is done in the model with very-low-minutes players, for example). If that’s the case, then it would be another example of why we should look at as much data as possible—because random methodological decisions can potentially have an effect on things. Given that Curry still looks as good or better even in the nbarapm “pure rapm”—for instance, he is ahead of LeBron in 4 of the 7 five-year RAPMs starting at 2013-14; 5 out of the 9 three-year RAPMs starting at 2013-14; and 4 of the 8 four-year RAPMs starting at 2013-14—while looking clearly better in the BasketballDatabase RAPM (as well as the old NBAshotcharts RAPM), the overall picture would still definitely be one of Curry looking superior in terms of RAPM with no prior.

[eminence]

I don't see the value in new all in one xRAPMs. How long ago was it JE came out with his xRAPM (https://xrapm.com/about_page.html), ~15 years? Get some new material folks.

[AEnigma]

… So rather than use the site you link in your original post, the one called “NBARAPM.com”, which explicitly lists “Pure RAPM” and is probably where most people would have first been exposed to the metric at subject in this thread (considering that it has been on the page for a few months now), you instead decided to reference a completely different site, with completely different values, which coincidentally all happen to give you the exact RAPM outputs you want. What an interesting approach.

Okay, so a few things here:

1. Your post was not suggesting I should’ve used a different source, but rather that I had misrepresented the source that I had claimed to use. For instance, you said “Do you just assume no one fact checks anything?” This clearly suggests I provided factually inaccurate information. You were 100% wrong about that, and so now you’re falling back to saying I should’ve used a different source.

No, I am saying you represented a definite impact advantage for Curry that is not actually present.

2. Your reasoning for why I should’ve used a different source is that I linked a different website in my OP. But I did not link that website to talk about RAPM at all, but instead linked it simply to note that it seems to have the original version of MAMBA.

I am saying you made an active choice to disregard “pure RAPM” on a site you had already been using and referencing in favour of something completely different which just so happened to skew much harder toward Curry.

3. As I just explained, I used BasketballDatabase’s RAPM because it was a discussion where I was specifically talking about RAPM with no priors and that website *very explicitly* says that its RAPM has no priors. The nbarapm website uses the label “pure rapm,” but it’s not actually clear whether that really means no prior or just means that it’s not a sophisticated all-in-one like the other measures it includes. To put a point on this, I can imagine someone calling Cheema’s RAPM “pure rapm” even though it does actually have a basic prior. With the BasketballDatabase RAPM, we know for sure that there’s no prior at all, so obviously that’s the one I used when making a specific point about what RAPM with no prior tells us.

Even “no priors” is not a set form. Is the postseason included? Is the postseason weighed differently? Is garbage time filtered out? No RAPM is the same, regardless of whether it has priors or not.

4. I believe the old NBAshotcharts website also did not use a prior for its RAPM (though I don’t recall if it was completely explicit about that like the BasketballDatabase website is—I think maybe it was, but can’t remember and the website doesn’t exist anymore). And, as I’ve posted about before (example here: https://forums.realgm.com/boards/viewtopic.php?p=114668707#p114668707), Curry was ahead of LeBron in every 5-year RAPM that NBAshotcharts had starting at the 2013-14 season. That completely matches the BasketballDatabase data, which has Curry ahead of LeBron in every five-year timeframe starting at 2013-2014 except the one that post-dates the NBAshotcharts website (i.e. 2020-2024). So I believe we have two examples of RAPM with no prior that are consistent with each other and both show prime Curry consistently ahead of LeBron.

5. It is of course *possible* that nbarapm’s “pure RAPM” also has no prior and just has different results than the BasketballDatabase (and NBAshotcharts) RAPM. If that’s the case, then it’d likely have to do with differences caused by methodological decisions that do not relate to whether there’s a prior (perhaps something like what is done in the model with low-minutes players, for example). If that’s the case, then it would be another example of why we should look at as much data as possible—because random methodological decisions can have an effect on things. Given that Curry still looks as good or better even in the nbarapm “pure rapm”—for instance, he is ahead of LeBron in 4 of the 7 five-year RAPMs starting at 2013-14; 5 out of the 9 three-year RAPMs starting at 2013-14; and 4 of the 8 four-year RAPMs starting at 2013-14—while looking clearly better in the BasketballDatabase RAPM (as well as the old NBAshotcharts RAPM), the overall picture would still definitely be one of Curry looking superior in terms of RAPM with no prior.

And generally I think that is fine to argue, because as previously discussed, most people are not overly focused on who was more likely the more impactful regular season player per possession over varying years. Fair to say that Curry generally has the RAPM edge in the regular season over a majority of those frames. I would not qualify it as a particularly clear impact advantage, and I continue to be curious to what would happen with further extensions past five year samples.

My issue is when your initial move is consistently to present that data as a lot more definitive than it actually is. Curry’s RAPM data consistently looks better in 2018, 2019, and 2022; when you say “RAPM” is inconvenient for Lebron, you really mean those three years specifically (and the MAMBA creator explicitly said as much). That is often enough to give the advantage across spans including those ranges when we also know that from 2014-17 they are neck-and-neck (either jumping ahead based on those aforementioned variations in approach). But in turn, 2020/21/24 strongly favour Lebron, and while you have previously taken the angle that you primarily care about pre-2020 Curry, I think you also recognise the general principle that it looks better for these players when they show an ability to maintain value across eras and roles and teams and rule changes and opponents rather than tie themselves to one specific period and set of circumstances.

All of which is to say, when the creator of this metric openly says that in most years Lebron is doing a lot defensively that does not show up in any particular non-impact stat and thus this metric consistently underrates him on that end, it is weird that your response to that comment is to immediately lean hard on one dataset that favours Curry.

[lessthanjake]

My issue is when your initial move is consistently to present that data as a lot more definitive than it actually is. Curry’s RAPM data consistently looks better in 2018, 2019, and 2022; when you say “RAPM” is inconvenient for Lebron, you really mean those three years specifically (and the MAMBA creator explicitly said as much. That is often enough to give the advantage across spans including those ranges when we also know that from 2014-17 they are neck-and-neck (either jumping ahead based on those aforementioned variations in approach).

…

All of which is to say, when the creator of this metric openly says that in most years Lebron is doing a lot defensively that does not show up in any particular stat and thus this metric consistently underrates him on that end, it is weird that your response to that comment is to immediately lean hard on one dataset that favours Curry.

Most of what you said doesn’t even really disagree with the point I made. So I’ll just address these two things.

1. The reason to use RAPM over multiple seasons is because you want to get a larger sample to reduce noise. RAPM over small samples is very noisy. Taking larger samples reduces that noise (but you also don’t necessarily want it to be too large, since that creates other issues—so there’s a balance). You identify smaller time periods in larger samples, where Curry does particularly well compared to LeBron. The implication is that Curry was less impactful when you take those years out. But that basically just amounts to wanting to cherry-pick out a smaller sample that is more advantageous for the result you want. And especially when smaller time periods are generally more noisy, that basically just means you’re wanting to index on positive noise by taking away the years where noise probably favors Curry. Noise will sometimes benefit Curry and sometimes benefit LeBron, so if you ignore the smaller parts of the sample where Curry does best and are left with other parts where they’re “neck-and-neck,” then that is actually very good for Curry. There’s not really a principled reason to reject multi-year RAPM in favor of multi-year-cherry-picking-Curry’s-most-advantageous-years-out-of-the-sample RAPM.

2. The creator of this metric has subjective views about players that are no more or less valid than anyone else’s. The reality is that raw RAPM doesn’t seem to support the point about LeBron’s post-Miami DRAPM over larger samples. I showed that using the BasketballDatabase DRAPM, but I’ll note that it’s also the case with the nbarapm data that you pointed to. If you look at how post-Miami LeBron ranks in DRAPM in the various timespans available on that website, it looks similar to what I posted from the BasketballDatabase website. For instance, in terms of five-year DRAPM, that website has post-Miami LeBron ranking 116th, 49th, 56th, 125th, 40th, and 55th. That actually looks a bit worse than what it looks like in the BasketballDatabase. So yeah, this isn’t a point that requires me to “lean hard on one dataset.” Given what we see with raw RAPM, I suspect the author’s point might be based on looking at multi-year RAPM with basic priors added. But, as I said, if that’s the case, then the argument doesn’t really follow, because it’d basically amount to saying a sophisticated prior must be wrong because it conflicts with a less sophisticated prior. In any event, the metric was later updated specifically to make LeBron look better, so it’s not even clear whether the creator himself still thinks the metric underrates him.

[AEnigma]

Regarding 1, I am not opposed to using larger sets and explicitly said I would like to see sets that extend past five years, because even five-year sets can be easily skewed. I recall Engelmann saying his preferred range was 8 years (not that he has made 8-year sets widely accessible).

Regarding 2, I strongly suspect that Ohayo is quoting the creator from an earlier iteration of MAMBA, because Lebron’s D-MAMBA is a lot better now than what it was when first added to that NBARAPM site. However, you calling the metric’s creator “subjective” by quoting databases that you do not know are being used is again odd; he was the one looking directly at the inputs and the one directly speaking with the designers of other similar metrics, so he would know where the skew occurs in his formula. If you think DRAPM says something different, then that should tell you the DRAPM you are using is not the same as what he is using.

[jjgp111292]

more evidence that harden was robbed of mvp in 2019.. almost like averaging 36ppg and being 8 points ahead of the second place scorer means something

Why do Harden fans bitch about the 2019 MVP when 2017 is right there? Giannis had one of the easiest conventional cases of this era.

[lessthanjake]

Regarding 1, I am not opposed to using larger sets and explicitly said I would like to see sets that extend past five years, because even five-year sets can be easily skewed. I recall Engelmann saying his preferred range was 8 years (not that he has made 8-year sets widely accessible).

Regarding 2, I strongly suspect that Ohayo is quoting the creator from an earlier iteration of MAMBA, because Lebron’s D-MAMBA is a lot better now than what it was when first added to that NBARAPM site. However, you calling the metric’s creator “subjective” by quoting databases that you do not know are being used is again odd; he was the one looking directly at the inputs and the one directly speaking with the designers of other similar metrics, so he would know where the skew occurs in his formula. If you think DRAPM says something different, then that should tell you the DRAPM you are using is not the same as what he is using.

Nothing much here to disagree with, so I’ll just say a couple things:

Not sure I actually agree with Engelmann about 8 years being preferable, since I do think the larger the sample the more it glosses over real changes in players’ quality over that timespan (for instance, it’s going to be unduly bad for someone playing with 2007 Shaq for the range to be so large that the model is lumping in 2000-2002 Shaq when controlling for Shaq’s presence on the court in 2007). But that’s an issue that is about judgment calls, which I don’t necessarily feel very strongly about either way.

As for your last sentence, I did in fact say that it told me exactly that. Specifically, I noted that I imagine the DRAPM the author was using for these purposes was probably one of many DRAPMs that have basic priors, rather than a raw RAPM. That seems very plausible (and, I’d say likely) to me, given what we see in raw RAPM and that we generally tend to see LeBron’s standing improve in RAPM with unsophisticated priors. But if that’s the case, then I don’t think it’s a great reason to think more sophisticated priors are wrong. If RAPM with no prior and RAPM with really sophisticated priors both generally say one thing, and RAPM with less sophisticated priors says another, I don’t think the natural conclusion is necessarily that the RAPM with less sophisticated priors must be the right one. But yes, I agree that that quote might not even be relevant anymore, now that MAMBA was updated.

[OhayoKD]

more evidence that harden was robbed of mvp in 2019.. almost like averaging 36ppg and being 8 points ahead of the second place scorer means something

Why do Harden fans bitch about the 2019 MVP when 2017 is right there? Giannis had one of the easiest conventional cases of this era.

2017 had 2 much better rs players and 1 better rs player who actually won the MVP

[jjgp111292]

more evidence that harden was robbed of mvp in 2019.. almost like averaging 36ppg and being 8 points ahead of the second place scorer means something

Why do Harden fans bitch about the 2019 MVP when 2017 is right there? Giannis had one of the easiest conventional cases of this era.

2017 had 2 much better rs players and 1 better rs player who actually won the MVP

I do think Kawhi had a better case but Hardens case for 2017 is much better than 2019...but the fact that you think Russ deserved it over Harden in the year where Harden actually did have a case for being robbed is just perfectly demonstrating how you're just box score watching

[ShotCreator]

more evidence that harden was robbed of mvp in 2019.. almost like averaging 36ppg and being 8 points ahead of the second place scorer means something

Volume alone means only so much. And defense is a thing.

Also, the 2019 Bucks won 7 more games than the Rockets, and Giannis WAS a 28/12/5/6 guy on 64.4% TS himself, 2nd in the DPOY race and all that.

"Robbed" is a big word, which is violently inappropriate. One can make an argument that Harden should have won the MVP that year, but "robbed" is very much not an accurate description.

Harden was a good defender in 2019 and 2020.

[ShotCreator]

If it improves the data overall, then it *likely* improves the data as it relates to comparing any two players (though there is of course a chance it won’t). This is in part because an individual player’s measured impact is highly dependent on how the model evaluates all the other players that are on the court with and without them, and so a model that improves the data “for 400 players” will be likely to improve the data as it relates to any individual player too, unless the prior is really off for that player.

In any event, logically, the argument you’re making leads to a conclusion that we should use pure RAPM with no prior, since your argument is that pure RAPM amounts to “results” and any prior isn’t really “impact.”

But the problem there is that we have pure RAPM with no prior on the BasketballDatabase website, and it too tells us that prime Steph had better impact than LeBron. Here is who is ahead in three-year and five-year pure RAPM for each time period that is from 2014 onwards:

Pure RAPM: Three-Year

2014-2016: Curry
2015-2017: Curry
2016-2018: Curry
2017-2019: Curry
2018-2020: Curry
2019-2021: LeBron
2020-2022: Curry
2021-2023: Curry
2022-2024: Curry

Pure RAPM: Five-Year

2014-2018: Curry
2015-2019: Curry
2016-2020: Curry
2017-2021: Curry
2018-2022: Curry
2019-2023: Curry
2020-2024: LeBron

In other words, Steph was clearly ahead of LeBron in pure RAPM in the last decade… The problem is that the pure RAPM says Steph is more impactful.

… Do you just assume no one fact checks anything?

On that site you linked, and starting in 2014, Lebron is ahead of Steph in 2014-15 RAPM (omitted), 2015-16 RAPM (omitted), 2014-16 RAPM (directly contrary to what you claimed), 2016-17 RAPM (omitted), 2015-17 RAPM (directly contrary to what you claimed), 2014-17 RAPM (omitted), 2019-20 RAPM (omitted), 2016-20 RAPM (directly contrary to what you claimed), 2020-21 RAPM (omitted), 2018-21 RAPM (omitted), 2017-21 RAPM (directly contrary to what you claimed), 2020-22 RAPM (directly contrary to what you claimed), 2020-23 RAPM (omitted), 2023-24 RAPM (omitted), and 2021-24 RAPM (omitted), in addition to the 2019-21 and 2020-24 RAPM you did bother to acknowledge and accurately represent.

Pathological.

What are you talking about? My post clearly referred to “the BasketballDatabase website,” which can be found here: https://thebasketballdatabase.com/index.html. I used this because the website specifically says the following, which makes very clear it uses raw RAPM without any prior (other websites are less completely clear-cut about this or definitely do have a prior): “The values presented on this site are "vanilla" RAPM - the result of a pure ridge regression without any priors to tell the regression what the results "should" look like.”

I referred to three-year and five-year RAPM on that website. You claim I was wrong about 2014-2016, 2015-2017, 2016-2020, 2017-2021, and 2020-2022. This is all false.

Let’s take those one-by-one:

1. For 2014-2016, BasketballDatabase RAPM has Curry ahead of LeBron: https://thebasketballdatabase.com/2015-16RegularSeasonPlayerThreeYearRAPM.html

2. For 2015-2017, BasketballDatabase RAPM has Curry ahead of LeBron: https://thebasketballdatabase.com/2016-17RegularSeasonPlayerThreeYearRAPM.html

3. For 2016-2020, BasketballDatabase RAPM has Curry ahead of LeBron: https://thebasketballdatabase.com/2019-20RegularSeasonPlayerFiveYearRAPM.html

4. For 2017-2021, BasketballDatabase RAPM has Curry ahead of LeBron: https://thebasketballdatabase.com/2020-21RegularSeasonPlayerFiveYearRAPM.html

5. For 2020-2022, BasketballDatabase RAPM has Curry ahead of LeBron: https://thebasketballdatabase.com/2021-22RegularSeasonPlayerThreeYearRAPM.html

Meanwhile, you refer to some other timeframes that I did not refer to and that do not even exist on this website (it has one-year, three-year, and five-year RAPM).

You obviously seem to be confused, which is fine but made a bit galling given how rude your post was (i.e. calling me “Pathological”).

It's an indicator, like any other statistic. But you clearly need more priors on a RAPM sample that doesn't have James Harden in the top 10, ever, and has 2019 Kyle Lowry and Chris Paul as #2 and #3.

[OhayoKD]

Why do Harden fans bitch about the 2019 MVP when 2017 is right there? Giannis had one of the easiest conventional cases of this era.

2017 had 2 much better rs players and 1 better rs player who actually won the MVP

I do think Kawhi had a better case but Hardens case for 2017 is much better than 2019...

I was referring to Lebron and Steph who were far more valuable in the regular season. Kawhi you could argue but he missed a bunch of games and it didn't help him the team did okay when he did.

but the fact that you think Russ deserved it over Harden in the year where Harden actually did have a case for being robbed is just perfectly demonstrating how you're just box score watching

Rejecting a case built on...box-score watching is box-score watching? Westbrook had a worse constructed roster, showcased significantly more impact up until that point in the rs and the playoffs including 2017, and had a case as the league's lead creator and had an outlier shooting year having been the thunder's MVP without that in 2016.

Harden had no case for being robbed. He was worse and would have lost even if Westbrook didn't have the triple double. The MVP race was over when he hit the buzzer-beater vs Denver and just like in 2016 Westbrook was clearly better in the playoffs.

Project elsewhere.