Best NBA players of the last six years per 538’s latest acronym

[Sam Ray Not]

https://fivethirtyeight.com/features/the-best-nba-players-of-the-last-6-seasons/
Smarter guys than me can parse the details of the “RAPTOR” methodology, but at a quick glance, the basic premise that adjusted plus-minus stats get better in the biggest possible samples (in this case, the six years in which they’ve been closely tracking data, both reg season and playoffs) is clearly sound.

Needless to say, the highest-rated player by far over the past seasons (reg season and playoffs) is Stephen Curry, with a 120.0 total WAR, ahead of #2 Harden (107.9) and #3 LeBron (93.2). The top 10:

1. Curry
2. Harden
3. LeBron
4. CP3
5. Draymond (??)
6. Durant
7. Kawhi
8. Lowry
9. George
10. Westbrook

In more recent news, based on preseason results, neither Curry nor Harden is showing any signs at all of letting up (though I obviously share Kliq’s concern about Steph holding up physically):

Curry 39.3 pts per 36 on .690 true shooting
Harden 38.9 pts per 36 on .653 true shooting

Ridiculous. (Preseason alert).

 

[chilidawg]

Kemba Walker over Kyrie makes me happy.

 

[mikeford]

I find it hard to take seriously a metric that gives James Harden that much credit when he literally does not and cannot play one side of the ball.

 

[mikeot]

Kemba Walker over Kyrie makes me happy.

This.

 

[Euclis20]

I find it hard to take seriously a metric that gives James Harden that much credit when he literally does not and cannot play one side of the ball.

More specifically, it says Harden has been more valuable on defense than Lebron, Durant and Klay, among others. Meh.

 

[benhogan]

I'm interested in seeing Draymond Green's RAPTOR after he goes from playing with 4 future HOFers down to 1.

 

[Ale Xander]

CP3 over Durant, and Lowry over entire NBA minus 7 players, seems sketchy

 

[lovegtm]

I don’t think this has more value than a Bill Simmons ranking of OMG BEST OF THE DECADE. Maybe less.

To the extent it looks reasonable, it’s because it’s just confirming what we know (Curry is awesome). To the extent it’s controversial, we have no way evaluating its rankings.

This smells like pure, unadulterated anti-knowledge.

 

[Jimbodandy]

This smells like pure, unadulterated anti-knowledge.

This is the best thing that I have read in quite a while. "Anti-knowledge" is just so much better than simply calling it "garbage".

It's a shame to see 538 double-down on the blended Scotch approach. When the results of your algorithms jump off the page as wildly inaccurate, mixing in a dash of another two or three metrics isn't the answer. Sure the resulting list isn't AS ridiculous as the original, but how about finding the bugs instead? I think that these guys could add to the conversation more than they are.

 

[Captaincoop]

I'm interested in seeing Draymond Green's RAPTOR after he goes from playing with 4 future HOFers down to 1.

These kinds of stats are so flawed.

 

[luckiestman]

I like the theoretical approach to adjusted plus minus stuff compared to nonsense like PER. There are shortcomings. These do seem to hold up a bit with player movement but using a guy like Draymond is tough. Let’s say his rating goes down this year. Is it because Klay is out or is it because he is older? Tbf, Draymond graded out well pre-Durant.

CP3 always grades out well and CP3 is awesome; I think people assuming a metric is wrong because of how it rates Paul are not making a good case.

Any ranking like this will be ordered but 538 should report tiers. I do t know that they would claim a significant difference between 1-2, 2-3, 3-4 etc

In some respects these rankings are only worthwhile for guys like Draymond where it might make you take a closer look at him as I don’t need some nerd to tell me Kevin Durant is good.

 

[amarshal2]

The biggest, obvious flaw in +\- no matter the sample is players who have correlated time on the court together. As an extreme example, pretend that Curry and Draymond play with each other 95% of the time. You’d know that together they are an incredible duo but you wouldn’t know if either of them is better than LeBron.

 

[Red Right Ankle]

The biggest, obvious flaw in +\- no matter the sample is players who have correlated time on the court together. As an extreme example, pretend that Curry and Draymond play with each other 95% of the time. You’d know that together they are an incredible duo but you wouldn’t know if either of them is better than LeBron.

Hope you locked your doors and windows 'cause @Sam Ray Not is coming for you about this.

 

[chilidawg]

The biggest, obvious flaw in +\- no matter the sample is players who have correlated time on the court together. As an extreme example, pretend that Curry and Draymond play with each other 95% of the time. You’d know that together they are an incredible duo but you wouldn’t know if either of them is better than LeBron.

Don't metrics like RPM adjust for who is on the court?

 

[slamminsammya]

Yes. Using ridge regression. Or maybe that was just RAPM.

 

[PedroKsBambino]

Don't metrics like RPM adjust for who is on the court?

They seek to, but are limited by sample sizes.

 

[bakahump]

I'm interested in seeing Draymond Green's RAPTOR after he goes from playing with 4 future HOFers down to 1.

This. Dray seems like a good or really good player but 5th?

Until RAPTOR or SCRUM or MERSH or WTF ever can realistically separate this kind of outlier then its gonna be questioned.

 

[bowiac]

Yes. Using ridge regression. Or maybe that was just RAPM.

RPM is expected RAPM, so uses all the same tools as RAPM, and extends them by adding box-score components. RPM knows who is on the court, and use box-score score data to disentangle guys players who are highly colinear like that Curry/Draymond example.

However, more generally, Curry and Draymond don't play 95% of their minutes together. That's what makes a metric like RPM possible - that very few teams have substitution patterns like that, and players get hurt, miss games for rest, change teams, etc... There are some isolated examples of players like this (Derek Fisher almost never played without Kobe for instance), but they're rare. Curry played 700 minutes without Draymond last year, and only 70% of his minutes were with Draymond on the court. Over the course of multiple years especially, you get big samples of players playing separately, and the model exploits those minutes to understand who was "driving the bus" during the times when they were playing together.

I don't know that RAPTOR is right or wrong, but this Draymond as the 5th best guy of the last six years isn't bananas. The casual eye-test will habitually underrate defense, and also underrate big man passing. Everyone needs to calm down, and be less confident that watching 15 Warriors games a year has given them the ability to divine if Draymond has been a more impactful guy than Kyle Lowry.

 

[Sam Ray Not]

They seek to, but are limited by sample sizes.

Exactly, which is why a full six-year sample starts to tell you at least something. I'd agree that its really hard try to account for synergy — but also somewhat pointless, since we're talking about a 5-on-5 team game in which synergizing with teammates is a huge part of what makes an effective player. My sense is that when people hear ratings like this they instinctively think first of "who's the toughest to guard one-on-one?" (and/or "who's the toughest to score on one-on-one?") which is not what plus-minus-derived measures are trying to capture.

I do suspect that that Draymond gets a boost from playing such a high percentage of his minutes with Steph (edit: though as Bowiac points out it's not as high as 95%), and from synergizing with Steph better than pretty much any player does. As far as Steph, though: I wish I could find it now, but there was a chart that came out last year that showed that over the last five years, even without any of the other big four, the Warriors with Curry on the floor played like a 60-win team. At this point don't think it should be particularly controversial to posit that Curry has been the best *five on five player* in the NBA over the past six seasons. Most of the available evidence points in that direction, under every available adjusting system. Whether that exactly jibes with one's criteria for best player is another question; but given that is a 5-on-5 sport, I do think silly how seldom Curry even gets considered at #1 when people drop their player rankings.

Anyway, I was secretly hoping all along that Bowiac would step in and put all of us in our places, so my work here is done.

 

[mauf]

RPM is expected RAPM, so uses all the same tools as RAPM, and extends them by adding box-score components. RPM knows who is on the court, and use box-score score data to disentangle guys players who are highly colinear like that Curry/Draymond example.

However, more generally, Curry and Draymond don't play 95% of their minutes together. That's what makes a metric like RPM possible - that very few teams have substitution patterns like that, and players get hurt, miss games for rest, change teams, etc... There are some isolated examples of players like this (Derek Fisher almost never played without Kobe for instance), but they're rare. Curry played 700 minutes without Draymond last year, and only 70% of his minutes were with Draymond on the court. Over the course of multiple years especially, you get big samples of players playing separately, and the model exploits those minutes to understand who was "driving the bus" during the times when they were playing together.

I don't know that RAPTOR is right or wrong, but this Draymond as the 5th best guy of the last six years isn't bananas. The casual eye-test will habitually underrate defense, and also underrate big man passing. Everyone needs to calm down, and be less confident that watching 15 Warriors games a year has given them the ability to divine if Draymond has been a more impactful guy than Kyle Lowry.

Draymond looked like he lost a step last season, but I can buy him as a top-10 player over the past 6 years. Lowry is the guy who jumped at me as overrated, but I’m sure I can count on my fingers how many regular-season Raptors games I’ve watched in the past 6 years, so you make a good point about overconfidence.