Real Plus/Minus - Where Do the Celtics Rank?

[PhoSita]

Personally I've found that WS is the metric that most consistently aligns with my own perceptions of how valuable players are.

It tends to excessively reward role players on good teams, but that's not entirely unreasonable.  It also values bigs more than other players, but with most stats it's best to compare within a general positional grouping anyway.

Celtics WS/48 among rotation players:

IT - .203 
Horford - .173
Amir - .147
Crowder - .140
Zeller - .107
Jerebko - .102
Bradley - .101
Olynyk - .088
Rozier - .076
Brown - .074
Smart - .032


Here we see IT and Horford as the clear-cut best players on the team. 

Amir, Zeller, and Jerebko all seem to benefit from the bias toward big men. 

This stat too shows that Bradley's value may not be as high as we might expect; indeed, he's never rated particularly well in Win Shares.  This is probably due to his lack of free throws and relatively low steal rate.

Smart rates very poorly by this metric.  Arguably, it doesn't take defense enough into account.  But I think it reflects the common sense line of thinking that a player who is as futile as Smart is offensively has to do a lot in other categories to make up for it, and this year his steal rate, free throw rate, and rebounding have gone down.

[ThePaintedArea]

RPM is a really cool stat and the math behind it is really interesting, but ultimately it is one of the more flawed advanced stats.

Care to elaborate?

[Chris22]

Bradley is one of the worst passers I have ever seen.

[ThePaintedArea]

RPM numbers continue to look unreliable, noisy, and often nonsensical in anything but a very large (one full season or more) sample.

The Bradley numbers are a great example of that.

About the sample size: the point is well taken.  Kevin Pelton rates Real Plus/Minus highly, but for himself he uses more than one year of data.

The case of Avery Bradley - well, as I pointed out, he's really the outlier on the team if you compare RPM to minutes played (I made the argument that Rozier was the other one, but that it was justifiable given the need to force-feed him developmental minutes).

So my question to you is, if the other players get reasonable minutes allocations, WHY is Bradley the outlier?  What is it about Bradley that RPM misses?

[ThePaintedArea]

Not sure what makes real plus minus any more or less useful as a stat to judge players overall games than other stats. I respect all stats and think they all should be used, to some extent, to judge a player but not all by itself. Stats like these tend to show some whacky results when you look at players rankings in them across the league.

https://espn.go.com/nba/statistics/rpm/_/sort/RPM

What makes Real Plus/Minus useful is that it's measuring a player's whole contribution.  Why it's better than the raw Plus/Minus that you find in the box scores is that it controls for who else is on the floor.

I'm with you on the point you make about using a range of stats, rather than getting seduced by a single one. 

As for "wacky" results: do you have any examples, and an argument as to why they do or don't make sense?

Amir, Kelly, Marcus, and Zeller are typically lightning rods on this board; there's a certain amount of tension for a lot of posters between how highly they value those guys vs. how highly Brad does.  In general, this stat tends to confirm Brad's choices - with the exceptions of Rozier and Bradley.

Whacky results?

How about Steph Curry isnt in the top 10 in RPM?
How about Patty Mills is 24th right behind Al Horford?
Lucas Nogueira is 31st?
Otto Porter higher than John Wall and is 21st in the league?

Sounds pretty whacky to me to use as an all around gauge of player performance.

Adding to that, reining DPOY winner Kawhi Leonard has a DPM of 0.56, fitting him between Jonas Jerebko and Jaylen Brown (and only 0.09 above elite defender James Young)

Watch many Spurs games? Kawhi finished the year last year at 3.88 defensive RPM, which is stellar; it wasn't tops in the league; I think that that beast Draymond Green's 5.43 corresponded to a lot of people's perceptions that he was more deserving of the award.  But the mano-a-mano athletic on-ball defender gets an advantage in consideration for such things, just because he's more visible in what he does.

As for the quarter-pole of this year: a lot of stars coast in the early part of the year, especially on the defensive end.  I don't think that it would be a surprise to most San Antonio observers that he's sacrificed some defense in taking on more responsibility for shot creation.  But I'm willing to bet that Leonard finishes the year closer to 3.88 than 0.56.

As I mentioned above, James Young's low minutes diminish how meaningful his RPM is.  But having said that, he's shown some nice tenacity and a willingness to skin his knees going for loose balls in his minutes this year - worth a Tommy point or two.  I think that his lack of acceleration limits his potential as a defender, but the defensive RPM number is certainly encouraging.

Not sure what makes real plus minus any more or less useful as a stat to judge players overall games than other stats. I respect all stats and think they all should be used, to some extent, to judge a player but not all by itself. Stats like these tend to show some whacky results when you look at players rankings in them across the league.

https://espn.go.com/nba/statistics/rpm/_/sort/RPM

What makes Real Plus/Minus useful is that it's measuring a player's whole contribution.  Why it's better than the raw Plus/Minus that you find in the box scores is that it controls for who else is on the floor.

I'm with you on the point you make about using a range of stats, rather than getting seduced by a single one. 

As for "wacky" results: do you have any examples, and an argument as to why they do or don't make sense?

Amir, Kelly, Marcus, and Zeller are typically lightning rods on this board; there's a certain amount of tension for a lot of posters between how highly they value those guys vs. how highly Brad does.  In general, this stat tends to confirm Brad's choices - with the exceptions of Rozier and Bradley.

It attempts to control for who else is on the floor.  It doesn't do a very good job of this, though, especially defensively, which gives a lot of wacky results.  It also has issues with properly correcting for who comes off the bench for them (so having a good defender like AB backed up by a good defender like Smart downplays their defensive abilities)

Only trouble is, mostly Smart is not coming in for Bradley.  The most common lineup with Smart and not Bradley has only gotten 23.1 minutes so far this season:

 Thomas-Bradley-Crowder-Johnson-Horford   112.9
 Thomas-Smart-Bradley-Johnson-Olynyk   73.1       
 Thomas-Smart-Bradley-Crowder-Horford   42.3    
 Smart-Bradley-Crowder-Johnson-Horford   38.4
 Thomas-Bradley-Brown-Johnson-Zeller   23.2
 Rozier-Smart-Brown-Jerebko-Olynyk     23.1

In his two most used lineups, Smart is coming in for Crowder, who DOES rank highly on defensive RPM.

I'm open to a more developed argument on this point - which I think is a critical question - but the specific case you cite has a faulty assumption, and therefore is not persuasive.

It's a stat that can be useful when paired with the eye test and a bunch of other advanced stats, but using it on its own to compare players' impacts isn't really useful

I would agree that you should not give up the eye test, and I can't think of a good argument to not look at other stats.  But it's hard to agree that it "isn't really useful" - if it weren't, the rankings would mostly contradict the conventional wisdom.  And the exceptions mostly prove the rule.  The casual fan, for instance, would not expect Otto Porter to have such a high ranking - but when you look at what he's actually accomplished this year it makes perfect sense, and his high ranking is confirmed by other stats.

Real Plus/Minus is measuring things that you can't find elsewhere; I would argue that not only is it "really useful" - it's indispensable because it is alone in measuring certain critical outcomes.

[ThePaintedArea]

I was literally just reading about how funky this stat is last night.

Perhaps you could provide a link so that we could evaluate it.  I mean, I don't think that you want us to accept that Real Plus/Minus is "funky" just because you read it somewhere.

It also seems biased towards bigs, especially defensively.

On what basis are you claiming this?  Of course, bigs DO have a bigger impact on defense; it's harder to move a big guy, and harder to shoot over him, and that's true of some pretty unskilled players, as long as they're in the right spot.  Isaiah Thomas' big issue as a defender is less effort than height - but more importantly weight.

What's interesting is that IT and Smart rank 14th and 15th respectively for point guards, iirc.

Yes, more or less, depending on which day you look at it.  Their offense/defense splits are about where we'd predict, too: Thomas is a big minus defender but a big plus on offense, while Smart is a modest minus on offense and a modest plus on defense.  That's an up-stat for him, by the way, from last year; as good as his on-ball defense has been from jump, his off-ball defense has been a work in progress (true for Bradley as well). His tendency to help one pass away, leaving his man open, was a persistent problem.  He also upped his defensive rebounding this year, probably making coach happy.

But you might spell out the point you're making, as it remains unclear.

Another example is to look at the significant difference between Paul and all other point guards in DRPM. He's at 3.17, which is 2.25 higher than Smart at number 2, who is at 0.92. Are we really going to say that Paul is having three times the effect on the defense as Smart? Or is it more to do with the fact that he plays with great defenders in DJ/Reddick/LMAM and is usually replaced by either Felton, Rivers, or Crawford, all of whom are terrible defenders?

I don't think it does nearly as good of a job on factoring out the contextualities as it claims.

It's actually less complicated than you present it.  In fact Paul is almost exclusively subbed by Felton (they've only played together about 15 minutes so far this year), who ranks 13th in overall RPM defense, which is not bad at all.

Your overall point seems to be that Paul's number is so high because his subs are so low; but the reality of the subbing pattern doesn't match that claim.  And surely you aren't going to claim that Chris Paul is not one of the top defenders in the league? He's one of the few who gets steals one-on-one (the Celtics have two of them, but that's pretty unusual), and he's been one of the league leaders through his career.  I could go on but let that stand for more.

A bigger takeaway here is the unsurprising confirmation from this stat that Points are generally the worst defensive players on their teams (Chris Paul being the outstanding exception).

[Ilikesports17]

RPM is a really cool stat and the math behind it is really interesting, but ultimately it is one of the more flawed advanced stats.

Care to elaborate?

Sure. The first problem with RPM is that it virtually impossible to find how espn actually calculates it. This makes me a bit skeptical, but you do know that it is based on basically running regressions on other raw +/- (which is apm) then takes into account some extra stuff like age home/away and back to back data etc.

The biggest problem with RPM is that it contains an error that ESPN does not show. You see a bunch of players within hundredths of points of each their and have no idea if statistically you can confidently state any of them are better than the others.

Also, as is always the problem with predictive stats in sports, players change rapidly. RPM expects player growth but it's simply impossible to correctly account for it. Also, the regressions will only really work if there is a substantial sample size of every player playing each other. Say the Celtics decide Kelly and Marcus will always play together because Brad wants to play them together until Danny realizes that he drafted a 6'3" big and a 7' pg by accident. If they only play together it's really hard to quantify each players affect on the team without introducing a lot of bias.

The next problem is that I'd say it doesn't account for role.

I'll go back to Marcus Smart. Smart is obviously a really good defensive guard. However for a few weeks this year he was asked to start at and cover opposing small forwards. this isn't his optimal role and thus his pg defense +/- will drop because he was unable to be a good SF.

as I said, sports are tough because guys changeZ they get hot they get cold they get better they get worse they get hurt they get traded. Any predictive measure like this has got to have error bands and I'd really like to see them.

I'm not sure if RPM factors in role or minutes but James youngster good rating indicates that it does not.

I suppose it's a bit better than DRTG which tends to overrate counting stars like steals and blocks at the detriment of team defense.

Again, RPM is very cool but when you build all hear regressions on top of each other you need a ton of data and at this point in the season I don't know if they have it. 9d be far more inclined to read into this data if it came with error bars.

[chilidawg]

This year's data is certainly a small sample size and the numbers will change over the season.  Still, it's interesting to me to find data that alters my conceptions of what I see.  For instance, RPM tells me that IT is a good scorer and a bad defender, no surprise there.  It's also telling me that Crowder is good but not great at both ends of the court, and very good overall.  That seems about right.  But when it tells me that Bradley is actually a poor defender, well I'm not going to take that as the gospel truth, but maybe I should pay a little more attention to his defense and not to the common conception that he's an elite defender.  That's how I think the stats should be used, to challenge your preconceptions of players.

The point about error bars is an excellent one btw.

[nickagneta]

Can anyone here post the formula for calculating RPM because without that equation, its [dang] hard to come up with an opinion of RPM other than to look at the RPM leaders and determine if they look funky.

[ThePaintedArea]

  But when it tells me that Bradley is actually a poor defender, well I'm not going to take that as the gospel truth, but maybe I should pay a little more attention to his defense and not to the common conception that he's an elite defender.  That's how I think the stats should be used, to challenge your preconceptions of players.

Well, he's an outstanding on-ball defender, though his capability drops when he's got to battle with a bigger guy.

Coach Daniel (see YouTube) shows examples of him ball-watching and other sins. But I think that it's also obvious that he can put an unusual amount of pressure on the ball, among other assets.

I'm less interested in whether the stat is invalidated by these individual examples than why it might or might not be catching something critical about them.

The point about error bars is an excellent one btw.

Wholeheartedly agree.

[Ilikesports17]

Can anyone here post the formula for calculating RPM because without that equation, its [dang] hard to come up with an opinion of RPM other than to look at the RPM leaders and determine if they look funky.

ESPN hides the formula. However it is really similar to APM which I will post the formula for ina sec.

Ok the exact formula was tougher to find than I expected but they basically take every single stint(period between subs) and then set that group of 10 equal to a net +/-. They then take all the stunts in the Entire season and solve the mega system of equations. That's APM.

RPM is supposed to account for being at home and playing back to back etc etc.

A TP from me if anyone can find the actual RPM formula but I am pretty confident no one outside ESPN had seen it.

Even Kevin Pelton doesn't know how it's calculated.