26 replies to this topic

[Sir Thursday]

So ESPN have just written up some stuff about a new stat they are touting called RPM. They have yet to reveal the formal details of how it actually works beyond some handwaving about 'improvements' over RAPM, but it provides the same Offensive/Defensive split in it's analysis of player performance. You can see the full rankings here. 

 

Anyway, there's one thing that really jumped out to me - nestled at number 14 in the list is none other than our own Patrick Beverley! He is the highest Rocket on the list (Dwight Howard is in second at #23) and the #3 ranked point guard behind Chris Paul and Stephen Curry. What's especially interesting about this is that the majority of his score does not actually come from his defence as one might naively expect, but from the offensive end. He is +3.15 offensively (21st in the league, apparently) and +1.42 defensively (92nd). Very counter-intuitive!

 

So perhaps we should be looking a bit closer at what Beverley does well offensively. Perhaps it is his safe handle on the ball? Perhaps it is all those corner threes he hits? I don't know. Probably need to see the methodology to check for things that might bias it before taking it as concrete evidence, but it's interesting to think about.

 

ST

[Cooper]

The list basically makes sense considering player roles but it will be nice to see the exact process behind rpm.

[timetodienow1234567]

Lin is better than Kyrie. So I like our PG tandem. Why upgrade? The only teams with a better duo are OKC and SAS.

[timetodienow1234567]

Sarcasm btw.

[thenit]

I get the sarcasm, Kyrie is amazing on the offense, but the lack of defence is scary. He puts up big numbers and actually doesn't have a terrible cast around him, still they have never been above .500 which is just sad, especially when you are in the East. He is overated as an overall player. One of the best PG scorers with good vision but just no lack of leadership or getting his team involved makes me wonder if he really is even a top 10 PG in the league. ( Counting defence and offence)

 

In terms of Bevs high number on offense a big part is because of our starting lineup being so potent and one of the best. He doesn't turn the ball over and plays a game that fits into the starting lineup as the fifth option. His defensive numbers lags because he is on the court with Harden and Parsons who don't play D and T Jones who kind of get bullied in the post if we face a Mansize PF. So he gets dragged down on the defensive side. I think individual D is so hard to measure unlike Offense where you can measure efficiency, points assists turnovers etc.

[Sir Thursday]

I get the sarcasm, Kyrie is amazing on the offense, but the lack of defence is scary. He puts up big numbers and actually doesn't have a terrible cast around him, still they have never been above .500 which is just sad, especially when you are in the East. He is overated as an overall player. One of the best PG scorers with good vision but just no lack of leadership or getting his team involved makes me wonder if he really is even a top 10 PG in the league. ( Counting defence and offence)

 

In terms of Bevs high number on offense a big part is because of our starting lineup being so potent and one of the best. He doesn't turn the ball over and plays a game that fits into the starting lineup as the fifth option. His defensive numbers lags because he is on the court with Harden and Parsons who don't play D and T Jones who kind of get bullied in the post if we face a Mansize PF. So he gets dragged down on the defensive side. I think individual D is so hard to measure unlike Offense where you can measure efficiency, points assists turnovers etc.

 

In theory this stat should be isolating Beverley's performance from those of the players he shares the court with, so his offensive and defensive numbers shouldn't be affected by their presence. That is the stated goal of the tool, although whether they've actually managed to achieve that is still up for debate until we know how they did it.

 

ST

[thenit]

@ST

Thanks for the clarification. I tried to rationalize and until they reveal the process we can't really do anything else, except to speculate.

 

I do stand by that its easier to measure offensive numbers for a player than the defensive numbers due to more factors and less stats to base an analysis on or new stat. You can easily use stats to support Player A's D is better than Player B's. But in another stat you can flip them by using something different. All these advanced stat are useful but I just think in terms of D its hard to measure since its hard to isolate a player in a team scheme.

[Chichos]

It would make sense if his contribution was put into context of what he was expected to produce based on the shot he takes or if his production was compared to a "replacement" player playing that position.

[dbd]

Yep. D-Mo's number is much better than TJ. So, D-Mo should start.

Bev is much better than Beard, Dwight, and Hair. So we should give the key to Bev and let him be No.1 option.

 

Lin, TJ, Garcia, etc should compete in D-league with Kyrie and co. 

[Sir Thursday]

@ST

Thanks for the clarification. I tried to rationalize and until they reveal the process we can't really do anything else, except to speculate.

 

I do stand by that its easier to measure offensive numbers for a player than the defensive numbers due to more factors and less stats to base an analysis on or new stat. You can easily use stats to support Player A's D is better than Player B's. But in another stat you can flip them by using something different. All these advanced stat are useful but I just think in terms of D its hard to measure since its hard to isolate a player in a team scheme.

 

I think 'measure' is perhaps the wrong word to use. I agree that with more metrics available, you have more indicators to be able to make a determination of how good a players is. But each of them is going to be weighted towards a different part of the game and at the end of the day you're going to have to make your own decisions about which ones are important/valid. It's less of a measurement and more of a fuzzy guesstimate .

 

ST

[shirtless]

I think I have an idea how they tried to control for team effects when factoring in RPM. I did something very similar last month when I tried to calculated real off/def ratings while controlling for team performance, strength of opponents, and position bias.

 

The thing about plus minus is that its an intra-team measure. It is not an inter-team measure, even though that's how it's being used here. Basically, you can statistically control for how others on the court affect an individual's plus/minus, but at its core a plus/minus rating is a measure comparing you to your teammates who are not on the court, especially the players who play your same position. 

 

Think of it this way, if Pat Beverley played on the same team as Lebron James, Beverley's plus/minus (both nominal and real) would actually go down. The reason for this is because he would inevitably spend some time on the court when Lebron James is off the court, at which point the team's performance would plummet and would make Beverley's plus/minus look bad. Because Beverley's plus/minus currently looks good, it means he's outpacing his teammates, particularly his backups--Lin, Canaan, Hamilton, and Garcia.

 

On a side note, I'm actually not surprised that Beverley's impact is more felt on the offensive end. I've been saying for some time that several defensive measures don't seem to indicate that Beverley's renowned defensive is as effective as we might think, at least not from a tangible perspective (as far as intangibles go, he's still the Rockets leader in my mind). When several statistical measures and models support the same conclusion, that's when it's really time to pay attention and evaluate any pre-formed opinion you might have. 

Edited by shirtless, 08 April 2014 - 08:03 AM.

[Red94]

New post: On Patrick Beverley's sparkling RPM
By: rahat huq

So this was interesting.  ESPN rolled out its new stat, RPM, just yesterday.  While it seems to be a descendent of RAPM, I'm not entirely clear on the distinctions.  In any event, here is the full list of rankings.  It's interesting to note that Patrick Beverley is #14 in the entire league, just ahead of Mike Conley.  What's fascinating to note, and quite counter-intuitive, as Rob Dover pointed out [in the opening post of this forum thread], is that Beverley's glossy rating is primarily from his offensive stat rather than his defensive one.  What's going on here?

One might initially surmise that this occurrence is simply a factor of playing next to James Harden and Dwight Howard in one of the league's most high octane units.  But on the contrary, the stress of RAPM was that it sought to mitigate such advantages; it also claims to account for intra-squad discrepancies such as, for instance, the boost seen from having a bad backup.  The way I always understood RAPM was that rather than measuring "how good" a player was, it was a measure of how good a player was at the role they were playing on their team.  As the statistic is gleaned from play-by-play data, rather than your stats in the box score, its a composite of the impact you are having while in the game as indicated by the ebb and flow of the scoreboard.

For Beverley, it might not really be too big of a surprise that he is so impactful offensively.  I noted upon his injury, incurring the ridicule of Jeremy Lin's faithful band of Twiter loyalists, that one of Beverley's greatest qualities was his decisionmaking abilities on offense.  While he's prone to over-agressiveness on defense (subsequently racking up stupid fouls), he's extremely calm and composed on the scoring side of the ball.  He protects the ball, rarely forces the issue, and also hits down wide open 3's.  Contrast this with Jeremy Lin (just as an example; the statistic is not meant as an intra-team replacement comparison) who often looks like a chicken with its head cut off and you can see how the overall offense could be effected.  There is certainly something to the hazards of indecisiveness.

I think that Beverley is great at his role because when he's on the court, everyone's role becomes more clearly defined.  You'll note that Harden trusts him to set the play up because he knows he'll get it right back (ala the Rafer Alston - Tracy McGrady dynamic).  Beverley also rarely looks to drive in unless he has a clear shot; this leads to less errant passes and a smoother machine.

Ultimately though, this is all conjecture as, until the formula for RPM is released, we don't really know exactly what is going into this ranking.  For now, we can just speculate upon the purported results.

[dbd]

You still can't explain why Bev is better than Dwight and Beard (per this new RPM).

[Cooper]

You still can't explain why Bev is better than Dwight and Beard (per this new RPM).

You can't use plus minus to compare players that have such different roles and who plays when said player is off the court, Dwight and Harden key the offense (and Defense for Dwight) for 38min when they are out there Bev just plays a limited but solid role on offense and plays hard on defense if you asked Bev to be Harden on offense he'd be awful.

[dbd]

You can't use plus minus to compare players that have such different roles and who plays when said player is off the court, Dwight and Harden key the offense (and Defense for Dwight) for 38min when they are out there Bev just plays a limited but solid role on offense and plays hard on defense if you asked Bev to be Harden on offense he'd be awful.

Take a look this new RPM again. It showed Bev was much better on offense than his defense and ranked as #14. That really confused me a lot.

[thejohnnygold]

OK, there is sooooo much to consider so I will just throw out bullet points and hopefully people can build on them or find good tangents.

 

It appears WAR (wins above replacement) is a key factor here as toggling between the two does little else then re-shuffle the order of players in the top 40.  Thus, I would guess they use WAR as a starting point, or platform, and then add in efficiency/production factors to sort the order.

 

Isolating Centers is interesting as their ORPM max out at 1.32.  Sorting them by ORPM shows a few interesting things.

 

One, it seems apparent that efficiency is key and that Off. rebs and turnovers are likely big factors.  Offensive centers with actual skill sets (and aren't just dunk machines) are lower down the list--except for Al Jefferson!  He is a special player in this regard.  Brook Lopez gets an honorable mention here, but having only played 17 games this year is excluded from consideration.  Joakim Noah's all around game garners a #7 ranking.

 

Dwight comes in at #20 in the ORPM list for centers.  My guess is that his high turnover rate and the fact that he often produces his own offense in the post forces him down the list.  I do think that if all centers were required to play the way he does he would find himself in the top 5, but that's not the deal.  I know his post game is not appreciated by many, but count me as one who does appreciate it--despite it's failings.  (The fact that the Rockets made it a point to throw the ball into Omer more than once for post-ups in the last few games shows me how much they value the play and the overall effect it has on the team and winning)

 

I don't understand the high level of confusion about Bev's score.  Clearly it is not a volume measure or Nick Collison wouldn't have found his way into the #6 slot.  Now, assuming WAR is a large component then he is benefiting from Jeremy Lin's sub-par offensive season.  Ultimately, When Bev is on the floor our own turnovers decrease, opp. turnovers increase, blocks are up, efg% is up, it's all up!  LINK  His fault or not, when he is on the court good things happen.

 

Going back to Collison, I can understand how he makes this list.  I have watched him since his days at Kansas and cursed him the entire time.  He is always making a play of some kind.  A box out, a put back, a rebound, taking a charge, making the extra pass--you name it and he does it well.  He seldom makes mistakes or has mental lapses.

 

As one sorts ORPM lists from position to position it is evident that eFG% is a key factor.  There is no bonus/penalty for free throw shooting--just overall shooting efficiency.

 

Just noticed this: the definition of RPM at the bottom of the page.

 

RPM: Player's estimated on-court impact on team performance, measured in net point differential per 100 offensive and defensive possessions. RPM takes into account teammates, opponents, coaches and additional factors

 

Interesting that coaches are factored in.  How the heck does that work?  (let the McHale bashing begin! )

 

All in all, this measure doesn't seem too different from any other in that it means very little without having a solid grasp of all of the players beforehand.  How can I differentiate D'Andre Jordan's offensive superiority to Al Jefferson?  By watching them play.

 

Oh, the top 4 SF's by ORPM are Lebron, Durant, Melo, and.......Matt Barnes!  OK then....oh, all he does is shoot open 3's (pretty well this season) and finish off fast breaks.....ok then.  Corey Brewer at #7 is the same--threes, fast breaks, and not much else make one very efficient.

 

Then there is weird stuff like Kyle Korver (shooting 48% on threes this season attempting 5.6/game  ) who comes in at #17 amongst SG's on ORPM.  Huh?  So, what is the penalty for having a .637 eFG% but not making the top 10?  Based on who is ahead of him I'd say usg rate is heavily factored as his is a relatively low 14.1% for a starter playing 34 mpg.

 

I won't nitpick them all, but we are left in the same situation as always.  Does it tell us anything?  Maybe a little bit.  Does it make us ask better questions? (something that some purport to be the real usefulness of these stats)  Maybe, if you weren't already looking in the right direction.

 

The best question I can think of is, "can a computer, formula, or motion tracking device ever surpass the human eye and common sense in evaluating a player's impact on the court?"  Sure, they remove emotional bias, but outside of that I say not really.  They can provide useful raw data that is difficult to collect, but trying to apply them in a meaningful way, for me, has not proven very successful so far.

[rm90025]

I always like analytics, but at first glance I don't see the added value of this stat. Would you really take Beverley over 37 other PGs including players like Conley, Westbrook, Lowry and Parker?  I think it's hard to quantify whether a player is adding value or is a beneficiary of other teammates and a favorable system.  Also, these stats don't seem to take actual wins and losses into account.  The primary metrics seem to be based on point differential. Basically, most good teams have a positive point differential and most bad teams don't, but you don't need advance analytics to tell you that.

[Fury]

Honestly, what I found the most interesting about the RPM release was not Bev's high ranking. I've been high on him for a while, and I'm glad to see him get some appreciation this year.

What shocked me is that, despite the fact that Harden is top 5 in ORPM (which should suprise nobody,) he is still outside the top 20 due to his awful defense. His ORPM is 5.88, 5th in the league and just barely behind CP3. He has the highest ORPM of any SG. However, his DRPM is -2.44. He is ranked 390 out of 435 players, and he is ranked 76 out of 80 SGs.

To be fair, the stat seems to discriminate against SGs in DRPM. There are a disproportionate amount of SGs with a negative DRPM. This is the first time that I have seen a stat that is quite so harsh against Beard, but it does satisfy the eye test. Harden is getting better at his D of late, but this is something that must be a major concern going forward. We've had no shortage of discussion on Harden's D on this forum, but this stat only serves to further the hyperbolic discrepancy between Harden's offense and defense.

Edited by Fury, 08 April 2014 - 05:52 PM.

[thejohnnygold]

Something that has been alluded to, but I neglected to mention (outside of Dwight) is a player's role on the team.

 

Do I want Kyle Korver being the leader of my team and initiating the offense/creating for himself?  Nope.  James Harden is much better suited for that.  The same goes for guys like Matt Barnes, Matt Bonner, and Corey Brewer who all scored very highly on their ORPM.

 

If Chris Paul passed the ball to D'Andre Jordan in the post 12-15 times a game how would that end up?  Does anyone think he would still hold that #1 ORPM ranking?

 

Going back to Nick Collison....is he a #1 option ever...on any team?  No way.  Yet, guys like him make teams with #1 options much better.  Amir Johnson, Chris Anderson, and Mike Dunleavy don't jump out as stars in any way; yet, they find themselves in the top 40 of all players for RPM.

 

Roles are crucial, and those players who can produce consistently, and efficiently, in #1 roles are the true stars of the league.

 

This harkens back to some ideas Richard Li noted in his last study.  Can the Rockets nab a fringe #1 or #2 player at a lower value (due to the perception that they are not good players) and plug them in at a #4 or #5 role and watch them transform into complete studs.  It has been put forth that Chandler Parsons benefits from this and I agree.  On our team and in this role he is very good.  Put him on Milwaukee and ask him to lead their offense.....a very risky proposition that likely leads to him becoming the poster boy of bad contracts.  Rudy Gay will send him a fruit basket.

 

If somebody wanted to take this RPM data, and as objectively as possible, assign value according to role on their team (like a 1-5 system based on perceived pecking order.  For instance, OKC would go Durant, Westbrook, Ibaka, Jackson, Lamb, Butler...or some such thing.) and weigh that against their RPM score I think it could balance it out a bit.

 

D'Andre Jordan probably grades out around 4 or 5 depending on injuries and where all those guards fall into place.  That should handi-cap his score down a bit while Dwight getting a #2 rank should elevate his score some.  In this model it appears to have the opposite effect.  Once again, Al Jefferson's score is excellent considering he probably garners a #1 ranking on his team and is still #3 on the ORPM list for centers.

 

For Houston, you're probably looking at a list like this: Harden, Howard, Parsons, Lin, Jones, Beverley, Asik, D-Mo and then it's a toss up game to game.  Thus, Beverley's score probably takes a big hit if we applied this weighting.  He is asked to play within himself and do what he does best.  It just makes sense that he would excel.  On the other hand, James Harden gets the ball with 10 seconds left down by three and knows his job is to get us to overtime--much more difficult.

[Sir Thursday]

OK, there is sooooo much to consider so I will just throw out bullet points and hopefully people can build on them or find good tangents.

 

It appears WAR (wins above replacement) is a key factor here as toggling between the two does little else then re-shuffle the order of players in the top 40.  Thus, I would guess they use WAR as a starting point, or platform, and then add in efficiency/production factors to sort the order.

 

Isolating Centers is interesting as their ORPM max out at 1.32.  Sorting them by ORPM shows a few interesting things.

 

One, it seems apparent that efficiency is key and that Off. rebs and turnovers are likely big factors.  Offensive centers with actual skill sets (and aren't just dunk machines) are lower down the list--except for Al Jefferson!  He is a special player in this regard.  Brook Lopez gets an honorable mention here, but having only played 17 games this year is excluded from consideration.  Joakim Noah's all around game garners a #7 ranking.

 

Dwight comes in at #20 in the ORPM list for centers.  My guess is that his high turnover rate and the fact that he often produces his own offense in the post forces him down the list.  I do think that if all centers were required to play the way he does he would find himself in the top 5, but that's not the deal.  I know his post game is not appreciated by many, but count me as one who does appreciate it--despite it's failings.  (The fact that the Rockets made it a point to throw the ball into Omer more than once for post-ups in the last few games shows me how much they value the play and the overall effect it has on the team and winning)

 

I don't understand the high level of confusion about Bev's score.  Clearly it is not a volume measure or Nick Collison wouldn't have found his way into the #6 slot.  Now, assuming WAR is a large component then he is benefiting from Jeremy Lin's sub-par offensive season.  Ultimately, When Bev is on the floor our own turnovers decrease, opp. turnovers increase, blocks are up, efg% is up, it's all up!  LINK  His fault or not, when he is on the court good things happen.

 

Going back to Collison, I can understand how he makes this list.  I have watched him since his days at Kansas and cursed him the entire time.  He is always making a play of some kind.  A box out, a put back, a rebound, taking a charge, making the extra pass--you name it and he does it well.  He seldom makes mistakes or has mental lapses.

 

As one sorts ORPM lists from position to position it is evident that eFG% is a key factor.  There is no bonus/penalty for free throw shooting--just overall shooting efficiency.

 

Just noticed this: the definition of RPM at the bottom of the page.

 

RPM: Player's estimated on-court impact on team performance, measured in net point differential per 100 offensive and defensive possessions. RPM takes into account teammates, opponents, coaches and additional factors

 

Interesting that coaches are factored in.  How the heck does that work?  (let the McHale bashing begin! )

 

All in all, this measure doesn't seem too different from any other in that it means very little without having a solid grasp of all of the players beforehand.  How can I differentiate D'Andre Jordan's offensive superiority to Al Jefferson?  By watching them play.

 

Oh, the top 4 SF's by ORPM are Lebron, Durant, Melo, and.......Matt Barnes!  OK then....oh, all he does is shoot open 3's (pretty well this season) and finish off fast breaks.....ok then.  Corey Brewer at #7 is the same--threes, fast breaks, and not much else make one very efficient.

 

Then there is weird stuff like Kyle Korver (shooting 48% on threes this season attempting 5.6/game  ) who comes in at #17 amongst SG's on ORPM.  Huh?  So, what is the penalty for having a .637 eFG% but not making the top 10?  Based on who is ahead of him I'd say usg rate is heavily factored as his is a relatively low 14.1% for a starter playing 34 mpg.

 

I won't nitpick them all, but we are left in the same situation as always.  Does it tell us anything?  Maybe a little bit.  Does it make us ask better questions? (something that some purport to be the real usefulness of these stats)  Maybe, if you weren't already looking in the right direction.

 

The best question I can think of is, "can a computer, formula, or motion tracking device ever surpass the human eye and common sense in evaluating a player's impact on the court?"  Sure, they remove emotional bias, but outside of that I say not really.  They can provide useful raw data that is difficult to collect, but trying to apply them in a meaningful way, for me, has not proven very successful so far.

 

I think you're making the mistake of assuming this comes from box score stats. It doesn't - it's based solely on the team's performance or lack thereof while that player is on the court. It doesn't matter if Kyle Korver shoots 48% if the team as a whole functions worse while he's out there than it would otherwise. It tries to control for the abilities both of the other players on the court at the time and the players who might otherwise be on the court instead, sure. But it is possible for a player to shoot a high percentage and yet not necessarily improve a team that much, and presumably that is the implication here.

 

I think I disagree with your suggestion about weighting via role, as I think it defeats the purpose of the exercise. Players with high RPM (assuming it's working the way it's supposed to) have a beneficial effect on the team's offence/defence, but it says nothing about their role on the court and rightly so. The inference that should be taken from a good RPM score is not that the player's role should change in any way, but rather that they should be on the court for more minutes, since the way they are currently being used makes the team better when they play. It's not trying to find the 'best' player, but rather the 'most effectively used' player.

 

ST

Edited by Sir Thursday, 08 April 2014 - 06:57 PM.