Plus/Minus Superstars (Shaq and Robinson)

[lorak]

We also have RAPM and Winston's APM and both these metrics show Stockton's elite impact. so minutes aren't problem here. He definitely wasn't role player who played 20 mpg (like Porter).

But "on" data (or total raw +/-) is pretty useless, because players on good teams would have good "on". We need net on/off.

[MacGill]

Shaq is indeed amazing. Honestly the more I look at his +/- data, the more I'm able to let go of the fact that I consider him about the stupidest, pettiest person on the planet.

I appreciate you saying this Doc

You know I am a Shaq fan, but I try to be a very rational one. And while I have always agreed with your points around how he handled himself personally, it never deterred me from his immense impact made on the court. Also, since I go back and watch actual games of him, it never made sense when poster's tried to label him as what his career became to be after 14 years in the league.

BTW, doesn't the bolded belong to Dwight Howard now?

[ElGee]

@Colts -- I'm using NBA.com data. They provide pace to two decimal points, which allows for accurate calculation of possessions played. (Again, if we assume NBA.com is accurate, which you are for this thread.) The off can be calculated once you know the on.

We also have RAPM and Winston's APM and both these metrics show Stockton's elite impact. so minutes aren't problem here. He definitely wasn't role player who played 20 mpg (like Porter).

But "on" data (or total raw +/-) is pretty useless, because players on good teams would have good "on". We need net on/off.

You (and Colts) and anyone else on this train has a major issue that you have not addressed, which is that of co-variance and selective minutes. It's a problem with APM as well due to multicollinearity. Also saying "both" RAPM, APM and on/off show Stockton's "elite impact" is redundant, since they are reductions of the same data.

Here are the players I mentioned and their MPG:
Porter 21 M (00-02)
Chalmers 25 M (11-13)
Collison 21 M(10-13)
Fisher 26 M (99-02)
Najera 19 M (02-05)

Stockton 29 M (98-00)
Robinson 30 M (99-03)

Compare that to their star teammates
Duncan 39 M (99-03)
James 38 M (11-13)
Durant 39 M (10-13)
Shaq 38 M (99-02)
Dirk 38 M (02-05)
Malone 37 M (98-00)

What you are asking everyone to believe is that, when compared to Robinson and Stockton, that a 5-10 MPG difference is substantial for the role players in the first group , but when comparing Stockton and Robinson to players like Malone, Duncan (and other superstars), where there is less selectivity in the minutes, that 8-10 MPG is not relevant.

If you have an explanation for this I'm all ears, but it seems to me you have to either start pimping everyone of these role players on good teams as "plus-minus superstars" (who just happen to lose that skill when the star in question isn't next to them), or stop touting Robinson and Stockton in that same vein.

[colts18]

@Colts -- I'm using NBA.com data. They provide pace to two decimal points, which allows for accurate calculation of possessions played. (Again, if we assume NBA.com is accurate, which you are for this thread.) The off can be calculated once you know the on.

We also have RAPM and Winston's APM and both these metrics show Stockton's elite impact. so minutes aren't problem here. He definitely wasn't role player who played 20 mpg (like Porter).

But "on" data (or total raw +/-) is pretty useless, because players on good teams would have good "on". We need net on/off.

You (and Colts) and anyone else on this train has a major issue that you have not addressed, which is that of co-variance and selective minutes. It's a problem with APM as well due to multicollinearity. Also saying "both" RAPM, APM and on/off show Stockton's "elite impact" is redundant, since they are reductions of the same data.

Here are the players I mentioned and their MPG:
Porter 21 M (00-02)
Chalmers 25 M (11-13)
Collison 21 M(10-13)
Fisher 26 M (99-02)
Najera 19 M (02-05)

Stockton 29 M (98-00)
Robinson 30 M (99-03)

Compare that to their star teammates
Duncan 39 M (99-03)
James 38 M (11-13)
Durant 39 M (10-13)
Shaq 38 M (99-02)
Dirk 38 M (02-05)
Malone 37 M (98-00)

What you are asking everyone to believe is that, when compared to Robinson and Stockton, that a 5-10 MPG difference is substantial for the role players in the first group , but when comparing Stockton and Robinson to players like Malone, Duncan (and other superstars), where there is less selectivity in the minutes, that 8-10 MPG is not relevant.

If you have an explanation for this I'm all ears, but it seems to me you have to either start pimping everyone of these role players on good teams as "plus-minus superstars" (who just happen to lose that skill when the star in question isn't next to them), or stop touting Robinson and Stockton in that same vein.

Elgee, when will the malware attack on your site get fixed? I tried visiting one of the old realgm top 50 peak threads and I couldnt go on it because your site was trying to attack my computer.

[SideshowBob]

FWIW, I've visited backpicks loads of times since the warnings started showing up (probably twice in just the last week), and I've never once had an issue.

[ElGee]

I'll probably get around to it at some point. I've been unable to separate the malware from the content, so it will probably have to be a manual reload when I get around to it.

[lorak]

Here are the players I mentioned and their MPG:
Porter 21 M (00-02)
Chalmers 25 M (11-13)
Collison 21 M(10-13)
Fisher 26 M (99-02)
Najera 19 M (02-05)

Stockton 29 M (98-00)
Robinson 30 M (99-03)

Compare that to their star teammates
Duncan 39 M (99-03)
James 38 M (11-13)
Durant 39 M (10-13)
Shaq 38 M (99-02)
Dirk 38 M (02-05)
Malone 37 M (98-00)

Stockton has clear advantage over Malone in APM.
Lets check your examples (not to mention difference in minutes between Stockton and Malone is smaller than rest of the players - usually twice times...)

Chalmers -0.6 RAPM in 2011 and -0.1 in 2012

Collison looks great, but he plays almost half minutes less than Durant.

Najera 0.7 in 2002, 1.8 in 2003, 1.4 in 2004, 1.1 in 2005 and also half minutes less than Dirk. SO obviously good role player.

For Fisher and Porter I don't have APM data.

So what's your point? Expect of Collison no one looks as good in APM (and no one is better than 1st scoring option) as Stockton (he is better than Malone according to APM), but Collison plays almost half minutes less than Durant, so he clearly is role player.

[colts18]

@Colts -- I'm using NBA.com data. They provide pace to two decimal points, which allows for accurate calculation of possessions played. (Again, if we assume NBA.com is accurate, which you are for this thread.) The off can be calculated once you know the on.

How do you reconsile with the fact that NBA.com has different numbers for on court +/- per 100 and Net Rating (O rtg-D rtg) when calculating the on/off. For example, 2000 Shaq is +11.2 on on court per 100 with NBA.com. Then if you click on the advanced tab, it says Shaq is +11.0 (106.3 O rtg-95.3 D rtg). Can you help me figure out how exactly they calculate their Net rating? It looks like the on court per 100 seems accurate. But for the Net Rtg, I'm not sure exactly how they are getting their numbers. I'm playing around with it and I can't get accurate on court numbers for the net Rating that would fit in with the on court per 100. I tried emailing NBA.com for an explanation on this but they never emailed me back.

[ElGee]

@Colts -- I'm using NBA.com data. They provide pace to two decimal points, which allows for accurate calculation of possessions played. (Again, if we assume NBA.com is accurate, which you are for this thread.) The off can be calculated once you know the on.

How do you reconsile with the fact that NBA.com has different numbers for on court +/- per 100 and Net Rating (O rtg-D rtg) when calculating the on/off. For example, 2000 Shaq is +11.2 on on court per 100 with NBA.com. Then if you click on the advanced tab, it says Shaq is +11.0 (106.3 O rtg-95.3 D rtg). Can you help me figure out how exactly they calculate their Net rating? It looks like the on court per 100 seems accurate. But for the Net Rtg, I'm not sure exactly how they are getting their numbers. I'm playing around with it and I can't get accurate on court numbers for the net Rating that would fit in with the on court per 100. I tried emailing NBA.com for an explanation on this but they never emailed me back.

Some sort of pace calculation issue I'd guess. I'm not sure about their accuracy...

According to the NBA.com calculations, places like 82games were using pace estimations to calculate team ratings in on/off statistics. NBA.com claims that they aren't using calculations and are counting possessions, which renders an accurate pace. This pace change changes the numbers across the board -- in general, most stars seem to play faster than using a team estimate (this makes sense intuitively -- weaker teams want fewer possessions) which is shifting down all the numbers we've previously looked at for the last decade or so.

For instance, the cap on offensive rating used to be around 120. It was hard to find a high-minute player who was on the court with an ORtg in this range. The best offense with a single player on the court in the last decade was Steve Nash on the court in 2005:

82games: 121.7 ORtg
BBR: 120.1 ORtg
NBA.com: 117.7 ORtg

BBR claims their pace was 96.2. NBA.com claims it was 98.7. Thus the difference in the ratings. Both claim they are counting possessions...Who do we believe?

At the team level,

NBA.com says the 2005 Suns played at a 98.62 pace.
Oliver's estimate says they played at a 95.9 pace.
BBR's counter says the 2005 Suns played at a 96.18 pace.

Without knowing much about how NBA.com is generating its numbers, I'm initially skeptical of their "counting possessions;" In order to generate the kind of difference in the above example, they are counting more than 200 extra possessions in a season. My guess is that there is a technicality difference in their counting. (eg Are they counting technical fouls as a separate possession? Still, 200 extra possessions is a lot...)

[mysticbb]

Some sort of pace calculation issue I'd guess. I'm not sure about their accuracy...

You can reduce the effect due to differences in terms of pace by using ratings adjusted for league average. For 2005 bbr has the league average at 105.8 (derived from the pbp data), NBA.com has it at 103.1. For your example with Nash that would make:

NBA.com: +14.6
bbr.com: +14.3

That cuts the difference down from 2.4 to 0.3. That isn't that bad. But I find it also weird, that NBA.com has such higher pace.

[SideshowBob]

Pace issue was the first thing I noticed when the new stats were put up.

http://stats.nba.com/leagueTeamGeneral.html?MeasureType=Advanced&PerMode=Totals&sortField=PACE&sortOrder=DES

http://espn.go.com/nba/hollinger/teamstats/_/sort/paceFactor

NBA.com's pace figures match up exactly with Hollinger's on ESPN (though Hollinger goes out to 1 decimal). This leads me to believe that they are indeed NOT counting possessions from PbP, but are just using the same (incorrect) formula that Hollinger/ESPN do.

[lorak]

Pace issue was the first thing I noticed when the new stats were put up.

http://stats.nba.com/leagueTeamGeneral.html?MeasureType=Advanced&PerMode=Totals&sortField=PACE&sortOrder=DES

http://espn.go.com/nba/hollinger/teamstats/_/sort/paceFactor

NBA.com's pace figures match up exactly with Hollinger's on ESPN (though Hollinger goes out to 1 decimal). This leads me to believe that they are indeed NOT counting possessions from PbP, but are just using the same (incorrect) formula that Hollinger/ESPN do.

I checked 1st Spurs game this season (31st October vs NOH). NBA.com has 96.78 pace for this game so it's almost exact as my manual calculation from pbp: 97 possessions.

[mysticbb]

I checked 1st Spurs game this season (31st October vs NOH). NBA.com has 96.78 pace for this game so it's almost exact as my manual calculation from pbp: 97 possessions.

If they are counting, the numbers should be exactly the same, so 97 would be the correct number (I just counted it myself per hand), if we count the possessions at the end of a quarter after a miss or with a few seconds left as full possessions as well. bbr is not counting those possessions, I also don't count that. bbv is counting that as well as possession. bbr is in agreement with Oliver here, who is also discounting that as possession, because by definition a possessions ends with either a FGM (or FT respectively), a DRB or a TOV, thus the defensive rebound at the end is not leading to a new possession, but just ended the previous one here.

[lorak]

I checked 1st Spurs game this season (31st October vs NOH). NBA.com has 96.78 pace for this game so it's almost exact as my manual calculation from pbp: 97 possessions.

If they are counting, the numbers should be exactly the same, so 97 would be the correct number (I just counted it myself per hand),

Yes, it's odd nba.com's number is not "full". Any explanation?

if we count the possessions at the end of a quarter after a miss or with a few seconds left as full possessions as well. bbr is not counting those possessions, I also don't count that. bbv is counting that as well as possession. bbr is in agreement with Oliver here, who is also discounting that as possession, because by definition a possessions ends with either a FGM (or FT respectively), a DRB or a TOV, thus the defensive rebound at the end is not leading to a new possession, but just ended the previous one here.

Without those kind of possessions pace of this game would be 96, so still different than b-r estimation (93.5).

And BTW, way ortg is better than play (from synergy)?

[mysticbb]

Yes, it's odd nba.com's number is not "full". Any explanation?

No, at least not yet.

Without those kind of possessions pace of this game would be 96, so still different than b-r estimation (93.5).

There are 3 uncounted possessions (at the end of the 1st, the 3rd and the 4th), which makes it 95.5. And when we use the standard estimation FGA+0.44*FTA+TOV-ORB for both teams, we get 95.3 as pace estimate.

And BTW, way ortg is better than play (from synergy)?

I would go by points per 100 possession over points per play (Synergy).

Oh, bbr is counting 94 possessions for that game for the Spurs and Hornets, if we go by the play index+ count. So, they obviously don't count something else either, because the minutes are adding up to 240 for each team.

[lorak]

Without those kind of possessions pace of this game would be 96, so still different than b-r estimation (93.5).

There are 3 uncounted possessions (at the end of the 1st, the 3rd and the 4th), which makes it 95.5. And when we use the standard estimation FGA+0.44*FTA+TOV-ORB for both teams, we get 95.3 as pace estimate.

We shouldn't count possession at the end of 1st as uncounted, because NOH started 2nd and that was their 23 possession (Spurs ended 1st with 23 possessions, NOH with 22). So we have only two "fake" possessions, so real pace is 95. Still significant difference from b-r 93.5 (why now you are using standart estimation?)

And BTW, way ortg is better than play (from synergy)?

I would go by points per 100 possession over points per play (Synergy).

But why?

[colts18]

Pace issue was the first thing I noticed when the new stats were put up.

http://stats.nba.com/leagueTeamGeneral.html?MeasureType=Advanced&PerMode=Totals&sortField=PACE&sortOrder=DES

http://espn.go.com/nba/hollinger/teamstats/_/sort/paceFactor

NBA.com's pace figures match up exactly with Hollinger's on ESPN (though Hollinger goes out to 1 decimal). This leads me to believe that they are indeed NOT counting possessions from PbP, but are just using the same (incorrect) formula that Hollinger/ESPN do.

Good find. that makes perfect sense. It looks like the Net rating figures come from Hollinger pace factor. But if you go to the base section of player/team stats and click on the tab to make everything per 100 possessions, that calculates a completely different possession number for some reason. I'm not exactly sure why they are using 2 different possession numbers and how they calculate each of those possession numbers.

[SideshowBob]

Pace issue was the first thing I noticed when the new stats were put up.

http://stats.nba.com/leagueTeamGeneral.html?MeasureType=Advanced&PerMode=Totals&sortField=PACE&sortOrder=DES

http://espn.go.com/nba/hollinger/teamstats/_/sort/paceFactor

NBA.com's pace figures match up exactly with Hollinger's on ESPN (though Hollinger goes out to 1 decimal). This leads me to believe that they are indeed NOT counting possessions from PbP, but are just using the same (incorrect) formula that Hollinger/ESPN do.

Good find. that makes perfect sense. It looks like the Net rating figures come from Hollinger pace factor. But if you go to the base section of player/team stats and click on the tab to make everything per 100 possessions, that calculates a completely different possession number for some reason. I'm not exactly sure why they are using 2 different possession numbers and how they calculate each of those possession numbers.

They use On Court possessions for that adjustment.

Player A plays game with 100 possessions. Player A is on court for 70 possessions, so his stats are adjusted for those 70 possessions, not 100.

[mysticbb]

We shouldn't count possession at the end of 1st as uncounted, because NOH started 2nd and that was their 23 possession (Spurs ended 1st with 23 possessions, NOH with 22).

That has nothing to do with it. The possession at the start of the 2nd went by default to the Hornets, not because they had the defensive rebound at the end of the 1st quarter. Their next possession is counted as the first possession in the 2nd quarter, that's how I counted it anyway.

So we have only two "fake" possessions, so real pace is 95.

Only two fake would mean 96, because that has to be devided by 2 to integrate that in pace. Pace tells us how many possession each team had, and it is per se an estimate.

Still significant difference from b-r 93.5 (why now you are using standart estimation?)

Just because I wanted to show the number, nothing else. Oliver has a reason for the more complicated formula, because it is an estimate of possessions for the whole season. The distribution of the free throws is usually different than in that specific game, which also explains that the estimate is far off.

But why? ;)

For prediction purposes, for estimations, etc. pp. It is easier to handle. And ppp is only useful for player to player comparison, not so much for team-wide comparisons anyway.

[colts18]

Good find. that makes perfect sense. It looks like the Net rating figures come from Hollinger pace factor. But if you go to the base section of player/team stats and click on the tab to make everything per 100 possessions, that calculates a completely different possession number for some reason. I'm not exactly sure why they are using 2 different possession numbers and how they calculate each of those possession numbers.

They use On Court possessions for that adjustment.

Player A plays game with 100 possessions. Player A is on court for 70 possessions, so his stats are adjusted for those 70 possessions, not 100.

explain it more. Why is that way in the team pages too? For example if you go to on court per 100 possession on the team stats page, that number is different than the Net rating. It doesn't make sense. I assume they are using a different possessions number for each one.