#6 Best Defensive SG of All-Time - Top 10 Defense at each position project

[mdonnelly1989]

Jerry Dumars

[KnickFan33]

Dumars

[PigsOnTheWing]

I've done a quick calculation for this run-off. I've used the defensive component of the three most important statistical plus/minus I know (PIPM, BPM, Estimated Impact). This approach allows to compare them in a fair way, which is impossible using purest plus/minus derivates like RAPM, because they are not computable for Robertson.
I blended these three stats with a weighted average. I found the coefficients in a retrodiction test some time ago and they are about equal to each other for the defensive side (not for offense though) so it's basically a arithmetic average. Then I multiplied the result by the total possessions to make the final result cumulative of all the regular seasons values.
Eddie Jones:
1995 - 80.7, 1996 - 103.1, 1997 - 150.5, 1998 - 115.7, 1999 - 104.8, 2000 - 155, 2001 - 114.9, 2002 - 177.4, 2003 - 79.4, 2004 - 78.4, 2005 - 141.6, 2006 - 125.0, 2007 - 53.4, 2008 - 40.0 - FINAL SCORE: 1520.1
Alvin Robertson:
1985 - 73.5, 1986 - 254.5, 1987 - 178.2, 1988 - 167.3, 1989 - 148.4, 1990 - 178.9, 1991 - 193.3, 1992 - 132.2, 1993 - 116.4, 1996 - 103.8 - FINAL SCORE - 1546.6

So, despite this being a cumulative calculation, Robertson comes up slightly ahead of Jones. He also has 4 seasons with better value than Jones' best. I obviously know that boxscore derivated numbers aren't enough to make a final judgement, primarily because they rely too much on steals and blocks (in which Alvin easily trumps Eddie). Fwiw, I've read (though I don't remember where) that a steal is worth about 1.4 points. So averaging 4.8 steals x100poss like Robertson did in his DPOY season provides your team about 7 points per game. As for gambling issues, they made me question if he was good enough to be #5, but ultimately it's a problem which can dammed with a good big men, while having a disruptive force on the perimeter is something that every coach would be glad to have. Consider this post my argument for Robertson, since I have already voted for him.

btw Bounce_9, is there any chance you'd be willing to compile and present this same data for several of the other candidates [as well as #1-4 already voted in]? I think it would be relevant to the discussion. I'd like to see these same figures for Jordan, Moncrief, Cooper, Allen, as well as Dumars, Doug Christie, Stacey Augmon, Kobe, Wade, maybe Danny Green and Andre Roberson, too.
EDIT: Jimmy Butler, too, if you're willing.

Ok, so I've compiled the data for all the players you mentioned. In the previous post I forgot to say that this is basically Total Defensive Value Over Replacement player so there is a replacement level which I set at -1.5.
I'm gonna present the data for overall career, best 5 years (prime) and peak. If you, or others, have any question about a particular year(s) of any player, just let me know.

OVERALL CAREER DEFENSIVE VALUE:
#1 Jordan 2009.6
#2 Robertson 1546.6
#3 Jones 1520.1
#4 Wade 1297.1
#5 Christie 1180.8
#6 Allen 1000.3
#7 Bryant 993.6
#8 Moncrief 897.7
#9 Cooper 874.2
#10 Green 870.3
#11 Augmon 799.2
#12 Butler 752.8
#13 Roberson 498.8
#14 Dumars 415.8

5 YEARS PRIME:
#1 Jordan 990.3
#2 Robertson 972.2
#3 Christie 817.2
#4 Jones 749.6
#5 Wade 704.1
#6 Butler 665.4
#7 Green 640.1
#8 Moncrief 628.2
#9 Bryant 558.8
#10 Cooper 523.5
#11 Roberson 498.8
#12 Allen 467.0
#13 Augmon 459.2
#14 Dumars 352.4

PEAK:
#1 Robertson 254.5 (1986)
#2 Jordan 240.7 (1988)
#3 Christie 216.4 (2003)
#4 Roberson 191.2 (2017)
#5 Jones 177.4 (2002)
#6 Green 173.4 (2016)
#7 Butler 173.0 (2014)
#8 Wade 166.0 (2009)
#9 Moncrief 158.0 (1982)
#10 Cooper 144.4 (1981)
#11 Augmon 142.1 (1994)
#12 Bryant 134.2 (2000)
#13 Allen 118.4 (2009)
#14 Dumars 85.0 (1988)

Now I really don't have the time to comment on the data, I will do it tomorrow, as well as voting.

[trex_8063]

I've done a quick calculation for this run-off. I've used the defensive component of the three most important statistical plus/minus I know (PIPM, BPM, Estimated Impact). This approach allows to compare them in a fair way, which is impossible using purest plus/minus derivates like RAPM, because they are not computable for Robertson.
I blended these three stats with a weighted average. I found the coefficients in a retrodiction test some time ago and they are about equal to each other for the defensive side (not for offense though) so it's basically a arithmetic average. Then I multiplied the result by the total possessions to make the final result cumulative of all the regular seasons values.
Eddie Jones:
1995 - 80.7, 1996 - 103.1, 1997 - 150.5, 1998 - 115.7, 1999 - 104.8, 2000 - 155, 2001 - 114.9, 2002 - 177.4, 2003 - 79.4, 2004 - 78.4, 2005 - 141.6, 2006 - 125.0, 2007 - 53.4, 2008 - 40.0 - FINAL SCORE: 1520.1
Alvin Robertson:
1985 - 73.5, 1986 - 254.5, 1987 - 178.2, 1988 - 167.3, 1989 - 148.4, 1990 - 178.9, 1991 - 193.3, 1992 - 132.2, 1993 - 116.4, 1996 - 103.8 - FINAL SCORE - 1546.6

So, despite this being a cumulative calculation, Robertson comes up slightly ahead of Jones. He also has 4 seasons with better value than Jones' best. I obviously know that boxscore derivated numbers aren't enough to make a final judgement, primarily because they rely too much on steals and blocks (in which Alvin easily trumps Eddie). Fwiw, I've read (though I don't remember where) that a steal is worth about 1.4 points. So averaging 4.8 steals x100poss like Robertson did in his DPOY season provides your team about 7 points per game. As for gambling issues, they made me question if he was good enough to be #5, but ultimately it's a problem which can dammed with a good big men, while having a disruptive force on the perimeter is something that every coach would be glad to have. Consider this post my argument for Robertson, since I have already voted for him.

btw Bounce_9, is there any chance you'd be willing to compile and present this same data for several of the other candidates [as well as #1-4 already voted in]? I think it would be relevant to the discussion. I'd like to see these same figures for Jordan, Moncrief, Cooper, Allen, as well as Dumars, Doug Christie, Stacey Augmon, Kobe, Wade, maybe Danny Green and Andre Roberson, too.
EDIT: Jimmy Butler, too, if you're willing.

Ok, so I've compiled the data for all the players you mentioned. In the previous post I forgot to say that this is basically Total Defensive Value Over Replacement player so there is a replacement level which I set at -1.5.
I'm gonna present the data for overall career, best 5 years (prime) and peak. If you, or others, have any question about a particular year(s) of any player, just let me know.

OVERALL CAREER DEFENSIVE VALUE:
#1 Jordan 2009.6
#2 Robertson 1546.6
#3 Jones 1520.1
#4 Wade 1297.1
#5 Christie 1180.8
#6 Allen 1000.3
#7 Bryant 993.6
#8 Moncrief 897.7
#9 Cooper 874.2
#10 Green 870.3
#11 Augmon 799.2
#12 Butler 752.8
#13 Roberson 498.8
#14 Dumars 415.8

5 YEARS PRIME:
#1 Jordan 990.3
#2 Robertson 972.2
#3 Christie 817.2
#4 Jones 749.6
#5 Wade 704.1
#6 Butler 665.4
#7 Green 640.1
#8 Moncrief 628.2
#9 Bryant 558.8
#10 Cooper 523.5
#11 Roberson 498.8
#12 Allen 467.0
#13 Augmon 459.2
#14 Dumars 352.4

PEAK:
#1 Robertson 254.5 (1986)
#2 Jordan 240.7 (1988)
#3 Christie 216.4 (2003)
#4 Roberson 191.2 (2017)
#5 Jones 177.4 (2002)
#6 Green 173.4 (2016)
#7 Butler 173.0 (2014)
#8 Wade 166.0 (2009)
#9 Moncrief 158.0 (1982)
#10 Cooper 144.4 (1981)
#11 Augmon 142.1 (1994)
#12 Bryant 134.2 (2000)
#13 Allen 118.4 (2009)
#14 Dumars 85.0 (1988)

Now I really don't have the time to comment on the data, I will do it tomorrow, as well as voting.

Awesome, thanks for contributing this. I'll review it all later (and also decide how I want to incorporate it; more info the better as far as I'm concerned though).
fwiw, I think this thread is about to end or turf to runoff, as it's been 48 hours. So if you want to vote, better do it really soon. (side note: EJ 3rd in career value, 5th for peak, 4th for 5-year........I realize this is excluding some older guys, but he's nonetheless again he's shaping out as a solid candidate).

[trex_8063]

Thru post #24, I have the following count (and it's been ~49 hours):

Eddie Jones - 4 (trex_8063, iggymcfrack, Dr Positivity, pandrade83)
Jerry Sloan - 4 (Samurai, cecilthesheep, Luigi, lebron3-14-3)
Joe Dumars - 2?? (KnickFan33, Odom Fan???)
Don Chaney - 1 (penbeast0)

.

You want me to turf this to runoff, and then you can take over starting with #7?

[trex_8063]

I made an executive decision to keep things moving, and have officially turfed this into runoff. If you didn't already vote for EJ or Sloan, please be heard as to your preference now.....

Spoiler:

a

a

a

a

a

a

a

a

a

a

a

a

a

.

.

a

a

a

.

I'

a

.

a

a

a

a

a

a

a

a

a

a

a

a

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

)

/

/

/

/

/

/

/

/

/

/

/

/

/

/

/

/

/

/

[Johnny Firpo]

Jerry Sloan.

[LA Bird]

I should point out that since BPM, PIPM and Estimated Impact are all more or less the same thing (ie, box score based metrics weighted to replicate RAPM), using an ensemble of them together won't necessarily improve the accuracy. A player with huge box score stats is most likely going to look great in all 3 metrics and the same applies for those with minimal presence in box scores.

Anyway, I will continue to vote Eddie Jones.

[penbeast0]

Jerry Dumars

or Joe Sloan

[penbeast0]

I'll go with Eddie Jones in this runoff. He was a very solid player, not disruptive like a Robertson or Sloan, but more fundamentally consistent and I will trust the stat heads that this measured into very strong positive impact.

[mdonnelly1989]

Lol really though.

Joe Dumars

[SinceGatlingWasARookie]

Vote Eddie Jones

https://www.basketball-reference.com/play-index/h2h_finder.cgi?request=1&player_id1_hint=Jerry+Sloan&player_id1_select=Jerry+Sloan&player_id1=sloanje01&idx=players&player_id2_hint=Earl+Monroe&player_id2_select=Earl+Monroe&player_id2=monroea01&idx=players

[OdomFan]

Dumars

[penbeast0]

Vote Eddie Jones

https://www.basketball-reference.com/play-index/h2h_finder.cgi?request=1&player_id1_hint=Jerry+Sloan&player_id1_select=Jerry+Sloan&player_id1=sloanje01&idx=players&player_id2_hint=Earl+Monroe&player_id2_select=Earl+Monroe&player_id2=monroea01&idx=players

Why is Sloan v. Monroe particularly probative?

[SinceGatlingWasARookie]

Vote Eddie Jones

https://www.basketball-reference.com/play-index/h2h_finder.cgi?request=1&player_id1_hint=Jerry+Sloan&player_id1_select=Jerry+Sloan&player_id1=sloanje01&idx=players&player_id2_hint=Earl+Monroe&player_id2_select=Earl+Monroe&player_id2=monroea01&idx=players

Why is Sloan v. Monroe particularly probative?

I don't have on off stats for Sloan. I also did not see Sloan. Other than awards, what can I use to Judge him by? All I have is his effect on other players.

Monroe was just the first guy I thought of. If Sloan did not guard Monroe because the point guard guarded Monroe then Monroe would be a bad choice. The Monroe Stats vs the Bulls are not particularly conclusive of anything. The Bulls.did not slow Monroe down.

Here is another guy. Charlie Scott. Scott is the 2nd guy I tried. Scott scored more than his average vs Sloan. It is more clear that Sloan would guard Scott. https://www.basketball-reference.com/play-index/h2h_finder.cgi?request=1&player_id1_hint=Charlie+Scott&player_id1_select=Charlie+Scott&player_id1=scottch01&idx=players&player_id2_hint=Jerry+Sloan&player_id2_select=Jerry+Sloan&player_id2=sloanje01&idx=players

[PigsOnTheWing]

I should point out that since BPM, PIPM and Estimated Impact are all more or less the same thing (ie, box score based metrics weighted to replicate RAPM), using an ensemble of them together won't necessarily improve the accuracy. A player with huge box score stats is most likely going to look great in all 3 metrics and the same applies for those with minimal presence in box scores.

Anyway, I will continue to vote Eddie Jones.

I'm gonna speak a bit about methodology, because posting numbers without explanation is a bit useless. To start I must say I've never liked BPM, which the most famous of these kind of stat, primarily because there are some interaction terms that can easily produce outliers in specific situations and contexts (like Westbrook, Harden, Paul and so on).
This effect is quite common but in BPM it is quite bad. It also hates Nash who is my favorite player of all-time, so I thought "well, maybe there are other boxscore based metrics with the same purpose of BPM, but which are a bit more accurate (and that have Nash in the top 5 ever possibly )". So I've found PIPM and Estimated Impact and a bunch of others that unfortunately are available for a too limited period. I could have also used RAPM, which has no biases towards anyone, since it hasn't a formula that uses boxscore numbers, but unfortunately it is available only since 1997.
Ideally, I'd like to create a stat that is reliable and it can be used for every single year since the shot clock invention. As you know, for the 60's it's almost impossible but from 1973 there are the necessary numbers to try.

Then, I took all the teams with <50% Roster Continuity and I blended those three metrics in order to have the lowest Mean Absolute Error (thus improving the accuracy) in estimating Team Pythagorean Wins with previous year players' stats. I don't know if it's clear because I'm not a statistician so the terms I've used might be wrong.
As for the results, surprisingly enough, the lowest MAE (5.848 with Pythagorean Wins and 6.517 with true Wins) was obtained assigning negative coefficient for OBPM and OEI and a super positive one for OPIPM, so I think it's safe to say it's preferrable to use the latter.
For defense the numbers are close (.3*DPIPM+.38*DEI+.32*DBPM).
It's important to note that post-2001 PIPM has in the formula also of/off and some luck-adjuested data, so it cheats the purpose. I think I should do the same test for pre-2001 period to find the coefficients for when PIPM is boxscore only.
Also, the methodology was pretty raw, for example I used replacement level (-2) for all the rookies and players who hadn't played in the previous year so I might try to refine it a bit.

Anyway, maybe you're right in saying that using three different stats that measure the same thing in basically the same way doesn't really make sense, but , geninuely asking, if doing so improves the accuracy isn't the final blend better to estimate a single player impact towards team wins?
Then, i know a stat can't be judged by a bunch of selected players but the values I've posted seem to fairly represent the actual ability on defense of those players (the only two who are probably underrated are Moncrief and Dumars, though I'm starting to question the impact the latter had on the Bad Boys Pistons defense).

I also forgot to mention that EI exists from 1952 and as unreliable as the pre-1973 values are, they are still better than nothing.
Jerry Sloan appears as a good candidate for this spot being always a positive defender (he ranges from +.5 to +2). It's also important to note that pre-1973 values are pretty low (i.e. Russell is between +2.5 to +4.5 on D when he was probably closer to +6.5/7, so a bit of mental adjustment is needed).
Despite this, for this runoff I vote for Eddie Jones who was a really impactful defender for a long time.

[SinceGatlingWasARookie]

Looking at the 1974 Bulls team with both Van Lier and Sloan being 1st team all defensive team, what do we see? They have a slow pace. They force more turnovers than any other team. They are a bad rebounding team. They are worse than average at opposing team FG percentage.

https://www.basketball-reference.com/play-index/h2h_finder.cgi?request=1&player_id1_hint=Pete+Maravich&player_id1_select=Pete+Maravich&player_id1=maravpe01&idx=players&player_id2_hint=Jerry+Sloan&player_id2_select=Jerry+Sloan&player_id2=sloanje01&idx=players

In 1973-74 Maravich is on a bad Hawks team and is playing strictly off guard. Maravich averages 27 per game at 45.7 Fg. We know the Bulls slow down the pace of games a bit.

In 3 games Maravich shoots 44.3 % and averages 25.7 points per game. Nothing conclusive there.
Charlie Scott averged 32 points a game at 48% in his 3 games vs those Bulls. Well above his average.

[LA Bird]

To start I must say I've never liked BPM, which the most famous of these kind of stat, primarily because there are some interaction terms that can easily produce outliers in specific situations and contexts (like Westbrook, Harden, Paul and so on). This effect is quite common but in BPM it is quite bad

IMO, the outliers in BPM highlight a bigger problem present in recreated historical plus minuses in that the fitted model will still struggle to deal with edge cases no matter how accurate it is in general. As long as a particular statistic has a positive coefficient, accruing a larger total in that category would improve the player's predicted plus minus, regardless if they are helping or hurting their team in the process. Not all individual stat have equal team impact and this quality is unfortunately omitted in box score recreated plus minuses unless some form of team level on off statistic is incorporated in the calculation as well.

So I've found PIPM and Estimated Impact and a bunch of others that unfortunately are available for a too limited period. I could have also used RAPM, which has no biases towards anyone, since it hasn't a formula that uses boxscore numbers, but unfortunately it is available only since 1997.
Ideally, I'd like to create a stat that is reliable and it can be used for every single year since the shot clock invention.

The problem with most of these metrics is the lack of a centralized and consistent source. BPM, despite its flaws, is readily accessible on basketball reference. The historic PIPM data set is not on the nylon calculus page as far as I am aware of and there is also the issue of having 2 different sets of PIPM as you mentioned. Estimated Impact seems ok but we have no knowledge about its methodology. JE published 2001~15 RAPM in one fell swoop but then it is unknown if the same methodology and priors was used by others for the 1997~2000 RAPM. RAPM in recent years have also been increasingly hard to find with RPM superseding it in mainstream media. Even if we had 1 consistent source of non box score impact statistics from 1997 to today, I could argue that extrapolating the regressed weights to the 70s/80s let alone the volatile 50s/60s may not be a good idea with the drastic changes in league environment over the years.

Anyway, maybe you're right in saying that using three different stats that measure the same thing in basically the same way doesn't really make sense, but , geninuely asking, if doing so improves the accuracy isn't the final blend better to estimate a single player impact towards team wins?

I am not a statistician either but based on my basic understanding of classifier ensembles, the results and errors generated by the metrics would need to be uncorrelated for the ensemble as a whole to work much better than a single classifier. BPM and PIPM were both regressed upon the same 15 year data set so I would imagine the two are highly correlated. It's nice to have more data in general but we also don't want to be overconfident just because we think we are using multiple different metrics.

Then, i know a stat can't be judged by a bunch of selected players but the values I've posted seem to fairly represent the actual ability on defense of those players

I still don't like the fact that it's 100% box scores but I think it looks better than most other defensive metrics. In particular, I will have to look into DPIPM more. I would also agree Sloan is a good candidate for this round (even though I prefer Jones).

[SinceGatlingWasARookie]

There is no decent defensive stat prior to 1997.

You can do what I was doing to look at man to man match ups but that tells us nothing about a player's contribution to team defense. Also who wants to do the math?

WOWY was interesting were it tries to use injury games to replicate on off stats.

[trex_8063]

Eddie Jones is now leading the runoff 8 to 5; calling it for him. Will leave it to lebron3-14-3 to reclaim his project for the #7 thread.