RealGM

For the RealGM Top 100 project I wanted to have a resource on the best teams since 1955.

I decided to calculate overall SRS over 3-year (seasons) windows, for the following reasons:

• Figured out some time ago how to perform the recursive calculation for SRS like it's done on basketball reference
• Grouping multiple seasons can be used to identify so called team dynasties
• Was inspired by Santserre (viewtopic.php?f=64&t=2012241), details on methodology below*

I included the Strength-of-Schedule (SOS) as well to provide an indication on the competition that was faced during the 3-season window.

Any thoughts? I like how some teams from the late 50's and 60's pop up in the top 15, and it's not all BOS with 66-68 PHI showing up as well. SOS really helps emphasize how tough the competition was LAL 00-02, as one would expect. 08-10 LAL had even tougher competition, relatively speaking.

Edit: Updated the results after correcting the initial calculations. 66-68 PHI drops all the way to 63 now

Here are the results limited to the top 100 "dynasties":

ABA only (top 10):


* Similar to Santerre I did the following:

• Weighed playoff games differently: A single playoff game is the equivalent of 10 regular season games
• Calculated standard deviations above the mean

Respect the effort but there are too many odd results — and we can just highlight the 1996-98 Bulls being placed directly next to the 1997-99 Bulls as an obvious reason how, although there are multiple other such abnormalities.

I don't see a way the '96-'98 Bulls can have the same result at the '97-'99 version.

That to me would suggest an error present somewhere.

WestGOAT wrote:• Figured out some time ago how to perform the recursive calculation for SRS like it's done on basketball reference

as in beating a team affects their SRS, which affects your SRS for beating them, which affects their SRS and you iterate until it stabilizes? or something different? that was one concern with sansterre's list. that it didn't iterate so series results could be unbalanced with the winner looking too good and the loser looking too bad.

eminence wrote:I don't see a way the '96-'98 Bulls can have the same result at the '97-'99 version.

That to me would suggest an error present somewhere.

I was just noticing how low the '15-17 Warriors are. Even the '14-'16 Warriors are higher. How were 2014 Warriors better than 2017 Warriors?

f4p wrote:

WestGOAT wrote:• Figured out some time ago how to perform the recursive calculation for SRS like it's done on basketball reference

as in beating a team affects their SRS, which affects your SRS for beating them, which affects their SRS and you iterate until it stabilizes? or something different? that was one concern with sansterre's list. that it didn't iterate so series results could be unbalanced with the winner looking too good and the loser looking too bad.

Correct, it's an iterative calculation and it stops once the new SRS values have marginal differences compared to the previous ones.

eminence wrote:I don't see a way the '96-'98 Bulls can have the same result at the '97-'99 version.

That to me would suggest an error present somewhere.

Yup, quality control failed me! The overall SRS looks fine, looks like something is going wrong during the grouped 3-season window calculation:

AEnigma wrote:Respect the effort but there are too many odd results — and we can just highlight the 1996-98 Bulls being placed directly next to the 1997-99 Bulls as an obvious reason how, although there are multiple other such abnormalities.

giberish wrote:

eminence wrote:I don't see a way the '96-'98 Bulls can have the same result at the '97-'99 version.

That to me would suggest an error present somewhere.

I was just noticing how low the '15-17 Warriors are. Even the '14-'16 Warriors are higher. How were 2014 Warriors better than 2017 Warriors?

Figured out what went "wrong", makes more sense now I think. Thanks for spotting!

That looks better. Still some weirdness — e.g. the 1984 Lakers being so close to the 1987 Lakers — but no longer anything totally glaring in methodology.

Any chance you could run '51-'54?

But looks nice at first glance, will take some more time later to look through.

AEnigma wrote:That looks better. Still some weirdness — e.g. the 1984 Lakers being so close to the 1987 Lakers — but no longer anything totally glaring in methodology.

Is it really that weird? I assume with 1984 you are referring to '83-'85 Lakers, or maybe '84-'86? What about 1987?
The average overall SRS over a 3-year window is being taken and then adjusted to the SRS of the other teams in the league.
Maybe this table below with individual overall SRS per season helps.



Despite winning the title in 1988, the overall SRS of the Lakers was lower than that of 1984. This might be bring down the overall scores for the windows that use 1988 score, such as '86-'88. For example, the average of 6.73 is based on (6.59 + 9.45 + 4.14) / 3.

eminence wrote:Any chance you could run '51-'54?

But looks nice at first glance, will take some more time later to look through.

I actually have the data for 1954, but the rest is not readily available. I'll look to include that in the future maybe. I took 1955 as a cut-off due to beginning of the shot-clock era.

Here is the average overall SRS included over the 3-year windows:

Teams in chronological order, as that's something I find useful.

The MDA Suns were the highest ranking non-champion, 2nd was the '72-'74 Bucks. '96-'98 Utah the second that didn't win a title in the immediately surrounding seasons.

I was surprised not to see the MDA Rockets on the list. Mind sharing their results? I imagine a very narrow miss.

'22-'24 Boston must be poised to join the list.



# of appearances by franchise:
Boston - 21
LAL - 16
San Antonio - 14 (Duncan appearing in more 3 year periods than any other player at all 14 of these)
Philadelphia - 8
Chicago - 7
Detroit - 6
Golden State - 6
Milwaukee - 5
Utah - 3
Cleveland - 2
Miami - 2
OKC - 2
Orlando - 2
Sacramento - 2
Seattle - 2
Dallas - 1
Phoenix - 1

eminence wrote:Teams in chronological order, as that's something I find useful.

The MDA Suns were the highest ranking non-champion, 2nd was the '72-'74 Bucks. '96-'98 Utah the second that didn't win a title in the immediately surrounding seasons.

I was surprised not to see the MDA Rockets on the list. Mind sharing their results? I imagine a very narrow miss.

'22-'24 Boston must be poised to join the list.

Code: Select all

Team   First   Last   3yr SRSz
BOS   1956   1958   1.89
BOS   1957   1959   2.62
BOS   1958   1960   2.41
BOS   1959   1961   2.85
BOS   1960   1962   2.93
BOS   1961   1963   2.61
BOS   1962   1964   2.25
BOS   1963   1965   2.43
BOS   1964   1966   2.73
BOS   1965   1967   1.96
PHI   1965   1967   1.5
PHI   1966   1968   1.68
BOS   1966   1968   1.52
PHI   1967   1969   1.54
MIL   1970   1972   2.37
MIL   1971   1973   2.24
LAL   1971   1973   1.56
MIL   1972   1974   1.84
BOS   1973   1975   1.49
BOS   1974   1976   1.55
GSW   1975   1977   1.64
PHI   1978   1980   1.64
PHI   1979   1981   1.56
BOS   1980   1982   2.26
PHI   1980   1982   1.76
LAL   1980   1982   1.51
PHI   1981   1983   2.08
BOS   1981   1983   1.7
MIL   1981   1983   1.64
LAL   1982   1984   1.8
BOS   1982   1984   1.68
PHI   1982   1982   1.61
LAL   1983   1985   2.16
BOS   1983   1985   1.54
BOS   1984   1986   2.53
LAL   1984   1986   2.47
LAL   1985   1987   2.51
BOS   1985   1987   2.06
LAL   1986   1988   1.95
BOS   1986   1988   1.83
LAL   1987   1989   1.83
DET   1987   1989   1.83
DET   1988   1990   1.9
LAL   1988   1990   1.5
CHI   1989   1991   1.65
LAL   1989   1991   1.57
DET   1989   1991   1.54
CHI   1990   1992   2.2
CHI   1991   1993   2.48
CHI   1992   1994   1.84
CHI   1994   1996   1.74
SEA   1994   1996   1.53
UTA   1994   1996   1.52
CHI   1995   1997   2.12
UTA   1995   1997   1.75
SEA   1995   1997   1.53
CHI   1996   1998   2.53
UTA   1996   1998   1.76
SAS   1998   2000   1.52
LAL   1999   2001   1.85
SAS   1999   2001   1.63
LAL   2000   2002   2.44
LAL   2001   2003   2.13
SAS   2001   2003   1.77
SAC   2001   2003   1.67
SAS   2002   2004   1.92
SAC   2002   2004   1.62
LAL   2002   2004   1.55
SAS   2003   2005   2.09
DET   2003   2005   1.55
SAS   2004   2006   1.93
DET   2004   2006   1.91
DAL   2004   2006   1.57
SAS   2005   2007   2.24
PHO   2005   2007   2.03
DET   2005   2007   1.61
SAS   2006   2008   1.68
LAL   2008   2010   1.99
ORL   2008   2010   1.64
BOS   2008   2010   1.53
LAL   2009   2011   1.76
ORL   2009   2011   1.55
MIA   2011   2013   2.06
SAS   2011   2013   1.68
OKC   2011   2013   1.61
SAS   2012   2014   2.38
MIA   2012   2014   1.76
OKC   2012   2014   1.58
SAS   2013   2015   2.24
GSW   2013   2015   1.49
SAS   2014   2016   2.21
GSW   2014   2016   1.84
GSW   2015   2017   2.88
CLE   2015   2017   1.98
SAS   2015   2017   1.78
GSW   2016   2018   3.04
CLE   2016   2018   1.66
SAS   2016   2018   1.58
GSW   2017   2019   2.87
MIL   2019   2021   1.83


# of appearances by franchise:
Boston - 21
LAL - 16
San Antonio - 14 (Duncan appearing in more 3 year periods than any other player at all 14 of these)
Philadelphia - 8
Chicago - 7
Detroit - 6
Golden State - 6
Milwaukee - 5
Utah - 3
Cleveland - 2
Miami - 2
OKC - 2
Orlando - 2
Sacramento - 2
Seattle - 2
Dallas - 1
Phoenix - 1

Rockets:


Taking 2018 as the peak, grouping the '17, '18, '19 gives you an average relative SRS of 1.47 (standard deviations above the mean), indeed just missing out the top 100 (lowest was 1.49). The blow-out losses against GSW in 2018 hurts their overall SRS. They should not be penalized against GSW for coasting in the RS, since the overall SRS of GSW is still pretty high in 2018:

The lack of 1969-73 Knicks representation stands out, even with a relative disappointment in 1971 and a middling SRS in 1972.

AEnigma wrote:The lack of 1969-73 Knicks representation stands out, even with a relative disappointment in 1971 and a middling SRS is 1972.

Looks like their 1971 season tanks them:


Probably their margin-of-defeat (MOD) against the decent, but not great, Bullets plays a big role:


They'd probably show up if we were doing 5-year windows though.

WestGOAT wrote:.

Another cool post WestGOAT! It's nice to have 3-year overall SRS in an easily accessible spot. Continuing our discussion of using value vs standard deviation from the other thread, it's worth mentioning the the final list that Sansterre presents (https://forums.realgm.com/boards/viewtopic.php?p=85943319#p85943319) is not in order of overall SRS (OSRS) value or standard deviation, but a combination of the two, along with a factor to show the number of series won (https://forums.realgm.com/boards/viewtopic.php?p=86687130#p86687130). He still gives people each team's OSRS and OSRS standard deviation (ORS_STD) rank in each article, but he wanted to combine these metrics for his overall team list.

Sansterre's final formula to rank his teams is thus: (OSRS + 0.7*(series_won + 1) )*(sqrt(OSRS_STD) ). OSRS is defined as overall SRS, giving playoff games a 7x weighting. series_won is defined such that you get a pass for skipping the 1st/2nd round in early eras if you're a higher seed, so all teams that lose in the first round have a series_won + 1 = 0 + 1 = 1, while all teams that win in the finals have a series_won + 1 = 4 + 1 = 5..

As you may remember, (since it looks like you were commenting in the original thread!), this wasn't a scientifically chosen combination or anything... Sansterre just tinkered with the formula until all the various team were closest to his prior.
(e.g. he wanted the usual best teams like the 96 Bulls and the 01 Lakers being in the top ~10 with the 96 Bulls slightly higher, which helped define the playoff game weighting; he wanted standard deviation weighed heavily enough that the Russell Celtics weren't too low; he wanted standard deviation weighed not too heavily such that the 76 Warriors and other non-dominant late 70s champions weren't too high; he wanted a series_won factor to reward teams that won a series with a negative MoV e.g. the champion Hakeem Rockets, but he didn't want this weighed too heavily such that you punished teams who were better but lost to a worse team due to bad luck / random noise).
Ideally, in a version 2.0, we would test which weighting for playoff games vs regular season games, OSRS, series_won, and OSRS_STD were most predictive. E.g. we run the calculation for the first round across all playoffs, then see what weighting produces the best fit for predicting the 2nd round victors / MoV of every playoffs. If you did this for every round of the playoffs simultaneously, you might get less arbitrary formula weighting.

A few random notes/qualifiers:
-Difference in Playoff Weighting between you and Sansterre: It looks like you had playoff games weighted 10x a regular season game, while Sansterre actually had them at 7x a regular season game. Not that one weighting is inherently better than another A Priori -- we'd have to run the optimization to see which was most predictive like I described above to find out. But it's worth mentioning there's this slight difference in calculating OSRS between you and Sansterre, so a 3-year OSRS using Sansterre's original formula might be slightly different.
-In case you haven't seen, Sansterre published a best 3-year Playoff SRS list (with various filters added like no overlapping runs and minimum performance in all 3 years): https://forums.realgm.com/boards/viewtopic.php?p=87339138#p87339138
-If time permits, I would be very interested in seeing the HM list of the teams that just barely missed the cut... say the top 110–125 or so. Similarly, if there's any other major champions besides the 70s Knicks that aren't on the list (i.e. the Hakeem Rockets who won twice, but were some of the least dominant champions ever), it might be interesting to see how far back they are.

With that said, if only to satisfy my curiosity, I've calculated
The Top 100 3-year Overall SRS (using the same ranking order Sansterre does, which combines OSRS, standard deviation, and series won)
1 16-18 GSW
2 15-17 GSW
3 17-19 GSW
4 96-98 CHI
5 91-93 CHI
6 60-62 BOS
7 00-02 LAL
8 85-87 LAL
9 59-61 BOS
10 12-14 SAS
11 84-86 BOS
12 70-72 MIL
13 71-73 MIL
14 64-66 BOS
15 90-92 CHI
16 84-86 LAL
17 61-63 BOS
18 95-97 CHI
19 13-15 SAS
20 14-16 SAS
21 57-59 BOS
22 58-60 BOS
23 63-65 BOS
24 11-13 MIA
25 15-17 CLE
26 08-10 LAL
27 80-82 BOS
28 62-64 BOS
29 01-03 LAL
30 85-87 BOS
31 83-85 LAL
31 88-90 DET
33 81-83 PHI
34 05-07 SAS
35 86-88 LAL
36 87-89 LAL
37 03-05 SAS
38 87-89 DET
39 92-94 CHI
40 14-16 GSW
41 96-98 UTA
42 12-14 MIA
43 72-74 MIL
44 86-88 BOS
45 19-21 MIL
46 99-01 LAL
47 09-11 LAL
48 04-06 DET
49 65-67 BOS
50 16-18 CLE
51 82-84 LAL
52 02-04 SAS
53 95-97 UTA
54 89-91 CHI
55 94-96 CHI
56 71-73 LAL
57 05-07 PHO
58 04-06 SAS
59 15-17 SAS
59 01-03 SAS
61 89-91 DET
62 80-82 PHI
63 81-83 BOS
64 89-91 LAL
65 08-10 ORL
66 88-90 LAL
67 08-10 BOS
68 11-13 SAS
69 06-08 SAS
70 82-84 BOS
71 66-68 PHI
72 11-13 OKC
73 12-14 OKC
74 99-01 SAS
75 56-58 BOS
76 03-05 DET
77 66-68 BOS
78 01-03 SAC
79 02-04 LAL
80 81-83 MIL
81 82-84 PHI
82 95-97 SEA
83 73-75 BOS
84 83-85 BOS
85 74-76 BOS
86 09-11 ORL
87 16-18 SAS
88 67-69 PHI
89 80-82 LAL
90 94-96 UTA
91 05-07 DET
92 98-00 SAS
93 94-96 SEA
94 13-15 GSW
95 02-04 SAC
96 75-77 GSW
97 79-81 PHI
98 17-19 HOU
99 78-80 PHI
100 65-67 PHI
101(?) 69-71 NYK

Note: This uses WestGOAT's slightly heavier playoff game weighting over Sansterre's. Also, the bottom of the list may be missing a few teams: Since I only had the top 100 3-year runs in Standard Deviation to work with from the OP, it's possible someone was e.g. 101st in standard deviation, but was ranked higher in OSRS or series_won such that they'd sneak into the bottom half of the list.

Overall, this gets closer to our prior (e.g. I'd definitely have the 3-year run of the 96-98 Bulls to be better than any 3-year run from the Russell Celtics), but since this formula was calculated to track to our prior, that makes sense (and doesn't make it any better necessarily, without calculating which weighting is most predictive).

DraymondGold wrote:.

No problem at all, it's just a matter of re-running some scripts. As requested, I expanded the list to include the top 300 teams so you can see which teams just missed out on the cut-off point, like mid '90s HOU and early 70's NYK. I removed entries for teams with overlapping seasons (both aforementioned HOU and NYK make the the top-100 cut now) and keeping the one with the highest score (I actually realized there was a mistake in the code so I will have to update the previous listings ):


The list if we include entries with overlapping season windows, like I did initially:


I do plan to post, like I did with the DRtg topic, also share the other [1,4,5,6,9]-season windows. First I will do the ORtg though!

Pretty neat set up. That GSW dynasty with Durant was something else.

Predictions on any current teams making the grade? Where would current Den rank here?

Nvm see 21-23 ranked as 198.

DraymondGold wrote:.

Edit, had to make a new post cause of character limit being reached.

Also I did consider a scoring system, but felt that it makes stuff too over-convulated and even more arbitrary than it already is. This is the scoring system I came up with:


If you then sort by the following order (1) # of titles won, (2) points scored, and (3) zSRS you get the following list:

This doesn't necessarily say much about how strong a team is, but rather how well they managed to achieve their objective, which is advance as far as possible in the playoffs and ultimately win a title.

If you are getting into game 7s you are not really that dominant. 1960s Celtics were clutch and kept on winning their game 7s but they got into too many game 7s to call them dominant.

You are not a serious person if winning eight consecutive titles is not “dominant”. It is quite literally the most dominant stretch of play in any North American league; no other professional team ever managed more than five consecutive, and even Wooden’s Bruins “only” got to seven consecutive.

Wring our hands because theoretically they were pretty close to maybe not winning every time, utter joke.