RealGM

How good are the best players’ teams at closing out games, relative to how good the teams are?

SRS, or Simple Rating System, is commonly considered the best box-score stat for measuring team dominance in the regular season. It's basically Margin of Victory, adjusted for the difficulty of opponent. Regular season SRS is more predictive of playoff success than regular season record. However, there is still valuable information in regular season record. For example, it is record that determines playoff seeding, not SRS, and of course seeding affects how difficult your playoff run will be.

There are teams that seem to consistently (slightly) outperform or underperform in their record vs SRS. These are teams that are better or worse at closing out close games, relative to their SRS. This can be found when a team’s actual record differs from their expected record based on their SRS.

The problem: since we’re dealing with small samples (1, 2, or 3 wins over/under the expected wins), most of the signal will be obscured by noise. But we can try to tease out the signal from the noise by 1) looking at years when there’s a significant difference between actual wins and expected wins then applying context, or 2) by looking at multi-year samples. Here, I’ve done the latter. Specifically, I’ve looked at the best players’ teams across 10-year primes to see how their actual record compares to their SRS-expected record.

Here are how the top players' teams look:
Player // # wins per season - # expected wins per season // # seasons outperforming expectations - # seasons underperforming expectations

Russell: +1.3 extra wins per season, 5 extra seasons outperforming expectations
Wilt: +0.1, 1
Kareem: -1.9, -6
Bird: +1.8, 5
Magic: +1.9, 6
Jordan: -0.1, 0
Hakeem: +1.8, 4
Shaq: +1.7, 6
Duncan: -1.6, -4
Garnett: +0.1, 2
LeBron: +2.0, 6
Curry: +0, 0

Putting players’ teams into tiers, we have
Teams perform better than expected: LeBron +2, Magic +1.9, Hakeem + 1.8*, Bird +1.8, Shaq +1.7*, Russell +1.3
Teams perform similar to expected: Wilt +0.1*, KG +0.1, Curry +0, Jordan -0.1
Teams perform worse than expected: Duncan -1.6, Kareem -1.9

*Hakeem rises above Magic if we shift his 10-year prime, falls below if not.
*Shaq rises if we shift his 10-year prime
*Wilt stays over KG if we use his younger 10-best years, falls below Curry if we use his older 10-best years.

The raw data: Wins vs Expected Wins

Comments

How much can we take from this? Well, probably not too much. As above, the primary uncertainty comes from the fact that this measurement is so noisy. Still, there are additional sources of error:
-Systemic Error: Blowout games are known to make SRS less accurate. Do blowout games limit accuracy here? For example, high-SRS superteams might be extra likely (relative to their record) to have blowout victories, over-inflating their SRS relative to their record. Likewise, low-SRS lottery teams might be extra likely (relative to their record) to have blowout losses, under-inflating their SRS relative to their record. If so, then players on the best teams might be underrated and players on the worst teams might be overrated here (or vice-versa if the opposite were the case). We’d need to look across a larger sample of teams to find out.

-Potential Error: do injuries affect SRS and record in different ways? For example, if a teammate gets injured for one game, does that drop SRS by a smaller amount but make it more likely to lose that one game? (This may relate to resolution)

-Measurement Error: Resolution limits. You can have 6.05 SRS but you can’t have 60.5 wins. This limits how accurately we can compare record and SRS.

For potential next steps, we might try to address these errors (e.g. curving down the effect of blowouts / treating injuries more carefully), consider playoff data (where some teams may improve or worsen), or look more specifically at close games only. Or perhaps everything is too noisy to believe anything.

Regardless, it was still interesting to look at! And it could be argued to mentally ever-so-slightly curve-up the teams of prime LeBron, Magic, Hakeem, Bird, Shaq, Russell; mentally maintain the teams of prime Wilt, KG, Curry, and Jordan; and mentally curve down the teams of prime Duncan and Kareem. Or… perhaps not. The biggest change is only ~2.5% increase or decrease in games won compared to the expected wins, so it’s pretty small regardless.

What do y’all think? Is there any signal to find in this noise? How good are the best players’ teams at closing out games, relative to how good the teams are?

If finding out whether teams were good at closing out games I feel like there are better options. Things like what their record was in close games or how often they lost games despite starting the 4th quarter with a lead (or the other way around of winning after being behind late).

Dutchball97 wrote:If finding out whether teams were good at closing out games I feel like there are better options. Things like what their record was in close games or how often they lost games despite starting the 4th quarter with a lead (or the other way around of winning after being behind late).

Honestly I agree. These numbers were just easier to find/calculate than those (though I'm open to suggestions if there's a nice way to access those more accurate numbers).

The one thing that makes these numbers interesting is that it has the potential to isolate the ability to win close games after correcting for the overall team ability. Whereas I might worry that some sort of stat like "record in games where the score was close at the start of the 4th" might just end up giving us the signal that the better teams are better and worse teams are worse. Which might be true! But then we'd just be measuring how good the teams are in general, not how good they are at the specific skill of closing games out. If we use one of the other stats, what would the best way be to correct for this?

Interesting look, thank you. I will have to look and think about this one a bit.

DraymondGold wrote:

Dutchball97 wrote:If finding out whether teams were good at closing out games I feel like there are better options. Things like what their record was in close games or how often they lost games despite starting the 4th quarter with a lead (or the other way around of winning after being behind late).

Honestly I agree. These numbers were just easier to find/calculate than those (though I'm open to suggestions if there's a nice way to access those more accurate numbers).

The one thing that makes these numbers interesting is that it has the potential to isolate the ability to win close games after correcting for the overall team ability. Whereas I might worry that some sort of stat like "record in games where the score was close at the start of the 4th" might just end up giving us the signal that the better teams are better and worse teams are worse. Which might be true! But then we'd just be measuring how good the teams are in general, not how good they are at the specific skill of closing games out. If we use one of the other stats, what would the best way be to correct for this?

Lots of interesting thoughts here. How about looking at performance against teams that were better, equal or maybe slightly worse to weed out the bad teams they should be beating anyway? You'd still need to account for injuries as well since a good team might not be as competitive with one or more key contributors out or even teams with bad records because injuries early in the season who are now much better than their record due to their guys being back. I've got to wonder though how many games a season would be left for the sample if being that specific.

Dutchball97 wrote:

DraymondGold wrote:

Dutchball97 wrote:If finding out whether teams were good at closing out games I feel like there are better options. Things like what their record was in close games or how often they lost games despite starting the 4th quarter with a lead (or the other way around of winning after being behind late).

Honestly I agree. These numbers were just easier to find/calculate than those (though I'm open to suggestions if there's a nice way to access those more accurate numbers).

The one thing that makes these numbers interesting is that it has the potential to isolate the ability to win close games after correcting for the overall team ability. Whereas I might worry that some sort of stat like "record in games where the score was close at the start of the 4th" might just end up giving us the signal that the better teams are better and worse teams are worse. Which might be true! But then we'd just be measuring how good the teams are in general, not how good they are at the specific skill of closing games out. If we use one of the other stats, what would the best way be to correct for this?

Lots of interesting thoughts here. How about looking at performance against teams that were better, equal or maybe slightly worse to weed out the bad teams they should be beating anyway? You'd still need to account for injuries as well since a good team might not be as competitive with one or more key contributors out or even teams with bad records because injuries early in the season who are now much better than their record due to their guys being back.

Good idea!

I've got to wonder though how many games a season would be left for the sample if being that specific.

Probably not many! Which may point us back to taking these large ~10-year samples, or perhaps this would only give us a handle on smaller stretches where teams drastically over-performed or underperformed.

...

One thing that might get mixed up in this signal is coasting. Take LeBron, who has a (well-deserved) reputation of being good in close games. However, his teams also have a reputation of coasting in the regular season, particularly in his 2nd Cleveland stint. Now teams that coast may have some reputation of not going fully all-out in regular season games, but occasionally "flip the switch" when games get close. This SRS-vs-Record signal might be sensitive to this kind of close-game "switch flipping". So how do LeBron's Cleveland teams look?

2015: 0 extra wins over expected
2016: +1 extra wins over expected
2017: +2 extra wins over expected
2018: +7 extra wins over expected

Wow, check out 2018! It definitely seems like in 2018 especially, LeBron's team were either 1) unusually good at closing out games vs their SRS, and/or 2) were coasting most of the time, but occasionally flipped the switch in close games.

Interesting stuff! So in theory, this kind of analysis should give us a signal either on 1) how good the team is at closing out (relative to their overall goodness) or 2) how much better the team might be in the regular season when not coasting.

Of course, the question then becomes how do we decipher between the two. Here, we'll almost certainly need to apply context. But it would be nice if we could get some approximate numbers to support our contextual analysis.

One path forward might be looking at games against teams of similar skill (like you suggest), but specifically including playoff games as well. This might increase our sample more (though it might be swayed by favorable or unfavorable matchups).

Fascinating work with a lot of potential.

What do I mean by that?

Well, my first parsing of the post made me assume you were talking playoff over-performance… because that distribution is extremely similar to the results you may see performing a similar postseason exercise — albeit reflecting slightly more harshly on Jordan, who does have postseason “over-performances” in 1993 (slightly) and 1989 (somewhat more significantly, although the Price injury clouds those results). So then my question becomes, is this therefore a decently correlative signal to a team’s or player’s ability to elevate and handle postseason adversity?

As you said, SRS is generally a better marker of teams than raw wins. If two teams are at home against the same opponent, you should comparatively favour the higher SRS team of the two. However, is that true if the higher SRS team is on the road? Would take a lot of work, but deeply curious whether it is more important to have a homecourt advantage or to have an SRS advantage — or otherwise, what is the approximate equaling point / SRS value of a team receiving homecourt (and this of course still significantly depends on the specific team and homecourt). Without considering team health, there are after all only so many series where the home team had a lower SRS value.

I think because of the 1998 Finals, Jordan at least has more road wins than SRS underdog wins. Same I think with Lebron (2011 Bulls and 2017 Celtics) and Russell (1969 76ers and Lakers). Hakeem I think has more SRS underdog wins than road wins (1994 Knicks and 1997 Sonics); gut feeling would be that is probably the rarer dynamic, and unless we think Houston had some incredible home advantage, that probably plays to that idea of Hakeem being a nearly unprecedented playoff riser (or alternatively, being an unprecedented regular season sandbagger lol).

Cool stuff to consider.

DraymondGold wrote:

Dutchball97 wrote:If finding out whether teams were good at closing out games I feel like there are better options. Things like what their record was in close games or how often they lost games despite starting the 4th quarter with a lead (or the other way around of winning after being behind late).

Honestly I agree. These numbers were just easier to find/calculate than those (though I'm open to suggestions if there's a nice way to access those more accurate numbers).

The one thing that makes these numbers interesting is that it has the potential to isolate the ability to win close games after correcting for the overall team ability. Whereas I might worry that some sort of stat like "record in games where the score was close at the start of the 4th" might just end up giving us the signal that the better teams are better and worse teams are worse. Which might be true! But then we'd just be measuring how good the teams are in general, not how good they are at the specific skill of closing games out. If we use one of the other stats, what would the best way be to correct for this?

Yeah, this is key. Any "clutch" argument is based on the idea of improved performance at certain points, relative to performance at other points, no?

If you just go by record without weighting im not seeing how thats any different than just assessing how good a team is

DraymondGold wrote:How good are the best players’ teams at closing out games, relative to how good the teams are?

SRS, or Simple Rating System, is commonly considered the best box-score stat for measuring team dominance in the regular season. It's basically Margin of Victory, adjusted for the difficulty of opponent. Regular season SRS is more predictive of playoff success than regular season record. However, there is still valuable information in regular season record. For example, it is record that determines playoff seeding, not SRS, and of course seeding affects how difficult your playoff run will be.

Well RS record plus for most of history divisional quirks. And there were some odd early formats. And for some time iirc 1-3, 2-4 seeding (better to be 2 than 1, and to be 4 than 3). Regardless this is not so much intrinisic to the record as influence of the record which seems separate from information "in" the record.

DraymondGold wrote:There are teams that seem to consistently (slightly) outperform or underperform in their record vs SRS. These are teams that are better or worse at closing out close games, relative to their SRS. This can be found when a team’s actual record differs from their expected record based on their SRS.

2 questions here
1) Given SRS neutralizes for opponents and record doesn't, doesn't it partially reflect who had a difficult or easy record? Or are you just doing points dif pythag wins (such as those at basketball-reference)? If not would then shouldn't this schedule stuff taint the data for the purposes you're using it for?
2) Are there teams that consistently outperform their points dif pythag wins (there should be ones that do so on SRS pythag wins, because conference balance typically won't wildly shift year to year, so on a version of SRS pythag wins posted on here apparently originally from Ben Taylor, the 80-88 Lakers consistently outwon their SRS pythag wins in a weak conference, though the margin might seem to suggest that something else is at play too [playing down to competition? conserving energy as frequent defending champs rather than stomping teams? good in close games? lucky? w-l somehow else capturing something SRS is missing])? Moreso than would be expected by luck? I don't know for sure either way (guess would I think be yes but marginal, noisy), but would want to know more before going too deep into this.

DraymondGold wrote:The problem: since we’re dealing with small samples (1, 2, or 3 wins over/under the expected wins), most of the signal will be obscured by noise. But we can try to tease out the signal from the noise by 1) looking at years when there’s a significant difference between actual wins and expected wins then applying context, or 2) by looking at multi-year samples. Here, I’ve done the latter. Specifically, I’ve looked at the best players’ teams across 10-year primes to see how their actual record compares to their SRS-expected record.

Here are how the top players' teams look:
Player // # wins per season - # expected wins per season // # seasons outperforming expectations - # seasons underperforming expectations

Russell: +1.3 extra wins per season, 5 extra seasons outperforming expectations
Wilt: +0.1, 1
Kareem: -1.9, -6
Bird: +1.8, 5
Magic: +1.9, 6
Jordan: -0.1, 0
Hakeem: +1.8, 4
Shaq: +1.7, 6
Duncan: -1.6, -4
Garnett: +0.1, 2
LeBron: +2.0, 6
Curry: +0, 0

Putting players’ teams into tiers, we have
Teams perform better than expected: LeBron +2, Magic +1.9, Hakeem + 1.8*, Bird +1.8, Shaq +1.7*, Russell +1.3
Teams perform similar to expected: Wilt +0.1*, KG +0.1, Curry +0, Jordan -0.1
Teams perform worse than expected: Duncan -1.6, Kareem -1.9

*Hakeem rises above Magic if we shift his 10-year prime, falls below if not.
*Shaq rises if we shift his 10-year prime
*Wilt stays over KG if we use his younger 10-best years, falls below Curry if we use his older 10-best years.

The raw data: Wins vs Expected Wins

Spoiler:

*note: shortened seasons are extended to an 82-win pace.
*alternate 10-year averages are included when the best 10 years are less clear

Bill Russell’s Celtics:
1957: -5
1958: 1
1959: 2
1960: 2.5
1961: 4
1962: -1.5
1963: 0
1964: 0
1965: 3
1966: 1
1967: 0
1968: 2
1969: -8
10-year average: 1959-1968: +1.3 extra wins per season, 5 extra seasons outperforming expectations

Wilt’s Warriors, 76ers, Lakers:
1960: 5
1961: 4
1962: 1
1963: -3
1964: -5
1965: -4
1966: 3
1967: 5
1968: 0
1969: 3
1970: 0 *injury
1971 : -3
1972: 0
1973: -3
10-year average: 1960-1969: +0.9 extra wins per season, 3 extra seasons outperforming expectations
10-year average: 1961-1969/1971: +0.1 extra wins per season, 1 extra seasons outperforming expectations
10-year average: 1962-1969/1971-1972: -0.3 extra wins per season, 0 extra seasons outperforming expectations

Kareem’s Bucks, Lakers:
1970: 3
1971 : -4
1972: -5
1973: -2
1974: -3
1975: -4
1976: -2
1977: 4
1978: -4
1979: -3
1980: 4
1981: 3
1982: 3
1983: 3
1984: 3
1985: 3
1986: 2
1987: 2
1988: 7
1989: -3
10-year average: 1970-1979: -2.0 extra wins per season, -6 extra seasons outperforming expectations
10-year average: 1971-1980: -1.9 extra wins per season, -6 extra seasons outperforming expectations
10-year average: 1972-1981: -1.2 extra wins per season, -5 extra seasons outperforming expectations

Bird’s Celtics:
1980: 0
1981: 4
1982: 4
1983: 0
1984: 3
1985: 4
1986: 2
1987: 0
1988: -1
1989: -3 *injury
1990: 2
1991: 0
10-year average: 1980-1988, 1990: +1.3 extra wins per season, 5 extra seasons outperforming expectations

Magic’s Lakers:
1980: 4
1981: 3
1982: 3
1983: 3
1984: 3
1985: 3
1986: 2
1987: 2
1988: 7
1989: -3
1990: -3
1991: 2
10-year average: 1982-1991: +1.9 extra wins per season, 6 extra seasons outperforming expectations

Jordan’s Bulls:
1985: -2
1986: -2 *injury
1987: -5
1988: -2
1989: 0
1990: 6
1991: -3
1992: 0
1993: -1
1994: 6 *retired
1995: -5 *semi-retired
1996: 2
1997: 1
1998: 1
10-year average: 1987-1993/1996-1998: -0.1 extra wins per season, 0 extra seasons outperforming expectations

Hakeem’s Rockets:
1985: 3
1986: 4
1987: -1
1988: 3
1989: 3
1990: -5
1991: 1
1992: 7
1993: 4
1994: 5
1995: -1
1996: 2
1997: 5
1998: 4
1999: 6
2000: -5
2001: -4
2002:
10-year average: 1986-1995: +2 extra wins per season, 4 extra seasons outperforming expectations
10-year average: 1987-1996: +1.8 extra wins per season, 4 extra seasons outperforming expectations
10-year average: 1988-1997: +2.4 extra wins per season, 6 extra seasons outperforming expectations

Shaq’s Magic, Lakers, Miami, Suns, Boston:
1993:
1994: -2
1995: -2
1996: 4
1997: 4
1998: 1
1999: 2
2000: 4
2001: 4
2002: -3
2003: 1
2004: 2
2005: 2
2006: 1
2007: 7
2008:
2009:
2010:
2011 :
10-year average: 1995-2004: +1.7 extra wins per season, 6 extra seasons outperforming expectations
10-year average: 1996-2005: +2.1 extra wins per season, 8 extra seasons outperforming expectations

Duncan’s Spurs:
1998: 5
1999: 0
2000: -5
2001: -4
2002: -1
2003: 3
2004: -4
2005: -3
2006: 3
2007: -5
2008: 0
2009: 3
2010: -5
2011 : 3
2012: 1
2013: -1
2014: -1
2015: -4
2016: 0
10-year average: 1998-2007: -1.1 extra wins per season, -3 extra seasons outperforming expectations
10-year average: 1999-2008: -1.6 extra wins per season, -4 extra seasons outperforming expectations

KG’s Timberwolves, Celtics:
1996: 0
1997: 4
1998: 3
1999: 1
2000: 1
2001: 1
2002: -1
2003: 3
2004: 0
2005: -2
2006: -3
2007: 0
2008: 1
2009: 1 *injury
2010: -1
2011 : 1
2012: 0
2013: 5
2014:
2015:
2016:
10-year average: 1999-2008: +0.1 extra wins per season, 2 extra seasons outperforming expectations

LeBron’s Cavs, Heat, Cavs, Lakers:
2004: 3
2005: 0
2006: 3
2007: -1
2008: 6
2009: 2
2010: 3
2011 : -2
2012: 0
2013: 6
2014: 1
2015: 0
2016: 1
2017: 2
2018: 7
2019: 0 *injury
2020: 1
2021: -1
2022:
10-year average: 2008-2017: +1.9 extra wins per season, 6 extra seasons outperforming expectations
10-year average: 2009-2018: +2.0 extra wins per season, 6 extra seasons outperforming expectations
10-year average: 2010-2018/2020: +1.9 extra wins per season, 6 extra seasons outperforming expectations

Curry’s Warriors:
2011 : 1
2012: -10 *injury
2013: 2
2014: -5
2015: 0
2016: 6
2017: -2
2018: 1
2019: -2
2020: -4 *injury
2021: 0
2022: -1
10-year average: 2011/2013-2019/2021-2022: +0 extra wins per season, 0 extra seasons outperforming expectations

Comments

How much can we take from this? Well, probably not too much. As above, the primary uncertainty comes from the fact that this measurement is so noisy. Still, there are additional sources of error:
-Systemic Error: Blowout games are known to make SRS less accurate. Do blowout games limit accuracy here? For example, high-SRS superteams might be extra likely (relative to their record) to have blowout victories, over-inflating their SRS relative to their record. Likewise, low-SRS lottery teams might be extra likely (relative to their record) to have blowout losses, under-inflating their SRS relative to their record. If so, then players on the best teams might be underrated and players on the worst teams might be overrated here (or vice-versa if the opposite were the case). We’d need to look across a larger sample of teams to find out.

-Potential Error: do injuries affect SRS and record in different ways? For example, if a teammate gets injured for one game, does that drop SRS by a smaller amount but make it more likely to lose that one game? (This may relate to resolution)

-Measurement Error: Resolution limits. You can have 6.05 SRS but you can’t have 60.5 wins. This limits how accurately we can compare record and SRS.

For potential next steps, we might try to address these errors (e.g. curving down the effect of blowouts / treating injuries more carefully), consider playoff data (where some teams may improve or worsen), or look more specifically at close games only. Or perhaps everything is too noisy to believe anything.

Regardless, it was still interesting to look at! And it could be argued to mentally ever-so-slightly curve-up the teams of prime LeBron, Magic, Hakeem, Bird, Shaq, Russell; mentally maintain the teams of prime Wilt, KG, Curry, and Jordan; and mentally curve down the teams of prime Duncan and Kareem. Or… perhaps not. The biggest change is only ~2.5% increase or decrease in games won compared to the expected wins, so it’s pretty small regardless.

What do y’all think? Is there any signal to find in this noise? How good are the best players’ teams at closing out games, relative to how good the teams are?

Would think injuries would hurt SRS more than W-L. W-L being binary you can still win even if lesser than previously, and if you lose you can only add one loss per game. SRS you can take a hit versus expectations even in wins and the downside is less limited. This otoh so could be missing stuff.

Would have been inclined to tag more years for injury to the star (possibly exclude?, don't know) e.g. Jabbar '75, '78 plays 65, 62 games, and is lower by 801 and 751 minutes than the preceding seasons. Olajuwon plays 56 games in '91 (a couple others around 70, fwiw). Shaq played an average of 2020 RS minutes per season 96-98 and from 2002 on never more than 2535, '57 Celtics are without Russell for, iirc, the first 24 games (of 72).

Feel like I'd want to know better what hypothesis you're testing to understand if this is the best way of looking at it (instinct in general would be to want more normal teams for a greater sample, but depends as I say on the question but maybe even just as a control; if you are especially interested in close games, inclination would be to look only at those as other factors may be at play for differences here).

How good are the best players’ teams at closing out games, relative to how good the teams are?

In general my understanding was the (statistically) informed consensus was worse in a statistical sense if not in an absolute one, because the tiny samples give you less opportunity to demonstrate your superiority. Much greater role of chance. So worst ever SRS '93 Mavs can go 5-1 in games decided by 3 or less, '92 Bulls among the strongest go 3-8 [both per STATS Inc]. Of course as before it depends what you want to measure and technically how well these teams "closed out" the games depends on what the position was at the start of the "closing out" time period (more of this type of stuff presumably available in PbP era).

Anyway, always interesting to see people put time and effort into this type of stuff, even if there isn't an immediate "takehome" message (and I see it's inspired some discussion since I started writing).

AEnigma wrote: As you said, SRS is generally a better marker of teams than raw wins. If two teams are at home against the same opponent, you should comparatively favour the higher SRS team of the two. However, is that true if the higher SRS team is on the road? Would take a lot of work, but deeply curious whether it is more important to have a homecourt advantage or to have an SRS advantage — or otherwise, what is the approximate equaling point / SRS value of a team receiving homecourt (and this of course still significantly depends on the specific team and homecourt). Without considering team health, there are after all only so many series where the home team had a lower SRS value.

I think because of the 1998 Finals, Jordan at least has more road wins than SRS underdog wins. Same I think with Lebron (2011 Bulls and 2017 Celtics) and Russell (1969 76ers and Lakers). Hakeem I think has more SRS underdog wins than road wins (1994 Knicks and 1997 Sonics); gut feeling would be that is probably the rarer dynamic, and unless we think Houston had some incredible home advantage, that probably plays to that idea of Hakeem being a nearly unprecedented playoff riser (or alternatively, being an unprecedented regular season sandbagger lol).

Cool stuff to consider.

Cool idea for a project! As you say, it might take a lot of work, but it might be informative.

OhayoKD wrote:

DraymondGold wrote:

Dutchball97 wrote:If finding out whether teams were good at closing out games I feel like there are better options. Things like what their record was in close games or how often they lost games despite starting the 4th quarter with a lead (or the other way around of winning after being behind late).

Honestly I agree. These numbers were just easier to find/calculate than those (though I'm open to suggestions if there's a nice way to access those more accurate numbers).

The one thing that makes these numbers interesting is that it has the potential to isolate the ability to win close games after correcting for the overall team ability. Whereas I might worry that some sort of stat like "record in games where the score was close at the start of the 4th" might just end up giving us the signal that the better teams are better and worse teams are worse. Which might be true! But then we'd just be measuring how good the teams are in general, not how good they are at the specific skill of closing games out. If we use one of the other stats, what would the best way be to correct for this?

Yeah, this is key. Any "clutch" argument is based on the idea of improved performance at certain points, relative to performance at other points, no?

If you just go by record without weighting im not seeing how thats any different than just assessing how good a team is

Good points! Re: your last comment, the idea (or perhaps hope) is that SRS (not Record) is better at assessing how good a regular season team is... and if so, differentials between record and SRS may contain some sort of information (noise, along with coasting, ability to close out close games, etc.).

Like you say, record is also closely tied with goodness, so we have to be careful how we get at that information. This post was more of a first attempt. But I can't help feeling there's information in the data somewhere, if we can just figure out how to measure it. (that's partly why I made this post... to brainstorm next steps!)

Re: clutchness, good points! I've often been a proponent that Clutchness is overrated in popular media. Thinking Basketball (the book) has made a far more in-depth argument of this than I will, but the crux of it boils down to the fact clutchness comes into play so few times in a season, that your overall ability to win games (mostly in the first 47 minutes) is the dominant part of where your value comes from.

Still, if we can get away from all the media baggage ("SkIp BaYlOr lists his ClUtChEsT athletes of aLl tImE!!" ), team/player clutchness is still a fun thing to discuss, and there is a real signal for it in the data somewhere! It just plays a relatively smaller role in the value/goodness of a team than talk shows tend to think.

i knew the hakeem data, but it arguably makes him more impressive compared to his already good reputation for underdog wins.
he won tons of road series, but only after he had already won more games than he should have.

AEnigma wrote:
I think because of the 1998 Finals, Jordan at least has more road wins than SRS underdog wins.

i'm getting 5 and 5 for road (1989 1st and 2nd, 1993 ECF/Finals, 1998 finals) vs underdog (1989 1st and 2nd, 1990 2nd, 1993 semi's and finals). weirdly, 1993 semi's against cleveland were actually a -0.1 SRS underdog.

Hakeem I think has more SRS underdog wins than road wins (1994 Knicks and 1997 Sonics);

1994 semi's against phoenix were also an underdog home win.

that probably plays to that idea of Hakeem being a nearly unprecedented playoff riser (or alternatively, being an unprecedented regular season sandbagger lol).

it's definitely hard to find someone better. probably only coasting lebron in his 2nd cleveland stint matches up (with lebronto being an unbelievable -6.7 SRS series) and lebron clearly didn't care about the regular season. i've often wondered if there was something to hakeem's game that made him underperform in the regular season or if he was just that good in the playoffs. the fact he won 2 DPOY's and almost certainly should have won in 1989 and 1990 means it's hard to argue he was slacking at the part of the game most people slack on in the regular season. and his 1993 and 1994 regular seasons are basically unimpeachably great and had arguably his best supporting casts, with great health for the team in both years, and the rockets still finished 6th and 6th in SRS. the evidence all points to insane playoff elevation.

AEnigma wrote:As you said, SRS is generally a better marker of teams than raw wins. If two teams are at home against the same opponent, you should comparatively favour the higher SRS team of the two. However, is that true if the higher SRS team is on the road? Would take a lot of work, but deeply curious whether it is more important to have a homecourt advantage or to have an SRS advantage — or otherwise, what is the approximate equaling point / SRS value of a team receiving homecourt (and this of course still significantly depends on the specific team and homecourt). Without considering team health, there are after all only so many series where the home team had a lower SRS value.

if i use:

Pythagorean formula with 16.5 exponent to calculate win%
Consider the DRtg of the road team to be 108 (basically league average for last 40 years)
Then add the SRS differential to 108 for the ORtg (i assume this basically works)
Homecourt advantage of 3 points (+1.5 ORtg, -1.5 DRtg)
7 game series

I ended up getting that an SRS differential of only 0.45 was all that was needed to overcome the homecourt advantage and get the road team to win 50% of the series. if i reduce the homecourt advantage to 1.5, the SRS differential only needs to be 0.22. Is this just homecourt advantage divided by 7? could be since the only uncanceled advantage is getting game 7 at home so it's just homecourt advantage for 1 game (game 7) spread over 7 games.

a series like the 1994 finals with the knicks as a 2.29 SRS advantage team should have been the knicks winning about 64.4% of the time in a 2-2-1-1-1 format and 65.0% in a 2-3-2 format.

Owly wrote:

How good are the best players’ teams at closing out games, relative to how good the teams are?

In general my understanding was the (statistically) informed consensus was worse in a statistical sense if not in an absolute one, because the tiny samples give you less opportunity to demonstrate your superiority. Much greater role of chance. So worst ever SRS '93 Mavs can go 5-1 in games decided by 3 or less, '92 Bulls among the strongest go 3-8 [both per STATS Inc]. Of course as before it depends what you want to measure and technically how well these teams "closed out" the games depends on what the position was at the start of the "closing out" time period (more of this type of stuff presumably available in PbP era).

probably the best "great teams don't win close games, they avoid them" example is the 72-10 bulls to end the season. they went 18-3 in their final 21 games. their first loss? by 1 point. their second loss? by 1 point. their third loss? by 1 point.

f4p wrote:

Owly wrote:

How good are the best players’ teams at closing out games, relative to how good the teams are?

In general my understanding was the (statistically) informed consensus was worse in a statistical sense if not in an absolute one, because the tiny samples give you less opportunity to demonstrate your superiority. Much greater role of chance. So worst ever SRS '93 Mavs can go 5-1 in games decided by 3 or less, '92 Bulls among the strongest go 3-8 [both per STATS Inc]. Of course as before it depends what you want to measure and technically how well these teams "closed out" the games depends on what the position was at the start of the "closing out" time period (more of this type of stuff presumably available in PbP era).

probably the best "great teams don't win close games, they avoid them" example is the 72-10 bulls to end the season. they went 18-3 in their final 21 games. their first loss? by 1 point. their second loss? by 1 point. their third loss? by 1 point.

It's certainly a nice bit of trivia. My issue would be that the sample is perhaps too small, functionally only 3 games since the win margins aren't considered. As such there's a greater chance that this type thing could come up by chance and could be found elsewhere by those determined to do so.

Fwiw, using that STATS inc definition, (decided by 3 points or less) 96 RS Bulls go 5-3 (though this is vulnerable to tampering/dependent on definitions, make it 2 points and it's 1-3, make it 4 points and it's 10-3).