Has anyone looked at team USG% trends close + late?

[TeamEd]

Looking at Memphis' consistently ridiculous record in close games (with Conley injured this year) has me thinking if missing the team's primary ballhandler in close + late situations might actually be a stealth advantage?

So far, I have done no work on it, but my thinking would be this: Intuitively, I expect a team's best players to shoot more frequently close + late. Intuitively, I also expect to see those players shoot less efficiently in those situations: as playing to to set up a clutch shot for, say, Damian Lillard, is 1. playing to the defense's expectations and 2. ignoring 2nd and especially 3rd and 4th options that are efficient and which a team might otherwise look for in a non-clutch situation.

Which gets me thinking: Is there a relationship between a change in team USG distribution close and late, and winning? Does a more unbalanced offence produce worse outcomes? What about a less unbalanced offence?

So, before I look into this, has anyone done a full look at close + late team usage trends? If not, I'll see if I can do it.

The end goal would be to see if I can find a correlation with teams with records that consistently outperform their overall NetRtg, like the Grizzlies have for years.

Here's how I would do it:
1. If it doesn't exist, create a measure of team offensive balance (think a gini coefficient https://en.wikipedia.org/wiki/Gini_coefficient for basketball, where 0 is perfect USG% equality and 100 is perfect inequality).
2. Apply that same measure to extreme close + late situations (like under 2 mins, +/- 3 points.)
3. Compare the two numbers, with an offense that becomes more unbalanced getting a + number and a team that becomes less unbalanced getting a - number. This numbers shows in which direction teams change their offense with the game on the line.
4. Correlate that with outcome, either measured in change in offensive efficiency vs. the team's overall rate or, simply, winning.
5. See what I get, and find out if a key to clutch success is to stay within your overall offensive pattern, or even to share the ball *more* than your team's typical offense.

Sample size would be extremely low, so I'd have to run it over several years...

Any thoughts?

My last project from a while back before I got a job was kinda similar, looking at change in player offensive trends while trailing: http://www.apbr.org/metrics/viewtopic.php?f=2&t=8885
Here are those Tableaus:
- Assists vs. expected while trailing: https://public.tableau.com/shared/5YN49 ... VizHome=no
- Shots vs. expected while trailing: https://public.tableau.com/shared/5SCFW ... VizHome=no
- Numbers from 96-97 to 2014-15: https://public.tableau.com/views/NBACar ... VizHome=no

[bchaikin]

you're talking about a team that is 18-9 but with the average per game point differential (+0.1 pts/g) of a team that historically has played close to .500 ball...

they are 1st in the league in defensive efficiency, but 3rd worst in offensive efficiency...

they've won 4 OT games and another 3 games by just 4 points total - that's 7 games with just a 4 point total differential after regulation time...

they've also lost games by 26, 19, 17, 15, 11, 10, and 9 points - that's 7 of 9 losses by an average of 15 points...

considering how big most of their losses have been, i would not ascribe their close wins to any definitive attribute - they've simply been lucky/fortunate in a small sample size of games played...

they are playing at a .667 clip (18-9). no team over a full season in nba history has played at even a .600 clip with an average per game point differential of worse than +1.8 pts/g...

[TeamEd]

you're talking about a team that is 18-9 but with the average per game point differential (+0.1 pts/g) of a team that historically has played close to .500 ball...

they are 1st in the league in defensive efficiency, but 3rd worst in offensive efficiency...

they've won 4 OT games and another 3 games by just 4 points total - that's 7 games with just a 4 point total differential after regulation time...

they've also lost games by 26, 19, 17, 15, 11, 10, and 9 points - that's 7 of 9 losses by an average of 15 points...

considering how big most of their losses have been, i would not ascribe their close wins to any definitive attribute - they've simply been lucky/fortunate in a small sample size of games played...

they are playing at a .667 clip (18-9). no team over a full season in nba history has played at even a .600 clip with an average per game point differential of worse than +1.8 pts/g...

Yeah. That's the point. That's the specific outlier, that's leading me to something I think I can investigate.

In the last few years, a rough glance says the Grizzlies have clearly outperformed NetRtg in, let me see... 2016-17, 2015-'16 and 2013-'14, and won about as often as you'd expect in 2014-15 and 2012-13, when they had 55-win teams.

That's pretty consistently better than point differential over a fairly long stretch of time. And yeah, it might be luck, but it might also point to something the team does that makes it perform well in situations that are coinflips for other teams.

Again, I haven't done anything to test this. But if no one else has done it already, I think there's fertile ground here.

[bchaikin]

In the last few years, a rough glance says the Grizzlies have clearly outperformed NetRtg in, let me see... 2016-17, 2015-'16 and 2013-'14...

yeah, it might be luck, but it might also point to something the team does that makes it perform well in situations that are coinflips for other teams...

the 13-14 and 16-17 memphis teams have different coaches and only 4 players in common - gasol, randolph, conley, and t.allen...

what could that something be, other than luck?...

[TeamEd]

In the last few years, a rough glance says the Grizzlies have clearly outperformed NetRtg in, let me see... 2016-17, 2015-'16 and 2013-'14...

yeah, it might be luck, but it might also point to something the team does that makes it perform well in situations that are coinflips for other teams...

the 13-14 and 16-17 memphis teams have different coaches and only 4 players in common - gasol, randolph, conley, and t.allen...

what could that something be, other than luck?...

Uh... again, that's the whole question: I have *no idea* if it's luck (or even if there is any signal there at all) but I do have a hypothesis to test, and an interesting way to investigate, I think. Luck is obviously the null.

First, though, I want to know if anyone has done it already.

[Crow]

Go for it Ed when time permits.

I looked briefly back at your referenced earlier threads. Will try to look them closer later.

Past and current query are worth doing.

[Mike G]

In games decided by 5 or less, the Grizz are 5-0 with Conley and 5-0 without him.

oops: They lost one last night; now 5-1 with MC

[TeamEd]

Hey, look: Tom Haberstroh is all over the Grizzlies' consistent clutch success.

http://www.espn.com/nba/story/_/id/1827 ... h-team-nba

Some interesting stuff happening in Memphis.

Meanwhile, all my old sheets are corrupt, so I gotta relearn how to scrape NBA.com to do this for real. Not sure I can actually pull it off without play by play data, however.

[Mike G]

I'm finding that team Ast/FG is positively correlated with point differential, and even more strongly correlated with Win%.
Additionally and consistently, Ast/FG ratio correlates with (Wins - PythWins).

W - PW is not always exactly equivalent to 'winning the close ones'. It also includes the fact that blowouts don't create extra wins. But they're generally the same thing.

Higher Ast/FG is also not quite the same as 'having your starting PG on the floor', but it might be good as a quick overview.

[Crow]

It is probably a combination of things. Perhaps including player age. Over last 4 seasons the Griz on average are the 3.7th oldest team minutes weighted and not younger than 6th oldest.

Relative consistency of main players may matter in general, in the most used lineup or perhaps a 3-4 man core, especially if it is used to close games.

[sndesai1]

maybe we should start having a contest next season to see who can best predict the difference between pyth win% and actual win%
i'd be curious to see how many people do much better than just guessing 0 for every team