A Team Greater Than The Sum Of Its Parts

[Guy]

The results were simply that offensive efficiency increased as the standard deviation of:

2P FGA increased,
3P FGA decreased,
Offensive Rebounds decreased,
Assists decreased,
Turnovers increased,
FTA decreased.
I also did a simple correlation looking at the relationship between the ratio of 2PFGA/FGA and offensive efficiency, as well as the relationship between 3PFGA/FGA and offensive efficiency, among other similar correlations, to find that we want more 3P FGA and fewer 2P FGA.

Although hardly a novel observation on my part, I think it's worth recalling here that "correlation is not causation." We cannot simply assume that the causal arrow runs from variance to efficiency, and not the other way around. At most, we can say that these tend to be attributes of good offensive teams.

For example, we can NOT say that a team which decides to spread responsibility for 3PA will become more efficient, or even that on average such teams will become more efficient. In fact, it is extremely likely that low variance in 3PA simply signals that a team has more competent 3P shooters than average. A team that lacked such players, but nonetheless asked more guys to take 3s, would quickly see their efficiency decline.

Similarly, having more players capable of grabbing offensive rebounds is clearly a good thing, but that doesn't mean it would help a team to assign that responsibility to more players. For that matter, this might be a case of spurious correlation -- perhaps low OReb variance isn't predictive of a higher Oreb% (or not only that), but simply means you have taller-than-average players who get to the rim more successfully.

And because 3PA is so closely tied to players' 3P%, we certainly cannot say that teams in general should be taking more 3PA. That recommendation rests on an unstated assumption, which is that additional (marginal) 3PA would have the same efficiency as the mean of current 3PA. We have no reason at all to believe that would be true, and many reasons to think it is not.

[schtevie]

And because 3PA is so closely tied to players' 3P%, we certainly cannot say that teams in general should be taking more 3PA. That recommendation rests on an unstated assumption, which is that additional (marginal) 3PA would have the same efficiency as the mean of current 3PA. We have no reason at all to believe that would be true, and many reasons to think it is not.

Could you please elaborate on these remarks? It would be in some ways comforting to believe that the NBA, writ large, has finally learned to multiply by 1.5, but I see no compelling evidence for this proposition.

It is true that in the last few years the ratio of 3PA to other scoring attempts has stabilized in the NBA. However, there has (as but one formulation of opportunity cost, indeed, the most obvious one) been no accompanying diminution in the wedge in the return between long 2s and 3PA. None.

And what this implies about the "unstated assumption" is that the marginal 3PA would only have to be higher than the marginal long 2A foregone. Is there any reason - any at all - to think, that in globally optimizing offenses, that the 14 points of eFG% that currently separates these two scoring opportunities would evaporate on the margin? To begin, it is a very strong assumption that the additional 3 would necessarily come at the expense of an above-average long 2. And if this is not made, what is the argument? That an extra 0.24 or 0.20 or 0.16 points per shot isn't worth acquiring?

There is also, to my understanding, no suggestion in the data that more 3PAs comes at the expense of the only better type of shot, those at the rim. Then we see the actual range of 3PAs taken across teams the NBA. It is very large. Much room for many to make a positive change.

What other objections could there be? That 3 point shooting skills are highly inelastic? Really?

[Mike G]

I'd say as long as there is a 24-second shot clock, there will always be a similar proportion of long 2's.
When you can't get a shot near the rim, and you can't get a shot from the arc, you can often still get a long 2.

[Crow]

Yes the 24 second clock is a significant factor. But has any NBA team seriously worked at trying to eliminate the long 2 by holding regular scrimmages where the long 2 is prohibited or heavily discouraged? I am unaware of any pre-season games where this was being tried in an extreme test scenario fashion. Tell me if you've seen otherwise. If NBA teams are not doing either of these regularly, let alone doing it in real games, then I do not think they are doing enough to reduce "discretionary" mid and long 2 attempts.

[Mike G]

Suppose that 1/3 of the time, you cannot get a good look at 3, nor can you get near the rim with a decent shot. So 1 of 3 shots is going to be a long 2, a worse shot, or no shot.

The best teams, knowing that they have to be good at making the less-desirable shots, are probably going to practice making their long 2's in every scenario.

[schtevie]

I'd say as long as there is a 24-second shot clock, there will always be a similar proportion of long 2's.
When you can't get a shot near the rim, and you can't get a shot from the arc, you can often still get a long 2.

I am perlexed. There will "always be a similar proportion of long 2's."? How does that square with the historical record? 3PAs are currently at about 23% of all FGAs. When the 3P rule was first introduced, their fraction of the offense approximated zero. Given this change how could the share of long 2's have remained constant-ish?

And then across teams in any given year, you don't see a similar proportion of long 2s. Last year these varied from 15.3% to 34.7% of FGAs, sez hoopdata. There isn't even a very similar proportion of long 2s + 3PAs.

The existence of the 24-second shot clock only implies that some long 2s will inevitably be taken (some of which, of course, are optimal in conception).

[Mike G]

Sorry, let me clarify. At present and in the foreseeable future, I don't see how the league average shot distribution will change much.
Schtevie, you noted earlier that frequency and accuracy of 3-pt shots has topped out over the last few years. Teams have demonstrated the effectiveness, and yet they still take those pesky long-2's.

Of course there are great variations between players, and between teams that have these varying players. Coaches also make a difference. But success breeds success, and the league average doesn't seem to be changing.

So, it may well be that (a) all teams should minimize their long 2's, and (b) better teams tend to be better at this.
And (c) is still that the average team will take 20-30% of their shots from long-2 range.

[Bobbofitos]

Here's another angle to consider:

The existence of the 24-second shot clock only implies that some long 2s will inevitably be taken (some of which, of course, are optimal in conception).

What do you suppose is the limit of long 2s? In practice, at least, to what extent can a team truly reduce their 2PA in favor of 3PA?

I doubt any team has maxed out in this regard, but how far away are they, really?

[EvanZ]

Denver took 12.5 16-23 ft jumpers per game. Orlando was second with 13.4. The next closest team was Cleveland with 16.3. The average was 19.9. Charlotte took 27.8.

Denver got there by shooting 34 "At Rim" shots per game. The average was 25. That's a good tradeoff, if you can manage it. Here's the 3-year adjusted inside shooting rate I did a while back:



With McGee and Iguodala on board, I suspect Denver will lead the league again. Los Angeles (which was only average before) should take a big leap up the board with Howard there.

[Crow]

Expanding a bit on what Evan said:

Lowest average 16-23 foot FGAs per game per team:

2006-7: 16.4
2007-8: 12 (Orlando, Stan Van Gundy's first season)
2008-9: 13.6 (Orlando)
2009-10: 12.6 (Orlando)
2010-11: 14.7 (Orlando)
2011-12: 12.5 (Denver, Orlando 2nd)

You can get it down to 12 and be good against current defensive efforts but it is very rare- 3 times in 6 seasons. I'd guess that 10 is possible. Below that, may be possible but only with great effort and design and probably not much lower. Would probably have to take an available 3 before the last 10 seconds to accomplish this. The trade off in lost inside shots and FTs that migh t b efound with more looking might not be worth it but turnovers and OR% also are factors in the optimization. The 3 ball is the only the third best scoring option.

The league average in 2011-12 was 20. Getting to 10, might be worth 3 pts if the average 3 if .15 pts more than the average long 2. Teams should move mountains to gain 3 pts per game.


Does anyone have average 3 pt FG% when under heavy pressure? Under heavy and within the last 5-10 seconds of the shot clock? Does Synergy share that? I'd think that NBA perimeters I(with most of the minutes being starters) could create / chuck 3s at least 30% under lack of other options, heavy pressure and a short clock but tell me if you have data on this.

[Crow]

For all the scrub bigs who get 2nd, 3rd, 4+th chances, does anyone else think Fesenko should another job and maybe a better chance to play? Trivia but in 2011-12 in extremely light minutes he was tops in the league on some things: Evan's link for adjusted inside shooting rate and winshares per 48 (at.374). Not good on some career average boxscore stats (i.e. shooting) but decent to good RAPM 3 years in a row. I know he looks clumsy of the court, is a goofball, and maybe he doesn't practice hard or say the right things to coaches but his performance is better than that.

Just over 1000 minutes in 5 seasons. Only 43 games with 10+ minutes. Only 5 for over 20 minutes.

His per 36 minutes stats for those 5 games over 20 minutes?

11pts, 9 rebs, 70% FG%, 2.6 Blocks, 0.5 steals, 1 assist, 1.7 TOs, 3.8 PFs, 30% FT.
Yes these may be his best games but so few real chances to get comfortable and play.

[schtevie]

Sorry, let me clarify. At present and in the foreseeable future, I don't see how the league average shot distribution will change much.

I suppose I could be persuaded that the league average shot distribution won't change much in the near term. But this wouldn't be based upon the absence of real opportunities. These still exist. Rather it would imply that the "new" NBA isn't that much different from the old.

It's been about 45 years since the ABA came into being and with it the 3 point shot. Thus, decades to capitalize on the opportunity for those both willing and able to multiply by 1.5 and understand the concept of opportunity cost. Never throughout this history, however, was this opportunity decisively seized. Painting with a broad brush, just incremental bits of tentative progress.

So now these increments have gotten tiny or disappeared. What is a more reasonable prior (if you had to choose one)? That this stabilization of the 3PA share in the offensive mix reflects a very recent and thorough league-wide adoption of best practice, butting up against real constraints, or that the dead hand of tradition that has ever dominated 3P decision-making in the NBA - from the front office, through the coaching ranks, down to the players themselves - is what has stopped progress cold?

Schtevie, you noted earlier that frequency and accuracy of 3-pt shots has topped out over the last few years. Teams have demonstrated the effectiveness, and yet they still take those pesky long-2's.

I actually noted that the frequency only has topped out. As for the accuracy, my comment was only about the wedge between long 2s and 3s, which remains huge for each and every team in the NBA. But yes, teams still take those pesky long 2s.

Of course there are great variations between players, and between teams that have these varying players. Coaches also make a difference. But success breeds success, and the league average doesn't seem to be changing.

Very strange, no, success not breeding success? Let's take a brief pass through last season's data to see what there is to see. In particular, lets focus on long-range offense as a whole - the combination of long 2s and 3s - and the implications of choices made in this area.

Why focus on this combination? Well, long range shooting (long 2s and 3s) is a distinct component of offensive organization - what becomes available to you as you move away from the teeth of the defense - and this fraction of overall shot attempts is quite similar across teams (with a standard deviation of 3.96 percentage points). Hence, the approximate share of long range offense that is to take place behind the 3 point line is a basic and fundamental strategic choice.

Organizing the data on this dimension we see at the two extremes the hopelessly hapless Bobcats (32.6 of its long range shots being 3PAs) and the envelope-pushing Magic (66.8%). For the purposes of this exercise, wanting to infer potential opportunities for the "representative" team, let's throw these two out of the mix (we can include them if one wishes; it only makes things a bit clearer than the truth). What do some naive, univariate regressions then suggest to us?

First, for every percentage point increase in 3PA share, long-range eFG% (the weighted average of long 2s and 3s) increases 0.136 percentage points. What this implies for the average team is that the marginal benefit of a 3PA (at the expense of a long 2) is an additional 0.272 points. This is not small (and, one recalls, does not include the premium owing to a higher expected OR%) and, coincidentally, is equal to what is implied by the average wedge between 3P eFG% and long 2 FG% for the teams in question: 0.279 points per 3PA.

But might this hypothetical gain reflect the fact that teams that shoot more 3s are somehow "innately" better at it? Maybe, but if so, it is being directly offset by compensatory defensive pressure, because such a notional skill premium isn't apparent. Regressing 3P eFG% on 3PA share, you get that an additional percentage point increase in 3PA share is only associated with an 0.013 percentage point increase in 3P eFG%. Essentially no effect over the possible, relevant range.

On the face of it then, the very significant gains from three point shooting are, still, just there for the taking.

So, it may well be that (a) all teams should minimize their long 2's, and (b) better teams tend to be better at this. And (c) is still that the average team will take 20-30% of their shots from long-2 range.

So, I would summarize the situation as: (a) it is sure that most if not all teams should switch thier long 2s to 3s on the margin, (b) this benefit is apparently available to teams regardless of quality, and (c) I have no confidence, given the observed history of the NBA, that the ultimate potential - whatever it is - will be realized anytime soon.

[Mike G]

Earlier (mid page 1 of this thread) I wrote:

Defenses (a) protect the paint and (b) chase shooters off the 3-pt line. The mid-range shot is what's left.

As NBA offenses have exploited the 3-point advantage, defenses have reacted to stifle it. For many years, players and offenses were not good enough to make it a viable option. And since they've improved, so have defenses.

Suppose wide-open 3's are made about 50% of the time. That the Cavs last year allowed 41% from the arc would then imply their opponents were un-covered at the arc less than 1/2 of the time.
41 is about halfway between league-best 31 and 50.

I'm sure even the 2012 Celtics left a few shots open at the arc, in allowing 31%. But the range of these opponent 3FG% -- from very advantageous to really not -- proves that defense has much to do with the effectiveness of the shot.

[schtevie]

Mike, the only relevant way to introduce defense into the argument is to present evidence that they have "collectively" evolved to such a degree that further 3PA are either physically impossible or that the shot opportunities presented are so poor that no self-respecting player would take them.

There is no such evidence, on either account.

And as good as the 2012 Celtics were at defending the 3P shot, this was of a piece of their being a really, really good defense generally. But, more on point, they still allowed a 10.3 percentage point premium over long 2s.

Finally, though the historical record is not directly relevant on this point, it simply has to be said that the idea that "For many years, players and offenses were not good enough to make it a viable option." is completely and utterly incorrect. From its first introduction in the NBA the 3P shot began as an above-average below average shot, i.e. surely better at the margin that the long range 2s we have been discussing. And there were, way back then, many players capable of raising the average faster than was realized would they only have been encouraged to do so. And then by the onset of the 1990s, if I recall, 3P eFG% reached parity with non-3 FG%.

Yes, defenses have been improving. And it may turn out to be the case in the tug of war between offense and defense that innovations in defense will eliminate the 3PA premium faster than offenses will avail themselves of the opportunities that currently exist or that future offensive innovations will afford. But that is a different issue than whether such a premium and opportunities currently exist.