Sunday, October 24, 2010

Run Hoyas, Run? Part 2

Last time, I discussed the distribution of Georgetown's possessions last season as a function of possession length.*  Today, I'm going to look as the efficiency of Georgetown's offense (and defense) as a function of possession length.  Then, I'll combine the two to find if the Hoyas were leaving points on the table.
*here, possession length = time until first action; see previous article for details.

A few years ago, Ken Pomeroy posted a plot of possession efficiency for each second of possession length, derived from five years of play-by-play stats for Division I college basketball.  I re-posted that figure last time, so here I'll just show my re-plot of his data, along with possession efficiencies for three-second aggregate bins (0-3s, 3-6s, etc.) as I did last time for possession length:

For now, just focus on the solid gray line, which represents Ken's original data - I'll come back to the bins further down the page.

As Ken discussed in his original article, there are three areas of interest on the figure:
  • For possession lengths less than 12 seconds, there is a large increase in scoring efficiency compared to possessions that last longer.  There is a sharp peak at 3-4 seconds (typically fast breaks after steals) where the average D-I team is scoring better than 1.2 points per possession (ppp) - and remember that a middle-of-the-road team will likely average right at 1.0 ppp overall.  However, the improved efficiency drops slowly from the peak, and finally reaches that 1.0 ppp baseline only at 12 seconds into the possession.  Teams benefit greatly from scoring off the break, but continue to benefit well into the possession time as the defense scrambles to get set.
  • For possession lengths of 12-30 seconds, there is very little variability in efficiency, as teams average 1.008 ± 0.010 ppp (yeah, I'm actually reporting a standard deviation here - get over it).  There is a slight reward for scoring earlier in the possession:  12-24 seconds into the possession, teams average 1.013 ± 0.005 ppp, while 24-30 seconds into the possession the average drops to 0.996 ± 0.007 ppp.  It's a subtle and not statistically significant difference.  Here, we've effectively reached an even match between the offense and defense.
  • For possession lengths greater than 30 seconds, efficiency decreases quickly with added time.  As the last few seconds wind off the shot clock, scoring efficiency approaches 0.8 ppp, which is a very poor number.  By now, the defense holds the advantage, as the offense loses its selectivity in an effort to get any sort of shot up at the basket.

So how did the Hoyas and their opponents fare compared to Ken's aggregate? 

(more after the jump)

Let's take a look using three-second time bins:

Looking first at the opponents' efficiency (the bottom half of the chart), for possessions less than 12 seconds teams were able to score at a very healthy clip against Georgetown, consistently above what an average team would expect.  And keep in mind here that the Hoyas, while certainly not world-beaters, had an above-average defense (ranked 47th nationally by Pomeroy).  Until Georgetown was able to set its defense, it was a lousy defensive team.

For possession lengths longer that 12 seconds, the Hoyas were an excellent defensive team, as they kept their opponents scoring efficiencies at or below the national average for most of the rest of the shot clock.  The uptick for the 30-33s bin (or the downtick at 27-30s) is probably nothing systematic, but rather just a bit of statistical noise.

When we take a look at Georgetown's offense, we see that the team excelled at two periods within the possession.  Up until about 15 seconds have elapsed, the Hoyas were very efficient - I'm not sure why the first bin (0-3 s) isn't keeping up with the rest, but generally Georgetown could punish a team early in the possession.

But more curiously, the Hoyas also were efficient late in the shot clock - from 24-33 seconds into the possession, the Hoyas score at a much higher clip than the average D-I team, and actually managed 1.13 ppp for that stretch (the average D-I team pulls 0.99 ppp for that same time period).  I believe this is the result of the notorious Princeton offense:  Georgetown was turning down "good" scoring opportunities in the half-court offense for "great" ones that could develop later in the shot clock.

But here's the kicker - if you refer back to the previous posting on possession time, you'll notice that a lot of Georgetown possessions end in the "dead" period in possession length (15-24s):  33% of all possessions as a matter of fact.  I think that we're seeing the effect of turnovers, more than anything else.  Simply, the Hoyas are not as likely to attempt a shot in the half-court until about 10s are left on the shot clock, so a greater of percentage of those possessions that did end during the "dead" time are the result of turnovers, rather than missed shots.

Meanwhile, the Hoyas were using about half as many possessions (18%) during the early stretch (6-12s) when they were scoring with relative ease (1.26 ppp).  Now certainly some of that can be attributed to opportunity, as teams can't always fast-break (Grinnell excluded) - but Georgetown was well-below average for that stretch, implying that strategy was also a factor.

We can summarize the last post and this one on a single, busy plot:  we'll simply take the difference, for either % possessions or possession efficiency, between Ken Pomeroy's average and Georgetown's actual offense and defense.  In each case, I'll refer to the difference as the marginal value.  For instance:

Marginal off. eff. (6-9s) = Off. eff. (6-9s) - KP average off. eff. (6-9s)

And here's the result:

Again starting first with the opponent (i.e. Georgetown's defense), teams were able to score above average efficiency early in the shot clock - now shown as the gray line (plotted on the right axes) which is above the zero line.  However, those same teams were only getting in extra possessions at the very earliest bin (0-3s) to take advantage of this.  On the other hand, teams were very inefficient at scoring in the 12-15 second time period, but also ended many fewer possessions then, minimizing the damage.  Looking at it from the Hoyas' perspective, while they were lousy defenders early in the shot clock, they also allowed fewer scoring attempts from 3-12 seconds into the possession to minimize the damage.

Now looking at Georgetown's offense, there is an exceptional trend early in the shot clock:  the Hoyas were far more efficient than the average team between 6-15 seconds into the possession, but used a lot fewer possessions during that stretch.  I can make this even clearer by multiplying the two marginal factors against each other for Georgetown's offense:

The value of the y-axis is the marginal gain or loss in the Hoyas' overall offensive efficiency, accounting for both possession usage and efficiency as a function of possession length.  The sum across all possession lengths is the overall margin, and is equal to -0.012 ppp. That is to say, Georgetown could have increased its unadjusted offensive efficiency about 0.012 ppp by using possessions in a more rational sensible manner.  That might not sound like a lot, but that margin would have improved Georgetown's offense from 8th to 5th in conference last year.

Now the above argument could simply be reversed, and stated as Georgetown's offense was more efficient than average because they used fewer possessions during that stretch - they were more selective. But the implication is the same either way, that Georgetown was too conservative during this stretch.  If the offense just could have moved possessions from later in the shot clock (e.g., from that "dead" period 15-24s) into the early stretch, it would have been significantly better.

Maybe Mike D'Antoni is right.


    1. Great post, as always!

      You mention it in the last paragraph (that there exists the possibility the offense was more efficient because they used fewer possessions), but I don't understand how that gets you to the same implication.

      By moving more possessions into that time period, wouldn't you "dilute" the gains? Furthermore, wouldn't that subsequently improve the 15-24s buckets because you are moving less efficient possessions earlier? Granted, my statements are assuming that any marginal possession movement is net negative for the offensive efficiency but I still think that is a distinct possibility.

    2. Hi Anonymous (you seem to be everywhere on the interwebs),

      I think your argument is valid, so long as less efficient possessions from later in the shot clock (e.g. turnovers) remain inefficient if used earlier.

      I just don't think that is necessarily the case for two reasons:
      1. Georgetown is using such a small percentage of possessions between 6-12s, relative to an average team, that I think they are capable of picking up the pace before the inevitable decline in efficiency
      2. If I'm right that the drop is due to a high rate of turnovers in the 15-24s time, you'd naturally expect a net gain simply from having less opportunity to commit a turnover in a shorter possession (the whole "7 seconds or less" theory I alluded to with the D'Antoni reference).

      But you're right that I'm perhaps a bit optimistic on the marginal returns of playing with a "normal" offensive pace distribution.

    3. So glad to have you back posting. Thanx for your efforts.