Run Hoyas, Run?

Edit:  Part 2 is now posted.

When John Thompson III took over the Georgetown Hoyas for the 2004-5 season, he installed his version of the Princeton offense which quickly developed the reputation for its plodding pace.

Whether Georgetown can (or has) increased the pace at which it plays has been discussed here before, but I thought I'd post a quick recap, looking at the JT3 era:

Season        Adj. Pace      Rank  
2004-5          60.4        322/330
2005-6          58.7        329/334
2006-7          59.9        328/336
2007-8          62.2        316/341
2008-9          63.6        285/344
2009-10         66.9        188/347

After placing in the bottom-10 nationally for the first three seasons, the pace of the Hoyas' offense has begun to ramp up, until last season the team was just a hair below the median for all of college basketball.  Call it the Roy Hibbert effect (early years) or the Chris Wright effect (later years), but last year's Hoyas are not your older brother's, at least in term of the speed of the offense.


I thought it might be interesting to dig a little deeper into last year's stats, to try and understand how Georgetown was picking up the pace, and whether it was paying dividends.  To do that, I'll refer to an old post by Ken Pomeroy, where he presented a nifty plot showing the distribution of time-of-possession for all NCAA teams and their efficiency at each time (click any figure to enlarge):

This figure was lifted from Basketball Prospectus

The red line, plotted on the right y-axis, shows that possession times are not normally distributed, but rather tend to skew to shorter times.  This is rational behavior, due to the info from the blue line, plotted on the left y-axis.  Shorter possessions have higher scoring efficiencies, mostly due to fast break opportunities (the big hump on the far left of the curve).  As possessions stretch out past 30 seconds, teams become much less likely to score. 

(how this relates to Georgetown after the jump)

I'll need to spend a few seconds here discussing Ken's definition of possession time, because it's not necessarily what you think it is.  Ken defines the possession time elapsed as the time it takes for an action to take place.  To wit:

. . . the time elapsed until what I call "the first action," which is either a turnover or a shot attempt by the offense. Therefore, when a team misses a shot five seconds into the possession, gets its own rebound and scores 25 seconds later, its score is placed in the five-second bin and not the one for 25 or 30. While these might not seem like fast-break points at first, they result from a possession that started with a fast-break attempt, thus it's only fair to include them when calculating efficiency from those attempts.

The explanation is straight-forward, although I normally define possession time to be the full length of the possession (which I've always thought to be reasonable).  Because of this, any possession with an offensive rebound will be longer in my normal stats than with Ken's, and it would make any intercomparison of my stats to his problematic.

So I've gone ahead and added this extra stat (time until first action) into my play-by-play parser, and re-run last season's game stats, looking at all games from vs. UW forward, excluding the Butler game (see here for further explanation). 

Here is a revised look at Ken's distribution of possession time elapsed again, now with time grouped into three second chunks (e.g. 0-3s, 3-6s, etc.):

You may notice that the % of possessions per bin are higher than for Ken's original plot above, but that's simply because I'm adding together possessions played during each 3 second interval.

I've grouped his stats into 3-second chunks because I'm going to compare his stats against the stats generated last season by the Hoyas.  Since his database has literally millions of possessions tracked, he can use a highly granular approach to his analysis.  In the meaningful games I'm looking at here, the Hoyas played in ~3570 possessions total, or about 1785 possessions on either offense or defense.  If I break things out to the second, the number of possessions per bin gets too small, and the results get very noisy.

This is all for the sake of setting up the next plot, looking at the distribution of possession length for Georgetown and their opponents:

In each case, I've overlaid Ken's original distribution to give you the cumulative Div-I NCAA rates.

Looking first at the opposition (bottom in red), we see a relatively good fit of the data to Ken's expected distribution.  Georgetown's opponents scored often on extremely short possessions, which can generally be thought of as the sum of back-court steals and late-game FT attempts (when the other team is trying to extend the game).  I'm not sure if this is primarily because Georgetown gave a lot of late-game fouls last season, but that wouldn't be my guess.

The middle of the distribution (10-15s) is a bit lower than on average - once teams were forced to run their offense, they weren't as likely to attempt a shot right away.  And the tail of the distribution (27+ s) shows a consistently higher percentage than average, as teams were forced (or chose) to play deep into the shot clock.

There's one caveat here, to explain why a good chunk of possessions - almost 3% for the Hoyas' opponents - are longer than the 35-second shot clock:  my game tracker starts counting seconds after the last possession ended.  After a made basket with more than a minute left in the game, the game time to inbound the clock is included by my stats even though the shot clock doesn't turn on until the ball is touched in play.  I'm not sure if Ken's stats are generated in the same way as mine, but I'm guessing they are so those higher bin totals for long possession times should be real.

Taking a look at Georgetown's distribution of possession times (top in blue) shows a very different result.  In fact, the Hoyas have a bimodal distribution, rather than the typical Poisson we see for college teams as a whole.

What's going on here?

It appears that the Hoyas had a disciplined game plan last year:  when the opportunity was there, they would attack the opponent in an attempt to score on the fast break.  If the break wasn't there, Georgetown set up and then ran their offense, which found shot attempts generally after 12 seconds ran off the shot clock.  This behavior is actually there in Ken's data, as well; although not nearly as strong - that little flat spot in the distribution around 11s, which I've been ignoring up to now.

What's missing is the secondary break shot attempts (6-12s into the possession), which I generally think of as the outside (read: 3-point) shot or trailing layup that is set up by the driving player kicking the ball back outside.  Now one could argue that some part of this is simply from Chris Wright's hell-bent attitude of getting his shot off or drawing a foul on the break rather than making a pass, rather than all resulting from of the game plan, but that's for another day. 

Edited to add:  I have no idea if that other notch, at 27-30s of possession time is significant or just statistical noise, but I'll ignore it here.


If you look back up to the top of the page to Ken's original plot, you also see the scoring efficiency as a function of possession length.  Ken has found that there is little difference for teams scoring efficiencies between 12-25 seconds of possession time, i.e. when teams run their sets, they tend to score at the same rate.  But the peak benefit of the fast break (0-6 s) does extend out to 11 seconds or so.  That is, teams on average are more efficient scorers on the secondary break, than when running their half-court offense.

So, did the Hoyas give away points by not running a secondary break last season?  The answer to that will be in the next post.