Sunday, November 7, 2010

Deep bench, short rotation

JT3 subs in five players - documented! (Joseph Silverman / The Washington Times)
Alan's excellent post about likely minutes for the upcoming season got me thinking about John Thompson III's reputation for playing a short rotation.  I think this rep was developed in his first two seasons, when the cupboard was mostly bare beyond the first 6 or 7 players on the roster.  But now that he's been coaching the Hoyas for six seasons, I wonder if this reputation is myth or reality.

But before I can dig into the question of whether JT3 goes with fewer players than he should, I need to work through some basics and find a reasonable measuring stick.

First, we need to decide what constitutes being part of the rotation for a player, rather than just playing garbage minutes.  Next, we need to have some meaningful comparison of the size of Georgetown's rotation versus what other coaches/teams run.  Finally, we'll need to consider what to do about injured/suspended players and their effect on the rotation.

More after the jump

I'll take these in reverse order:

Injured players
I can easily account for changes in player time due to injury for Georgetown during JT3's time on the hilltop - I've kept track for the past four seasons and can refer to the illustrious archives of and HoyaTalk to puzzle out the rest.  But I'm really not interested in doing the same for other teams.  So instead, I'm going to ignore the impact of injuries and suspensions.  Why?  My reasoning is that, over the course of six seasons, injuries should start to even out between teams and, frankly, the ability to juggle a rotation without a major player is an important part of coaching (shorten the rotation or bring in an untested player).  As for suspensions, you can insert your obligatory UConn or Syracuse (or West Virginia) joke here, but I'll argue they also are negligible.

While I can make a table of minutes played for Georgetown, I can stare at it until I'm red in the face and won't learn a lot unless I can compare it to something else.  Here, I've decided to limit the intercomparison to the Big East.  Specifically, I'll only use the regular season conference games to figure out how big the rotation was.  And I'll limit the comparison to only those teams that have had the same coach during the past six seasons - after all, I'm looking for coaching patterns here, not team patterns.  That gives us UConn (Calhoun), Notre Dame (Brey), Pitt (Dixon), St. John's (Roberts), Syracuse (Boeheim) and Villanova (Wright).  To pad it out, I'm also going to use Louisville (Pitino), even though they were still in Conference USA in 2004-5.

Rotation players
The statistic I'm using here is simply minutes per conference regular season game (courtesy of StatSheet).  The question is what constitutes being a regular player - the size of the rotation will have an effect upon this.  That is, a team that uses only 6 players regularly will have rotation minutes averaged around 33 mpg (200 min / 6 players), while a team that goes 11 deep with have rotation minutes averaged around 18 mpg (200 min / 11 players).  If I declare 15 mpg to be the cut-off, it would work well for the first team described, but would probably underestimate the rotation size for the second team.  A player that gets 15 or 20 minutes of playing time each game is clearly part of the rotation, while one getting only 3 minutes a game is not, so we'll work over this range.

Let's take a look at this with a plot, limited to two teams from last conference season (2009-10).

The dashed black line represents the average across all eight teams, and the Hoyas and Cards data points are color-coded.

The results are not a surprise.  Georgetown was notorious for running a short bench last year: only seven players averaged more than 5 minute per game in conference last season, but those same seven all played at least 12.5 mpg.  In fact, the Hoyas were using less than average rotation players for any qualifying time you'd choose, except at 12.5 mpg - the perception that Georgetown was not deep was, in fact, reality.

Meanwhile, Louisville was getting more than 8 mpg for eleven (!) of their players last season, with a steep drop-off in playing time between 10 and 15 mpg.  While the Cards appear to have given run to the whole roster last year, what's interesting (to me, at least) is that if we were to choose a high qualifier for mpg → rotation player, Louisville would also have been considered a shallow team, and actually using just as many players as the Hoyas.

To describe the difference between the two teams, I'll resort to one of my usual tricks.  We can simply re-plot the figure above, but instead use marginal rotation players [= rotation players - average rotation players], which is simply the difference between each team's data and the average line (again, this is for the 2009-10 conference season only).

Now to summarize each team's rotation last season, I'll use the integrated area under each curve (this is statistically more sound than simply averaging).  For Georgetown, we'd end up with -0.9 marginal rotation players; that is to say, last season JT3 had a rotation that was just about one player less than the average used by our gang of eight coaches.  Louisville's curve gives the result of +0.9 marginal rotation players for last season.  This may seem a bit low for the Cardinals from looking at their curve, but we're accounting not just for the exceptionally deep number of players getting some minutes, but also for the few players getting a lot of minutes.

One additional comment before completely jumping off the statistical deep-end:  for the marginal rotation player calcs, I'm actually using the average players per qualifying time over all six seasons, not just per season.  This should help to smooth out some of the idiosyncratic noise from year to year.


The reputation of John Thompson III for playing a short bench is well-founded.  He typically uses about one less player than the average across our gang of eight coaches.  The only season that Coach ran out more than the average number of players in his rotation, his team was in free-fall for most of the year.  The 2008 season is an interesting one, since it was relatively successful [Big East reg. season champ, #2 NCAA seed, 7th by KenPom], and may have added another body to the rotation if Chris Wright had been healthy.  But, as I said at the top, we're going to ignore the effect of injuries here.  Oh, and last year's team did indeed have a really short rotation (shortest in JT3's regime).

Other short-rotation teams

Three other coaches in the Big East also tend to go with shorter rotations than the average for our group.  Two you probably already knew about:  Jim Boeheim up in Syracuse has used his vaunted 2-3 zone to save energy on defense and allow his rotations to stay short, while Mike Brey at Notre Dame - the coach who uses the fewest players of all in the group - has an apparently well-founded reputation for grinding his starters over the course of the season.  The surprise for me is that Jim Calhoun is also in this group; I usually think of UConn as loaded with top recruits and extremely deep, but Calhoun actually uses his bench slightly less than average.

Deep-rotation teams

I picked the Cardinals for a contrasting example to the Hoyas for a reason: Rick Pitino brings in top recruits, and he plans to use them.  Norm Roberts had tried to emulate that plan in Queens, but without the recruits.  The team he left behind for Steve Lavin is loaded with seniors - I would have expected it to be deeper than any I've looked at here, but with a new coach I have no idea if that will translate.

Middling but trending?

Jamie Dixon has been coaching at Pittsburgh one season longer than JT3 has been at Georgetown.  His player usage is easily the most obvious for showing a trend as he's replaced his predecessor's players with his own - he's gone from a very deep rotation to a modestly short one.  It will be interesting to see if this continues.  Jay Wright at Villanova is trending the opposite direction, as he brings in wave after wave of top recruiting class and tries to make sure those players get enough time on the court.

To summarize, here's the eight teams averaged across all six seasons:
.               Rotation
Team        Players   Marginal  95% CI
UConn         7.54     -0.25    (0.59)
GTown         7.32     -0.47    (0.58)
LVille        8.39      0.60    (1.00)
NDame         7.13     -0.65    (0.44)
Pitt          8.00      0.22    (0.60)
St Johns      8.31      0.53    (0.92)
Syracuse      7.54     -0.25    (0.63)
Villanova     8.06      0.28    (0.82)
---------     ----
Average       7.79
The values in parenthesis are the confidence intervals about each average - if this number is smaller that the absolute value of the marginal rotation, we can say that the team [read: coach] is running a short or long rotation with statistical confidence (here, we'd be right 19 times out of 20).

Mike Brey is certainly using fewer players than the group on average, and significantly so; JT3 is also using a smaller rotation than the group, but not at a statistically significant level.  Boy, that was a lot of work to state the obvious.  None of our eight coaches used a significantly deep rotation over the past six seasons, with Rick Pitino and the recently departed Norm Roberts coming closest.

The drawback of the summary table is that it glosses over any trend - which seems evident for Pitt.


There are (at least ) two remaining questions to be answered:
  1. Is there a relationship between the size of a team's rotation and its efficiency?
  2. Is JT3 burying someone on the bench each season that could have improved his team?
We'll tackle the first question in graphical form, after a few provisos.

To consider if there is a trend between rotation size and efficiency, we need to define a reasonable test.  We're looking for changes in coaches behavior here, rather than an intercomparison between coaches/teams.  That is to say, when JT3 increases the number of players in his rotation, do the Hoyas become more/less efficient either offensively or defensively (or net)?

To make this work, I'll look at only how each team performed relative to it's own six-year average, be it offensive, defensive or net efficiency.  For example, the Hoyas posted an offensive efficiency of 103.9 in conference in 2008; over the six seasons we're looking at here, they averaged 106.7.  Therefore, for this comparison we'll call the relative off. efficiency for 2008 to be -2.8 [= 103.9 - 106.7].  Obviously this is simplistic as it doesn't account for the relative change in talent on a team, but it's the best I can do right now.

Also, since we're looking for changes in a coach's substitution behavior, I'll also look at changes in marginal rotation size only within each team's six year average.  Again returning to the 2008 Hoyas, their rotation was 7.75 players, while they've average 7.32 players over the 6-season stretch examined here.  That gives us a relative marginal rotation of 0.43 [= 7.75 - 7.32].

So here are three plots, showing relative net, offensive and defensive efficiency against relative marginal rotation size, respectively.  In each case, I've shown all other teams as a single color with the Hoyas data overlaid in home gray.

If you can see any trends there, you've got better eyes than me.  All three plots show that the general trend over all teams is no trend - there just isn't a relationship between rotation size and efficiency in any form.

If you want to squint very hard, you might try to convince yourself that there is a trend for the Hoyas' offensive efficiency and the size of the rotation.  The decrease in offensive efficiency with increasing rotation size would likely imply that Coach Thompson tends to increase the size of the rotation when the team struggles to score.  But the trend is tenuous at best.

Finally, I'll return to the original question that begat this monstrous post:  does JT3 use fewer players than he should?

Now that we've got a means of determining marginal rotation size, we have found that Georgetown runs a tighter rotation than the average team.  This leads one to wonder who was the best available player that didn't make the rotation.  But I want to be fair here and not look at future performance from players on the bench, but rather: who was the best player that season that couldn't get regular minutes?

Here comes the last table, followed by some pithy comments:
Season    Rot. size     Best avail.
2004-5        7.2      Tyler Crawford 
2005-6        7.2       Marc Egerson  
2006-7        7.0      Vernon Macklin
2007-8        7.6      Tyler Crawford
2008-9        7.9       Henry Sims?
2009-10       6.8       Vee Sanford
Some of these were harder than I expected.  Comments by season:
  • 2004-5:  I was thinking Ray Reed, but he picked up enough minutes late in the season as a defensive stopper to be considered rotational, meaning that Crawford [fr] was about the only one left on the bench (along with the oft-injured RaMell Ross).
  • 2005-6:  Jessie Sapp was getting just enough time to be considered a rotation guy, so Egerson [fr] is the easy pick.  I still consider him the biggest loss to transfer in the JT3 era.
  • 2006-7:  Another easy pick.  I actually wrote an article that season decrying the Big Ticket's [fr] lack of playing time.
  • 2007-8:  I'm excluding Chris Wright here, since he wasn't available.  About everyone useful on the bench got playing time, so Tyler [sr] is the default answer.
  • 2008-9:  Yeeshh.  When Nikita Mescheriakov becomes a starter halfway through conference play, there really isn't likely to be much left.  The easiest answer is that Sims [fr] should have gotten more playing time, but he averaged about 8 mpg in conference, so might not be eligible.
  • 2009-10:  Another easy one:  Vee Sanford [fr].  He was simply squeezed by having three very good to great guards ahead of him, but he still should have been in more than 3.3 mpg. 
So who could have improved the team?  Egerson, Macklin and Sanford are the three players who could have contributed more than they did, and improve the team in the process. All three were freshmen, whom Coach Thompson is loathe to rely upon.  The first two are the two best players to transfer out of Georgetown during Coach Thompson's career on the hilltop.  How Sanford will cope with tight playing time - and how he will be handled by the coaching staff - will be an important story this season.


    1. Do you really think JTIII is loathe to rely upon frosh? He definitely seems to have a "prove you deserve PT" kind of philosophy, but most of these decisions are in the margins -- he has a pretty long history of starting frosh if they are good enough.

      I don't think it's frosh, he just doesn't really believe in game PT for development purposes (or that's my take).

    2. I know he has been quoted as saying he doesn't like to play freshmen, but I'm too lazy to find it.

      It's also clear from the table that when he's actually had a player good enough to use but still sitting [2006, 2007, 2010], it's been a freshman.

      I just think he's more comfortable with short rotations. The trick is to recruit and develop players appropriately if that's the case.