Saturday, October 25, 2008

Big East Growth Charts

About this time last year, I posted about possession usage, based upon an article by Ken Pomeroy about effective usage. Possession usage, or possession percentage (%Poss) is yet another in a litany of stats that you'll find here and other site run by fans with a predilection towards numbers and too much time on their hands. It is simply the ratio of possessions that a player ends divided by total possessions played. A player can end a possession by:
  1. scoring
  2. turnover
  3. missing a shot that is rebounded by the defense.
On a team that was truly democratic (socialist?), everyone would have a %Poss = 20%, meaning that team possessions were used equally. In the real world, %Poss for Big East basketball players typically ranges from 10 to 30%, with a median value around 18%*. In other words, the majority of players choose to use less than number of possessions you'd expect.
*Edited to add: Where did this come from? See the end of the article.

Why? Simply, because someone else on the court is using possessions at a disproptionately higher rate. There are a few reasons why this could be (here comes another list):
  1. the coach has instructed the team to allow certain player(s) to use more possessions
  2. certain player(s) have decided on their own that they should use more possessions
  3. the player in question thinks other player(s) should use more possessions
Whatever the reason, the logic of the 18% player is sound so long as he is giving away those possessions to someone more efficient (i.e. with a higher Off. Rating) than his own. However, when a player has a very high Off. Rating (e.g. Darrel Owens, Colin Falls) but uses significantly less than his allotted 20% of possessions, he may by hurting his team. Of course, the counter-argument can be made that by using more possessions, he'd become less efficient, but there would still be some marginal returns until this player's efficiency is comparable to his teammates - and yes, I've just lapsed into economic theory.

Ken Pomeroy's - remember Ken? we started out by talking about him - thesis is "role players don't usually become go-to guys from one year to the next, or at any point during their careers."

To prove his point, he provided a couple of nice charts, showing the change in %Poss from one year to the next, or one year to two years later. I'll reproduce them here (as always, click any image to enlarge).

There are 4 types of fitted lines on these charts.
  • The thick black line represents the best fit of the data (all 2005, 2006 and 2007 college players, I think).
  • The thick dark blue lines indicate the 50% prediction interval around the fit - that is, 50% of all players should be between these two lines. Another way to think of the lines is that 25% of all players will fall below the lower line, and 75% of players will fall below the upper line (and therefore 25% of players will be above the upper line).
  • The thin black lines are 95% prediction intervals - only 2.5% of all players should fall below the bottom line, or lie above the upper line.
  • Finally, there is the dashed line, which the is the 1:1 line. A player lying on this line would have no change in his %Poss from year 1 to year 2 (or 3).
There's actually a bit more information we can glean from these plots. For example, you'll notice that the 1:1 line is below the best fit line until %Poss ≈ 22.5% for year 2 vs. year 1. What this means is that, for players who used less than 22.5% of their possessions in a given year, more will increase their usage the following season than not. For the year 3 vs. year 1 plot, this point of intersection is ~23% - very similar.

This is not to say that all players will eventually become 22.5% possession users, but rather that this is the point where increasing possession usage becomes more difficult than not, likely due to increased competition with teammates for available possessions.

So, is there a point to all of this?

Since KenPom has %Poss data available back to 2005, I decided to plot all Big East players from the last 3 seasons (or 2 seasons) on his charts, to see if the Big East behaves as the rest of college basketball with respect to changes in usage.

Here's season 2 vs. season 1 (you'll need to click to expand to see things clearly):

A bit of explanation is in order (if there's one thing I want to be famous for, it's busy charts).
  • I've sized the markers by Season 1 %Min (% of available minutes played), so that end-of-bench players wouldn't swamp regular players on the scatter plot.
  • I've color-coded the markers by Season 2 Off. Rating. When I initially ran this analysis, I expected that the rate of increase from year 1 to year 2 would be strongly related to how well the player performed in the 2nd year, but this is obviously not the case - the color appears random.
  • I've also color-coded Georgetown players as gray rather than on the color scale, so they'd stand out. Nothing of exception with this group.
  • I've added tags identifying a few outliers.
  • Finally, I've added a horizontal and vertical line at 22.5%, indicating the point where more possessions become scarce.
The analysis by KenPom seems to apply very well to the Big East - the data almost entirely within the 95% prediction interval, and follows the trend line.

What's of most interest to me are the two points in the upper left quadrant: James Holmes and Draelon Burns. These two players represent the exception to KenPom's rule, in that they made the leap from role players (%Poss = 18.7 & 19.5, respectively) to go-to players (27.5 & 28.4) in a year. Since I wasn't paying much attention to either team at the time (or now), I'll leave it to someone else to explain what happened in each case.

On to the two-year gap (season 3 vs. season 1):

Many fewer data points here (n=80 here; n=274 for the previous plot), but again the analysis by KenPom seems appropriate for the Big East.

The players of interest here include Daryll Hill, who fell from go-to to role player due to injuries, and three rising seniors: Anthony Mason, Levance Fields, and Georgetown's own Jessie Sapp. Only Mason has made the leap into true got-to status (17.1 to 23.0 to 26.9), but Jessie Sapp has made an extraordinary rise from pass-only to important cog (12.2 to 18.7 to 22.6). It will be interesting to watch these three to see if they can continue to absorb possessions.

Edited 10-27-08, 10:00pm to add:

While playing with the data, one thing I did look at was the distribution of %Poss for Big East players. Here are the histograms for all players (n=611) from 2005-2008, and also for those with %Min > 40% (n=395). What's interesting is that majority of players who play less than 40% of available minutes also use less than 18% of available possessions (note that KenPom has his own filter of %Min > 10% on the data, so players with very little playing time [< 4 min / game] are already dropped from the data set).

I suspect that the subset of players with 10% < %Min < 40% group has a significant number of freshmen who are slowly being introduced to their coach's respective systems. I used the median value of the entire population in the discussion above, since many starters began their career in this role as bench / role players (e.g. Jessie Sapp).

I hadn't thought to plot %Poss vs. %Min before tonight, but the histograms above imply a relationship. Here it is:

The data points are both colored and sized by offensive rating, but there doesn't seem to be much trend in that variable. The slope of the line is ~0.085; in other words, an increase in %Min by 23% will increase %Poss by 2%, on average.

You've got to love D. Caracter's freshman season at Louisville.


  1. What I find interesting is that it really isn't all that unusual to for a player to go from 20% to 25%. Yes, they don't regularly become a huge focus -- the 28-30+% guys, but it isn't strange to see someone become a primary scorer.

    I wonder how this relates to GU, given that our offense isn't likely to have a 30% anytime soon. If 25% the equivalent of going to a 28%? We've actually had Jeff, Roy and DaJuan all make leaps from around 20% to 25%.

  2. I don't know the answer to that, but it is the sort of question that I should be able to answer with the dataset I've got from KenPom. I'll try to take a look at it later this week.

    Thanks for reading!

  3. Terrific work as usual. The explanation for Holmes can be found by reviewing the Bull's roster for 2005 and comparing it to the 2006 roster. Holmes and Jones are among only 4 "returning" players. And both took a year-over-year jump in possessions and shots, though Jone's jump was not as eye popping (Poss% 16.4 to 23.0; Shot% 13.7 to 21.9). Pomeroy stated in that article, "...Any major changes in a player's usage are usually the result of filling the hole left by a departing possession eater...", and it appears Holmes (and Jone's) "rise" had much (more?) to do with roster turnover (and injuries -- more on that in a bit) than talent/skill. Of the eleven players on the Bull's 2004-05 roster, only four appeared on the 2005-06 roster. The others either graduated or transfered. Of the four, Jones, Capko (a walk-on) and Holmes remained injury-free for the season. The fourth player, Collin Dennis, was indeed one of six players who pulled up injured for part (or most) of the season. Dennis returned briefly the next year. There were so many injuries that season that only eight players (of the 12 listed on the "official roster") appeared in 10 or more games for the Bulls. McCullum was so desparate for a point guard that he used a walk-on, Chris Capko, for most of the season.

    Burns followed the same path as Holmes, albeit on a healthier squad. The player whose role Burns appears to have stepped into was Quemont Greer of Milwaukee, Wisconsin, whose Poss% and Shot% numbers for 2004-05 appear to be about the same as Burn's numbers in 2005-06 (high 20s).