Counting the Baskets

Defense is Easier When You Don't Have to Run Back

(The Pacers are Good at Defense, Run on the Heat, the Lakers are Weird and Other Lessons from Prior Event Data)

If you listen to Doc Rivers, and many other NBA coaches, they will tell you that the key to the offense is their defense and getting stops. There are a couple of reasons for this, I think, all of which are good. Part of it may be just motivational, for basketball players at every level it is sometimes harder to motivate effort on the defensive end as many popular gifs of James Harden will testify. Defensive effort is also something players and coaches can control unlike whether shots fall, though they can maximize the expected efficiency of those shots, bounded by what the defense gives them. Lastly, it's probably true.

Using the data from NBAWowy.com, I aggregated the effective field goal percentage for each team defense after different 'Prior events' taken by NBAwowy from the Play by Play data. The events tracked are made baskets, defensive rebounds, offensive rebounds, steals, other turnovers, time outs, and non-shooting fouls. The data only tracks the last event before an attempts basket, so it doesn't tell us about free throw attempts or possessions that end up in turnovers before an attempts is made. But, even without that layer of detail, I think the Prior Event data tells us some interesting things about game flow and defense. In fact, the possessions that lead to the highest eFG% such as following steals probably lead to the highest number of free throw attempts given that those possessions lead to attempts at the rim most often.

Below are the Prior events to a field goal attempt with the eFG% of every team's opponents with data through 11/28 (note that the league average for offense should be the same):

FGM/FTMeFG%	DRB eFG%	ORB eFG%	STL eFG%	TOV eFG%	FOUL eFG%	TIME eFG%
47.2	49.9	53.1	62.0	46.0	50.9	45.0

 

Those numbers are all very similar to ones compiled by Evan Zamir, the creator of NBAWowy, last year. This year's numbers already represent over 37,000 field goal attempts, so I think we are getting some real signal in the data. Of note is that steals or live ball turnovers are the by far the most efficient start to an offense and the toughest position for the defense, while attempts immediately after time outs were the least advantageous for the offense and the best for the defense.

Coming off a time out the offense's eFG% is 4.9% lower than following a defensive rebound. Yet noticeably too many coaches call time outs after a defensive rebound in late game situations, which probably speaks of both unfamiliarity with the stats, and a desire to attain the illusion of control over the situation.   But, the coach should trust his offense and let the players play it out.

Looking at the data for the league and on a team level, it struck me that a useful way to group the Prior Events is as 'Dead Ball' events, or, Out of Bounds Events; made baskets, turn overs, fouls and time outs, and 'Live Ball' events; defensive rebounds, offensive rebounds, and steals. I grouped the opponents' eFG% weighted by the number of possessions after each Prior Event and calculated the average for each team and then league wide. The league wide averages showed about a five percent gap between shots after stopped ball events and live ball events favoring the defense.

	Stopped Ball	Live Ball
Average eFG%	47.3%	52.4%
Std Deviation	3.0	3.2
Maximum	53.8%	59.7%
Minimum	39.2%	45.4%

 

In terms of individual team defense there are a couple of interesting bits to pull out of the split between Stopped Ball and Live Ball events:

• The best team off of any Stopped Ball event, and second best on Live ball events so far is the stifling defense of the Indiana Pacers with opponents shooting at eFG% of 39.2% after any out of bounds situation and 45.6% after Live Ball prior events.
• Most teams, 26, are allowing opponents to shoot a higher eFG% after Live Ball events, with three teams allowing a marginally higher eFG% after Stopped Ball events.
• The Lakers are weird.  They are allowing the highest eFG% after out of bounds events and the lowest after Live Ball events.  That is driven by a league high 54.5% Opponent's eFG% after made baskets and a league lowest 39.7% after defensive rebounds.  That seems like random noise to me, since I have no coherent explanation.
• The Heat have the largest gap between stopping opponents after Live Ball events and out of bounds events, allowing 45.1% eFG after a Stopped Ball event and a porous 58.2% coming off a Live Ball event.
• The Pistons have been the worst team defense coming off of a live ball event, particularly off of defensive rebounds, allowing opponents to hit at 59.7% eFG.
• There is only a small variation in the number of Live Ball to Stopped Ball events between teams. But the Bucks have been putting their defense in the toughest positions with a near even split in Stopped Ball and Live Ball events.  While the Raptors, to my surprise, have put their defense in the best position with 56.8% percent of their opponents shots coming after an out of bounds situation where the defense has had time to set.

Below the break I have the table for Opponent's eFG% after Live Ball and Stopped Ball events by team:

 

Why Have the Celtics Been Allowing so Many Points in the Paint?

Previously I looked at the Celtics under Brad Stevens primarily on the offensive side using Ian Levy's Expected Point Per Shot (Xpps).  The same principles that guide Xpps on offense (shooting from the most efficient locations) can be applied in reverse on defense, denying one's opponent from shooting from those locations and not allowing them to get to the line.

 

So, I am taking a look at the Celtics defensive Xpps.  Below is an image from Hickory-High with the Celtic's Actual Point Allowed and Xpps over the last seven seasons. 

 

 

You can see that this year's team is allowing shots from more efficient spots than any of the Kevin Garnett era teams (the triangle in the upper right is the tanktastic 2007 season).  Opponents have converted at a slightly smaller rate than the Xpps, but both are just over the NBA averages (the reference lines on the chart).

 

The achilles of the Celtics defense to date has been points in the paint, according to data from NBA.com, the Celtics have allowed the second highest percentage of their shot attempts in the restricted area, at 37.7% with only woeful Utah worse at 39.4%.

 

But this appears to be more closely related to turnovers and transition defense than rim protection or half court defense.  This year's Celtics have seen both the points off of turnovers and the point off of fast breaks edge up compared to the last few years.  The table below shows the percentage of Celtics' opponents points off of Turn Overs (TOP), in Fast Break Points  (FBP) and Points in the Paint (PITP).  There is considerable overlap between the three categories, the majority of fast break points come in the paint, many turnovers lead to fast breaks and so on.

 

By Year	Opp Pts off of TO Pct of Scoring	Opp Fast Break Pct of Scoring	Opp Pts In ThePaint Pct of Total Scoring
2013-14	18.8%	13.7%	46%
2012-13	17.9%	12.8%	45%
2011-12	18.3%	13%	45%
2010-11	16.7%	12.4%	41%
2009-10	16.9%	12.2%	40%
2008-09	19%	11.5%	39%
2007-08	18.1%	10.2%	38%
2006-07	18.5%	12.6%	42%

 

 The point is that they're related, and what looks like a 'rim protection' problem is as often a live ball turn over problem and transition defense problem.  A quick simple regression of opponent's Fast Break Point percent and Opponent's PITP percent for the Celtics over the last ten years indicates that around 46% of the variation in PITP comes from a change in fast break points.

 

One more way to show this is to look at the Play by Play data via NBAWowy.  One of my favorite NBA data sites.  NBAwowy tracks the eFG% of each team and the eFG% allowed just after certain prior events.  For example, after giving up a steal or an offensive rebound.  Here's how the Celtics fare defensively after each prior event.  

 

	FGM/FTMeFG%	DRBeFG%	ORBeFG%	STLeFG%	TOVeFG%	FOULeFG%	TIMEeFG%
Celtics Opponents	410 (45.9)	313 (51.4)	149 (57.7)	80 (55.6)	72 (36.8)	77 (50.6)	47 (37.2)

 

The first number is the attempts after that event type with the eFG% in parathensis.  The Celtics have been much better stopping their opponents after they make a basket 45.9%, off of dead ball turnovers 36.8% and out of time outs 37.2%.  And much worse after giving up offensive rebounds  or defensive rebounds or after live ball steals (not all turnovers are equal).  That's very typical, though the Celtics actually allow a slightly lower eFG% offf of made baskets than the NBA average last year and are allowing more after both offensive and defensive rebounds.

Essentially, the data suggests, I think ,that Brad Stevens has been able to instill a really solid half court scheme holding opponents to a low eFG% on all stopped ball situations, but, in turnovers and transition situations have a long way to go.

 

 

A Look at Brad Steven's Celtics through Xpps

Brad Stevens is already becoming famous for stressing process over results.  The idea being that if players play the right way most of the time the results will come, but that one has to accept it when they do not and just keep playing the right way.

It is a philosophy that fits very well on a rebuilding team.  Without the star players and an unbalanced roster results are going to be difficult to achieve, but establishing the correct process, one hopes, will lead to better results down the line. Of  course, judging how a coach is performing without simply looking at the wins and losses more difficult even if it is the proper way to do it.

In terms of coaching, I think there are a couple of big things we can look at to evaluate the coach's impact. Briefly:

• Line Ups; in terms of playing the best players and in the right combinations.  Stevens has had to scramble on this one without an experienced healthy Point Guard or an experienced healthy Center.  I think he's moved in the right direction since moving Avery Bradley off ball and putting Jordan Crawford in as Point and finally putting Jared Sullinger in the starting line up (He's simply your best player right now).
• Defensive effort and execution, where I think Stevens has done very well other than the Houston game, (Of which we will never speak again).
• Shot selection, probably the biggest contribution of the analytics movement to date has been to change the perceptions about shot selection to the detriment of the mid-range jump shot.  This is the one I want to focus on here, using Ian Levy''s Expected Points Per Shot (Xpps) to look at how well the Celtics under Stevens have done so far.   The evidence is decidedly mixed there.

Levy's Xpps calculates the expected points of a shot based on its location on the court according to NBA averages over the last twelve years.  Getting the ball to the rim or to the corner three are the most efficient locations (Or the foul line, though that is heavily related to attacking the rim, as I found here).

 

The image below shows both the Xpps and Actual Points Per Shot of the Celtics teams since the start of the Kevin Garnett era, via Hickory-High.

 

 

This is the table of the same data:

 

Team	Season	Actual PPS	XPPS	Difference	ORTG
Boston Celtics	2008	1.138	1.065	0.073	107.64
Boston Celtics	2009	1.142	1.058	0.084	108.07
Boston Celtics	2010	1.128	1.064	0.064	105.44
Boston Celtics	2011	1.122	1.043	0.079	104.00
Boston Celtics	2012	1.071	1.024	0.047	98.87
Boston Celtics	2013	1.085	1.033	0.052	101.10
Boston Celtics	2014	1.049	1.039	0.01	97.33

 

Astute observers will notice the pattern, this Celtics team has under performed the Doc Rivers lead big three teams in terms of shot selection and making.   However, the trend had been decidedly down before the regime change, probably reflecting the declining abilities of the Boston players, as well as, stagnation in River's offensive sets.

That brings me to two other points,  while the coach can influence shot selection, he can not control it.  Many NBA players would like to get the ball to the rim, but the defense is just as intent on stopping them.  Three point shots are similar, when I looked at Shot Creation previously, I noted that off the dribble threes are fairly rare.  A point that has been confirmed by the SportVu data, where I found that 19% of all Pull Up attempts were for three.  Pull Up corner three point attempts are even more rare given that the player has to move sideways to the basket to get to his spot for the shot.  The Celtics are currently without an NBA experienced natural Point Guard, it is not certain how many corner threes they can set up per game.

To look at the Celtics roster as it exists, the table below shows the Actual Points each player had generated as of 11/17/13, compared to the league average as well as their shot selection:

Player	Minutes	Average PPS	Actual Points Per Shot	Difference	XPPS	Shot Making
Kelly Olynyk	245	1.047	0.948	-0.108	1.055	-0.107
Avery Bradley	328	1.047	0.994	-0.007	1.001	-0.007
Kris Humphries	67	1.047	1.237	0.183	1.054	0.183
Brandon Bass	327	1.047	0.993	0.004	0.989	0.004
Courtney Lee	171	1.047	1.230	0.228	1.002	0.228
Gerald Wallace	309	1.047	1.146	-0.051	1.197	-0.051
Jared Sullinger	188	1.047	1.142	0.093	1.049	0.093
Jeff Green	361	1.047	1.100	0.021	1.079	0.021
Jordan Crawford	292	1.047	1.075	0.089	0.986	0.089
MarShon Brooks	28	1.047	0.735	-0.245	0.981	-0.246
Phil Pressey	109	1.047	0.680	-0.313	0.993	-0.313
Vitor Faverani	213	1.047	1.073	-0.053	1.126	-0.053

 

Khris Humphries and Courtney Lee, who should be getting more playing time, were leading the team in actual point per shot, while Gerald Wallace, Vitor Faverani and Jeff Green lead the team in Xpps.  Phil Pressey can't shoot, which is a problem, teams are starting leave him unguarded.  The Celtics guards as a group struggle with getting looks from high Xpps spots, which I expected since there aren't many three point  shooters on the team.

Finally to see if there has been any Stevens effect, I compared the players who were in the league last year with their Xpps this year and last.  I found essentially no Stevens effect.

Player	2013 XPPS	2012 XPPS	Xpps Year to Year Difference
Avery Bradley	1.001	1.029	-0.028
Kris Humphries	1.054	1.073	-0.019
Brandon Bass	0.989	0.959	0.030
Courtney Lee	1.002	1.039	-0.037
Gerald Wallace	1.197	1.154	0.043
Jared Sullinger	1.049	1.065	-0.016
Jeff Green	1.079	1.081	-0.002
Jordan Crawford	0.986	1.001	-0.015
MarShon Brooks	0.981	1.046	-0.065

 

At this point I think Stevens has been focusing on line ups, where he's made progress and defense where the results are mixed.  Similarly, I would say that getting the most out of this offense in a work in process.

The Value of Line Up Stretch

This post originally appeared at Hickory-High.com on 11/06/2013

The value of the three-pointer is a staple of basketball analytics. It is often referred to as 'advanced,' but really it is mostly the result of pretty simple algebra, a team has to hit 50% more two point shots to get the same number of points as they would get off of three point shots. Away from the rim, most players struggle to do that.

Of course, there is more to it than that, all kinds of nuances can effect the percentages of any shot. Defenses can do the same math and prioritize guarding the three point line. When misses do happen there are different expected percentages of offensive rebounds. It is easier to for the opposing team to score off of defensive rebounds, which add up with more misses from distance. Driving to the rim results in more free throws, which is a very efficient way to score.

Then there is the issue of 'spacing,' simply for its own sake. Kirk Goldsberry, in a Sloan paper, estimated the scoring area in basketball is 1280 square feet based on where the vast majority of shots are taken. From a geometry stand point taking shots from further out increases the percent of that area the five defenders have cover and increases the space between defenders, opening lanes to the basket. As Houston GM Darryl Morey or any old time basketball hand would probably tell you, driving to the rim and spacing are complimentary. Players need space to open lanes and drive to kick it back out.

A while ago I ran a number of K-means Cluster analyses based on shot type using data from the suddenly defunct, HoopData. The analysis grouped Guards, Wings and Bigs based on where they shoot in relation to the basket. The analysis grouped both the guards and the wings into groups of Stretch and Slasher types, scoring from the three point line and at the rim respectively. Meanwhile, Bigs fell into three clusters, Bigs that shot almost exclusively at the rim, those with more varied offensive games and stretch bigs, who added a three point shot.

For example, Stretch Wings averaged 56% of their shots from three point range and 19% at the rim, while Slashing Wings averaged 21% from three and 35% at the rim, according to my analysis.  The numbers for Stretch and Slashing guards were both very similar. The cluster analysis allows me to group player types together to look for patterns in the data, for example, I found some interesting patterns in terms of style of play with Stretch back court players and wings have higher eFG% percentages, but Slashers get more offensive rebounds and more free throws. Again, I will use the Wings as my example.

Group	COUNT	Avg TS%	Avg eFG%	Avg ORB%	Avg DRB%	Avg AST%	Avg TOV%
Slash Wing	39	52.7%	49.1%	5.84	16.09	10.79	11.95
Stretch Wing	25	53.7%	51.2%	3.20	13.08	10.03	11.15
Grand Total	64	53.1%	49.9%	4.81	14.91	10.50	11.64

But to take the analysis to the next step I wanted to look for an independent value of spacing, and its limits. I think that is another value of using categorization, Stretch players are measured as players that take more three point shots, not necessarily make them at a higher rate (though there is a correlation there too). The methodology adopted was both to chart the Stretch players in five man line ups against offensive rating (ORTG) and, secondly, to calculate an expected ORTG, (points per one hundred possessions) based on the individuals in a line up and compare it to the actual ORTG generated by that five-man lineup, then add Spacing configurations to the mix. (Similar to an analysis done here by Jacob Frankel using a different definition of stretch.)

I used line up data from the last two years, courtesy of NBA.com's media stats site. The chart below uses a cut off of five-man lineups that played at least 250 minutes together in the either of the last two years, however, I ran the same chart multiple times with different minute filters and got largely the same results.  With the five man line up at this cut off, Tyson Chandler, Carmelo Anthony, Jason Kidd, Raymond Felton and J.R. Smith had the highest ORTG with 119.4 points per 100 possessions, and Brandon Knight, Jason Maxiell, Greg Monroe, and Rodney Stuckey had the lowest at 93.6 points per 100 possession.

 

I used a polynomial line to visualize the trend.  Consistently, with different minute filters the trend showed an increase in offensive rating going from 0 Stretch players in a line up to 1 and 2, and then flattened.  The same pattern was seen with Wing and Guard Stretch players only, as well as, using effective field goal percentage. Then I ran the expected ORTG for each line up against the actual ORTG the line up had, with the 250 plus minute filter the R^2 was a decent 0.434.  The chart below shows the number of Stretch shooters in each line up against the residual error on the expected ORTG.

Again the same pattern can be seen with an increase in the offensive rating with two stretch players on the court and a then leveling off indicating diminishing returns to spacing.  Using that information I made a 'dummy' variable for each lineup that had two or more Stretch shooters and ran that variable in a step wise regression along with the Expected ORTG and linear count of Stretch shooters.  In every minute filter the Dummy Stretch came out as the most significant contribution in addition to the Expected ORTG. The result of the model run with the 250 minute filter indicated that the value of having at least two Stretch shooters on the floor is 2.44 points per hundred possessions.

	Unstandardized Coefficients		Sig.	95.0% Confidence Interval for B
Variable	B	Std. Error		Lower Bound	Upper Bound
(Constant)	13.29	12.53	0.29	-11.74	38.31
Expected ORTG	0.83	0.12	0.00	0.60	1.07
StrechersDummy	2.44	0.89	0.01	0.67	4.21

 

The results were similar with other filters on minutes, with the independent value of having two Stretch shooters in a line up centered just above two points per hundred possessions.  It might not seem like a lot, however, this value represents the value above the simple algebraic shooting percentage multiplied by three points. And two net point per game equal about 5 wins over the course of the season.

The Rookie Escalator

It's no secret that Anthony Bennett is having a rough start to his year and performing well below expectations, certainly below Cleveland Cavaliers' GM Chris Grant's expectations. And, of course, Michael Carter-Williams has exceeded almost all expectations outside of his mother's probably including Sam Hinkie's expectations.

But, anyone paying attention in the pre-season would have at least a little better prepared for both. Both Bennett's struggles and Carter-Williams success were strongly presaged by their performances in preseason play. At the beginning of the season I did an analysis using a linearized Bayes rule formula that indicated that Steven Adams, Michael Carter-Williams and Vitor Faverani were among the upward movers and Bennett, Andre Roberson, and Peyton Siva were among the biggest downward movers.

As of Friday we had about as much data from the regular season as from preseason, so I decided to download it to do some more analysis. I was interested in a couple of things, first how well my Updated Expectations looked after the first set of games and weigh the sample size to date against my prior beliefs after preseason.

I also added a few more of the rookies from Europe, using their preseason numbers with a rookie mean reversion factor

First the big movers up the Rookie Escalator. My prior expectations were set combining the rookies numbers coming into the NBA and their performance in preseason, you can see more on the methodology here. Again Michael Carter Williams and Vitor Faverani move up using Win Score as the arbiter.

For Carter Williams the unexpected excellence has been nearly across the board, including shooting, assists, steals and blocks, though the model expects some mean reversion, particularly in his shooting. Faverani continues to raise expectations in rebounding and blocks. Also raising expectations were Giannis Adetokunbo (aka the Greek Freak) and Solomon Hill. Giannis due to his better than expected shooting and Hill because of his rebounding.

Rebounds and blocks are shown per 48 minutes. Stats via RealGM.

Player	Team	FG%	Prior FG%	Updated Exp FG%	3P%	Prior 3P%	Updated Exp 3P%	REB	Prior Reb	Updated Exp Reb	BLK	Prior Blk	Updated Exp Blk
Giannis Adetokunbo	MIL	52.60%	39.10%	45.5%	50.00%	24.00%	37.5%	6.63	9.14	8.32	0.55	2.40	1.56
Michael Carter-Williams	PHI	38.90%	33.80%	35%	39.50%	26.70%	33.4%	7.04	6.776	6.86	0.98	0.23	0.57
Solomon Hill	IND	30.80%	38.00%	36.3%	11.10%	26.60%	18.5%	9.74	5.787	7.08	0.57	0.40	0.48
Vitor Faverani	BOS	46.00%	49.80%	48.9%	28.60%	21.70%	25.3%	13.9	11.846	12.52	3.6	2.59	3.05

 

On the other side of the spectrum going down the Rookie Escalator of Expectations, are Anthony Bennett, Dennis Schroeder, Phil Pressy and Nate Wolters. For Bennett the issue is primarily his atrocious shooting, though there isn't an area that he over performed in to offset the shooting woes. Note, the model still thinks that Bennett's shooting will improve, but can't ignore his numbers so far. Schroeder has been a disappointment to the model to start the season, where, unlike Carter Williams he has enhanced the concerns about his shooting prowess. In fact, shooting is the most common theme in those moving down along with rebounds and a lack of assists for Pressey

 

Player	Team	FG%	Prior FG%	Updated Exp FG%	3P%	Prior 3P%	Updated Exp 3P%	REB	Prior Reb	Updated Exp Reb	BLK	UpPrior Blk	Updated Exp Blk
Anthony Bennett	CLE	14.30%	45.80%	38.40%	8.30%	32.70%	20.00%	11.02	11.87	11.59	0.92	1.268	1.11
Dennis Schroeder	ATL	33.30%	40.60%	35.30%	14.30%	22.30%	18.10%	4.56	5.92	5.47	0	0.536	0.29
Nate Wolters	MIL	38.20%	40.70%	40.10%	12.50%	40.90%	26.10%	4.77	5.72	5.41	0.73	0.41	0.56
Phil Pressey	BOS	27.30%	32.50%	31.30%	25.00%	22.40%	23.80%	5.06	5.34	5.25	0.51	0.215	0.35

 

Finally, some of the rookie have been roughly as expected coming out of preseason, or at least, haven't moved their expectations going forward. Few of the changes were significantly different than expectations after the preseason.  Kelly Olynyk is a slight exception, where his shooting difficulties were neutralized by his better than expected rebounding, assists, and the model's expected mean reversion.

Player	FG%	Prior FG%	Updated Exp FG%	Change FG%	3P%	Prior 3P%	Updated Exp 3P%	Change 3P%	REB	Prior Reb	Reb St Dev	Updated Exp Reb	BLK	Prior Blk	Updated Exp Blk
Archie Goodwin	34.40%	39.00%	37.9%	-1.1%	11.10%	11.50%	11.3%	-0.2%	6.24	5.30	2.657	5.61	2.84	0.16	1.38
Ben McLemore	38.30%	44.40%	43%	-1.4%	36.70%	36.30%	36.5%	0.2%	5.83	7.11	2.657	6.69	0.58	0.69	0.64
Cody Zeller	41.30%	50.10%	48%	-2.1%		0.00%	0%	0%	10.21	11.88	2.657	11.33	0.87	0.84	0.85
Gal Mekel	41.20%	38.70%	39.9%	1.2%	22.20%	26.70%	24.4%	-2.4%	4.31	6.91	2.657	6.06	0.39	0.67	0.54
Gorgui Dieng	50.00%	47.20%	47.9%	0.7%		0.00%	0%	0%	12.94	12.94	2.657	12.94	5.88	4.08	4.9
Kelly Olynyk	41.20%	57.10%	53.4%	-3.7%	17.60%	27.80%	22.5%	-5.3%	11.67	10.14	2.657	10.64	0.42	1.35	0.93
Steven Adams	56.70%	53.00%	53.9%	0.9%		0.00%	0%	0%	12.31	13.59	2.657	13.17	2.77	2.78	2.77

 

I plan to update the rookie numbers again in the future, at that point I will also be able to add in some rookies that just missed my minute cut off like Jeff Withey and Jamaal Franklin.  Also I will be able to look at how the model at various stages performed.

Jeff Green Should Listen to Tommy Heinsohn

Over at Hickory High, I did a post giving my first look at the new Drive stats being generated by the SportVu cameras installed in the catwalks of every NBA arena this year.  In that piece I found that so far the NBA is averaging 1.1 points per drive play, with only 59% of the points off a drive being scored by the driving player.

Given that the NBA's definition of drive plays excludes fast breaks, drives are a very efficient play well above the average points scored in a half court set.

With those totals in mind, I wanted to look at the Celtics roster on drive plays so far this year.  The data is via NBA.com with the first nine games through 11/13/13.   I pulled out the the total number of drives so far,  points the player scores per drive and the points generated by the team on the player's drives.  I also calculated the the number of Drives each player had per 48 minutes and the ratio of points the player gets compared to those for the team off of assists.

Player	Total Drives	Drives per 48 minutes	Player Pts Per Drive	Team Pts Per Drive	Player/Team Ratio
Avery Bradley (BOS)	18	3.2	0.72	1.1	65.7%
Brandon Bass (BOS)	10	1.8	0.8	1.55	51.8%
Courtney Lee (BOS)	8	2.5	1	1.33	75%
Gerald Wallace (BOS)	7	1.2	0.57	1.25	45.7%
Jeff Green (BOS)	36	5.7	0.89	1.15	77.3%
Jordan Crawford (BOS)	36	7.5	0.64	1.18	54.4%
Keith Bogans (BOS)	0	0.0
Kelly Olynyk (BOS)	8	1.6	0.25	0.67	37.5%
Kris Humphries (BOS)	0
Phil Pressey (BOS)	11	6.4	0.36	0.94	38.5%
	134		0.65	1.15	55.74%

 

The table shows that Jordan Crawford is the most prolific driver on the team with 7.5 drives per 48 minutes that he plays.  Green has the same number of total drives, but isn't actually as active with 5.7 drives per 48 minutes.  Both have been more efficient on the drive than the NBA average so far, which already a relatively efficient play, with Green generating more of his own points with Crawford getting his teammates more involved.

The Celtics are going to struggle on offense this year, probably even after Rajon Rondo comes back from his knee injury.  They're going to need to use every above par weapon they have as often as possible with no disappearing acts.

Asik/Howard II: Picturing Negative Space

Last post I looked at some evidence of the effect of spatially challenged front courts, with an eye towards the Rockets erstwhile starting front court of Omer Asik and Dwight Howard.  In that post I looked at the performance of some similar non-spacing front courts over the last two years and found no degradation in offensive performance.

On the other hand, the Rockets have performed poorly thus far with that line up.  Part of the issue is that Asik/Howard are clearly clogging the middle, through the first six games, Harden, Chandler, and Lin were all shooting a lower percentage of their shots at the rim than without the tandem on the court., via NBAwowy. And, the Rockets as a whole were getting fewer shots at the rim with the twin towers in than out.

One answer maybe that the Asik/Howard front line is more spacing challenged than most of the other eight line ups I found.  Last year Asik took 88% of his shots in the restricted area and Howard took 75% according to NBA.com/Stats.  Here is the visual of the shot distribution for each this year.

First Asik:

Open court means don't bother guarding him if he stands out there.

Howard Shot Distribution:

That's a lot of pretty court.

As I repeat always, shots at the rim are a good thing, which both Asik and Howard do .  But, any scouting report is going to scream sag off and discount the jumper.  And, thus far the net effect has been to reduce the shots at the rim for their team.

The only true comparable last year I found with significant minutes was the combo of Amare Stoudemire and Tyson Chandler in 2011-2012.  Stoudemire shot 75% of his shot at the rim last year, while Chandler took 90%.  Chandler and Stoudemire were featured in two of the under performing five man man line ups with two Negative Spacers, which is somewhat suggestive, but nothing more than that.  In 2012, the first year we have NBAwowy data, Stoudemire/Chandler did not have the same negative effect on their teammates getting to the rim for the most part.  Carmelo Anthony and Raymond Felton had the same percentage of their shots at the rim, while JR Smith showed an 8% drop off.

Another answer may Net Spacing, which is a measure I have played around with that essentially penalizes line ups for Negative spacing front court members and adds for Stretch players.  In the analysis Net Stretch showed a positive independent contribution to team ORTG.  But the measure is still a bit conceptually muddled, for example, it was unclear whether the under performance was being driven by the stretch players or the negative effect of Negative Spacers.  Other than it is probably not an ideal line up to have two bigs that need to hug the basket and a back court that relies on drives to the basket to score, but we knew that right?

Can Asik and Howard Work Together? Maybe!

(The Non-Negative Value of Negative Spacing)

There's a well known bias in academic publishing that favors papers that find statistically significant results over any study that doesn't. That is unfortunate, because sometimes it is as important to find evidence to rule out relationships or causation as to find evidence to support them.

Of course, blog posts don't go through a committee or reviewers; but there is still a natural bias I think to report research that finds significant relationships. This isn't going to be one of those posts.

OK, throat clearing over, I have been doing a whole series of posts on shot location in the NBA here and at Hickory-High. The last one looked at the independent value of of having 'Stretch' shooters on the floor in addition to the improved efficiency of the three point shot. In that study I found that having at least two Stretch shooters on the floor appeared to improve a line up's offensive efficiency as measured by points per 100 possessions (ORTG), but adding an additional Stretch shooter did not aid offensive efficiency.

In my shot location Cluster Analysis series I found three groups of Big players; players who shot almost exclusively from arms reach of the basket, stretch Bigs that shot three point shots and Hybrids that had a mid range game.

This is how the shot selection and statistics looked for each type:

		AVG Att at the Rim	% Att at 3-9 ft	% Att at Mid Range (10-15)	% Att at Long Two(16-23)	% Att of Three Pointer
At the Rim Big		67.31%	21.24%	3.58%	7.62%	0.36%
Hybrid Big		40.31%	16.47%	10.94%	29.88%	1.96%
Stretch Big		26.10%	9.32%	5.83%	19.36%	38.91%

 

The 'At the Rim' shooters displayed almost no spacing value, and the majority of Bigs fell into the middle category.

In terms of interior spacing, there has also been a good deal of discussion about the viability of line ups with two bigs with limited jumps shots, such as Dwight Howard and Omer Asik in Houston. Both scored as 'At the Rim' shooters, and have displayed limited range on their jump shots.

Given that, I decided to add to the spacing analysis with some negative space analysis of Bigs. I coded every 'At the Rim' big as a negative -1, then added all of the negatives together, finding this distribution for Line Up combos over the last two years:
.

Statistics
RimBig	Count	Offensive Rtg	TmTOV	AST Ratio	Net Rtg
2 Negative Spacers	8	103.04	16.33%	17.59	3.53
1 Negative Spacer	50	104.42	15.62%	17.85	2.72
No Negative Spacer	36	104.76	14.50%	18.95	3.51

 

There were only eight line ups with 2 Negative Spacing Bigs that played over 200 minutes together in the last two years, indicating coaches certainly believe in interior spacing's value.

In the examples found there was little absolute degradation in offensive rating for the handful of line ups the featured 2 Negative Spacers without a distance or mid-range game. In fact the Net offense rating minus defense rating of the 2 Negative Spacers line ups were better than the 1 Negative Spacer line ups and virtually identical to the no Negative spacer line ups.

Looking at the difference between expected offensive rating ORTG based on the average of players in the line up and the actual ORTG achieved, there was also little evidence of a negative spacing effect.  There was no statistically significant effect from having negative spacing in the front court in any filter of minutes played together.

In fact one of the line ups that worked well last year and continues this year, is the starting Toronto line up with Jonas Valanciunas and Amir Johnson, who no one comments on because they play in Canada.  That line up had an average ORTG but a very good defensive rating for a Net of +12.9 per 100 possessions over 354 minutes last year.

Returning to Dwight Howard and Omer Asik playing together, this study found no negative effect of playing two Negative Spacers on the floor together.   All that suggesst that Asik/Howard could work together. However, even without an independent effect from negative spacing bigs, Howard and Asik had the lowest ORTG of the Rocket's current starting line up last year, both being below the league average.

And to date it hasn't worked that well in Houston.  Part of the issue is that Asik just isn't playing well so far, his eFG% through six games is down ten points to 45.5% and his turn over percentage is up to 28.8% from 18.8% last year.  But neither can be blamed on Howard, as Asik's eFG% is higher and TOV% is marginally lower when on the court with Howard according to NBAWowy.

It also appears that having Asik and Howard on the court is clogging the lanes for their teammates. Harden, Parsons and Lin are all taking a lower percentage of their shot attempts at the rim when the Asik/Howard tandem is on the court than when it is not, again via NBAWowy.

Player	% of Shots 0-3 feet WITH Asik/Howard Combo	% of Shots 0-3 feet WITHOUT Asik/Howard Combo
James Harden	20.8	33%
Chandler Parsons	34.6	60%
Jeremy Lin	33.3	39%

 

In fact, so far the team as a whole has taken a higher percentage of their shots inside three feet when the Asik/Howard combo is not on the court than when they are.

Part of the problem in addition to opponents sagging off of them may be contained in the high turnover and low assist numbers for both Howard and Asik, neither of whom are particularly good interior passers, and that's generous for Asik.

But the evidence I found suggests that line ups with two space clogging bigs can work, whether Asik and Howard will work is too early to tell.

(Note light editing added 11/9/13 for clarity)

Something Ugly This Way Comes? Celtics v Grizzlies.

The 0-3 Celtics take on the 1-2 Memphis Grizzlies tonight, 11/4, in Memphis.  Here's how I see the match up by the some of the numbers.

When I ran my initial Plausible Win Projections,  I had the Grizzlies pegged as a 50 win team and the Celtics at 33. If we take those seriously, and include home court advantage that projects to something like a 13% chance of a Celtic victory, without including the fact that the Celtics are coming off a back to back.

Good news out of the way, let's look at the roster stats Grizzlies, using the 2012-2013 stats.  The final column, ASPM, is Daniel Meyers' advanced all in one statistic, which I projected using two years of data and an age asjustment.

Pplulayer	TS%	Ast%	TOV%	Steal%	ORB%	DRB%	Blk%	Two Yr Ave ASPM
Ed Davis	55.45%	5.6	10	1.25	11.15	22.8	4.95	0.99
Jerryd Bayless	51.40%	24.9	15.1	1.8	1.4	10.5	0.80	0.14
Jon Leuer	49.35%	6.9	12	1.7	7.9	15.4	0.00	-1.14
Kosta Koufos	58.50%	2.6	8.7	1.2	13.3	21.2	4.40	2.22
Marc Gasol	55.90%	19.1	13.5	1.6	7.7	18.9	4.10	3.08
Mike Conley	54.90%	29.5	15.1	3.4	1.8	7.9	0.70	3.52
Mike Miller	59.60%	16.2	12.9	1.2	2.6	18	0.40	-1.18
Quincy Pondexter	56.50%	7.6	11.2	1.5	4	8.6	0.40	-0.80
Tayshaun Prince	48.15%	12.1	9.15	1	3.1	13.15	0.75	-1.78
Tony Allen	48.90%	7.6	11.5	3.1	6.6	13.8	1.70	1.47
Zach Randolph	50.60%	7.3	11.5	1.3	13.8	25.1	1.00	0.82
Grand Total	52.41%	10.66	11.21	1.55	6.99	16.47	1.82	0.08

 

In addition, the Grizzlies have rookies Nick Calathes and Jamaal Franklin  this year with Calathes seeing significant minutes as back up point guard so far this season.

Generally, speaking over the last couple of years the Grizzlies have been a strong and gritty defensive team, but struggled to generate points.  So far this that's been the case this year as well.

My eFG Versatility Index, which compares each player's eFG% to the league average eFG% in each shooting zone according to HoopData, also doesn't show any dominate scorers on the Griz, though Gasol is surprisingly versatile and effective in the mid range game.  Tayshaun Prince actually comes out higher from three point land than I expected.  (Also, completely ignore the Jon Leur row due to lack of sample size).

	Total FGA	At Rim Ratio	Short Ratio (3-9)	Mid Ratio (10-15)	Long Two Ratio (16-23)	Three Ratio	FTM/FGA Ratio	eFG Versalitiy Index
Ed Davis	482	1.39	0.83	0.81	0.68	0.00	1.12	0.67
Jerryd Bayless	621	1.31	0.52	1.26	0.76	1.08	0.86	0.97
Jon Leuer	52	1.41	0.67	1.51	0.86	0.00	0.00	0.77
Kosta Koufos	508	1.34	0.99	0.67	0.50	0.00	0.60	0.57
Marc Gasol	869	1.35	0.92	0.99	0.94	0.21	1.62	1.01
Mike Conley	940	1.16	0.83	0.93	0.88	1.02	1.31	1.02
Mike Miller	231	1.31	0.29	1.00	0.82	1.22	0.18	0.91
Quincy Pondexter	299	1.27	0.82	0.40	0.60	1.20	1.04	0.89
Tayshaun Prince	820	1.43	0.88	0.64	0.77	1.22	0.62	0.93
Tony Allen	638	1.18	0.56	0.74	0.66	0.38	1.11	0.77
Zach Randolph	1024	1.19	0.90	0.66	0.76	0.26	1.16	0.82
Grand Total	560	1.33	0.76	0.90	0.76	0.56	0.95	0.84

 

I think it is worth putting the Celtics eFG Versatility chart in for comparison, here Jeff Green comes out as the Celtics' key option, primarily because of his ability to score at the Rim, from three and get to the line.  If he's playing at the two, he'll be guarded by defensive stopper Tony Allen for much of the night.

	Total FGA	At Rim Ratio	Short Ratio (3-9)	Mid Ratio (10-15)	Long Two Ratio (16-23)	Three Ratio	FTM/FGA Ratio	eFG Versalitiy Index
Avery Bradley	478	1.03	0.67	0.67	0.86	0.97	0.41	0.77
Brandon Bass	589	1.28	0.70	0.89	0.92	0.00	1.15	0.82
Courtney Lee	530	1.32	0.56	0.73	0.90	1.12	0.62	0.88
Gerald Wallace	458	1.11	0.75	0.00	0.56	0.85	1.39	0.78
Jared Sullinger	225	1.38	0.42	1.23	0.66	0.60	1.10	0.90
Jeff Green	806	1.33	0.61	0.82	0.72	1.15	1.40	1.01
Jordan Crawford	725	1.20	1.03	1.02	0.77	1.01	0.94	0.99
Keith Bogans	284	1.51	1.20	0.00	0.92	1.02	0.20	0.68
Kris Humphries	310	1.15	0.68	0.63	0.68	0.00	1.65	0.80
MarShon Brooks	335	1.38	0.91	0.89	0.62	0.81	1.08	0.95
Rajon Rondo	465	1.25	0.83	0.76	0.96	0.72	0.68	0.87
Grand Total	494	1.26	0.78	0.72	0.78	0.77	0.96	0.87

 

This has the shapings of a potentially pretty ugly game from a couple of perspectives for the Celitcs.  Neither team has an aesthetically pleasing half court game, it is the Celtic's first back to back of the year, on the road, and the Celtics are a long way off from establishing the kind of team defense the Grizzlies already have.

Mascot Match Up

Often overlooked, but important piece of analysis, is the hypothetical match up between real versions of each team's mascot.  About fifty percent of the time the team with the more imposing mascot wins in the NBA.  

I always find the human mascot vs animal mascot's to be a complicated issue.  Physically, clearly twelve Grizzlies would overwhelm twelve Celtic people.  But, the opportunity for cooperation and tool use by the Celtic people could tip the balance,  making that one too close to call.

Omnibus of Rookie Expectations

The beginning of this post first appeared on Hickory-High.com on October 29th, with additional posts from the same series added after the break. This is meant to gather them together in one semi-coherent post for archive purposes, so I added some light editing to reduce redundancy

Earlier, I did some analysis and created a visualization of the correlation between rookie preseason and rookie regular season performance on a number of stats. Before the sell by date of the analysis passed, basically as soon as the regular season starts and the preseason is expunged from our collective memory, I wanted use that data to update my expectations of this year's rookies.

At this point one could run a couple linear multiple regressions with college stats and preseason stats, or maybe use one of the single number metrics. But, there is a huge amount of colinearity between the college stats and preseason. Guys who rebound well in college, for example, tend to do so in preseason, meaning many of the variables end up getting confounded, especially dealing with relatively small sample sizes.

In any case, Bayesian analysis actually lends it's self much better to updating our expectations using sequential data, small data sets than multiple regression, and it's analytically 'on trend.'

A key to Bayesian analysis is to establish a prior belief, or starting point. To do that I used a combination of general rookie averages and the incoming draft class's college stats and ranks.  As an example Celtics number thirteen pick Kelly Olynyk had an effective field goal percentage of 63.1% last year using data from Draft Express. I don't start my expectations that he will be able to do that in the pros. On the other hand, the fact that he was significantly above the 54% eFG% college average of his fellow rookies (2.075 standard deviations above) in that category is important information.

In the last three years rookies with over 200 minutes of playing time in the regular season had an eFG% of 45.8%, four percent below the NBA average via data from RealGM. So, taking that information, my Prior Expectation for Olynyk's eFG% was formed as below:

Rookie Average + (Standard Deviations from Class * Standard Deviation) = Prior Expectation

In this case: 0.458 + (2.075 .0506) = 56.3%

One could argue for tempering the adjustment away from the Rookie average, particularly for shooting. However, given that I am going to update these expectations with preseason observations, I think it forms a good informative starting point.

In eight games over the preseason Olynyk had an eFG% of 55.1%. Unfortunately, for demonstration purposes that means the Updated Expectations won't move too much.  To update my expectations I had to weight my prior expectation and Olynyk's preseason performance using a Bayesian weighting formula (shown below) coming out as 56.11%.

So, to look at one stat where Olynyk did move the expectations meter in the preseason, we can look at personal fouls. On personal fouls Olynyk already committed more fouls than average for his class, putting his expected fouls per 48 minutes above the rookie average of 4.9, at 5.63. But the big young Canadian well-surpassed expectations in the preseason getting called for fouls at a rate of 8.01 per 48 minutes. Weighing the two together Olynyk's Updated Expectations come to 6.81 fouls per 48 minutes (meaning they don't let you play 48 minutes in one game).

I applied the same methodology to all the 2013 rookies with college experience and significant playing time this preseason. Below are the preseason numbers, my 'Prior' belief and the Updated Belief for eFG%, Rebounds per 48 minutes, and Assists per 48.

Player	Team	Pos	Pre eFG%	Prior eFG%	Updated eFG%	Pre REB	Prior Reb	Updated Reb	Pre ASTS	Prior Asts	Updated Asts
Anthony Bennett	CLE	PF	0.413	49.7%	48.31%	12.05	11.68	11.87	1.96	1.7	1.83
Victor Oladipo	ORL	PG	0.447	57.7%	55.5%	10.27	8.88	9.58	8.45	3.3	5.87
Cody Zeller	CHA	PF	0.443	48.5%	47.8%	12.87	10.88	11.88	3.41	2.2	2.81
Ben McLemore	SAC	SG	0.546	51.4%	51.96%	7.84	6.38	7.11	2.32	2.9	2.61
Kentavious Caldwell-Pope	DET	SG	0.375	45.6%	44.25%	8.92	8.28	8.6	1.96	2.5	2.23
Michael Carter-Williams	PHI	PG	0.388	35.9%	36.35%	8.07	5.48	6.78	7.58	8.7	8.14
Steven Adams	OKC	C	0.622	49.1%	51.27%	16.39	10.78	13.59	1.46	1.5	1.48
Kelly Olynyk	BOS	PF	0.551	56.3%	56.11%	9.3	10.98	10.14	3.87	2.9	3.39
Tony Snell	CHI	SG	0.362	42.9%	41.79%	5.85	3.38	4.62	5.85	4.1	4.98
Mason Plumlee	BRK	PF	0.409	52.3%	50.39%	14.63	11.48	13.06	2.72	2.6	2.66
Solomon Hill	IND	SF	0.24	46.3%	42.58%	5.09	6.48	5.79	3.52	3.7	3.61
Tim Hardaway Jr.	NYK	SG	0.482	42.8%	43.72%	5.13	5.18	5.16	1.56	3.2	2.38
Andre Roberson	OKC	SG	0.273	43.5%	40.83%	12	13.38	12.69	0.92	2.1	1.51
Archie Goodwin	PHX	SG	0.415	37.7%	38.33%	4.82	5.78	5.3	1.75	3.8	2.78
Tony Mitchell	DET	PF	0.733	40.1%	45.63%	12.62	10.38	11.5	1.1	1.3	1.2
Jamaal Franklin	MEM	SG	0.468	37%	38.66%	8.69	11.28	9.99	4.97	4.4	4.69
Peyton Siva	DET	PG	0.404	37.6%	38.03%	2.53	2.98	2.76	9.49	7.7	8.6
Erik Murphy	CHI	PF	0.293	54.8%	50.56%	9.69	8.38	9.04	1.79	2.5	2.15

 

The positive movers are in yellow and the negative in red in terms of changes in expectations.  Effective field goal percentage didn't move much for anyone, primarily because the high variability of scoring and high standard deviation of eFG% in preseason.  Though Steven Adams moved up in eFG% somewhat, while Erik Murphy moved down.  On rebounds, Steven Adams looked like a beast on the boards all preseason, enough to significantly move his Updated Rebounds expectations, and Victor Oladipo playing a facilitating role with Orlando moved his Updated Assists up as well.

In addition I ran the numbers for free throws, turnovers and personal fouls, all per 48 minutes.

Player	Team	Pre FT%	Prior FT%	Updated FT%	Pre TOV	Prior TOV	Updated TOV	Pre PF	Prior PF	Updated PF
Anthony Bennett	CLE	0.692	66.4%	67.8%	6.17	2.83	4.5	10.37	5.53	7.95
Victor Oladipo	ORL	0.839	71.4%	77.7%	5.94	3.33	4.63	4.11	5.63	4.87
Cody Zeller	CHA	0.548	72.5%	63.7%	3.68	3.13	3.4	4.73	5.13	4.93
Ben McLemore	SAC	0.692	83.8%	76.5%	3.19	2.73	2.96	3.77	4.43	4.1
Kentavious Caldwell-Pope	DET	0.7	76.5%	73.3%	2.18	2.53	2.35	5.44	4.63	5.03
Michael Carter-Williams	PHI	0.647	66.2%	65.5%	2.44	4.03	3.23	5.38	4.73	5.05
Steven Adams	OKC	0.529	41.1%	47%	4.39	2.03	3.21	7.9	4.93	6.41
Kelly Olynyk	BOS	0.7	74.8%	72.4%	3.62	3.83	3.72	8.01	5.63	6.82
Tony Snell	CHI	0.714	81.1%	76.3%	1.72	2.73	2.22	2.75	3.83	3.29
Mason Plumlee	BRK	0.593	64.9%	62.1%	4.76	3.43	4.09	3.74	5.03	4.38
Solomon Hill	IND	0.75	73.4%	74.2%	1.96	2.83	2.39	3.52	4.73	4.12
Tim Hardaway Jr.	NYK	0.615	66.3%	63.9%	1.56	2.23	1.89	2.9	4.33	3.62
Andre Roberson	OKC	0	51.9%	26%	4.62	3.03	3.82	8.31	5.13	6.72
Archie Goodwin	PHX	0.588	60.5%	59.7%	4.82	4.03	4.42	3.07	5.83	4.45
Tony Mitchell	DET	0.5	63.7%	56.9%	1.1	3.23	2.16	7.68	5.73	6.7
Jamaal Franklin	MEM	0.8	75.8%	77.9%	3.73	4.23	3.98	4.55	5.33	4.94
Peyton Siva	DET	0.833	83.5%	83.4%	6.64	3.53	5.08	5.38	5.43	5.4
Erik Murphy	CHI	0.5	75.2%	62.6%	1.08	2.23	1.65	9.69	6.33	8.01

 

Again the most significant negative Updated Expectations are in red and positive in yellow.  It looks like I might let my parochial New England interests bury the lede as Anthony Bennett struggled enough with turnovers and personal fouls to worsen his Updated Expectations. That said, coming off an injury and by all accounts not being in game shape, I could probably temper my expectations adjustment somewhat. And I am not a 'Bash the Number One Pick Because He isn't the Best Guy in his Draft Class' kind of guy, I wouldn't have taken him there, but that's on Cleveland GM Chris Grant not Anthony Bennett.

I also tracked blocks and steals, but neither showed much impact on expectations. As promised above, here's the expectations updating formula I used. It is designed to diminish the impact of measures with high variability as measured by standard deviation. When trying to formulate a Bayesian continuous update formula this weekend, I ran across this one via Daniel Myers.

Updated Expectation= (Prior Expectation / Prior StDev^2 + New Observation/ New Observation StDev^2) / (1/Prior StDev^2 + 1/New Observation StDev^2).