Counting the Baskets

Revealed Coaching

I wanted to see how different one number player metrics lined up with how NBA coaches actually coach. So, I ran a series of simple linear regressions comparing the score on various player metrics to minutes player per game last year.  Since each metric takes a slightly difference approach, it could tell us a bit about what NBA coaches value. The idea is similar to a Ari Caroline piece over at Wages of Wins last year comparing how their metric, Wins Produced, stacked up with minutes. Here I have added PER, ASPM, xRAPM and Win Shares.

The results are very preliminary but too interesting not to share. Among the caveats are that currently the WP number is a a three year average while the others are for current the year.  Though  I ran the 2012-2013 Win Produced numbers on a slightly different data set and got very similar results.  Also, the two year ASPM actually did better than the single year, suggesting, sensibly, I think, coaches have prior beliefs coming into the season and adjust somewhat slowly to unexpectedly good or poor performances.

In any case, in the table below I have R squared numbers, or,  goodness of fit, from each simple regression I ran.

	R2 Player Metric and MPG
	WP48	WS	PER	ASPM	xRAPM	2 Yr ASPM
PG	0.335	0.262	0.441	0.494	0.359	0.535
SG	0.093	0.148	0.345	0.388	0.266	0.459
SF	0.095	0.068	0.215	0.348	0.311	0.391
PF	0.004	0.057	0.325	0.247	0.214	0.213
C	0.145	0.144	0.394	0.335	0.348	0.339
ALL	0.088	0.101	0.298	0.333	0.24	0.34

 

The yellow highlighted background cells are the highest correlation for that metric.  Astute observers may notice a pattern; all of the stats agreed most with the minutes allocations given out at the Point Guard position. This supports the idea that the Point is the best defined and understood position in the game right now. Since I made the claim that's how I am going to interpret it.

The red highlighted lines are the position each stat agreed least with the minute allocation. In this case, the Power Forwards are the opposite of the Point Guards where each metric save PER had their lowest fit on that position.  For PER the worst fit was at small forward.

Looking for the best fitting metric by each position, Daniel Myers' ASPM in both the single year and two year weighted average correlated the best to minutes assigned for Point Guards, Shooting Guards, and Small Forwards. While John Hollinger's PER fit the best for both Power Forwards and Centers.

By the way, my unofficial outlier awards goes to Ben Gordon, whose game all of the metrics hate and yet still managed to see 21 minutes a game.  And the hyper efficient Jeremy Evans, loved by all except xRAPM and Ty Corbin getting just under six minutes a game.

So what does that say about the advanced metrics and about coaches?

Not all of the metrics are created equally in terms of preditive ability (or retrodictive ability). A blog called the Sports Skeptic did a retrodictive test a while ago, using the advanced stat scores from the year before applied to the playing minutes the next year. The author found that of the ones I am looking at PER had the lower predictive abilities.  But confusingly ASPM had one of the highest predictive abilities.

So, coaches can judge guard's abilities, other than Ben Gordon, but not bigs? Probably too early to say that.

Also, I am currently working on some multiple regressions, which will include more years.  So far, the plus/minus stat xRAPM has come in as a positive combination with each of the other box score based metrics, the next step will be to break up plus minus into its offensive and defensive components to  see if my hunch that it is the defensive side of xRAPM that is adding value is correct.  Also, games played, which I included as an indication of injuries, has come in has a positive contributor improving each metric's score.

As for the metrics, it looks like analytics movement still has a ways to go in gaining actual influence, Wins Produced and Win Shares in particular.

Should Brad Stevens Start Jordan Crawford Over Avery Bradley at Point Guard?

(I can't believe I am writing this)

To the surprise of no one, on the Sports Hub yesterday, Danny Ainge confirmed that, Rajon Rondo is likely to miss the first part of the season.

Given that Rondo is both the best player on the team and the only natural Point Guard with NBA experience, I have been thinking about how the C's should start the season's back court line up, as I suspect has Brad Stevens and his staff.  The issues in picking a replacement back court also ties in nicely to some work I have been doing on shot creation and shot location in the NBA., in terms of finding the most complimentary mix of ball handing, passing, floor spacing and defense.

The shot creation pieces are particularly relevant to playing at the Point and the skills the Celtics will need to replace while Rondo recovers.  The Point on offense is probably the best defined position of the five cardinal positions.  He handles the ball, sets up his team mates, and gets his own shot when needed.

To take those skills in reverse order, most Point Guards ranked as what I called Shot Creators in my cluster analysis with only a quarter rating as Spot Up Guards, who score disproportionately off of getting assisted, including Avery Bradley.  In fact, of all the players defined as a Point Guards by Hoop Data with more than two hundred minutes played last year, Bradley ranked 11th from the bottom last year, or the 12th percentile, in his propensity to get his own two point shot. (Two point shots are the most telling since so few three point shots are unassisted, players with a good handle and a nice three stroke get artificially lowered in their shot creation when threes are included).  Rondo, on the other hand, rated highly, creating 77% of his own two point shots, putting him in the 73% percentile.

Bradley also rates low in getting assists.  Third from the bottom on that same list with 11.6%. Only Royal Ivey and the remains of Derek Fisher scored lower.  By contrast, Rondo led all NBA Point Guards with a 49.3% assist percentage, and Jordan Crawford, playing the two had a 19.4% assist rate in his time with the Celts last year, and higher in DC.

Ball handling is harder to measure directly.  Certainly the eye test says to most observers that Bradley looked uncomfortable as a ball handler last year, looking to rid himself of the ball as soon as possible to the nearest guy with a green jersey.  We can also probably look at the numbers above to make some inferences that Bradley wasn't comfortable with the ball in his hands

Bringing the shooting guards into the list; here are the Celtics guards in terms of getting creating their own shot, assists and turnovers, the signature stats for Point Guards.

	Pct Two Point Offense Assisted	Total Field Offense Assisted	AST Pct	TOV Pct
Avery Bradley-BOS-2013	49.68%	62.88	11.60	12.60
Courtney Lee-BOS-2013	52.13%	65.73	11.70	13.00
Jordan Crawford-BOS-2013	29.03%	46.64	19.40	15.50
Jordan Crawford-WAS-2013	20.93%	41.01	25.50	14.90
Keith Bogans-BKN-2013	54.18%	86.70	8.10	14.40
Marshon Brooks-BKN-2013	33.57%	39.99	14.60	15.50
Rajon Rondo-BOS-2013	23.22%	24.01	49.30	22.60
Phil Pressey- MO- 2013	4.72%	19.54	37.80	23.20

I included rookie Phil Pressey here, even though I do not have his shots assisted data from the NCAA as of now. (Update, per commenter Adrien, I have added Pressy's numbers from Hoop-Math). Apparently there was no else on the team with the slightest ability to throw a pass in Missouri last year.  Jordan Crawford's numbers look like a combo guard, he gets his own shots (not always to his advantage) and assists at a high rate with a higher assist percentage than turnovers.  Bradley, Lee and Bogans look like spot up Shooting Guards, not Point Guards, with Brooks playing the role of slashing Two Guard.

Last year the Celtics were able to pair Lee and Bradley as defensive pit bulls despite their ball handling and distribution weaknesses primarily because of Paul Pierce's ability to play Point Forward.  That's a luxury they won't have this year.  Some of this may sound like I am picking on Bradley's game, but I am not.  His defense is amazing, as is his effort.  He's just not a Point Guard, and playing him as one does him no favors.

As I wrote above, in addition to ball handling duties floor spacing and defense have to be considered in matching up a line up.  Offensive spacing does not just come from the back court, but the front court doesn't have any spacing specialists like a Kyle Korver or Dorrell Wright (unless they put Bogans in as a Small Forward).

 

Player- Team- Yr	Pct of Shots at Rim	Pct of Shots from Three	Three Point Percentage
Avery Bradley-BOS-2013	30.04%	26.46%	31.7%
Courtney Lee-BOS-2013	32.75%	30.59%	37.2%
Jordan Crawford-BOS-2013	26.46%	37.57%	33.7%
Jordan Crawford-WAS-2013	15.99%	34.62%	33.7%
Keith Bogans-BKN-2013	7.02%	86.67%	34.3%
Marshon Brooks-BKN-2013	35.60%	17.34%	27.3%
Rajon Rondo-BOS-2013	40.44%	11.11%	24%
Phil Pressey- MO- 2013		38.2%	32.4%

In terms of shot selection, Bogans is the only pure Stretch Guard taking nearly 90% of his shots from three, with Crawford rated on the borderline of being what my analysis considered a Stretch Guard.  In actual efficiency only Lee had a three point field goal percentage above the average for a Guard of 36.3% last year, though Bradley was down considerably coming off of the double shoulder surgery.  Defenses are likely lie well off both Rondo and Brooks outside the arc.

Putting it together in terms of combos, if we want to have one Shot Creator and one Stretch Guard on the court most of the time, our options look something like this without Rondo:

 

Creators	Stretch	Rim Attack	Defense
Crawford	Lee	Brooks	Bradley
Pressey	Crawford	Lee	Lee
Brooks	Bogans	Crawford	Bogans
Bradley	Bradley	Bradley	Pressey
Lee	Pressey	Pressey	Crawford
Bogans	Brooks	Bogans	Brooks

 

Not all of the those divisions are equidistant (Bradley and Lee's defense is probably significantly better than the next Guard available).  But then it's Steven's job to balance those trade offs setting his line ups for the first month.

If Your Point Guard Doesn't Create His Own Shots, He Might Not Really be a Point Guard

This the second in my Shot Creation series. The first post introduced my take on shot creation and started to look at creation at one position, Shooting Guard. This time I am looking at Point Guards.

For the record, I am only considering shot creation for one's self and not considering foul shots, since passes leading to foul shots aren't tracked in the traditional data (and the correlation to getting baskets one's own baskets from the floor and getting fouled on attempts with no set up is likely to be very high).

Last time out I found that shot creation is very position specific, it may be the item traditional position designations get the best. Because, one of the truisms of basketball analysis, along with the importance of the possession and the limited real estate near the hoop, is that there is only one ball on the court at a time. The Point Guard is traditionally the player on offense that handles the ball most often, and consequently has fewer of their shots assisted, as it is hard to catch a pass when you already have the ball.  Making assists is also heavily correlated to self-creation, since both require having the ball in your hands.

That said, there are, I think, some interesting differences in the ability and propensity of Point Guards to Create Shots.

Again I only used shots inside the arc because there is so little variation in assist rates on three point shots. The unassisted three is not really a thing. Including three point assists hides the ability of some players to create their offense and, in a sense, punishes them for being very good at hitting threes.

Like the shooting guards the cluster analysis found two distinct, groups. Unlike the Shooting Guards the groups were very imbalanced in terms of cluster size. Again I labeled them as Creator PGs and Spot Up Point Guards.

Here are some of the assisted stats for the Point Guard groups using three years of players:

Group	Count	Two Pt Shots Assisted	Total Floor Offense Assisted	Att at Rim per 48	Pct Assisted at Rim	Pct Assisted Long Two	Pct Assisted on Three Pointers
CREATOR PG	205	24.44%	40.47	4.46	26.10	27.61	71.86
Std. Deviation		0.08	12.47	2.02	9.77	14.21	17.16
SPOT UP PG	47	51.83%	65.91	2.69	50.20	57.51	80.73
Std. Deviation		0.09	12.62	1.91	18.77	15.68	23.18

 

The one quarter of Point Guards that were termed Spot Up PG's seem to play a different role in the offense compared to the rest of the Point Guards. They are assisted on twice as many of their two point baskets and get to the rim much less often. For reference the Creator Shooting Guards were on average assisted for 37.8% percent of their twos, while the Spot Up Shooting Guards were assisted 65.5% of the time.

It helps to focus on what the numbers mean when we look at some individual players. I pulled out a list with Point Guards that scored the most effective points in 2013.

Player	CREATOR TWO GUARDS	SPOT UP TWO GUARDS	Group
Andre Miller	1	0	Creator
Brandon Jennings	1	0	Creator
Brandon Knight	1	0	Creator
Chris Paul	1	0	Creator
Damian Lillard	1	0	Creator
Darren Collison	1	0	Creator
Deron Williams	1	0	Creator
George Hill	1	0	Creator
Goran Dragic	1	0	Creator
Greivis Vasquez	1	0	Creator
Isaiah Thomas	1	0	Creator
Jameer Nelson	1	0	Creator
Jarrett Jack	1	0	Creator
Jeff Teague	1	0	Creator
Jeremy Lin	1	0	Creator
John Wall	1	0	Creator
Jrue Holiday	1	0	Creator
Kemba Walker	1	0	Creator
Kyrie Irving	1	0	Creator
Luke Ridnour	0	1	Spot Up
Mike Conley	1	0	Creator
Monta Ellis	1	0	Creator
Nate Robinson	1	0	Creator
Randy Foye	0	1	Spot Up
Raymond Felton	1	0	Creator
Russell Westbrook	1	0	Creator
Stephen Curry	1	0	Creator
Tony Parker	1	0	Creator
Ty Lawson	1	0	Creator
Tyreke Evans	1	0	Creator
	28	2	30

 

The Creator Point Guards dominate this list with 28 out of the top thirty scoring PG's. The only exceptions are Luke Ridnour, who has generally been a legit Point Guard but shared the back court with Ricky Rubio for part of the year last year and Randy Foye, who according to 82 games played many more minutes as a small forward than at Point in Utah last year.

Since I was now fascinated by these statistical odd ducks, I pulled a few more of the Spot Up Point Guards. Some of the more prominent Spot Up Points are mostly old and/or slow guys like Derek Fisher, Chauncey Billups, Earl Watson or they were tweeners that have not found a true guard position, like Jimmer Freddette, Alek Burks and Tony Douglas.  (Note: The difference between combo guard and a tweener guard; a combo guard can play either guard spot, a tweener can't play either).  Also, included is Boston's Avery Bradley, who, I would argue, is simply tragically miscast as a Point Guard.

Looking at the stats, it is clear that the Spot Up Point Guards, as a group, at least, are an inferior option to Creator Points.  Below are the averages for True Shooting, Effective FG%, Offensive and Defensive rebounding percentages, assist  and turnover percentages and Advanced Statistical Plus Minus and Offensive Rating.  All of the red highlighted stats are the worse of the two, if you look close you may see a pattern.

Groups	Count	AVE TS%	AVE eFG%	AVE ORB%	AVE DRB%	AVE AST%	AVE of TOV%	AVE ASPM	AVE ORtg
Creators PG	65	52.41%	48.31%	2.16	9.16	28.05	16.54	-0.30	105.15
Spot Up PG	21	51.40%	47.46%	2.09	8.65	18.08	15.30	-2.24	102.14
Grand Total	86	52.16%	48.10%	2.15	9.04	25.61	16.23	-0.77	104.42

 

The Spot Up PG's scored worse than the Creator Shooting Guards in every area and worse than the Spot Up Shooting Guards in every except assist percentages.

I also cross referenced the Creator clusters with the Shot Location clusters I ran earlier.

Location Grp	Assist Grp	Count	AV TS%	AV eFG%	AV ORB%	AV DRB%	AV AST%	AV of ASPM
Stretch Guard	Creator PG	28	53.36%	49.95%	1.95	9.11	26.37	-0.49
	Spot Up PG	9	52.88%	49.28%	1.51	7.92	17.10	-2.08
Slasher Guards	Creators PG	37	51.69%	47.07%	2.33	9.21	29.32	-0.15
	Spot Up PG	12	50.30%	46.10%	2.52	9.19	18.82	-2.36
Grand Total		86	52.16%	48.10%	2.15	9.04	25.61	-0.77

 

This break out primarily confirms what I saw in the other one, Spot Up Points, generally, do not alook like a  good option.   Eventually, I would like to look at this with line up data to see if there is an independent, positive or negative, effect of using low creation Point Guard.

Next, though, I plan to look as Small Forwards and shot creation assuming I can find a stable cluster analysis (where players tend to stay in the same group from year to year).

Location Visualization

Below is the latest and greatest visualization for shot location, percentages based on percent of all shots taken at the rim or from three.. It defaults to show Wings, but you can change the filter for the Guards types and by year.  One can also see the Bigs, if you include everyone, but (spoiler alert!) they don't shoot many threes so there's less to see.

 

Learn About Tableau

Guard Location Viz

Below the break is a visualization of Guards' shot location broken out by my cluster analysis groups, Slashing Guards and Stretch Guards based on their percent of attempts at the rim versus from three...

Shot Creation and the Dog that Barked Rather Quietly

The term shot creation is both ill defined and much debated. I am not a big fan of the term myself, mainly because it's so ill defined. To those who are proponents of its importance, it is that certain je ne sais quoi that turns broken possessions into baskets and breaks defenses down. While opponents somewhat ungenerously term it chucking the ball in the general direction of the basket.

Trying to put my own definition on it I thought it would have to at least start with the ability and propensity to take shots without getting assisted. I also thought that the ability to get unassisted shots at the rim was a good indication of positive shot creation since those are both efficient shots and more likely to garner an offensive rebound if they don't go in the basket. Both are convenient definitions since I was staring at that very data broken out by distance from the basket from the last three years via Hoop Data.

I decided to again use cluster analysis to see if there were distinct groups present in the data that would at least point the way for further inquiry.

There was, at least, to some extent.

But, first let me say what where I didn't find it, because that's often almost as important for statistical inference and clarification of one's thoughts.

• Bigs aren't shot creators. No surprise, but that meant that for any Cluster to work beyond telling me who the Bigs were they needed to be eliminated.
• Self created three point shots are not really a thing.  Almost no one not named Steph Curry shoots threes unassisted. So the question of whether to include three point shot in looking at shot creation is a good question.  From my view it appeared that players with the ability and propensity to shoot off the bounce were labeled Non-creators or Spot Up guys simply because of their high rate of three point shots.  In other words, a guy that can hit threes and also drive the lane win a defender runs out is very valuable.  Also there is almost no correlation to unassisted threes and unassisted shots at the basket.  So, I based the study on two point shots only.
• Unassisted shots are very position dependent, because Point Guards have the ball in their hands much of the time. You can't get an assisted shot if the ball is already in your possession. Comparing Point Guard creation and Center's is pretty much pointless, like comparing strike outs thrown by pitchers and short stops.

So our world is getting a bit smaller here, but now we can at least look at some like groups, shooting guards, small forwards and point guards.  Within the positions there are, indeed, some cohesive groupings of players based on their assisted rate for shots. For now I am calling the two groups Spot Up Shooters and Creators.

For right now, I will look at Shooting Guards, here are the two groups using Shooting Guards over the last three years with at least 200 minutes in terms of assists by location:

 

Group	Count	Two Pt Shots Assisted	Total Floor Offense Assisted	Att at Rim per 48	Pct Assisted at Rim	Pct Assisted Long Two	Pct Assisted on Three Pointers
CREATOR TWO GUARDS	105	37.8%	54.54	4.25	40.95	39.36	81.43
Std. Deviation		0.10	13.43	1.89	12.27	19.85	17.08
SPOT UP TWO GUARDS	108	65.5%	75.90	3.33	63.61	73.55	91.14
Std. Deviation		0.08	8.81	1.74	11.86	14.56	11.77

 

What we find is that the shot location that shows the biggest difference in assist level is actually the dreaded Long Two from 16 to 23 feet. The bane of all basketball analytics everywhere. Not a a great start for our Shot Creators. However, they do , as group, take more shot at the rim.  And one can also see how little gap there is in three point shots.

To bring the groups I found into better focus, I pulled up some of the individual players with the most effective points in 2013.

Player	CREATOR TWO GUARDS	SPOT UP TWO GUARDS	Group
Andre Iguodala	0	1	Spot Up
Arron Afflalo	1	0	Creator
Ben Gordon	1	0	Creator
Danny Green	0	1	Spot Up
DeMar DeRozan	0	1	Spot Up
Dion Waiters	1	0	Creator
Dwyane Wade	1	0	Creator
Gerald Henderson	1	0	Creator
Gordon Hayward	0	1	Spot Up
J.J. Redick	0	1	Spot Up
J.R. Smith	1	0	Creator
Jamal Crawford	1	0	Creator
James Harden	1	0	Creator
Jared Dudley	0	1	Spot Up
Joe Johnson	1	0	Creator
Kevin Martin	0	1	Spot Up
Klay Thompson	0	1	Spot Up
Kobe Bryant	1	0	Creator
Lance Stephenson	1	0	Creator
Marcus Thornton	0	1	Spot Up
O.J. Mayo	1	0	Creator
Ray Allen	1	0	Creator
Vince Carter	1	0	Creator
Wesley Matthews	1	0	Creator
	15	9	24

One thing to notice is that Creator Two Guards are over represented in the top point scoring two guards with fifteen of the twenty-four.

For the most part the list looks pretty intuitive, with players like James Harden, Dwayne Wade, Kobe Bryant, and Joe Johnson as well as irrational confidence guys JR Smith and Jamal Crawford as Creators. But the list includes OJ Mayo and Ray Allen as a Creators.

This is where my definition matters, for Mayo 55% of his offense was insisted but only 37% of his inside the arc points were assisted. For Allen, while 76% of his offense from the floor was assisted only 49% of his two point offense is assisted putting him marginally in the creator camp last year. It also looks like the change to Miami affected his game as a far higher percentage of all of his shots in Boston were assisted and he rated a Spot Up Guard there.

On the Spot Guard list players like Danny Green, JJ Redick, Klay Thompson, Jared Dudley and Kevin Martin show up where we would expect. But, players like Iguodala were somewhat surprising, Again the way I am defining things matter, Iggy had about 62% of his overall offense assisted, while 55% of his offense inside the arc was a assisted, more than Allen.

Then I looked at some stats for the two groups as they were rated last year.

Groups	Count	AV TS%	AV eFG%	AV ORB%	AV DRB%	AV AST%	AV TOV%	AV ASPM	Av ORtg
Creators Twos	33	53.19%	49.35%	2.40	11.06	18.08	13.07	-0.64	104.70
Spot Up Twos	41	52.06%	49.10%	2.60	10.05	10.70	11.23	-1.53	103.56
Grand Total	74	52.56%	49.21%	2.51	10.50	13.99	12.05	-1.14	104.07

 

Pretty much across the line the Creators came out better.  That isn't a surprise when one looks at the list of Creators up above.  But, we can not, at this point, call this any sort of victory for shot creation.  It is, at least, as likely that superior two guards are given the ball more often in the offense.  IE, they get the opportunity to create because of their talent.  It could be that the offense would be better off with more ball movement and less, 'creation.'

Lastly, I also wanted to cross reference this analysis with my last analysis on Shot Location, expecting to see Shot Creators line up with Slashing Guards, and Spot Shooters with Stretch Guards.

And they did only to a small extent, there was only a slight bias in favor of Stretch Guards being in the Spot Up Group, because of  the fact that I gave the Stretch Shooters a pass on their threes. Again, using Shooting Guards from 2013:

 

Location Grp	Assist Grp	Count	AV TS%	AV eFG%	AV ORB%	AVE DRB%	AVE AST%	AV of ASPM
Stretch Guard	Creators	20	53.54%	49.97%	2.32	11.01	16.68	-0.78
	Spot Up	27	52.94%	50.12%	1.91	9.47	10.43	-1.41
Slasher Guards	Creators	13	52.66%	48.41%	2.53	11.14	20.24	-0.43
	Spot Up	14	50.34%	47.11%	3.94	11.17	11.23	-1.76
Grand Total		74	52.56%	49.21%	2.51	10.50	13.99	-1.14

 

Spot Up Slashing Guards come out the worst with scoring efficiency and overall ASPM rating.  No surprise, since they seem to have neither elite ball handling or shooting skills.

I will follow up with the Point Guards next.

 

Highly Plausible Win Projections

Over at Detroit Bad Boys, Ben Gulker asked last week if the Detroit Pistons can win fifty games.  The Wins Produced projection had them between 53 and 46 wins.  Gulker's post reminded me that I had done some initial projections myself.  I actually cribbed off of WoW-ster Arturo Galleti's minutes projections as a starting point.

But, in addition to WP, I added Myers' ASPM, an xRAPM projection by Nathan Walker and  Win Shares.  Then I normalized the projections to make sure that they all add up to 1230 wins.

The point of doing this is only in part to project records.  Projecting a team or the league's record is only partly related to the player metric, or whatever system you're using.  It also involves the playing time allotment, injuries and luck.  Sometimes everyone just has a career year at the same time. Using four metrics with the same playing time allotment helps me understand how the different metrics see different teams and players.  And how they view the game.

Plus, everyone else seems to do it.

Here's what I came up with:

Team	WP Adj	ASPM Adj	WS Adj	xRAPM Adj	Conf	Ave Wins	Low	High	Ave Rank
MIA	57	60	63	57	E	59	57	63	1.0
BRK	37	51	51	51	E	48	37	51	4.3
ATL	48	48	46	44	E	46	44	48	3.7
IND	40	48	43	47	E	44	40	48	5.0
DET	48	45	37	45	E	44	37	48	5.0
CHI	48	41	42	43	E	43	41	48	5.3
NYK	34	43	47	41	E	41	34	47	7.0
TOR	43	37	40	41	E	40	37	43	7.0
WAS	39	37	36	39	E	37	36	39	9.3
CLE	36	37	36	38	E	37	36	38	9.3
MIL	38	32	36	33	E	35	32	38	9.7
BOS	37	31	33	31	E	33	31	37	11.3
PHI	27	31	26	32	E	29	26	32	13.3
ORL	28	24	25	27	E	26	24	28	14.0
CHA	26	26	21	29	E	26	21	29	14.7
LAC	61	58	60	57	W	59	57	61	1.7
SAS	57	59	59	51	W	56	51	59	2.3
HOU	69	52	48	55	W	56	48	69	4.0
OKC	50	55	60	56	W	55	50	60	3.3
MEM	50	49	53	49	W	50	49	53	4.3
DAL	45	47	49	38	W	45	38	49	6.0
DEN	42	45	52	36	W	44	36	52	6.7
GSW	39	41	41	47	W	42	39	47	8.7
MIN	40	41	38	43	W	41	38	43	9.0
NOP	36	38	32	38	W	36	32	38	11.3
UTA	43	31	35	35	W	36	31	43	10.0
POR	34	35	32	38	W	35	32	38	11.7
LAL	27	32	34	31	W	31	27	34	12.3
SAC	28	29	31	33	W	30	28	33	13.7
PHO	23	27	25	26	W	25	23	27	15.0

 

Using Detroit as an example the range was between 37 and 48 wins, with 44 as the average.  That seems plausible!  Or Boston, the range is between 37 and 31 with an average of 33.  Also, plausible!

On the surprising side Dallas comes out ahead of Golden State, everywhere except xRAPM and everyone loves Brooklyn except Wins Produced.

I will likely update the projections after training camp in a slightly more rigorous manner.  Again, the everyone does it factor.

But, I am also going to do a wins over replacement estimate using the metrics to get a power and depth rating for each team based on the Wins over Replacement per 36 of their top seven players in terms of expected minutes and the depth rating on the rest of the team.  The idea being that the power and depth ratings would  give a somewhat more realistic idea of what we know about a team and, also, some idea which teams are able to withstand injuries.

No Threes for Us, We're Celtics (Keith Bogans Excepted)

I have been doing a series of studies on shot location in the NBA, breaking it down for guards, wings and big men. Naturally I could not help looking at how this applied to the Celtics.

Briefly, the series looks at shot location by position in the NBA with the help of a statistical technique called clustering, which looks for grouping patterns in data. (You can read about here if you are so inclined, but it's really not necessary).  The analysis gave a some nice intuitive results with 'Good' cohesion that showed that, yes, there are fairly distinct groups of Slasher Guards and Wings as well as Three Point Specialist Guards and Wings. For the front court players I found three groups, guys who only shoot at arms length from the rim, hybrids that add a mid range game and Stretch Bigs.

The study is primarily intended to look at big groups in terms of analysis. But, it is interesting to see how it applies to individual players and rosters plus it provides a nice test of reasonableness bfore I go further.  All that said and done this Celtics group is not a very stretchy roster, based on shot location last year. (For now I am ignoring the rookies).

The analysis rated only one of the current roster members as a 'Stretch' at their position, and that was Jeff Green because the HoopData.com data listed Green as a Power Forward.  Of course, this year Green will play more of his time at small forward, so I changed his position and reran the Cluster analysis where Green was rated as a Slash Wing.

For reference, the Slash Wings average 35% of their shots at the rim and 21% from three, while the Stretch Wings took 19% of their shots at the rim and 56% from three point land.

PlayerName	Pos	Pct Shots at Rim	Pct Shots 3-9 Ft	Pct Mid Range (10-15)	Pct Shots from Long Two (16-23)	Pct Shot from Three	Group
Jeff Green	SF/PF	38.61%	11.93%	5.23%	19.97%	24.26%	Slash Wing
Keith Bogans	SF	7.02%	1.75%	0.00%	4.56%	86.67%	Stretch Wing
Gerald Wallace	SF	45.43%	7.31%	.91%	9.13%	37.12%	Slash Wing

 

Crash is a slasher and that's a good thing given his poor three point shooting.  When Green is considered a small forward he also rates as a Slasher.  These numbers are based on his overall shooting tendencies, but looking at NBA WOWY Green shot a lower percent of his shots from three without Pierce on the court last year than with Pierce on the court, indicating he is likely remain a Slash Wing.

Hoop Data also listed Bogans as a small forward last year.  I kept him there because he may see more time there than in the crowded back court, especially if Wallace or Green miss any time.  In any case, Bogans is pretty much exclusively a three point shooter at this point.

Next let's look at the Guards. The break downs on the Guards were similar to the Wings.  Slashing Guards took 33% of their shots at the hoop and 23% from three, while Three Point Specialist took 19% at the rim and 49% from downtown.

PlayerName	Pos	Pct Shots at Rim	Pct Shots 3-9 Ft	Pct Mid Range (10-15)	Pct Shots from Long Two (16-23)	Pct Shot from Three	Group
Avery Bradley	PG	30.04%	4.71%	4.71%	34.08%	26.46%	Slashing Guard
Courtney Lee	SG	32.75%	6.27%	6.47%	23.92%	30.59%	Slashing Guard
Jordan Crawford	SG	26.46%	12.17%	8.99%	14.81%	37.57%	Slashing Guard
Marshon Brooks	SG	35.60%	20.43%	10.53%	16.10%	17.34%	Slashing Guard
Rajon Rondo	PG	40.44%	11.33%	6.44%	30.67%	11.11%	Slashing Guard

 

No secrets here.  The Celtics guards, as a group, are Slashers.  Jordan Crawford was the only Guard on the fence of Stretch Shot selection last year. Let that sink in...

In addition, only Courtney Lee  at 37.8% hit from three was above the league average for Guards of 36.3% per the NBA Geek.  A Rondo and Brooks back court would have spacing issues, to say the least, as they were justifiably shy shooting threes hitting only 24% and 27% respectively.  Bradley also struggled with his threes last year, though he had been much better the year before his shoulder surgeries.

Finally the bigs.  

PlayerName	Pos	Pct Shots at Rim	Pct Shots 3-9 Ft	Pct Mid Range (10-15)	Pct Shots from Long Two (16-23)	Pct Shot from Three	Group
Brandon Bass	PF	26.67%	8.83%	18.38%	45.77%	0.00%	Hybrid Big
Jared Sullinger	PF	49.55%	17.27%	5.91%	25.00%	2..27%	Hybrid Big
Jeff Green	SF/PF	38.61%	11.93%	5.23%	19.97%	24.26%	Slash Wing
Kris Humphries	PF	48.03%	15.46%	12.50%	23.36%	0.00%	Hybrid Big

 

Brandon Bass is interesting because he's close to being a Stretch Big without ever shooting from beyond the arc once. The Celtics also have rookies Kelly Olynyk and Vitor Faverani joining the front court this year, both of whom have some range, though I don't see Faverani as a true Stretch Big.  They also have the option to go small with Green or Wallace at the Four, though it's not an option I love, so forget I said it.

I don't really have the data at this point to indicate the actual importance of 'stretch' in offensive efficiency.  But, it looks as though the Celtics may be preparing to give it an interesting test, especially in the back court.  So, that's kind of cool, right?

Are Stretch Bigs A Good Idea?

Ok, I am not really going to answer that definitively. But, at first glance, it doesn't look it.

This is the third in my series on shot location in the NBA. The first two looked at guards and small forwards.  In this one I grouped Centers and Power Fowards together as 'bigs.' I used a Two Step K-Means cluster anaylsis to group Bigs based on their shot locations, without regard to shot making.

The idea of the 'stretch big' or stretch 4 is actually what first got me thinking about the project. Also the arguments over the value of guys like Tyson Chandler who almost never shoot from more than a few feet away from the basket and the ability of Omer Asik and Dwight Howard to play together.

The cluster analysis creates as many groups as seem to be present in the data, for both Guards and Small Forwards the analyses came up with two in each, one group centered around three point shooters and the other around shots at the rim. For both I named the two groups Slashers and Stretches.

For Bigs I came up with three groups that I am calling At the Rim Bigs, Hybrid Bigs, Stretch Bigs, at least until I think of better labels. This is the first analysis where mid-range shots and long twos were a significant factor in grouping players, because some of the bigs basically don't take any.

 

Here's how the clusters settled out in shooting location:

 

	COUNT	AVG Att at the Rim	AVG Att at 3-9 ft	AVG Att at Mid Range (10-15)	AVG Att at Long Two(16-23)	AVG Att of Three Pointer
At the Rim Big	35	67.31%	21.24%	3.58%	7.62%	0.36%
Hybrid Big	90	40.31%	16.47%	10.94%	29.88%	1.96%
Stretch Big	26	26.10%	9.32%	5.83%	19.36%	38.91%

 

The 'At the Rim Bigs' on average shot over 67% of their attempts within two feet of the basket, and over 87% within nine feet.  Don't bother guarding these guys at the elbow.

The best way to think of the Hybrids are as bigs with a little mid-range and long two game. The defining characteristic of the of the Stretch Big is the three ball.

Below I have some of the more prominent Bigs picked out by sorting for top ePts (effective field goal percentage times shots).  I had to add a couple at the end of the list just to get another At the Rim Big onto the list other  Dwight Howard.

 

PlayerName	Pct of Shots at RIm	Pct of Shots Lng 2	Pct Shots from 3	Group
LaMarcus Aldridge	24.12%	42.94%	1.18%	Hybrid Big
Blake Griffin	42.31%	26.15%	3.08%	Hybrid Big
David Lee	44.29%	25.00%	0.71%	Hybrid Big
Al Jefferson	27.33%	38.00%	1.33%	Hybrid Big
Brook Lopez	45.00%	20.71%	0.00%	Hybrid Big
Al Horford	37.14%	37.14%	0.71%	Hybrid Big
Ryan Anderson	22.31%	16.15%	53.08%	Stretch Big
Carlos Boozer	37.86%	26.43%	0.00%	Hybrid Big
DeMarcus Cousins	42.86%	28.57%	2.14%	Hybrid Big
Chris Bosh	31.67%	40.83%	8.33%	Hybrid Big
David West	27.69%	31.54%	2.31%	Hybrid Big
Tim Duncan	36.15%	35.38%	0.77%	Hybrid Big
Zach Randolph	43.85%	21.54%	2.31%	Hybrid Big
Dwight Howard	68.00%	4.00%	1.00%	At the Rim Big
Roy Hibbert	40.00%	14.55%	0.91%	Hybrid Big
Kevin Garnett	21.82%	48.18%	1.82%	Hybrid Big
Derrick Williams	38.00%	23.00%	26.00%	Stretch Big
Jeff Green	40.00%	21.11%	25.56%	Stretch Big
Tristan Thompson	50.00%	6.67%	0.00%	At the Rim Big
Dirk Nowitzki	10.00%	39.23%	22.31%	Hybrid Big
Nikola Pekovic	67.27%	3.64%	0.00%	At the Rim Big

 

Besides the fact that At the Rim Bigs don't, as a rule, seem to shoot much, a couple of things are kind of interesting. Two of the three stretch Bigs are pseudo-bigs, Jeff Green and Derrick Williams, both of whom should see a lot of time at the three this year.  Green because of the trade of Paul Pierce, and Williams because of the expected return of Kevin Love.

Also, Dirk just missed the Stretch category.  In fact, when the cluster was limited to two groups, he was the only player to switch groups becoming a Stretch with the Hybrid option eliminated.

Outside of volume of scoring the At the Rim Bigs dominate most of the efficiency stats.  They score more efficiently, in contrast to the Wings and Guards, where the Stretch groups scored more efficiently.  The At the Rim Bigs also rebound more and block more shots.  The Stretch Bigs lead in assist percentage and lowest turnover percentage.

Row Labels	Count	Av TS%	Av eFG%	Av ORB%	Av DRB%	Av AST%	Av BK%	Av TOV%
At the Rim Big	29	54.9%	53.53%	11.89	21.68	5.45	3.96	15.83
Hybrid Big	78	52.8%	49.12%	10.07	20.37	8.47	3.04	13.23
Stretch Big	24	52.4%	49.68%	5.95	16.76	8.65	1.77	11.17
Grand Total	131	53.2%	50.20%	9.72	20.00	7.84	3.01	13.43

 

The advanced one number player metrics for the most part value the At the Rim big guys more than the other Bigs groups.  The stretch Bigs are not well liked by the player metrics.  As with the guards, I am using Nathan Walker's xRAPM projection and a three year average Wins Produced per 48 from Arturo Galleti, while Win Shares and Daniel Myers' ASPM is a two year average weighted by minutes.

Row Labels	Count	Avg Proj xRAPM	Avg Wghtd 2 Yr ASPM	Avg WS/48	Avg WP 3 yrs
At the Rim Big	29	0.77	0.09	0.117	0.144
Hybrid Big	78	0.29	0.05	0.114	0.099
Stretch Big	24	-1.11	-1.42	0.080	0.058
Grand Total	131	0.14	-0.21	0.108	0.102

 

Since I combined the Centers and Power Forwards, it was worth it to look the two cardinal positions separately. There are some differences between the positions, especially the relative lack of Stretch Centers, which appears to be a good thing because the stats of the two I found weren't very good. The ratio of At the Rim and Hybrid Bigs is significantly closer.

The At the Rim Centers lead in most efficiency categories, including a gaudy 55.44 average effective field goal percentage.

Positon

Center

 

Row Labels	Count	Av TS%	Av eFG%	Av ORB%	Av DRB%	Av AST%	Av BLK%	Av TOV%
At the Rim Big	19	56.3%	55.44%	11.89	21.07	5.34	4.27	15.91
Hybrid Big	36	52.5%	48.98%	10.52	20.85	8.40	3.46	14.29
Stretch Big	2	53.4%	51.40%	5.10	17.00	12.55	1.45	14.90
Grand Total	57	53.8%	51.22%	10.78	20.79	7.53	3.66	14.85

 

The metrics all agree, Stretch Centers are a bad idea.  Though the fact that one of them is Byron Mullens doesn't help.  Nor does it that other, Boris Diaw isn't really a Center.

Position

Center

Row Labels	Count	Avg Proj xRAPM	Avg Wghtd 2 Yr ASPM	Avg WS/48	Avg WP 3 yrs
At the Rim Big	19	0.97	0.48	0.127	0.150
Hybrid Big	36	0.35	0.09	0.110	0.092
Stretch Big	2	-1.80	-2.02	0.065	0.010
Grand Total	57	0.48	0.14	0.114	0.108

 

The At the Rim power forwards are less dominant in scoring efficiency, but still dominate the boards and blocks.

Position

Power Forward

 

Row Labels	Count	Av TS%	Av eFG%	Av ORB%	Av DRB%	Av AST%	Av BK%	Av TOV%
At the Rim Big	10	52.1%	49.90%	11.90	22.84	5.65	3.38	15.67
Hybrid Big	42	53.1%	49.24%	9.69	19.96	8.53	2.68	12.33
Stretch Big	22	52.3%	49.53%	6.02	16.74	8.30	1.80	10.83
Grand Total	74	52.7%	49.41%	8.90	19.39	8.07	2.51	12.33

 

The metrics for the Power Fowards are very close between the Hybrid Bigs and At Rim Bigs, the Stretch Bigs still don't fare well:

Position

Power Forward

	Values
Row Labels	Count	Avg Proj xRAPM	Avg Wghtd 2 Yr ASPM	Avg WS/48	Avg WP 3 yrs
At the Rim Big	10	0.38	-0.64	0.098	0.134
Hybrid Big	42	0.24	0.01	0.117	0.106
Stretch Big	22	-1.05	-1.36	0.082	0.062
Grand Total	74	-0.12	-0.49	0.104	0.096

 

The next step, eventually, is to look into whether Stretch has an independent effect on offensive efficiency. The basic idea will be to look at line up Stretch against expected efficiency.

Two Ways to Wing It

In my last post I looked at shot selection and Guards.  This time I am going to look at Wings, AKA small forwards.

Again, I used a two step Cluster analysis based purely on shot locations without regard to how good the player was at making shots from any particular spot.  I only used players that played more than 150 minutes in the 2012-2013 season to limit complete noise.  At some point I will probably need to redo the cluster analysis with more years of data if I ever want to do more analysis with it.  But, like the Guard analysis this one gave me some interesting starter results.

Just like the Guards, Wings can be said to fall into clusters of players who try to get to the rim, Slash Wings, and three ball shooters, or what I am calling Stretch Wings. For the most part the cluster analysis output was pretty intuitive. Here are a couple prominent Slash Wings and Stretch Wings selected by top ePTS (Effective Field Goal percentage times shots).

PlayerName	Pct of Shots at Rim	Pct of Shots From 3	Cluster
LeBron James	40.59%	19.41%	Slash Wing
Kevin Durant	25.29%	24.12%	Slash Wing
Carmelo Anthony	29.55%	28.18%	Slash Wing
Paul Pierce	24.29%	35.71%	Slash Wing
Chandler Parsons	38.33%	43.33%	Stretch Wing
Thaddeus Young	50.00%	0.83%	Slash Wing
Danilo Gallinari	25.83%	42.50%	Stretch Wing
Nicolas Batum	20.91%	55.45%	Stretch Wing
Kyle Korver	3.75%	70.00%	Stretch Wing
Martell Webster	23.75%	53.75%	Stretch Wing
Jimmy Butler	40.00%	21.67%	Slash Wing

 

LeBron shot over 40% of his shots at the rim in the regular season according to HoopData, no wonder Popovich tried to pack the paint during the finals. Durant is also more of a Slasher than one might think, though clearly less than LeBron.  Durant took 25% of his attempts at the rim and 24% behind the arc.  By contrast Martell Webster, Kyle Korver and Batum shot over 50% of their attempts for three. Paul Pierce is probably the closest to switching clusters, which may actually happen next year as he gets a year older and is a little less able to convert at the rim.

The amount of 'stretch' appears to be pretty consistent between Guards and Wings.  Stretch Wings shot an average of over 50% of their shots from outside the arc, very close to the Three Point shooting Guards.  Slash Wings took 34% of their shots at the rim, just like the Slasher guards.

	COUNT	AVG Att at Rim	AVG Att at 3-9 Ft	AVG Att of Mid Range (10-15)	AVG Att of  Long Two (16-23)	AVG Att of Three Pointer	AVERAGE of eFG
	382	33.13%	12.15%	7.42%	23.30%	23.35%	48.12%
Slash Wing	51	34.97%	11.12%	7.70%	24.14%	21.35%	49.71%
Stretch Wing	31	19.45%	4.60%	3.18%	15.21%	56.03%	51.27%
Grand Total	459	32.50%	11.58%	7.20%	22.90%	25.11%	48.48%

 

There were also similarities in the non-shooting statistics the Stretch Wings had significantly better effective field goal percentages and better true shooting percentages, though difference isless dramatic because Slash Wings get to the line more often. The Slash Wings had better rebounding numbers particularly on the offensive end.

 

	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

 

As far as the advanced 1 number metrics go all of them liked the Slash Wings just a bit more, but overall they were fairly even. As with the guards, I am using Nathan Walker's xRAPM projection and a three year average Wins Produced per 48 from Arturo Galleti, while Win Shares and Daniel Meyer's ASPM is a two year average weighted by minutes.

	COUNT	Avg Proj xRAPM	Avg Wghtd 2 Yr ASPM	Avg WS/48	Avg WP 3 yrs
Slash Wing	39	0.17	0.07	0.102	0.109
Stretch Wing	25	-1.00	-0.70	0.090	0.105
Grand Total	64	-0.29	-0.23	0.097	0.108

 

Of course, the fact that both LeBron and Durant fall into the Slash Wing cluster helps the group scores on the player metrics.

I think both the Guards and the Wings look like there are probaby two fairly distinct styles of back court play.  One leaning on shooting ability and stretching defenses, the other more on raw athleticism.

Next I will write up my findings on the Bigs.  The Bigs are interesting since the Cluster analysis indicated three clusters, that I dubbed At the Rim, Hybrids, and Stretches.