Counting the Baskets

Curse of the Mid Range Shooter

(And The Shooting Competence Curve)

Last October I unveiled my Scoring Versatility Index at Hickory High.  The idea at the time was to create an index that graded a player's efficiency scoring in a variety of ways including from different spots on the court and from the line, with the theory that a player efficient from more locations would be harder to guard.  

The immediate problem that I found was that as one carves up the court into smaller pieces one creates more data sets with smaller sample sizes, and the less reliable data.  It is a problem, as I have noted, with the shot chart era in general.  Carving up the court into pretty pictures can sometimes give us pretty unreliable impressions. Looking to account for this has been an interesting challenge and lead me to a few somewhat unexpected directions.

Players with only a few shots from any zone will have averages unreflective of their 'true' talent level, and, in fact, with three point shots the inherent high variability it takes a fairly large number of shots to determine true talent with any confidence.  

Statistics and a couple of assumptions can help us out here.  Over a given period players will shoot above or below their 'true' ability based solely on random variance (compounded, perhaps, by opponent variation, injuries, confidence levels and so on).  It's a tried and true assumption that after these streaks, over the longer term the player's shooting percentage will revert or regress toward their 'true' average, with reversion taking longer with a higher variability.

That leaves open the question of what the player's true shooting ability.  

My assumption also was that a player that rarely takes a three point shot is likely to be  a 'replacement' level three point shooter rather than an average one.  That's convenient for estimating the Versatility Index as crediting every player with average percentages with few if any shots would not help in measuring difference in scoring versatility.  When I looked at the data is also happened to be true.

Using the NBA.com shot zone data I charted each of the NBA's five shot zones with each player's shooting percentage against the number of shots taken in that zone. (The zones are the restricted area, paint outside of the circle, mid-range, corner three and above the break three).  For the three jump shooting zones in the mid range and the threes the pattern was pronounced, players who shoot more mid range jumpers, for example, make them at a higher rate.

The number of attempts from the mid range conveys two pieces of information, the amount of time the player is on the court and the degree of comfort the player and his coach have in him shooting in the mid range, ie whether that is one his spots.  Contrary to some opinions coaches have some idea who their better players are and players have some idea where they shoot better. The same is true of corner and above the break three point shots. 

Curses and Blessings

The problem, for mid range shooters, is that even for good mid-range shooters the shot isn't terribly efficient.  That can be seen in the image below along with the increasing expected competence with more attempts via a logarithmic trend line.  

Only the most efficient mid range shooters like Chris Bosh and Dirk Nowitzki approach league average efficiency  as measured by eFG% (the dotted line) from mid-range.  Players like Carmelo Anthony, DeMar DeRozen and LaMarcus Aldridge last year were fairly good (and liberal) with a not good shot.

Via Ian Levy's work on Xpps, the average point per shot for a mid-range attempt is .793, or an eFG% of 39.6%.  The competence curve on mid range shots last year indicates that players who attempted fifty to sixty mid-range shots were more likely to average near 36%, while players with just over 150 attempts the expected competence is closer to league average and high volume mid-range shooters the curve rises to only 42% to 43%.

In contrast, most of the more prolific shooters of the above the break on the three hit for above average efficiency.  And we also see the same increasing competence curve, leaving players on the right side of the image taking a allot of an efficient shot and being pretty good at it.

 

All of the most prolific shooters of above the break threes hit is with above league average efficiency, with only Kobe Bryant and Goran Dragic, last year, hanging around league average on those shots.  The curve indicates that a player taking 50 above the break threes as an expected eFG% of about 44% on those shots, higher than the more prolific and skilled mid range shooters.  The players higher on the competence curve shooting above the break threes show an expected eFG% of aproximately 58%.

Summing Up

To be clear, the train of causation here is primarily that players are taking the shots they are more comfortable, and skilled at, more often.  And that those shots are more likely to be within the flow of the offense than Chris Anderson's random above the break heaves or DeAndre Jordan's stray fifteen footer.

On the mid range shots, hitting a higher percentage is certainly better than shooting many and hitting fewer.  In addition, part of the impetus for the Scoring Versatiliy Index was the idea that being efficient from more spots on the court puts more pressure on the defense than a specialist like Kyle Korver or DeAndre Jordan.  

Currently two of the NBA's most efficient offenses, the Houston Rockets and Portland Trailblazers, featue very different ideas about spacing with the Rockets focusing entirely on three point shots and the Blazers featuring a mix that includes Aldridge's  mid range game in addition to highly efficient three point shooting.  So clearly an effective mid range game can be a part of a larger efficient scheme, if only to open the floor to other, better shots.

 

Do Transition Points Translate?

This post was originally published at Hickory-High on 1/17/2014

We know that buckets in transition before the defense gets set are the easiest points to score in basketball and that it is easier to score off of possessions that start with a live ball. Every new coach, for this reason, pledges to run an up-tempo brand of basketball as soon as they are hired.

It is also why "runs like a deer" is one of the more frequent cliches we hear about potential NBA draft picks. The ability to run the floor and get transition buckets is a valuable skill for any team and is deservedly coveted. However, when evaluating college players looking to move up to the NBA, it is important to consider whether the player will be able to convert those plays against a higher level of competition. With transition baskets the question is simply whether those same opportunities will be there as a pro, and whether they are reliant on the 'easier' baskets to up their efficiency and point totals.

I spent some time looking at that using data from the website Hoop-Math that splits scoring for each college player into transition and non-transition baskets with the effective field goal percentage (eFG) for both scenarios.  Hoop-Math creator Jeff Haley  defines transition field goal attempts as shots in the first ten seconds of the possession, not including second attempts. At this point Hoop-Math has three years of data making any definitive predictions about the carry-over impossible, but it was enough to find some indications. I matched the transition/non-transition Hoop-Math numbers with the rookie true shooting (TS%) numbers from NBA.com to look at which of the two gave a better indicator for  scoring efficiency in the NBA.

I came up with 65 players with enough minutes to compare and ran the numbers a few different ways, including by year, separating bigs and back-court players, and using a statistical method known as  'Boosting' to draw a number of random sub-samples for comparison to try to test the relationships and check for overfitting. In all methods the non-transition eFG% was a better indicator for NBA efficiency than transition eFG%.  The strength of the bias showed up particularly strongly for wings and guards, in fact, the percentage of a player's points generated in transition was a negative factor in predicting NBA efficiency. Here's the correlations and significance looking at the non-big men.

Wings and Guards	Correlation	Significance
NBA TS%	1
College non-trans efg	0.345	0.014
College trans efg	0.121	0.225
TOT EFG	0.321	0.02
trans_pts_ratio	-0.358	0.011

 

The numbers for the entire sample were similar, though the transition eFG% does become statistically significant.

There are a couple of reasons to think the non-transition eFG% should have a stronger relationship to efficiency at the next level  than transition efficiency. The first is simply sample size, the average player in the study had more than twice as many non-transition field goal attempts than in transition making non-transition the number with the least expected luck factor. Second, transition plays favor raw athleticism more than the half-court, where future NBA players are likely to have an advantage in college hoops that won't be there in the pros. Third, there is the question of whether the ability to get into transition in college translates to the NBA, ie. do those deer runners generate running offense or benefit from opportunities that will not be available in the pros?

An example is a player like advanced stat favorite, Will Barton, who produced above average efficiency numbers in college primarily because a disproportionate percentage of his total offense came in transition, while his non-transition eFG% was a below average 48.8%.  In Portland last year Barton was not able to produce score efficiently with their sub-par bench.

You may be aware that there's an NBA draft this June, as there is every summer. There is unusual interest and scrutiny on this class due to the combination of supposed talent available in this draft and the continued value the CBA gives to players on their rookie contract. Focusing on the top back court and wing players, because that is where I think splits might have the larger implications:

• Andrew Wiggins is under performing in non-transition scoring.  His overall eFG% is somewhat below average at 49%, and his non-transition eFG% is only 47%, well below average for a projected first round back-court player.
• Jabari Parker is currently looking better with a 53.4% eFG% in non-transition. He also relies less on transition points as a percentage of his scoring -- higher percentage of transition scoring showed up as a negative in my numbers, though I am not fully confident in that result.
• Jahii Carson and Gary Harris have the lowest non-transition numbers. Carson relies heavily on transition scoring to boost his percentages with 44% of his scoring coming in transition, well above the average for wings and guards.
• Rodney Hood is the most efficient of the possible first round back court players in non-transition with a 58.8% eFG, he's efficient in transition too, he just doesn't do a great deal else on the court.
• Zach LaVine is an interesting case, efficient in non-transition so far, 55.7% eFG, but relies on transition the most for his offense, scoring over half of his points in transition, nearly twice the average of other prospects.

Given the size of the data set we can only draw tenuous conclusions, but I would take look twice at players like Carson that have difficulty scoring when the defense is set. I am really eager to hear what other observers think, are we under-valuing degree of difficulty or is getting in transition a skill for wing players and guard?

Slight Shot Chart Caveat

I was just revisiting Sabermatrician Consultant Sandy Weill's NBA shot model, thanks a Jacob Frankel piece on Hickory High earlier this week.

In short Weill used both the SportVu video tracking technology and the NBA's play-by-play data on 6,700 shots to create a model for expected shot value using a number of different variables, including shot type, the distance from the basket, defender proximity, and catch and shoot situations.

A couple of side notes from the model stood out to me as worthy of highlighting.  One of the effects that the data did not show any value for was angle to the basket.  In other words, straight on, right  side, or left side did not matter, nor was there a difference between 45 degree or 90 degree shots.  

That is really interesting information in the shot chart era.  Shot charts are fantastic tools to convey a huge amount of information very quickly and elegantly.  However, the more we slice and dice the court into pieces the smaller the sample size gets for each little piece, and inherently the less reliable the data.  

For example, looking to update at my Scoring Versatility Index, I calculated that, based on the variation in three point shooting, it takes approximately 175 three point attempts to reach the player's 'true' shooting percentage (or at least the point where the sample has the same variation as the population).  When we look at shot charts that divide three point shots into five zones including the two corners and three above the break three zones it is difficult to tell if the different colors right to left represent actual variation in a player's ability to hit shots or pretty random noise.

For example, here we see that Chandler Parson's is much better from the right side of the arc and particularly worse from the top of the key in his 2013-2014 shot chart via the great Vorped site.

Chandler Parson's Shooting 2012-2013

Or, maybe not.  Looking at Chandler's shot chart from the previous 2012-2013 season, his worst location was from the right side and his best above the break location was from straight on the basket.  The exact opposite of the current year.  Maybe there is a plausible causation, for example, Parson's could be getting the ball in a different way or through different plays.  But, really we should be cautious going beyond random variation without significant evidence.

 Chandler Parson's Shooting 2013-2014

On the other hand, we do know that corner threes are, league wide, more likely to go down.   And the Sabermatrician model gives us a pretty clear, if unsurprising, reasons why, it is shorter and more likely to be off of a catch and shoot,  two factors that would account for the most of the advantage in field goal percentage for the corner three.

The shot model and write up is below the break.

Measuring Turnovers via Touches Using SportVu

This post originally appeared on Hickory-High.com on 1/2/2014

There is no controversy that turnovers are bad. They are probably the pet peeve of every coach, as by definition, it's a wasted possession, even more than a very badly chosen shot.

There is also a widespread recognition that turnovers by a player need to be scaled to the opportunity to commit turnovers. Just as scoring scoring twenty points on twenty-five possessions is not as valuable as scoring eighteen points on fourteen possessions, a starting point guard with four turnovers in thirty-two minutes is less egregious than a back up center letting the only four passes near him go through his hands.

The two most popular ways to scale turnovers are assist to turnover ratio and turnover percentage. Assist to Turnover ratio is the number of assists a player is credited with divided by the number of turnovers the player is charged with committing. Players who make more assists generally have the ball in their hands more and errant passes are one source of turnovers.

Turnover percentage is generally considered the most advanced way to scale turnovers and is calculated as percentage of the possessions a player 'finishes.' The formula is effectively turnovers as a percentage of shots and trips to the free throw line. SportVU Player Tracking data gives us another way to scale turnovers, with player touches of the basketball data. The SportVU touch data tracks and counts every time a player touches the ball as well as the time that he has possession of the ball. I matched the SportVu data from the NBA website with the turnover count from their standard data for each player then calculated the number of turnovers each player committed per 100 touches. Of players with over 150 minutes of playing time, Utah rookie Rudy Gobert tops both the turn over per 100 touches  list and the TOV% list.  The top twenty offenders are listed below, with those also appearing in the top twenty based on TOV% highlighted in green.

Player	TOV/100 Touches
Rudy Gobert	7.59
Kobe Bryant	7.43
Kent Bazemore	7.04
Archie Goodwin	6.8
Ryan Hollins	6.69
Jermaine O'Neal	6.64
Dion Waiters	6.39
Anthony Randolph	6.34
Rudy Gay	6.29
Elliot Williams	6.19
DeMarcus Cousins	6.03
Nazr Mohammed	6.02
Reggie Evans	5.96
Kevin Seraphin	5.92
Timofey Mozgov	5.8
Jonas Jerebko	5.74
Dwight Howard	5.73
James Harden	5.65
Michael Kidd-Gilchrist	5.62
Amar'e Stoudemire	5.56
Gerald Wallace	5.52

 

The Falsely Accused

Comparing the different measures overall, TOV% and TOV per 100 touches correlate pretty well with an R^2 of around 0.50 for players with over 150 minutes of playing time. However, there are what look like systematic differences. The turnovers by low-usage players are highlighted while gunners' turnovers get de-emphasized.

The table below has the players with the biggest fall in turn overs when switching from TOV% to TOV per 100 touches.  Both measures were standardized based on the standard deviation, then the standardized measures were compared.  Players that do not shoot often, many of them big men, get tagged with a higher turnover rate using TOV%.

Player	Std TOV/100 Touches	Std TOV Pct	TOV %- TOV per Touch Diff
Reggie Evans	0.86	2.13	1.26
Kendrick Perkins	0.52	1.7	1.18
Jan Vesely	0.13	1.29	1.17
Jeff Ayres	0.03	1.02	0.98
Lou Amundson	0.4	1.31	0.9
Omer Asik	0.19	1.06	0.88
Brandon Rush	0.41	1.26	0.85
Garrett Temple	0.23	1.06	0.83
Gustavo Ayon	-0.63	0.14	0.77
Nick Calathes	0.19	0.89	0.71
Gerald Wallace	0.68	1.35	0.66
Daniel Orton	0.59	1.25	0.66
Nick Collison	-0.43	0.21	0.64
Diante Garrett	0.01	0.65	0.64
Cody Zeller	-0.25	0.39	0.64
Rudy Gobert	1.53	2.15	0.62
Alonzo Gee	-0.12	0.49	0.61
Ian Mahinmi	0.37	0.97	0.6
Ekpe Udoh	-0.15	0.44	0.59
DeAndre Jordan	-0.27	0.31	0.59
Pablo Prigioni	-0.88	-0.3	0.58

  Kendrick Perkins by TOV/100 touches turns the ball over at a slightly above average rate, but because of his low usage rate they are magnified to 25% of his possessions, 1.7 standard deviations above average. Players like Jan Vesely and Omer Asik go from approximately average fumblers to significant ones. While players like Nick Collison and DeAndre Jordan go from below average offenders per touch to above average turn overs using TOV%.

Getting Away With It

On the other side of the equation some players with a quick trigger finger may have had their turn over propensities ignored to some extent. The list of players with the largest increase in their turnover propensity is mostly high usage players including old stats friend Rudy Gay at third on the list. Gay turns over the ball on 13.2% of the possessions he finishes, but because he finishes so many possessions by throwing the ball toward the hoop that may credit him with better ball handling than he has. He turns the ball over 6.3 times per 100 touches, well above the league average of 3.8.

Player	Std TOV/100 Touches	Std TOV Pct	TOV %- TOV per Touch Diff
Klay Thompson	0.54	-0.46	-0.99
Dion Waiters	1.04	0.09	-0.95
Rudy Gay	1	0.05	-0.95
DeMarcus Cousins	0.89	-0.02	-0.92
Eric Gordon	0.47	-0.38	-0.85
Kevin Durant	0.51	-0.31	-0.82
Nick Young	0.13	-0.6	-0.73
James Harden	0.74	0.03	-0.71
Jamal Crawford	0.28	-0.38	-0.67
Brook Lopez	0.1	-0.56	-0.67
CJ Miles	0.22	-0.44	-0.66
Luol Deng	0.32	-0.33	-0.65
Stephen Curry	0.53	-0.1	-0.64
Dwyane Wade	0.63	-0.01	-0.63
DeMar DeRozan	0.14	-0.48	-0.62
Elliot Williams	0.96	0.34	-0.62
Andrew Bynum	0.34	-0.28	-0.62
Evan Turner	0.63	0.06	-0.57
Nikola Pekovic	0.07	-0.5	-0.57
Tyreke Evans	0.21	-0.35	-0.56
O.J. Mayo	0.56	0.01	-0.55

  Klay Thompson actually shows the largest increase in turn over propensity, followed by high usage players like Dion Waiters, Gay, and DeMarcus Cousins.

Summing Up

Briefly, assist to turnover ratio doesn't come out looking great as measure judging by TOV /100 touches. The two are not well correlated.  Because assists are so position related, most of the players that look better by Assist/TO are point guards, lead by Chris Paul, and most of the players that look worse by it are front court players, lead by Rudy Gobert (who doesn't need anyone to make his turnovers look worse) and Ryan Hollins. Because of the positional bias assist to turnover ratio is probably only helpful comparing players at the same position or role.

Turnover percentage will, and probably, should still continue to be used as the advanced turn over metric given that it measures a player's tendencies for turn overs fairly well and we have over thirty years of data using that measure for any league where basketball is played. That gives us a good base for both analysis and bar stool discussions, but having turnovers per touch gives us a new level of testing especially for players with very different offensive roles.

*All data via NBA.com as of 12-30-2013

Turnovers and the Celtics (and Gerald Wallace)

The Celtics offense is sputtering right now, and has been the chief culprit in giving up a couple of big leads in the last few games.  Among the problems all year have been turnovers, the Celtics are third highest in the league in turnovers per possession with 16.9 per 100 possessions, according to NBA.com. 

Over at Hickory-High.com, I wrote an analysis piece comparing different methods of tracking and quantifying player turnovers.   Specifically I was interested in comparing turnover rates as measured by assist to turnover ratio, turn over percentage (TOV%), and using 'touches' of the ball as captured by the NBA's SportVu cameras.

Given their turnover issues, the Celtics make a good case study for looking at player turn over tendencies using the various measure and where we might want to point the finger of blame.

On a league wide basis, the average TOV% was 12.8%, the average Assist/TOV ratio was 1.41, and turn overs per 100 touches was 3.84.

Below are the turn over stats for the Celtics players:

 

Player	MIN	Season TOV	TOV/100 Touches	TORatio	AST/TO	TOV/Touch Percentile	TORatio Percentile Rank	Asst/TOV Ratio
Avery Bradley	930	52	3.58	10.56	0.74	51%	40%	24%
Brandon Bass	889	33	2.5	8.4	1.33	18%	16%	54%
Courtney Lee	448	24	3.93	10.77	1.13	61%	42%	44%
Gerald Wallace	680	57	5.52	22.69	1.3	90%	95%	53%
Jared Sullinger	753	45	3.63	9.76	1.07	53%	30%	42%
Jeff Green	1006	59	4.11	10.78	0.79	66%	43%	26%
Jordan Crawford	912	63	2.82	10.5	2.59	26%	39%	91%
Kelly Olynyk	395	39	5.27	18.41	0.79	88%	88%	26%
Phil Pressey	323	17	2.23	11.54	3.24	13%	52%	97%
Vitor Faverani	418	34	4.37	18.53	0.38	73%	89%	7%

 Jordan Crawford has committed the most turn overs this year (as 12/30), slightly ahead of Jeff Green.  But as the starting point guard for the vast majority of the year, Crawford has had far more opportunities to turn the ball over, while Green has lead the team in minutes.

Turning to the more advanced measures, Crawford is actually arguably the most resposible ball handler on the team, followed by Phil Pressey.  As noted by the green highlights above, Crawford turns the ball over less per 100 touches, less as a percentage of possessions he finishes as captured by TOV%, and has better assist to turnover ratio than most of the league.  Pressey is only rated as average using TOV%, which is due to his lack of field goal attempts (which is a seperate problem).

Brandon Bass also has taken care of the ball fairly well, only performing average on assists/TOV, the measure I consider the weakest as it is so position dependent.

Gerald Wallace, Vitor Faverani and Kelly Olynyk are a different story, as well as Courtney Lee and Green to a lesser extent.  All of them turn the ball over significantly more than league average per 100 touches, with Wallace and Olynyk near the 90th percentile for highest turn over rates.  Wallace fairs even worse with traditional TOV% due to his low usage this year.

Green and Lee , on the other hand look much worse using TOV/100 touches than TOV%, as they look to score more often.

The other issue with turnovers as a statistic is that all turn overs are not equal.  Looking at Play-by-play data from the website NBAwowy, I found in a study that shots after steals are the highest effective field goal percentage shot at 62%.  Shots following non-steal turn overs are  much closer to a shot after an opponent's made basket or free throw, at 46%.

Again using NBAWowy data, for the Celtics it looks like Wallace, not only has the highest turn over rate, but his are disprportionately the most damaging kind. As shown in the table below:

Player	% of TOV's Steals
Avery Bradley	51.8
Brandon Bass	48.7
Courtney Lee	44
Gerald Wallace	65.6
Jared Sullinger	27.1
Jeff Green	40.3
Jordan Crawford	63.4
Kelly Olynyk	37.5
Kris Humphries	57.9
Phil Pressey	84
Vitor Faverani	47