Monday, February 2, 2015

Best & worst 5 man rotations: 2007/08 to 2014/15

After looking at top two man rotations, I was interested in understanding the best and worst five man rotations in the NBA. NBA stats collects rotational data (see an example here) going back to the 2007/08 season. From NBA stats, I pulled data on all five man rotations that played at least 100 minutes in a season together. For the strike shortened 2011-12 season and the current on-going 2014-15 season I prorated the number of minutes necessary to qualify based on a the games played in that season (so it was 80 minutes for 2011-12 and 56 minutes for 2014-15 when this data was pulled on January 30, 2015).

This left me with a population of 1,239 rotations over the last 8 seasons. I chose to measure the efficacy of a rotation based on offensive, defensive and net rating (basketball-reference.com provides excellent detail on these ratings in its glossary). Below I created a histogram offensive rating, defensive rating, and net rating to show how these 1,239 rotations performed on each.







The fact that each histogram looks like a bell where most rotations were clustered around a central score with smaller groups gradually declining from each side shows that each rating is relatively normal distributed. Below I calculated the mean, standard deviation (sd) and the mean +/- three sd. I calculated the +/- 3 sd to find those rotations that were exceptionally anomalous. Conventionally, any value that falls outside of 3 sd is considered a significant outlier, or in the context of this data an exceptionally efficient/ deficient rotation.

Offensive ratingDefensive ratingNet rating
Mean105.86102.972.8826
SD8.04137.519711.157
SD +3129.98125.5336.353
SD -381.73280.414-30.59

Below are the rotations that fell outside three sd (those outside four sd are highlighted in yellow) for offensive, defensive and net rating over the last 8 seasons. With respect to offensive rating this season we are witnessing both one of the best rotations in recent NBA history on the Raptors an exceptionally deficient rotation on the Pacers. In terms of defensive rating, a current Nuggets' rotation is outside four sd thus indicating that they are a historically poor defense. The best defensive rotations over the last 8 seasons came from the 11-12 Bulls and the 09-10 Thunder.

Finally, when looking at net rating, which simply takes into account offensive and defensive rating) two of the three rotations outside three sd are from this season. The Hawks have a rotation that is the best in the last 8 seasons and the Pacers have the worst with an 08-09 Kings' rotation not far behind. The 14-15 Pacers' historically low net rating is driven by their historically anemic offensive output.

Rotations with ratings falling 3 s.d. above or below the rating mean
**yellow highlighted cells reflect ratings that are 4 s.d. outside of mean
Offensive rating above (129.98)
Hansbrough,Tyler - Patterson,Patrick - Ross,Terrence - Vasquez,Greivis - Williams,Lou (14-15)135.1
Offensive rating below (81.732)
Hibbert,Roy - Hill,Solomon - Miles,CJ - Scola,Luis - Sloan,Donald (14-15)71.5
Davis,Glen - Duhon,Chris - Howard,Dwight - Redick,JJ - Richardson,Quentin (11-12)79.8
Carter-Williams,Michael - Mbah a Moute,Luc - Noel,Nerlens - Sims,Henry - Thompson,Hollis (14-15)80
Defensive rating above (125.53)
Brewer,Corey - Foye,Randy - Gomes,Ryan - Jaric,Marko - Jefferson,Al (07-08)126.5
Afflalo,Arron - Chandler,Wilson - Faried,Kenneth - Hickson,JJ - Lawson,Ty (14-15)134.2
Defensive rating below (80.414)
Asik,Omer - Deng,Luol - Gibson,Taj - Korver,Kyle - Lucas,John (11-12)77.5
Collison,Nick - Durant,Kevin - Harden,James - Ibaka,Serge - Maynor,Eric (09-10)79.3
Net rating above (36.353)
Horford,Al - Korver,Kyle - Schroder,Dennis - Scott,Mike - Sefolosha,Thabo (14-15)42
Net rating below (-30.59)
Hibbert,Roy - Hill,Solomon - Miles,CJ - Scola,Luis - Sloan,Donald (14-15)-38.6
Hawes,Spencer - Martin,Kevin - Salmons,John - Thompson,Jason - Udrih,Beno (08-09)-33.1

Friday, January 9, 2015

Best backcourt in the NBA?

There's been much talk about Steph Curry and Klay Thompson representing the best backcourt in the NBA in years (see here, here, here, etc.). I was interested in seeing how the two performed relative to their peers this season as well as the last few seasons to get some broader context of their success.

I pulled data on two-man lineup from NBA stats for the 2011/12, 2012/13, 2013,14 and 2014/15 (as of games played on January 6, 2015) seasons. As there are many combinations of two-man lineups I filtered the population by lineups that averaged playing approximately 20 minutes per game together during each seasons*.  This filter also enabled me to focus on high volume playing time lineups and eliminates the potential for anomalous performance. For each season this resulted in a population of 79 (2011/12), 100 (2012/13), 94 (2013/14), and 145 (2014/15) two player lineups.

Of this population of 418 lineups, I first focused on Net Rating, which subtracts Offensive Rating (points per 100 possessions) from Defensive Rating (points allowed per 100 possessions). The larger the Net Rating the better. The chart below shows the 20 best and 10 worst (for some context) two-man lineups in terms of Net Rating in the NBA over the last four seasons. Klay and Steph's Net Rating of 22.6 is third out of this population of 418 lineups and only trails (and is followed by) two other 2014-15 GSW two-man line-ups. We can interpret this as meaning that this seasons GSW starters dominate their opponents relative to the rest of the NBA over the last few seasons.


The reason for this exceptionally high Net Rating may surprise you. While the "Splash Brothers" are known for the shooting efficacy, they are actually only the 7th best two-man lineup in terms of Offensive Rating over the last four seasons (they trail 6 combinations of the 2014-15 LA Clippers starting lineup). The next chart shows the 20 best and 10 worst two-man lineups but this time just in terms of Defensive Rating over the last four seasons. Unlike Net Rating, with Defensive Rating the lower the number the better.



Again, Klay and Steph are the third best and only surpassed by two other 2014-15 GSW lineups (according to the Defensive Rating metric Kobe and Jeremy Lin combine to produce the worst defense in the NBA in recent memory, i.e., the last four seasons). So not only are Klay and Steph exceptional shooters but they're some of the better defenders in recent history (along with two GSW teammates).

What an aggregate measure, like Defensive Rating, misses are intervening factors (in this case teammates) that may contribute to Klay and Steph's defensive efficacy. In other words, their teammates (Green, Barnes) may be exceptional defenders and Steph and Klay may benefit statistically from the teammates being on the floor. (That being said, Klay can undoubtedly strap and Steph has elected to entertain the notion of there existing a defensive half of the basketball court this season so it is unlikely that their low/ great Defensive Rating is purely explained by omitted factors, like teammates' defensive efficacy).

In light of their defensive success, one cannot ignore Steph and Klay's offensive efficacy. In this last chart I plotted true shooting %, which is a measure that takes into account the value of free throws and three pointers in determining shooting efficacy, and pace, which captures the number of possessions per 48 minutes. These two are relevant measures of offensive efficacy because if a line-up shoots particularly well and they have a high number of possessions relative to the rest of the league then one would anticipate disproportionate offensive output.

In fact, of the 418 two-man lineups in this population, Steph and Klay were 9th in true shooting % (59.9%) and second in pace (102.69 possessions per 48 minutes). The chart below exhibits Steph and Klay's exceptional pace and shooting. In short--the hype is legitimate. Based on high frequency minute two-man line-ups over the last four NBA seasons, it is safe to argue that Steph Curry and Klay Thompson represent the best backcourt in the NBA.


*This amounted to 1,320 minutes in the strike-shortened 2011/12, 1640 minutes in 2012/13-2013/14, and 600 minutes in 201/15, which calculates to 20 minutes per game over 30 games (most teams had played about 34-35 games as of January 6). 

Tuesday, December 23, 2014

Special guest post: TERP ALERT!!!!

Like the Championship team of 2002, this year's Maryland Basketball team is big and deep.  The interesting thing is that it has the potential to be as talented.  Think about it, big, deep and possibly talented like our championship team. Wow.

Let's take a quick look at a specific player who will help make that possible. We have Melo Trimble, the pg county freshman who is playing like one of the best point guards in the country who can be an NBA guy. But I'm not talking about him. Or Dez Wells, the Naismith 50 finalist who might be an NBA guy. Or Jake Layman, the 6'9 crazy athlete with a nice jump shot who was co-Big Ten player of the week who can be an NBA guy. Or our 7 foot freshman,Michal Cekovsky, who is better now than Alex Len was at the same time in his career (Len was an NBA lottery pick).

Orrrrrr . . . Richaud Pack, the seasoned transfer playing his last year of eligibility and  silently guiding this team with defense and calming smarts. Or Dion Wiley, another Prince George's County freshman guard who can shoot, has size, and is confident. Or Evan Smotrycz, a 6'9 shooter....or Damonte Dodd, our starting 6'11 center who defends the rim...or 6'9 hustle guy Jonathan Graham.

I'M TALKING ABOUT JARED NICKENS.  The 6'7 freshman who's been starting at the three while Dez Wells has been hurt. I'm not saying he's better than Dez overall, but he handles the ball better and has a better stroke than him right now (note his Juan Dixony release). Here's a video from a weekend tournament last year against solid high school opposition over 3 games:



If you don't love him throwing the subtle jab at the west coast while talking about how it's "fun" to play against the best competition, then you don't love hoopin'.
While Dez is physically gifted, he gives occasional glimpses of season's past with a spazz out here and a wtf was that there.  Nickens is a glimpse of the future, filled with mega cool and specific to basketball talent.   

That's not to say that we shouldn't look forward to Dez returning.  He's a beast and we need him.   But the more I see Nickens either making 3's or almost making 3's coupled with his lack of turnovers and enjoying defense, I'm starting to wonder how bad of a thing to the program Dez's injury is.
And he was supposed to be our best player.

Whoa - Maryland is back in a big way.  

Go terps.

Special guest post by the luminary John Beckham

Monday, December 8, 2014

Relationship between advanced stats and winning

While a growth in advanced metrics provides a more robust picture of a player's value, the true measure of a metric should be it's ability to correlate with his team winning. I took six advanced metrics from basketball-reference.com for the 2014-15 season* and focused on only players that had played a minimum of 100 minutes this season through the December 5 games (this resulted in a population of 342 players). 

I compared each player's team's winning percentage (win %) with i) usage %, ii) player efficiency rating (PER), iii) true shooting % (TS%), iv) box plus-minus, v) value over replacement player (VORP) and vi) win shares per 48 minutes (WS/ 48). The definition for each are pulled mostly from basketball-reference.com's glossary. Below each definition the specific matrix is plotted on the vertical/ y axis and win % is plotted on the horizontal/ x axis. 

Each blue dot represents one of the 342 players that qualified. I included a yellow linear regression line to exhibit the direction of the relationship. If there is a very strong relationship between the metric and win % then the yellow line will begin at the bottom left of the chart and rise steeply to the top right of the chart, which would indicate that as the metric increases so too does win percentage. I include an R-square to indicate how well the model fits the data (or how well the metric explains win %). The lower the R-square means that the metric does a poorer job explaining win %.

Usage %Usage Percentage (available since the 1977-78 season in the NBA); the formula is 100 * ((FGA + 0.44 * FTA + TOV) * (Tm MP / 5)) / (MP * (Tm FGA + 0.44 * Tm FTA + Tm TOV)). Usage percentage is an estimate of the percentage of team plays used by a player while he was on the floor.



PER: A per minute rating that adds accomplishments and subtracts failures (see formula here).



TS%True Shooting Percentage; the formula is PTS / (2 * TSA). True shooting percentage is a measure of shooting efficiency that takes into account field goals, 3-point field goals, and free throws.



Box plus-minus: A box score estimate of the points per 100 possessions a player contributed above a league average player, translated to an average team.



VORP: A box score estimate of the points per 100 team possessions a player contributed above a replacement-level player, translated to an average team.



WS/ 48: An estimate of the number of wins contributed by a player per 48 minutes (league average is approximately .1000).


The above shows that while all six of these advanced metrics has a positive correlation with team win%, usage, PER and TS% are weakly related (they have relatively flat lines). Box plus-minus, VORP and WS/ 48 have a stronger positive relationship but the r squares indicate that they only explain ~15-18% of the relationship between the specific metric and team win %. However the includes a player's team's wins are included as a component in the formula for ws/ 48, thus the stronger positive correlation and larger R-square with ws/ 48 are expected. As a result, box plus-minus and VORP, which includes box plus-minus in its formula, should be considered the best metrics in terms of correlating with team win % that don't also include a team's wins in the formula (based on this small sample).

*A major limitation of this approach is that it focuses only on one incomplete season (2014/15). I will try to run a similar analysis on these metrics over several completed seasons, which should give a more accurate picture of the correlation with winning.