Wednesday, March 27, 2013

NBA Stats: Opponent FG%

The NBA's newish resource for statistical analysis, NBA Stats, illustrates why despite the attention that Moneyball,Sabermetrics et al garner baseball as sports' source of quantitative analysis, progress in the NBA's data collection techniques(and presentation) signal the league's appreciation for analysis. The facts that there are fewer players in basketball and that more of the game is quantifiable (especially now that spatial factors can be captured) will likely arouse greater interest in NBA analytics.

The main problem with NBA Stats is that they don't share the data so you're stuck with either using their standard reports or their limited filters on larger data sets. Of course, it makes sense that the NBA doesn't share the data because they are expensive to collect (and, thus, like most other organizations that collect vast amounts of data demand compensation for sharing) and because individuals could take the data and run their own analyses and draw attention away from NBA Stats, which the NBA presumably created simply to draw more traffic to NBA.com (as well as to build their brand with more quantitatively inclined consumers and fans).

A great set of individual defensive metrics that the NBA provides is on opponent's field goals (so how well all opponents shoot when being guarded by an individual player). It represents an intuitive measure of an individual's defensive success because it isolates individual defensive performance in a way that plus-minus or turnovers, which both largely reflect team defense, fail too. The metric is limited by the fact that a defender may still benefit from exceptional help defense, especially in the paint, and by the fact that some defenders' presence actually dissuades field goals from ever being taken in the first place. The data is organized in 5 foot increments so you can determine which player is a better defender closer to the basket (i.e., less than 5 feet) or on the perimeter (20 to 24 feet). 

I looked only at the 2012-13 season and focused only on defense close to the basket (less than 5 feet) and defense close to the three point line (20 to 24 feet). It's necessary to qualify based on some minimum number of FGA otherwise bench players who happened to be in when their man missed three of four close attempts appear to have the lowest opponent FG% (the lower the % suggests better defense). In fact, Terrell Harris, a 6'4" guard, has the lowest percentage of any player with opponents who have taken more than 10 FGAs (37 total for Terrell), which may reflect his superior interior defense but more likely suggests that Anthony Davis and co. help protect the net. Interestingly, of the top five defenders with opponent's that have taken at least 100 FGAs, four are on the Pacers--Roy Hibbert (48.8%), Orlando Johnson (49.4%), Gerald Green (49.6%), and Jeff Pendegraph (50.0%). (In addition, Pacers' C Miles Plumlee allows only 44.4% but he's only faced 36 FGAs.) Ronny Turiaf is the only non-Pacer with over 100 opponent FGAs within 5 feet, not on the Pacers, and allowing a FG% below fifty (49.6%). 

The Pacers' impressive interior defense likely reflects one of two scenarios. They're either the best coached and do an excellent job of executing sound interior defense, or one individual defender that gets a lot of playing time (i.e., Roy Hibbert) is such an imposing interior defender that his presence and help defense decrease not only his opponent's FG% but also his teammates' opponents' FG%. Here lies an issue with the NBA not sharing the data. To measure these two scenarios you would want to compare the other Pacers' opponents' FG% with Hibbert in the game and without him in the game. If the percentages hold in both scenarios then it suggests exceptional coaching and play. If his teammates' opponents' FG% increase without Hibbert in the game then Hibbert's value as an interior defensive player is immense.

For perimeter defense, we expect opponent FG% to be much lower as the shots 20 to 24 feet from the basket have a lower likelihood of going in than ones within 5 feet. In fact, 110 NBA players thus far in the 2012-13 season have held their opponents to under 35% FG on shots from 20 to 24 feet from the hoop. Reviewing the list of players that accomplish this low opponent FG% shows that many of them are power forwards and centers. Ideally (and NBA Stats does not let you filter by position in this table), we would want to look only at PG, SG and SF because presumably opposing PFs and Cs aren't accurate to begin with from 20 to 24 feet so low FG% may not reflect exceptional perimeter D, but rather that those PFs and Cs benefit from being able to guard poor perimeter shooters.

Three of the top five guards and SFs (not Ellington and Prince) with close to 100 opponent FGAs or more from 20 to 24 feet are surprising: Jordan Crawford (28.4%), Eric Maynor (28.8%), Wayne Ellington and Marquis Teague (both 30.7%), and Tayshaun Prince (32.2%). Interestingly, the worst perimeter defenders are dominated by rookies (Doron Lamb, Josh Selby, and Cory Joseph [second year but little playing time]) and lower-tier centers (Viacheslav Kravtsov, Joel Freeland, and Hamed Haddadi).

Thursday, March 14, 2013

Measuring Defense

As DBSF has exhausted much of his take on the peculiarities of the NBA and the styles and idiosyncrasies of certain players and their subcultures, it's time the blog goes in a different direction. Absent an Adam Morrison return to the NBA or JR Smith and Andray Blatche teaming up to lead a team to a first round, four loss, zero win playoff series in which each is suspended at least once for conduct detrimental to the team, this move will have been long over due.

The new focus will rely more on analytics, and for the foreseeable future the focus will be to identify the best defensive player in the NBA. This approach will demand more original analysis, which will also be more time consuming and result in less frequent posting. Finding the appropriate measure(s) of defense won't be accomplished in a single analysis and likely demands re-configuring and fine-tuning previous models. But--and depending upon the available data (this is key)--the goal is to establish some measure that in addition to a range of conventional defensive statistics as well as the other existing 'advanced' defensive statistics grants greater insight on a player's value. (Admittedly, this is no way a novel endeavor.)

But why defense? And, don't we already know who are the best defenders? To answer the first question, it's 50% of the game and gets maybe 5% of the attention on a good day. To answer the second, the NBA All-Defensive team is certainly a start. However we cannot be confident that All-Defensive nominees don't benefit from having superior defensive teammates and, thus, help defenders. Consider Dwight Howard at his prime. He was such a defensive presence in the paint that guards on the Magic could assume far greater risk on the perimeter and attempt to challenge more threes and perimeter shots because even if their man gets by them he still had to convert in the lane with Howard, arguably the best shot blocker in the last decade, contesting. In other words, Dwight Howard's aptitude as an interior, help defender might have made all of his teammates appear as better defenders.

Then what about conventional measures of defensive achievement, like steals or blocks? Both are undeniably important as one ends the opponent's offensive possession (steals) and the other potentially ends the possession or at least negates a potentially made field goal (blocks). But both demand the defender to assume risk, where undesirable outcomes include fouling or missing the steal and granting the opponent more direct access to the basket or missing the block and being out of position for a defensive rebound, which is a critical conventional measure of defensive aptitude because the defensive rebound absolutely ends the opponent's possession.

Take Allen Iverson, a three-time NBA steals leader, who was All-NBA first team 3 times, second team 3 times, and third team once but was never voted to an All-Defensive first or second team. There are likely many factors that explain this but consider Iverson's success in steals. While his great quickness would be an asset on perimeter defense, his 6'0" 165 pound frame becomes a liability the closer the offensive opponent draws him to the basket. As a result, it would be in Iverson's best interest to assume risk and attempt more perimeter steals (and, thus, increase his steal average) at the cost of sacrificing  interior defense where his physical size would prove ineffective.

So was Iverson a good defender or not? That's what the following posts will hopefully uncover. It may be that he was an excellent perimeter defender but a mediocre to poor defender inside the three point line--so we need to know where Iverson was on the court to answer if he was a good defender or not. Unfortunately, although NBA teams are collecting the spatial data necessary to conduct such analysis, they are not sharing it.

As a result, most "advanced" measures of defense reflect formulas that attempts to standardize one or more conventional defensive statistics, or they look at things, like the points allowed while the player was on the court. The latter is certainly an important indicator of all around defense however it is necessary that the measure accounts for the defensive aptitude of a player's teammates and the offensive ability of his opponents. (In other words, if a defensive player is stuck with four JR Smiths for teammates and is playing against the Heat and Thunder every night, then it's not fair to say he's a bad defender because his team gives up a lot of points.)

So what categories should define a defensive player? One was mentioned--defensive rebounds because it effectively ends a possession. Second, opponents' turnovers while the player was on the court. Many steals or poor offensive decisions (resulting in TOs) occur because of otherwise non-quantifiable defensive plays, like hustle, active hands, and quick help defense. Third--and this is likely the most difficult data to access--time left on the opponent's shot clock per their possession. This would have to be standardized by opposing team because some teams aim to shoot earlier in the shot clock than others. This measure would show which players (or teams) cause opponents to take more time to shoot than they do on average, which likely captures successful on-ball and help defense. All of this has to be standardized by minutes and position (and maybe size) as we wouldn't expect similar defensive rebounds from a center and a PG.

One limitation that's worth recognizing is that even if some useful measure of defense is determined it will only allow comparisons within cohorts of players. So, you could compare LeBron with Ibaka but you couldn't compare LeBron with Jordan. Besides the fact that rule changes over the years can affect a player's output (there is no better example of this than in the NFL where the restrictions on defensive backs have enabled QBs to shatter records in recent years), it would require comparing a line of cohorts to establish some inter-generational standard.

In other words, you'd have to do something, like compare a cohort of young players that played with Jordan (say Rasheed Wallace and Kevin Garnett) to a next level cohort (say Kenyon Martin and Tyson Chandler) and then compare how LeBron played against them while controlling for age and other factors. So you could say based on how LeBron plays against Tyson Chandler and based on how Chandler played against Rasheed Wallace and based on how Wallace played against Chris Webber and based on how Webber played against Dan Majerle and based on how Majerle played against Jordan we could expect LeBron and Jordan to match/ not match defensively? Obviously too much uncertainty (or, "noise") arises across generations so any attempts to standardize defensive performance will be relative to probably a 3-5 NBA season period.

Wednesday, March 6, 2013

Spencer Hawes--Harden'ing

The problem with when white bros go for the James Harden look--which Harden very well may have appropriated from some musician or entertainer, bear in mind that outside of HGTV the NBA serves as this blog's sole connection to pop culture--is that rather than look Urban Lincoln hip, white bros really just resemble a  rural terrorist. Take Spencer Hawes, his Harden beard makes gives him an air of Amish extremist. Like--no he does not like your values and yes he does hate technology with the exception of when it is used to create IEDs. The type of guy that is reaaaallllyyy holding out on believing that the President is an American citizen until he gets that real life birth certificate in his own personal hand. (But, heck, even then what with computers, Netscape and America On Line any hack can make a duplicate, right?) It's probably not to fun to be a rookie on the 76ers and on every road trip get coerced into boarding with the teammate that all the veterans say is a nephew of Ted Kaczysnki.