20 Second Timeout is the place to find the best analysis and commentary about the NBA.

Thursday, April 30, 2009

The Difference Between Measuring Defense in Basketball and Baseball

The April 6, 2009 issue of Sports Illustrated contains an article by Albert Chen titled Baseball's Next Top Models; Chen describes how baseball teams are using advanced statistics to ascertain which players are the best fielders at each position. The Tampa Bay Rays won the 2008 American League championship largely because they tremendously improved their defense by using advanced statistics as the basis for various personnel moves and for deciding how to most effectively deploy the players on their roster to maximize their defensive skills (for instance, they moved Akinori Iwamura from third base to second base not only because his defensive statistics are better at the latter position but also to make room for Evan Longoria to be called up as the new third baseman). Baseball statisticians have access to data that pinpoints where every single batted ball went and whether or not the fielder converted that play into an out. Although there are at least 10 players on a baseball field at any given time (one pitcher, eight fielders, one batter--assuming that there are no men on base), virtually everything that happens when the ball is in play can be broken down into a series of discrete, one on one actions: the pitcher throws the ball, the batter swings and, if he makes contact, a fielder attempts to catch the ball. Therefore, if one gathers together a large enough sample size of data, it is possible to create reliable models regarding pitchers, batters and fielders.

Obviously, basketball is a much more fluid and complex sport than baseball, at least in terms of constructing meaningful statistical models: even during an "isolation" play ostensbily involving only one ballhandler and one defender the other eight players on the court all have the potential to affect what will happen--the other four defenders may end up trapping and rotating, while the other four offensive players (depending on their size and skill sets) may be called upon to set a screen, cut to the hoop, spot up for an open jump shot or grab an offensive rebound. The play may result in an offensive rebound tip dunk or a made three pointer that never would have happened if the original ballhandler had not been talented enough to attract extra defensive attention but in the box score that original ballhandler may either receive credit for nothing (if he passes the ball and the recipient then swings it to a player who ultimately makes a three pointer) or he may even record a negative statistic (a missed field goal attempt) despite the fact that his actions directly led to the opening that created the putback opportunity.

Clearly, it is difficult for basketball statistics to fully capture what happens offensively; progress has been made in this regard but it is far from an exact science--and it is even more challenging to accurately measure basketball defense, particularly on an individual level. A perfect example of why individual basketball defense is tough to quantify took place in the first quarter of Boston's 106-104 game five overtime victory versus Chicago: Kendrick Perkins caught the ball on the left block versus Tyrus Thomas, spun baseline and scored a layup. TNT's Doug Collins noted that Thomas had positioned himself by Perkins' left shoulder (i.e., overplaying Perkins to force him to go to the baseline) because Perkins' best move from that spot is to go to the middle and shoot a jump hook; Thomas was supposed to receive help on the baseline--on an earlier play, help defender Derrick Rose stole the ball so easily from Perkins it looked like Rose was receiving a football handoff--but this time the help never arrived. How would a basketball "stat guru" evaluate that play in terms of Thomas' individual defense? Thomas' defensive rating would indicate that he allowed Perkins to score against him. Plus/minus data would award Perkins a +2 and Thomas a -2 and would also "indict" the other Bulls' defenders who were on the court at that time but would not reveal who was really at fault. A knowledgeable basketball observer would understand--as Collins immediately explained to the viewers--that Thomas did what he was supposed to do but that the help defender never arrived. Multiply this type of scenario over thousands of plays during the course of a season and it is easy to see why someone who watches basketball with understanding may come to a completely different conclusion about a player's value/skill set than someone who relies on nothing but numbers.

What about the success that Houston's General Manager Daryl Morey has had using advanced basketball statistics, as detailed in a New York Times article that I discussed here? If basketball statistical analysis is truly science and not pseudoscience, then it has to be based on the principles of the scientific method:

One "hypothesis" mentioned in the New York Times article is that Daryl Morey and his staff of numbers crunchers can use advanced basketball statistics to devise a game plan to slow down Kobe Bryant, the 2008 MVP and a two-time scoring champion. If we consider the four games that Bryant's Lakers played against Morey's Rockets this season to be the "experiment," then the "data" show not only that Bryant's Lakers won all four contests (with a convincing 13.0 ppg differential) but that Bryant averaged 28.3 ppg versus Houston while shooting .530 from the field and .533 from three point range, exceeding his overall regular season averages in all three categories; oddly, Bryant's free throw percentage versus Houston was only .680 (well below his .856 regular season average) but I doubt that even the most ardent advocates of basketball statistical analysis will claim that this is a result of Houston's "free throw defense." So, the results of this "experiment" show that advanced basketball statistics have yet to enable Houston to defend Bryant more successfully than other NBA teams--and this is despite the fact that the Rockets have two of the best one on one perimeter defenders in the NBA (Ron Artest and Shane Battier) plus a 7-6 shotblocking center (Yao Ming).

Don't think that I am picking on Morey or Houston; as I wrote in my PBN article cited above, I appreciate that Morey is fully aware of the current limitations of basketball statistical analysis:

It cannot be emphasized strongly enough that Morey is not merely looking at spreadsheets and randomly assigning arcane values to certain combinations of numbers; statistics give him an indication of what to look for when he watches game film but he still has to watch game film to determine why players are putting up the numbers they do and to figure out what exactly those numbers mean.

In other words, Morey appears to understand the limits of a purely mathematical approach to the game and thus uses numbers to confirm what his eyes tell him -- and vice versa. This is a completely different approach from the one taken by far too many stat gurus who are so enamored with their formulas that they dismiss the importance of actually watching games -- perhaps because they are in fact not truly capable of watching basketball games with any real understanding of what is happening on the court.

It is a laudable goal for basketball statisticians to strive to analyze the sport as effectively as baseball statisticians evaluate baseball but when "stat gurus" and their buddies in the writing business act as if basketball has already been "solved" from an analytical/statistical standpoint they are actually hurting their cause more than helping it, because intelligent observers can plainly see that such claims are false. As Cleveland General Manager Danny Ferry recently told me about basketball statistical analysis, "to just make decisions off of statistics would be a mistake but it can be an important part of the equation in basketball." It would be foolish for an NBA GM to not look at statistical data but it would be even more foolish for him to rely solely or even primarily on such data at this juncture; in the Perkins/Thomas example, it is much more useful for a GM or coach to know that Thomas did what he was assigned to do--and to find out which player missed the help assignment--than to get a spreadsheet filled with numbers detailing how many times Perkins scored in the post with Thomas as the primary defender, because without the proper context that data could be dangerously misleading if it influenced the GM or coach to make a negative evaluation of Thomas' defense.

Labels: , , , , , , ,

posted by David Friedman @ 4:48 AM


links to this post


At Thursday, April 30, 2009 12:32:00 PM, Blogger FreeCashFlow said...


The day someone devises a statistical system that accurately evaluates individual soccer players will be the day that I believe that someone can do the same for basketball.

At Thursday, April 30, 2009 5:15:00 PM, Blogger David Friedman said...

Free Cash Flow:

I don't know if it is possible to devise such a statistical system for basketball (or soccer) but it is important for people to understand the differences between baseball and basketball in this regard. The "sabermetric revolution" in baseball will be much more difficult to accomplish in basketball. If you read Morey's comments (or Danny Ferry's comments in my recent interview with him), it is obvious that NBA decision makers understand the limitations (and strengths) of basketball statistical analysis much more clearly than some "stat gurus" and many writers and fans do.

At Friday, May 01, 2009 12:46:00 PM, Anonymous Philippe said...

As a fan of basketball and of the subtle ways that coaches and players impact the game through both talent and execution, I agree with your general sentiment that statistics in basketball cannot explain or predict game outcomes on their own. However, your thought experiment in this blog post is not complete.
The hypothesis that you are experimenting with is that the statistical analysis that Morey employs and that Battier would seem to execute does not in fact lead to poorer performance from Byrant during head to head matchups. (If that is not the hypothesis that you are writing about than excuse the following analysis...)
Battier has been using this type of analysis for at least as long as he has been a member of the Rockets. This did not begin in the 08-09 season but in the 06-07 season. So, to take the full set of data over that time Bryant shot 43.5% overall and 38% from 3
vs an average of 46.3% and 35% over the same period of time. All other traditionals stats are worse for Bryant in the Rockets games vs the average of all games (less steals, assists, rebounds, and more TO's). He also shot the ball more in the games vs the Rockets, so he took up more offensive positions with an overall lower field goal %.
So my conclusion would be that Battier and whatever he learns from scouting and statistical reports has lead Bryant on average over the last 3 years into poorer in-game performance.

At Friday, May 01, 2009 3:25:00 PM, Blogger David Friedman said...


Even if we assume that Battier has been using such data as a tool in his matchups with Kobe since 06-07 and even if we assume that Battier was wholly responsible for Kobe's performance in the games that the teams have played against each other since that time, if nothing else it must be said that this season Kobe apparently made some kind of adjustment to nullify whatever Battier has been doing. In that regard, it will be very interesting to see what the Rockets/Battier come up with versus Kobe in a playoff series, because in a playoff series the value of preparation is heightened due to the length of time between games and the fact that you can focus completely on just one team.


Post a Comment

Links to this post:

Create a Link

<< Home