View Full Version : a pretty good piece on what wins championship

11-26-2003, 05:26 PM
Moving Beyond "Defense Wins Championships"
By Kevin Pelton
Nov 26, 2003, 13:30

Since the dawn of NBA statistical analysis, whenever, alas, that was, a popular topic has been the relative value of offense and defense. It is a subject of some general debate amongst basketball fans, mostly spurred on by that ubiquitous but unattributed quote, "Defense wins championships". As well, it has the advantage of being easy to conceptualize and measure. Since 1974, when the NBA began tracking team turnovers, we have had all the data necessary to answer questions about offense and defense.

The problem, in my opinion, is that the discussion has not moved beyond asking the question, "Does defense actually win championships?" Maybe it has, but if so, I have yet to see that study. These studies, looking at the relative offensive and defensive quality of NBA champions (as John Hollinger did in Pro Basketball Prospectus) or whether the superior offensive or defensive team wins in playoff matchups (as Dean Oliver did in Basketball on Paper), have certainly provided some important insights. To the extent that either offense or defense appears to be more important in these studies, offense shows an advantage. However, the primary conclusion these studies reach is not a new one -- both offense and defense are important. Teams that win championships must excel in both categories. (To be fair, Hollinger also looked at the performance of teams who led the league in either offense or defense.)

For two-thirds of the teams in the NBA each season, whether offense or defense wins championships is not of great concern. What I'm interested in looking at is whether the Golden State Warriors, or the L.A. Clippers, or any other non-contender can derive a significant advantage by focusing on either offense or defense. Does either appear to be more important in general, and, if so, is it undervalued in the NBA market?

To answer this question, the first step is to determine how to quantify offensive and defensive prowess. As I imagine readers of this column would know, points per game is inadequate for this purpose. The reason for this is that team pace goes a long ways toward determining per-game averages.

The best example for how pace can confound things last year was provided by the Detroit Pistons and the New Jersey Nets, the East's top two regular-season teams and the conference's finalists. The matchup was correctly perceived as a matchup of a defensive-minded team against a balanced one, with one problem. The Pistons were supposedly the offensively-challenged defenders, based in large part on New Jersey outscoring Detroit 95.4 to 91.4 in terms of points per games. Unnoticed in that assessment, however, was that New Jersey averaged far more possessions per minute than Detroit, 2.211 against 2.062.

This brings up an important point, which is that we have to define possessions. On this count, I differ from Oliver. For him, a possession only ends when the ball changes hands. To me, a possession is ended by any kind of field goal attempt, a last free throw, or a turnover. Functionally, the difference is that an offensive rebound starts a new possession for me, but it doesn't for Oliver (in whose parlance what I call a possession is a "play").

Based on this definition, the formula for an offensive rating (or a defensive one) is PTS/(FGA+(.44*FTA)+TO). I've probably used these ratings in this column without defining them, and the functional form is largely the same is for true shooting percentage, which I regularly use, but it's useful to formally define the ratings before using them.

Returning to my Detroit-New Jersey example, using per-(100) possession ratings instead of per-game ratings makes a big difference.

Team PPG PAPG Off Def
Detroit 91.4 87.7 91.3 87.7
New Jersey 95.4 90.1 89.5 85.8
In fact, it was New Jersey that was the better defensive team, third in the NBA behind Sacramento and San Antonio by this measure. Detroit was a very good defensive team, but not a great one, while the team that supposedly couldn't score was actually one of the NBA's ten best offenses on a per-possession basis (the Nets were a below-average offensive team).

The next step is to introduce a measure of how "biased" an NBA team is towards offense or defense. This requires us to compare a team to league average for that season. Naturally, the league average for offense is the same as for defense; they're flip sides of the same coin. Thus, the formula is (Off. Rat. - League Avg.) + (Def. Rat. - League Avg.). A positive mark means the team is more offensive-minded, and the magnitude reflects how imbalanced they are.

Returning to our Detroit-New Jersey example, the Pistons were -0.9, the Nets -4.4. Both teams were defensive-biased, but New Jersey was much more even than were the Pistons.

We can use this method to evaluate championship teams. My team database currently goes back to the 1989-90 season, so I'll show the champions going that far back.

Year Team Bias
2003 San Antonio -1.9
2002 L.A. Lakers 4.2
2001 L.A. Lakers 3.7
2000 L.A. Lakers -3.1
1999 San Antonio -5.3
1998 Chicago -4.1
1997 Chicago -0.5
1996 Chicago -1.0
1995 Houston 2.2
1994 Houston -2.2
1993 Chicago 0.7
1992 Chicago 3.3
1991 Chicago 3.8
1990 Detroit -3.3
Here is some evidence, though weak, that defense indeed wins championships. Eight of 14 teams are defensive-biased, and the teams average a bias of -0.25. It's not much, but it's something.

As I said at the start of this column, Iím interested in looking at the teams that werenít elite. My first way of doing this will be to separate the most extreme teams, either offensively- or defensively-biased, and compare their performances.

Team Year Win% Bias Team Year Win% Bias
Dallas 2002 0.695 10.3 Golden State 1999 0.420 -7.5
Milwaukee 2003 0.512 9.8 Philadelphia 2000 0.598 -7.6
Phoenix 1995 0.720 8.9 Vancouver 1996 0.183 -8.0
Indiana 1991 0.500 8.1 Chicago 2000 0.207 -8.2
Charlotte 1996 0.500 8.0 New Jersey 1991 0.317 -8.3
Utah 1998 0.756 8.0 New Jersey 1990 0.207 -8.4
Indiana 1999 0.660 7.6 New York 1993 0.732 -8.5
Denver 1999 0.280 7.4 New Jersey 1996 0.366 -9.0
Utah 1995 0.732 7.3 Golden State 1998 0.232 -10.1
Indiana 1990 0.512 7.3 Denver 2003 0.207 -11.6
Average 0.587 Average 0.347
Quite striking results, arenít they? Even the most extreme teams arenít necessarily what weíre interested in, however, so letís extend the results a little longer (this time, weíll avoid listing the teams).

Offense Defense
Top 10 0.587 0.347
Top 25 0.588 0.421
Top 50 0.583 0.475
Top 100 0.552 0.476
All 0.525 0.476
Wow. Honestly, I didnít see that coming at all. I was pretty confident the extremes would show the difference I did, but I had no idea it would extend as far as 100 teams in either direction. Or, for that matter, all the way Ė a 50-point difference in winning percentage is certainly something to ignore. I will deliver my scheduled caveat, which is this: The extreme offense teams were planned, the extreme defense teams generally accidental. In other words, the Mavericks went around trying to get a whole bunch of shooters, the Bucks went around trying to keep their bunch of shooters. The Nuggets werenít aiming for good defense last season per se, they were just trying to do the best they could while aiming for this season. (Philadelphia and New York are the notable exceptions on the defensive list.) Why are such horrid offensive teams able to play at least average defense, when we donít see the same on offense?

The answer, in my opinion, is that most people believe defense is less tied to talent than offense. A bunch of hard-working, smart players may not be able to produce a league average offense, but, with the right coach, they can produce an average defense. Therefore, the fact that the extreme teams were so much worse isnít necessarily meaningful. The same, however, cannot be said of the nearly 200 teams each that make up the entire offensive- and defensive-biased samples.

If thatís not enough, Iíve got more evidence for you. The next exhibit is the correlation between winning percentage and bias. Correlation, for those of you who didnít suffer through statistics in college, measures the strength of a relationship between two things on a scale from Ė1 to +1. The correlation between bias and winning percentage is +0.206, which is not very strong but is still meaningful in that there is a correlation and it is positive.

Wait, thereís more! We can use multiple regression analysis to compare the importance of offense and defense. Regression is a slightly more complicated version of correlation, where not only is the relationship strength analyzed, but also the magnitude of change. We can also look at several different variables to see their combined effects. What Iím using are the four primary measures I use to evaluate teams Ė offensive ratings, defensive ratings, offensive rebound percentage, and defensive rebound percentage. I standardized each of those relative to the league that season so that we can compare the relative importance of each category.

What I found was this. Combined, these four factors explain about 92% in the variation in winning percentage (also standardized). This should come as no surprise, as they basically make up point differential, which is an extremely strong predictor of team performance, and perhaps a better measure of a teamís ability. Improving an offense by one standard deviation improves the teamís winning percentage by 0.654 standard deviations. By comparison, one standard deviation of defensive improvement only improves winning percentage by 0.460 standard deviations. (For the sake of posterity, the same improvement in offensive rebounding percentage produces a 0.204 deviation improvement, in defensive rebounding percentage a 0.158 deviation improvement.) This method shows offense to be about 40% more important than defense.

The last argument I can make is this: Year in and year out, the standard deviations for offense are significantly higher than those for defense. What does this mean? My interpretation is that offense, more than defense, determines how a matchup between two teams goes. You know the baseball saying, ďGood pitching beats good hittingĒ? Thatís basically saying a similar thing. When a good offensive team and a good defensive one meet up Ė in theory Ė the offenseís rating should be higher than the value weíd expect strictly looking at the two ratings. Is that important? Of course it is, because it means a bad defense doesnít hurt you as much as a bad offense, or vice versa. This is a theory that I would like at a later date to test using actual games.

In summary, virtually all the available evidence suggests that even if offense doesnít win championships, itís much more effective at winning regular-season games, which is what most teams are primarily concerned with. This has hugely important implications. A team that is essentially neutral, has even offenses and defenses, can in theory improve itself by trading a defensive-minded player for an offensive-minded one of equal value. There are limits, of course, to the value of offense. The Mavericks, for example, would be better-served by the defensive specialist. In general, however, putting a premium, if a slight one, on offense makes sense.

© Copyright 2003 by HOOPSWORLD.com, a Basketball News Services Exclusive

11-27-2003, 12:33 AM
Interesting piece though it is mistitled. He lends a credible argument as to what aspect of the game is more valuable in regular season games for non elite teams but the only research he does regarding playoff games or chamipnship games for elite teams shows a slight bias towards the other end. I'd also like to see him apply this analysis to teams in the 70's and 80's.

11-27-2003, 02:27 PM
Yeah, I thought that was a cool piece. As for the defense wins championships bit, 8 out of 14 is way to small a difference, and way to small a sample size to draw any conclusions. If anything, that stat seems to indicate that whether you have an offensive or defensive bias is rather inconsequential, though it is worth noting that the list is relatively free of extreme bias scores, which would seem to indicate that while it's okay to hang your hat more often on either offense or defense, it has not usually been the case that teams were so dominant in one or the other respect that they could ignore the weaker aspect and still win a championship.