Offensive/Defensive Weight Part 1: Correlations Between Orientation and Success
By: Ian Geertsen
(The mid three-peat 2000-2001 Lakers are the most offensively weighted team to win a championship in recent history—the image above is from the Lakers’ 2001 finals vs the Iverson-led 76ers.)
We’ve all heard the sayings before: “offense wins games but defense wins championships,” or something in that vein. Today, we will see if there are any truths to these adages. Looking back at data from the last four decades, parts one and two of this piece will explore the relationship between teams’ offensive/defensive weight and their success. It is unlikely, however, that we find any large correlations; it’s in the sport’s nature that different teams will win through different avenues and with different focuses. Still, I believe this analysis reveals some interesting insights concerning offensive and defensive weightedness, both historically and looking forward.
Offensive and Defensive rating data is available going back to the 1983-1984 season, although this raw data itself can only tell us so much. The league has changed a lot over the last 30+ years; playing styles, rules, and officiating have all been altered dramatically. This means that we can not really compare a defensive rating of 100 if the 80s to a defensive rating of 100 now. But what we also know—and what is more valuable to us—is how teams performed relative to the rest of the league. So you had an offensive rating of 111.82 in 1984. What was the league average? What was the standard deviation? Tell me that you had the best offensive rating of any team in the 1983-1984 season, though, and now I know something about your team. As a point of comparison, the league’s average offensive rating in February of 2021 is 111.8.
Offensive/defensive weight is based around this concept; by determining how a team’s offense and defense rank compared to their peers, we can more objectively compare teams across different eras. First is still the first, whether in 1989 or 2019. The calculations for determining a team’s weight are pretty simple: take their offense rating rank for that season and subtract their defensive rating rank. Let’s look at last season as an example: the reigning champion Los Angeles Lakers finished 10th in regular season offensive rating, although they finished 3rd in their regular season defensive rating rank, giving them an offensive/defensive weight of 7. By using this system, we know that receiving a positive offensive/defensive weight value means you have a defensively weighted team, as their defense performed better relative to the league than their offense On the flip side, any negative weight value means you have an offensively weighted team. This also tells us the magnitude of their weighting as well; an offensive/defensive weight of -2 means that their offense was just slightly better relative to the league than their defense, while a weight of -12 means that their offense was much, much better than their defense. It is important to note, however, that this metric does not measure performance in any way. If a team did have a defensive weight of -12, they could theoretically be the best or the worst team in the league that year.
Let’s look at the 2019-2020 season as an example. In this season, the correlation coefficient between wins and offensive/defensive weight was 0.1022, while the correlation coefficient between SRS and offensive/defensive weight was 0.0425. SRS, or the Simple Ratings System, is a way of quantifying team success over a season using margin of victory and strength of schedule data; it makes team comparisons easy, as a team with an SRS of seven is expected to be five points better than a team with an SRS of two. Here is what that relationship looks like visualized—keep in mind that negative weight values are offensively weighted teams, and positive weight values are defensively weighted.
One of the first things to stand out is how much further the offensive side of the plot extends than the defensive side. This is largely due to the unique performance of the Portland Trailblazers—finishing 3rd in offense and 27th in defense, the Blazers are uniquely skewed towards offense. While not so extreme, the Mavericks (-18), Spurs (-15), and Wizards (-13) were also heavily geared towards offense over defense. This season did not see such extreme results among defensively weighted teams, as the team with the largest offensive/defensive weight, the Orlando Magic, only had a value of 14. While the correlation between offensive/defensive weight and success is small, it is positive, weakly suggesting that both win totals and SRS may be positively correlated with defensive weightedness. This means that, as teams go from offensively to defensively weighted, their wins and their SRS go up slightly. Or, thinking about it the other way, as teams improve in their win totals and SRS they also become more defensively weighted. Had these correlation coefficients been negative, we could have seen the inverse relationship, where wins and SRS would be correlated with offensively weighted teams.
Still in the 19-20 season, we can also compute that offensively weighted teams averaged 39.29 wins, while defensively weighted teams averaged 40.96. This data remains in line with the patterns we see above, as it shows that defensively weighted teams on average performed over a win and a half better than offensively weighted ones. On the other hand, the average offensive/defensive weight of playoff teams in 2019-2020 was -1.313, meaning that the average playoff team is slightly offensively weighted. Using calculations like these, we will examine the relationship between offensive/defensive weight and success from both a generalized and historical lens. As we can see from this example, the correlations we see will often be small, which will only become more true as we increase the sample size; that is not necessarily a bad thing, however. This piece is not meant to be definitive by nature, or to answer any specific question. If you think that my brain, my laptop, and basketball reference are going to answer the question “what is more important, offense or defense?” I’m afraid you might be disappointed. Still, I know we can glean some interesting and valuable information by looking at this data. So let’s get started!
While we want to use the tools we have to look at the relationship between offensive/defensive weight and success, it is important to address some confounding variables in our story. The most obvious of these variables is very simple—how have the rules of the game changed? There have been some undeniably significant rules changes to the game of basketball over the years, although the effects of these changes can be hard to discern. In the 88-89 season the number of refs officiating each game increased from 2 to 3. This kind of change clearly has an impact on the court, but in this context that impact isn’t easy to find. You might think that with more refs on the court more fouls would have been called, invariably making things a little bit easier for the offense. In reality, though, less fouls per game were called in the 88-89 season than the preceding one, and in the 89-90 season even fewer whistles were blown. In the 92-93 season the shot clock rules were changed, so that the clock would only be reset if the ball made contact with the rim, and not the rim or the backboard. This clearly gives a slight advantage to the defense, but just how big is this advantage? While we know a rule change like this can only make ripples so big, it is still impossible to say. Some rules changes made more than just ripples, though, the first and foremost of which being zone and illegal defense. Let’s take a look at the illegal defense rules instituted in the 81-82 season:
A. Weakside defenders may be in defensive position within the outside lane [16'] with no time limit but within the inside lane [12'] for no longer than 2.9 seconds.
B. Defender is allowed within inside lane as long as he is closely guarding a player adjacent to the 3- second lane.
C. Player without the ball may not be double-teamed from weakside.
D. Any offensive player may be double-teamed from strong side.
E. Offensive player above foul line and inside circle must be played by a defender inside dotted line.
F. If offensive player is above the top of the circle, defender must come to a position above foul line or remain in either outside lane.
G. Defender on cutter must follow the cutter, switch, or double-team the ball.
The first use of illegal defense would result in a shot clock reset, although any subsequent instances of illegal defense would result in a free throw and possession for the offense. In 96-97, illegal defense was eliminated in the backcourt, and in the 99-00 season, illegal defense was no longer applied to defenders on the strong side (the side with the ball) of the court. In 01-02, illegal defense rules were removed in their entirety. Getting rid of illegal defense in 2001 gave an undeniable edge to defenses moving forward, allowing players to defend with more freedom and coaches to scheme with more creativity. Take Nick Nurse’s famous Giannis wall or his box-and-one on Curry in the 2019 finals; both of these defenses would have been considered illegal in the 80s and 90s. On the other hand, the controversial rules introduced in the 04-05 season cut back on hand-checking and introduced the defensive three seconds rule. These rules were infamously designed to “open up the game,” and while they were far from the most popular rule changes, open up the game they did. From the 03-04 to the 04-05 seasons, the league average offensive rating spiked 3.2 points, going from 102.9 to 106.1, giving league offenses a clear boost.
Changes in rules and officiating of this magnitude have effects that are clearly present but impossible to perfectly measure. As we look back in history, it is important to keep in mind just how much the league has changed over the last forty years, as this will inform the lens with which we view historical league trends.
Looking at every team from every season since 1983-1984, the correlation coefficient between wins and offensive/defensive weight was -0.0379, while the correlation coefficient between SRS and offensive/defensive weight was -0.0447. This lack of correlation might not be that surprising in this setting—one wouldn’t expect much correlation when looking at every team that has played since the 83-84 season—but if you are anticipating any massive correlations or huge insights, I would put a damper on those expectations. Given the nature of this sport and of sports in general, there are always going to be teams that find success through offense and teams that find success through defense. Obviously there is no “correct way” when it comes to designing an offensively—or defensively—primed roster; still, looking at this data allows us to see the relationship between offensive/defensive weight and success, observe how this relationship has changed over time, and possibly even predict what this relationship will become in the future.
As we can see, the plot showing offensive/defensive weight and SRS is a bit cleaner and more uniform than the one showing the relationship between offensive/defensive weight and wins. This is to be expected, as using SRS tends to remove some of the randomness and luck that makes its way into win totals every season. Both graphs also demonstrate a clear oval effect, where the most and least successful teams tend to have weights at or around zero. This also makes sense, as the worst of the worst teams are likely to be bad on both offense and defense, while the best of the best will likely dominate both sides of the floor. In the shortened 2011-2012 season, the Charlotte Bobcats won just seven of their 66 games—adjusted to an 82 game season, that would be good for a whopping 8.69 wins. As you might imagine, this team ended up with an offensive/defensive weight of 0 after ending up 30th in both offensive and defensive rating. On the flip side, the 1995-1996 Chicago Bulls also finished with a weight of 0 after finishing first in both offense and defense.
This next graph visualizes the correlation coefficients of offensive/defensive weight and wins or SRS by decade, letting us observe how trends in the league changed over time. First, it is important to note that the ‘80s’ decade is incomplete, as there is no data for the 80-81, 81-82, and 82-83 seasons. Also, the 19-20 season is included in the ‘10s’ decade, but data from the current 20-21 season was left out.
While the overall correlation coefficient between offensive/defensive weight and success was very minimal—just -0.0379 for wins and -0.0447 for SRS—this does not hold true when broken up into decades. This chart shows that the eighties saw the largest correlation between offensive weight and success, while the aughts (or the 2000s) saw the largest correlation between defensive weight and success, though at a small rate. We will see if this pattern holds true when looking at the rest of our data.
Before we move on, though, a quick aside. The data on correlation presented above, and throughout the rest of the analysis, uses the Pearson correlation coefficient not the R2 coefficient of determination; this may seem unideal, as the R2 value provides a percentage variation in the dependent variable which is explained by the independent variable. I chose to use the R correlation coefficient instead of the R2 coefficient for two reasons: for one, the correlation values we are working with are very small, and using R2 would make them even harder to work with. More importantly, however, is the fact that directionality is key in this analysis, as a negative correlation coefficient means there is a greater correlation between offense and success than defense, and visa versa. We would not be able to observe the directionality if we squared every correlation value, hence the use of the correlation coefficient.
Correlation: Playoff Teams
While useful, the data above begs the question: “does all of this data really matter?” What is most important is the relationship between offensive/defensive weight and success at a high level; if there is a correlation between wins and weight among the worst teams in the league, that information might not be very valuable to us, and might end up muddling the rest of the data. One way we can limit our data to only include higher performing teams is by just looking at playoff teams. Among playoff teams since the 1983-1984 season, the correlation coefficient between wins and offensive/defensive weight was -0.0784, while the correlation coefficient between SRS and offensive/defensive weight was -0.0729. Again, the correlation is slightly negative, although this time both correlations are slightly stronger. The correlation between offensive/defensive weight and wins actually more than doubled when taking away non-playoff teams, showing that removing poor performing teams only exaggerates the patterns which were already present.
These graphs only show teams who earned a playoff berth so unsurprisingly the shape of our graphs has morphed from an oval to more of a triangle, as we essentially cut out a large swath of the low win and low SRS teams. The fact that the negative correlation already exhibited only became more pronounced shows that the correlation between offensive weight and team success should increase as you observe higher levels of play, although we will look further into this later.
The correlation between offensive/defensive weight and success over time follows a similar pattern when only including playoff teams as when including all teams. There are, however, a few glaring changes. For one, the correlation is negative in each of the four decades, including the 2000s which had a positive correlation when observing all NBA teams. The magnitude of the negative correlation also increased in the 80s and 10s, although this magnitude did decrease in the 90s. As we saw with the correlation spanning every decade, only focusing on playoff teams tends to make the negative correlation between offensive/defensive weight and success even stronger.
This graph appears to show quite a bit of evidence linking offensive weight to success. It is important to note, though, that even the largest correlation shown here—being the negative correlation between weight and wins in the 80s—is barely stronger than 0.2 and is still very minimal. There always have been great offensively weighted teams and great defensively weighted teams, and nothing will ever change that. This data is relevant not necessarily because it is significant, but because it can reveal directionality and patterns that haven’t been explored before.
Correlation: Championship Teams
When our data is only limited to recent league champions, we start to run into a problem that we have yet to encounter: the problem of sample size. Instead of dealing with hundreds of data points we are now given under forty; the patterns we see are still relevant, but this is an important thing to remember. When looking at the data, championship winners since the 83-84 season saw a correlation coefficient between wins and offensive/defensive weight of -0.0276, while the correlation coefficient between SRS and offensive/defensive weight was 0.0436.
Only observing championship caliber teams gives us our two least negative/most positive correlation values, suggesting that winning at the highest echelon might take some defense after all. I will admit, however, that this way of measuring the data is inherently flawed. We are looking at the relationship between wins or SRS and offensive/defensive weight, but all of these teams already won a championship, so does it really matter how many regular season wins they earned? Does winning more regular season games or having a higher SRS necessarily make you a better team if we are talking about teams that won a championship? I still see this as a valid tool of measurement, as the greatest of the greats will still have stand out performances—ie 72 wins on an 11.8 SRS for the 95-96 Bulls—but it begs an interesting philosophical question about the game.
Once again, sample size presents itself as a problem in this data. Split up into four groups, each decade has only ten teams to serve as the data, not nearly enough to call for any significant findings. This in part has led the magnitudes of the correlations to seem extreme when compared to the visualizations including playoff teams or all other teams.
It is interesting to note that in the 80s, the decade which had the most extreme correlation between offensive weight and success, now shows the reverse to be true. Michael Jordan and the Bulls may be responsible for the negative SRS correlation seen in the 90s, as they put up a few strong seasons with offensive-leaning teams. It is also interesting to note the opposing directionality of the 10s wins correlation and SRS correlation, likely another product of the suboptimal sample size.
Looking at the data we’ve seen from a bird’s eye view, it does seem that there may be a very weak correlation between winning and offensive weight. This correlation, however, is beyond insignificant; what is more interesting is observing how the league has changed over time. The 80s exhibited a still weak correlation between offensive/defensive weight and success, a correlation which diminished in the 90s, diminished further and even reversed in the 00s, only to become stronger again in the 10s.
Additionally, the offensive/defensive weight metric has some flaws when dealing with the highest and lowest caliber teams. Because the system relies on comparing your rank relative to the league, this rating system is not very effective at comparing offense to defense when a team has both an offense and defense playing at an all time level. Take the 95-96 Bulls, a team I’ve referenced multiple times already: they finished first in offense and defense, giving them an offensive/defensive weight of 0. Clearly they were very effective on both sides of the floor, but because they were so dominant this system does not register the difference in their dominance, and ends up rating their offense and defense equally when this may not really be the case. The same applies to the worst teams in league history as well, although this is obviously not nearly as relevant or important.
It is also interesting to note how the correlation between offensive/defensive weight and success tends to follow the same pattern as the pace—or an estimate of the number of possessions per 48 minutes—of the league. In the eighties, the correlation was its most negative while the average pace was the highest. The correlation became weaker in the 90s as the pace rose, and when the pace of play was the lowest in the 00s the correlation was its most positive. This makes sense, as the high pace game of the 80s was likely conducive to more offensive success, while the slower paced 00s were characterized by isolation offense, low scoring games, and more defensively-focused play. It is worth remembering, though, how weak these correlations are; it is possible that this pattern could simply be a result of noise in the data.
We will continue to look at this same data—but from a few different perspectives—in part two of this analysis.
GitHub Repository: https://github.com/Iangeertsen/Offensive-Defensive-Weight.git