What wins in the NBA: Offense or Defense?

By Julian Ryan

What matters more for winning games in the NBA: offense or defense? This question vastly oversimplifies the game and is somewhat moot as of course they both matter. More interesting, for GMs building rosters on a limited budget, is the question of where is the most value available: on offense or on defense? Put another way, would you rather be number one at offense or at defense?

One can look at a similar question in the NFL, for example by examining historical DVOA ratings from Football Outsiders. DVOA ratings are calculated by comparing the result of a play to an expectation of outcome, adjusted for the opponent, to assign degrees of success and failure to each play. Over the years there has been persistently greater spread in offensive DVOA, and offensive players’ capability to outperform expectation as compared to their defensive counterparts suggests that there is more value to be mined there.

In Dean Oliver’s “Basketball on Paper”, one of his findings is similar for the NBA, that offense wins playoffs and championships more often that defense. Now with the work done at inpredictable.com and their win probability model, we can apply a similar methodology to that above for the NFL, and at least attempt to quantify the relative available value of offense and defense in the NBA.

To do so, I split the game conceptually into three parts: offense, rebounding and defense. This is extremely broad and ignores causal links between the three (e.g. crashing the glass) but for this analysis I have assumed independence between the three.

I defined offense is how a team does in stuff that ends possessions when it has the ball – shots, getting to the line and the resultant free throws, and turnovers. Inpredictable.com’s win probability added (WPA) metric encompasses these contributions player by player. Just as DVOA compares outcome to an expectation, WPA documents changes in win probability as compared to the expectation of no change. Rebounding WPA (RWPA), based on the example outlined here, is the extra win probability gained from extra possessions at the offensive and defensive ends of the court. Defensive WPA added is defined as everything else which is not captured by the former two changes in probability.

For example a good screen, which allows your point guard to drive to the basket and improve your chances of scoring, will be eventually attributed to whoever takes the shot, gets fouled or turns it over. In this sense, WPA absolutely does not assess a player’s total offensive contributions (for example ignoring passing) but if we aggregate WPA at the team level, eventually someone on your team benefits from your good passing or screening at the end of the possession. Hence we can use team aggregate WPA to evaluate the total offensive win probability added over the course of the season

After a possession ends (on the offensive or defensive side of the ball), if your team rebounds the ball, it gains win probability by having an increased possession. It is again not ideal as a measure of individual rebounding contribution as it ignores boxing out among other things but again, aggregating at a team level accounts for this as eventually someone on your team gets the credit.

Defense is now my catch-all for everything else and I calculated a team defensive WPA (DWPA) as a team’s regular season wins minus its WPA from offense and rebounding. Good defensive teams will win more games than bad ones given the same WPA and RWPA because their opponents will score fewer points on the same number of possessions, and DWPA also captures the win probability added from turnovers forced by good defensive play. This is by no means perfect and none of these team stats are much use in isolation, especially as RWPA is not compared to an expectation due to difficulties in being able to credit/debit rebounders. However, if we compare each team’s stats to the league average, we can get a sense of the wins gained or lost from 41 (a .500 record) for each team in each of these three areas.

The data is only available for the past season and for players that switched teams, the site does not make clear the WPA/RWPA gained at each team, so I assumed that players accrued WPA/RWPA at a constant rate and split the player’s individual total between the teams he played for by games played for each team (e.g. Luol Deng’s totals count towards Chicago and Cleveland). Summing each team’s players’ WPA and RWPA and thus DWPA, I then subtracted the league average to see where teams got there wins this season in terms of offensive, rebounding and defensive win probability added over average (WPAoA).

East WPAoA

West WPAoA

Indiana stands out as a team that gets all its extra wins from defense, playing .500 ball otherwise in terms of offense and rebounding, while Portland is their antithesis with below average defense but elite offense and rebounding (best in the league) seeing them to 54 wins. It is also of note that Miami and San Antonio, two big contenders for the championship, both have extremely poor rebounding, costing them 3.4 and 5.4 wins below .500 respectively, as compared to an average rebounding unit.

Now we have a framework to see in which of the three phases there is the most variation and thus the most wins to be garnered:

Phase Standard Deviation Interquartile Range
Offensive WPAoA 7.98 13.81
Rebounding WPAoA 2.64 3.48
Defensive WPAoA 7.31 9.26

Dean Oliver’s result does indeed appear to hold for the regular season, there is greater offensive spread implying that a strong offensive team will usually outperform a similarly strong defensive team in the regular season. However, the difference is not that substantial and the success of teams such as Indiana and Memphis show there is more than one path to success in the NBA.

This conclusion is based on only one season of play and is thus not very robust. In addition, the WPA stats are not adjusted for opponent as DVOA is, though strength of schedule will matter less in the NBA than the NFL. As inpredictable.com provides data for future seasons and potentially improves its model, we will see if the result holds up.

This post has also treated offense and defense as black-and-white blobs, there is almost certainly variation within those categories. For example, I would wager that more wins come from a quality rim protector than a quality perimeter defender, given the high number of high percentage shots taken from within the paint. If true, that conclusion boosts the value of players such as James Harden, who is elite at offense (phase with many wins available) but abysmal at perimeter defense (area which costs Houston very few wins), and similarly reduces the value of Tony Allen types, who is the reverse in those categories.

Analytics is beginning to scratch the surface of evaluating defensive contributions from players. Ultimately though, offense appears to still matter slightly more. Regardless, no one wants to watch this when you could watch that.

About the author

harvardsports

View all posts

Leave a Reply

Your email address will not be published. Required fields are marked *