Which NBA Defenses Have Been the Unluckiest This Season?

One of the great things about using stats to analyze the NBA is that things are fairly stable.

Sure, upsets and surprises happen all the time, but in the long run, performances are more consistent in the NBA than other American professional sports, and good teams tend to win more often.

Each NBA game contains over 90 possessions, so sample sizes build up quickly, and crucially, there is not one killer, unpredictable variable in basketball like there is in other sports.

Hockey might be the most unpredictable because shooting and save percentages are prone to very high levels of variance. In baseball, roughly 70% of plate appearances end with a ball being put in play, an inherently random event.

Turnovers in football are significantly rarer than baseballs in play or hockey shots on goal but have a huge impact on the outcomes of games. Good luck predicting them though. Last season, the top three teams in turnover margin in the first half of the season combined for a +25 turnover differential; in the second half, these three teams had a -5 turnover margin.

The bottom three teams in turnover margin in the first half were -27; in the second half they were -5, the same as the best three teams in the league.

Teams that win the turnover battle in the NFL have won 76.4% of their games since 2013.

There is no event in basketball with the same combination of volatility and impact. That said, if we searched, we’d find the closest threat to the league’s aforementioned consistency beyond the three-point arc.

Three-Point Shooting Variation on Defense

Last season, the team that outshot its opponent from three-point range went 836-377, good for a 68.9% winning percentage (there were 17 games in which the two teams had an identical three-point percentage). It’s not as high of a winning percentage as NFL teams that win the turnover battle, but it is still notable.

There is also evidence three-point percentage is prone to a high level of random variation, particularly on defense. Dan Feldman found evidence of this at Basketball Prospectus, and Kevin Pelton said this randomness is magnified in small samples.

Since NBA teams have just passed the 25 games played mark this season, it would make sense to look at how certain stats after 25 games correlate with themselves over the next 57.

Three-point percentage allowed in a team’s first 25 games correlated with that percentage in the following 57 at just 0.24 last season (a correlation coefficient of 1 implies a perfect relationship and one of zero implies no correlation).

Contrast this with two-point percentage allowed (0.66), and three-point attempt rate allowed (0.74). These correlation coefficients support the inferences that teams exert little control over opponents’ three-point shooting and the best way to prevent three-pointers is to prevent attempts from beyond the arc (Ken Pomeroy found this to be the case in college basketball as well and has written extensively on the subject).

Predicting Future Performance

Despite the apparent randomness of three-point shot defense, the stability of three-point attempt rate and two-point percentage make effective field goal percentage stable itself (effective field foal percentage modifies traditional shooting percentage to account for the fact that three-pointers are more valuable than shots inside the arc).

A team’s effective field goal percentage allowed after 25 games correlates with itself over the next 57 games at 0.68.

Other elements of defense also show notable consistency, including turnovers per play (0.72), defensive rebound rate (0.66), and free throw attempt rate allowed (0.39).

Overall, points allowed per possession after 25 games correlated with itself over the next 57 at 0.56 last season.

This is a fairly strong relationship, despite the fact that one of its most important elements shows little consistency: three-point percentage in the first 25 games correlated with points per possession in the same span at 0.64, which was nearly as strong as two-point percentage (0.71), and stronger than rebounding (0.41) and free throws made per field goal attempt (0.56).

Opposing free throw percentage also plays a role, though this is obviously something a defense has literally no control over. Like with three-pointers, the most reliable way to prevent an opponent from making free throws is to stop them from taking free throws.

If we account for the inherent randomness in three-point shooting and opposing free throw percentage, we can predict future defensive performance more accurately.

We can start by taking the volume of three-point attempts a team has allowed and multiplying it by the league average three-point shooting percentage. This gives us the number of three pointers a team “should have” given up, or “Expected Three Pointers.”

Then, we repeat the process, but do so with free throw attempts and the league average free throw shooting percentage to get an Expected Free Throws total.

The methodology here is similar to xFIP in baseball, which subs out a pitcher’s actual home run per fly ball rate with the league average rate; this is because the theory holds that a pitcher can control how many fly balls he allows but has significantly less control over whether or not they leave the park.

Finally, we can multiply “Expected Three Pointers” by three and the number of actual two pointers the team allowed by two, and add the products to each other and the “Expected Free Throws” total. This sum is an Expected Points Allowed total.

Last season, Expected Points Allowed Per Possession after 25 games correlated with actual points allowed per possession at 0.67, outperforming points per possession itself (which had a correlation coefficient with future points per possession of 0.56, as mentioned earlier).

This means we can use xPPP to measure which defenses have been most and least affected by random variation and as a tool to project future performance.

Applying the Results to This Season

So how do this year’s defenses stack up in terms of Expected Points?

Here are the results, ranked by greatest difference between their actual points per possession numbers and their expected rate through December 20th.

The first thing you should notice is that while there are some teams, notably Dallas, with wildly disparate numbers in the "Rank" columns, most of them do not have huge differences in terms of their actual and expected ratings. Half of the league's expected rating is within a point per 100 possessions of its real rating, and 24 teams are within 1.5 points per 100.

Dallas does have the biggest difference in actual points allowed per possession and Expected Points Allowed Per Possession, looking like a bottom-10 defense in terms of the former and a top-five defense in terms of the latter.

This makes sense, given that the Mavericks have been burned by opponents' three-pointers, allowing a league-worst 39.8% three-point percentage, and have also allowed the 10th-highest free throw shooting percentage. They haven’t been much better inside the arc (they are tied for 25th) and do have a high three-point attempt rate (4th-highest) but have excelled elsewhere on defense.

They are the top defense in the league in terms of turnover rate and are fifth in defensive rebounding rate.

It is also not as if Dallas is allowing a ridiculous amount of wide-open threes. According to tracking data on NBA.com, the Mavericks have allowed the fifth-fewest “wide-open” three-point attempts per game in the league (attempts in which the closest defender is six or more feet away). Opponents are shooting 41.9% on these shots, which is tied for the fifth-highest rate and is above the league average of 38.4%.

They are also in the middle of the pack in terms of three-point shots allowed when there is a defender within four to six feet but have surrendered a league-worst 42.0% shooting percentage on them (the league average is 35.1%).

These percentages should start to regress towards the league average, meaning that if Dallas can continue their solid work on the defensive glass and turnover department, the results will get better on defense.

The Golden State Warriors are at the other end of the spectrum and have only allowed a 32.2% three-point shooting percentage. There is nothing obvious in the tracking data that suggests why, but the Warriors also allowed the second-lowest three-point percentage last year, so it’s possible this goes beyond random variation.

In any case, as you can see by the chart, even if Golden State allowed a league-average shooting percentage on threes, they would still have one of the 10 best defenses in the league, thanks to an elite two-point percentage and above average turnover and free throw attempt rates.

Also, take notice of the Trail Blazers and Suns. Aside from getting burned by three-pointers, the teams have both allowed 20 more points than would be expected on free throws alone. This comes out to about 0.8 points per game for both teams, who have both played in an above-average number of games decided by one bucket.

So while the NBA is more stable than the continent’s other major sports, random variance can still loom large.