NBA Week 5 in Review: Giving Thanks

During Thanksgiving week, some intriguing storylines continued to develop across the NBA. There is the ongoing demise of the depth of elite teams in the West, the Warriors are still undefeated. The 76ers, however, remain winless, as some terrible inverted beast of Golden State. But the league has been consistently entertaining, and for that every fan should be giving thanks.

Spurs Lying in Wait

While the Warriors hog all the glory, San Antonio has quietly been dominant, which is the quintessence of the team anyway. They have a point differential of a team that would win 66 games, and they’re 14-4 after Monday night’s loss to Chicago, which prorated over a full season would translate to roughly 64 wins. Many NBA analysts expected San Antonio to roll over the league like this, but the rationale was usually the signing of LaMarcus Aldridge. So far he doesn’t look like the primary cause.

Right now, San Antonio’s defense looks like one of the best ever, and that might surprise people. In fact, their defensive rating is nine points better than the league average, which is better than even their early 00’s defensive squads. At first glance, the Spurs have probably been getting lucky on opponent three-point percentage, as I discussed last week, but factoring in opponent FT% too and they’re still 8.7 points per 100 possessions better than the league average. They’ve been succeeding by controlling the defensive boards and giving up few free throws, ranking first in both categories. The team has been sensational on rebounds with Tim Duncan on the court. Per NBAWOWY, their DRB% has been 83.8% with him on the court. For reference, the best mark in NBA history was 79.3% from last season’s Hornets. It’s not just Duncan though because most of their wings can rebound well, from Kawhi Leonard to Kyle Anderson to Danny Green. The team’s core should indeed be great on defense, but it’s surprising how good they’ve been and there’s a legitimate question of how much their numbers will hold over the rest of the season.

However, San Antonio’s offense has been worse than last season, and LaMarcus Aldridge specifically has not been shooting well. He’s slumping on his long two-pointers, and he’s even shooting a lower percentage at the rim. He’s taking fewer shots overall, which isn’t surprising, but his efficiency hasn’t increased. He’s not taking more free throws, and his three-point experiment from Portland last season hasn’t carried over. Unfortunately, he hasn’t given their team a significant bump on offense in any measurable way[1. Using NBAWOWY again, I looked at how the team did with and without Aldridge, controlling for poor bench units by filtering out lineups with Jonathon Simmons, Ray McCallum, Boban Marjanovic, and Matt Bonner. The team was slightly better on offense without Aldridge.]. Aldridge might have the most value purely as a specialized tool the Spurs will use against Golden State because he has the foot speed and awareness to guard smaller guys on the perimeter and the size to destroy those same players in the post. Over the rest of the season, the Spurs might regress on defense but improve on offense, and they could be the one true weapon that works against Golden State.

Skillball

With the Warriors dominating the league with a fresh championship trophy in their hands, there is a lot of talk about smallball sweeping through the NBA. It’s a copycat league, and we’ve had stories like Indiana dismantling their frontcourt and insisting Paul George is their power forward and Washington benching Nene in favor of a shooter. But saying this is “smallball” conjures some negative thoughts. A label like skillball is more descriptive of the primary goal: it’s not about getting smaller; it’s about putting more shooting, passing, and quickness on the floor. Golden State is steamrolling the league with skill, not a lack of height.

North by North-Westbrook

During a normal start to an NBA season, Russell Westbrook would be receiving heavy media coverage for his play. After a superhuman performance last season without Kevin Durant, he’s extended his game with more efficient scoring and more assists. Many box score metrics love his play because he’s shooting as often as a young Michael Jordan while lighting up just about every traditional stat. What’s most remarkable is that people consider him to be the team’s best player now, not Kevin Durant, an MVP-holder and future Hall of Famer in his prime. Via NBAWOWY, the team is much better with both superstars on the court, suggesting a high level of synergy for these high usage players: an offensive rating of 117.6 with both of them, and just 109.6 with only Westbrook[2. Minutes only with Kevin Durant are limited, but they’re scoring a paltry a 99.6 points per 100 possessions just with him and no Westbrook, which is another sign that OKC is indeed his team now. But their 3PT% was abnormally low and, again, the minutes were limited.].

One may point to Westbrook’s historic amalgamation of stats, but there’s something in particularly he’s really improved upon: shots between 3-10 feet, which used to be a weakness. The problem he had in the past is that he’d be moving so fast that if he wasn’t near the rim to dunk, he was stuck in that in-between space close to the rim in an awkward stance. Here’s an example of a play from last week where he maintains balance on a drive and finishes over a playing standing under the rim. Teams are going to clog the lane against him, so he’ll need to keep working on his touch like this push shot in the paint. Here’s another play where he uses the backboard to finish over Robin Lopez. Normally, his modus operandi near the rim when it’s crowded is to use himself as a human bowling ball to get to the line, but his improved touch is helping to boost his efficiency. Overall, his numbers are jaw-dropping, but there’s one area to temper that enthusiasm: box score metrics are overrating his defense because he lights up the typical stats like rebounds and steals but he’s neutral there at best due to his wavering attention and tendency to take plays off. But this is legitimately MVP-worthy.

Boston’s Inconsistency

The Celtics, loved by advanced stats projections across the land, have barely been keeping themselves above 0.500, and some of their games have had curious results. They just got destroyed by the Hawks and beaten soundly by the Magic while narrowly escape defeat from the hapless 76ers. Yet they decimated the Wizards and have a healthy point differential on the season. When Boston wins, it’s usually been by a large margin. Research over the years has been pretty consistent on this matter: point differential is important, and discounting large margins of victory does not lead to better predictions. This is a team with a good defense and they’ve done it with much of an effective Marcus Smart, who had a disappointing start to the season before being sidelined by injury. But this season is mostly just for practice: the future includes multiple picks from the Nets. Rejoice!

Explaining RPM

Since RPM was recently released on ESPN, there’s been a lot of discussion about what the stat is and what it can’t or can do. There are some important distinctions to clear: one, ESPN displays single season RPM only, which is much less stable than a multi-season model, especially this early in the season. If you think, say, Andrew Wiggins is being underrated because RPM is a “trailing indicator,” you’re thinking of a different version of RPM or RAPM. How Wiggins was rated last season has no impact on how he’s being rated now. Secondly, the model is a mix of countable stats and plus/minus ridge regression with some demographic information like height. High scoring efficient players who fill up the box score in multiple ways look better in RPM than in pure plus/minus stats. There are also a number of non-traditional stats used from blocks that are rebounded to live-ball versus dead-ball turnovers.

One common question, or complaint, about RPM concerns standard deviation, which would display the statistical variation for every number. This would give people an idea about the likelihood a player is X “good,” via the rating. Why isn’t it shown? Because of the bias introduced by the ridge part of the regression, a simple standard deviation isn’t possible. One may bootstrap a standard deviation for RPM, however, which actually has been done before, but it’s an imperfect method. There was a lot of discussion last year about error rates in regard to Andrew Wiggins, who people believed would be the one true savior but had pretty ugly advanced stats. There’s sometimes criticism about how rookies are treated, but there’s no malicious intent or a systematic error here, especially with the single season version on ESPN. Rookies rarely contribute positively to a team.

Finally, there’s been one recent change to RPM: luck has been factored in. First free throw attempts  are replaced with an expected value based on FT%. Naturally, defenders have no control over how well a player shoots that free throw. This is related to the research I posted last week, and by sheer coincidence the change was announced a day later.

Doe Defense

Milwaukee was one of the most intriguing, pleasant surprises last season, making the playoffs with Jason Kidd as its coach and an aggressive switch-heavy scheme that led to one of the league’s best defenses. With Greg Monroe replacing Zaza Pachulia and the loss of Jared Dudley, a reasonable expectation would include a modest step-back on defense. But their defensive rating is the worst in the NBA, as they’ve fallen a shocking 12 points per 100 possessions. More specifically, they’re an awful defensive rebounding team (ranked last) whose opponents shoot well and get to the line frequently. The difference from last season is a matter of degrees on every factor, as they’re generating fewer turnovers too, but the major damage was caused by opponent effective field-goal percentage.

Turning to video, the Bucks defense has disintegrated and they’re making mistakes like this play, where no one communicates properly and Rudy Gay is left as wide open as possible for a three-pointer. Later during the game, Giannis Antentokounmpo picks up Rudy in transition, and haphazardly follows him and ineffectually falls down near the rim, either being pushed over or failing to draw a charge. A switch-heavy defense has its merits, but there are issues like Michael-Carter Williams being bulldozed by Rudy Gay. You can select virtually any game and find a number of errors. Not having Jared Dudley hurts, and Jabari Parker, while a promising scorer, is definitely worse on that end of the court. I don’t believe they’re as bad as their defensive rating suggests — for one, the team has probably been unlucky with opponent 3PT% and FT% —  but there’s a legitimate structural problem here. The East is deeper now, and they’ll need to resolve their issues quickly.

Most Unpredictable Teams

When one experiments with NBA metrics and writes out and explains win predictions, the uncertain nature of the league is apparent. Some hindsight bias probably obscures the large degree of unpredictability, but even the best NBA analysts are generally very wrong about several teams every season. Think about Houston this season; they’re significantly below 0.500 and in danger of missing the playoffs after a summer where they made the conference finals and picked up Ty Lawson. Why would anyone sane see this season coming? But large, wild changes happen, and the best numbers are powerless to avoid those whiffs.

But just how unpredictable are NBA teams, and which ones were the most unpredictable? For a systematic appraisal, I’m using basketball-reference’s BPM to predict how teams do in a given season with their actual minutes distribution with a player’s corresponding BPM value from the previous season. I also use some mean reversion and set rookies at -2, which is pretty standard.

Without further ado, the results are in the two tables below. The first table shows teams that were much better than expected, per BPM. The most unpredictable team in that sense was the 1989 Phoenix Suns, who went from 28 wins to 55. It was their first season with Tom Chambers and their first full season with Kevin Johnson, but it was still a massive turnaround. The next two teams share something similar: the Spurs with rookie Tim Duncan and the Celtics with rookie Larry Bird. Rookies rarely contribute much to a team’s bottom-line, but, obviously, those two legends provided a lot more. Most teams in that table were quite excellent — they were better than expected, so that makes sense — but there are a few mediocre teams who “should” have been worse. The Blazers in 2007, for example, were terrible the season before and had to rely on a few rookies, but some of those guys, like Brandon Roy, developed quickly. They actually had the worst predicted rating in the data-set. A more recent example is Utah last season, as sophomore Rudy Gobert had a breakout season and Gordon Hayward played at a fringe all-star level.

Table: over-achieving teams

For under-achieving teams, the sorry Bobcats from the lockout season are unchallenged. The Heat from 2008 didn’t have a ton of talent, but Wade played nearly 2000 minutes and they had a few minutes from players like Shaq, Shawn Marion, and Alonzo Mourning. But Wade was slowed down by injuries, and that’s something the numbers couldn’t see. The Knicks from last season are a recent example: they were supposed to be bad, but not quite that bad. There are only two decent teams in the table, and both are interesting: the star-packed 2003 Lakers, complete with Gary Payton, Kobe Bryant, Karl Malone, and Shaquille O’Neal; and the 2011 Miami Heat, who projected to have the second best team ever with LeBron James, Dwyane Wade, and Chris Bosh[3. The best projected team ever was the 1997 Chicago Bulls since they had a team intact coming off a year with 72 wins.].

Table: under-achieving teams

The future is curled in a mystery and clouded by a dense fog, and an appropriate use of statistics leads to genuinely better insights. But even with the best stats there are still complete surprises. We can make minor adjustments due to usage or shooting luck, but how do you predict an absolute collapse, like Houston so far, or a huge resurgence, like the surprising Phoenix team in 2014? The mechanisms here appear to have feedback loops where one issue leads to another and yet another and it snowballs to a point where a team with good players is deep in the lottery. That’s difficult to model, and there’s a one universe problem here: perhaps if we got to start the season again and again with minor tweaks, we’d see different results. But alas, we’re stuck with our one universe, and we’re stuck with these giant errors. At least the unpredictability makes the league more interesting.