Nylon Calculus Week 14 in Review: Standing up and team 3-point defense

I know that for many NBA fans it can be hard to concentrate on basketball with larger concerns in the world, but there is an overarching motivation when I write about the league and stats. Even though analytics and the scientific method have become a respected quantity around some sports leagues, with front offices staffing heavily with people who can pull and analyze numbers, that’s not entirely true with the rest of the world. There’s a growing anti-intellectual sentiment in the world, tied to the anti-globalist and anti-immigration movements, and people are ignoring the opinions of scientists and experts in all disciplines.

This phenomenon is not lobbyists and other figures who are against things like climate change science because it would hinder their industry, but this mindset has permeated in lots of other places as well.

In the NBA, an anti-information and anti-intellectual movement doesn’t do too much harm, self-contained in message board posts about how dumb new stats are or off-the-cuff remarks about silly numbers during a telecast. But out in the real world, ignoring data, science, and objective evidence has consequences much more severe than arguments about Real Plus-Minus. So please, stop blindly attacking experts and evidence-based methods, in the NBA and otherwise. If we can get more people to accept a thoughtful approach to science and data in the NBA, maybe that can spread that to other areas. It’s needed.

The NBA standing against trump

The notion of “stick to sports” has been killed by an extremely troubling political environment in the US, which of course has immense effects on the rest of the world. Donald Trump’s first few days in office have led to numerous responses from the NBA community, from Gregg Popovich calling out his maturity and the morality of the people defending him to players like Enes Kanter commenting on the discrimination of the immigration ban. Steve Kerr has been a vocal critic too, and he knows better than anyone the cost of terrorism, as his own father was killed in Lebanon where a jihadist claimed responsibility.

"“As someone whose family member was a victim of terrorism, having lost my father, if we’re trying to combat terrorism by banishing people from coming to this country, by really going against the principals of what our country’s about and creating fear, it’s the wrong way to go about it,” Kerr said. “If anything, we could be breeding anger and terror.“I think it’s shocking. I think it’s a horrible idea. I feel for all the people who are affected. Families are being torn apart and I worry in the big picture what this means to the security of the world.”"

Source.

The NBA’s role here is interesting. As a whole, they’ve been fairly progressive and inclusive where many figures have been allowed to speak their minds, and now there’s a broad-scale travel ban on seven nations: Iran, Iraq, Yemen, Syria, Syria, Ethiopia, and Sudan. As a league with a still-growing international influence, the NBA will be yet another organization trying to make sense of the administration and its orders. No current players are affected, but only by a razor-thin margin. Hamed Haddadi last played for the Suns in 2013, and he’s a citizen of Iran. Former University of Oregon star Arsalan Kazemi had a few flirtatious encounters with the NBA, but never played in an official game — he too is from Iran. Luol Deng and Thon Maker, the Bucks’ first round pick and project big man, were both born in the Sudan, a banned country, though where they were born is now technically in the country of South Sudan. The latter is a very recent country whose inception is, roughly speaking, because only part of the northern part of the state was majority Muslim.

Read More: Nylon Calculus — Appreciating the global NBA

Both Deng and Maker are refugees from the southern portion of that war-torn region. You can see how this could have resulted in catastrophe, but those two were lucky they had citizenship through different countries — and even if you are obviously not a terrorist, like an old grandmother in a wheelchair, or work, or will work, at a high-profile, important position like a scientist recruited for a research lab, you could still be detained and banned. For people who closely follow the league, we know players like Deng are not terrorists and are in fact model citizens. In many places in the US, you may not have to interact with refugees or others from nations like Syria, but through the NBA you can see quite clearly some of the potential unjust and unfair consequences.

The world needs more Andrew Wiggins commentary

I’ve noticed that people are doing some extreme mental gymnastics in believing Andrew Wiggins is the star and future superstar everyone expected him to be. We can dismiss the DeMar DeRozan and Rudy Gay similarities and instead point to defensive marvels who took some time in developing. But Wiggins’ weaknesses are components that are usually remarkably stable over a player’s career — his rebounding and steal rates do not befit a superstar wing, especially a defensive one. If you want to assert your impact on defense as a perimeter player, it’s tough to do so without causing a lot of turnovers. I understand that because of his hype, because of all the exposure he got while in high school and his number one pick status, we all want to believe he’s the next MVP-caliber young player. But the evidence isn’t on his side[1.].

Rookie-sophomore rosters announced

We’ve all spent an inordinate amount of time discussing All-Star snubs and selections and what it all means, but the rosters for the rookie/sophomore game have been posted and discussions out there are thin. The game was remade so it’s the US versus the rest of the world, which should make things quite interesting politically. Traditionally, the problem was finding enough rookies who even deserve to be there, which is partly why the old format was scrapped, but there’s still a lack of balance with the rosters and, of course, perhaps a few snubs.

Going by my Dredge metric, the most worthy qualified player who wasn’t invited was Montrezl Harrell. A power forward for Houston, he’s been an extremely efficient scorer so far, which alone has made him into a productive player. By most standards, he’s been better than Jahlil Okafor, who likely only got invited because he’s a more well-known player — Philadephia figured out how to be a good team by benching him. It’s why others like Buddy Hield and Brandon Ingram were chosen too; they’re known. Marquese Chriss is another weak addition; he’s a rookie and by no objective measure has been productive. But if you don’t like Harrell, there are some other solid candidates for the US team: Sam Dekker, another Houston player, and T.J. McConnell, among others. The biggest World snub was probably Salah Mejri, but it’d be tough fitting him onto the roster. The World team is full of talented big men and weak on perimeter guys. This is probably because if you’re super tall in another country, you could be funneled into basketball, while smaller guys play the more popular sports in their respective countries. The selection process won’t ever be perfect, but if we get to see a lineup with Nikola Jokic, Kristaps Porzingis, and Joel Embiid, I won’t complain.

Your weekly Process update

Joel Embiid hasn’t played much lately, and there’s still no definite time-line on when he’ll return to the lineup, but he did play in a game last week: a nationally televised one against the Rockets. Embiid was sensational, scoring 32 points in 28 minutes and filling up the rest of the stat sheet. In one stretch, he tracked a James Harden fast-break and swatted away the ball for a LeBron-style chase-down block, and he then recovered the ball and ran it down-court to post-up Nene Hilario, a strong veteran post defender, where he scored nearly immediately in dominating fashion.

Embiid’s combination of size and skill is overwhelming — and excuse the comparisons, but this is what made legends like Wilt Chamberlain, Shaquille O’Neal, and Kareem Abdul-Jabbar so effective. It’s like watching a bunch of middle-school aged kids try to guard a high-schooler. It’s unfair. There are no easy counters to Embiid, besides swarming him, even the popular ones deployed on big men don’t work because he can hit shots reliably from behind the arc and at the free throw line. Stay healthy, Joel. Please.

The Oklahoma City trio

As the years fly by, regret must be growing with Thunder fans who have to watch other super-teams wreck the league while remnants of their own are propping up the contenders. According to HBox, my box-score metric, the player with the best cases for MVP were all once teammates: James Harden, Kevin Durant, and Russell Westbrook. You can view HBox for the current season here, by the way, along with some other stats — it’s a quick way to compare players.

James Harden sticks out because of his high Morey Index; he’s over 1, which is pretty rare historically. We all know how Westbrook is unique — he’s even rebounding the ball on offense at an above average rate. But Kevin Durant’s candidacy should not be forgotten just because his teammates are excellent too. His scoring efficiency is ridiculous, and his peripheral stats are good too; his defense has been definitely above average as well. We can’t say he’s shooting at unsustainable rates either because he’s done this before, and likewise we know his stats aren’t being propped up by a good team. He’s done this before. So please, don’t forget him when you discuss possible MVPs. He deserves to be in the discussion at the very least.

Pretty good

Jon Bois, a sports writer I’d recommend reading even if only for his epic Death of Basketball article, has a video series of esoteric events and oddities, and to my delight his newest one covered one of my former obsessions: super-long field goal attempts. I’ve covered this before, and we both noticed Andre Miller’s willingness to wreck his field goal percentage with halfcourt heaves — and we also noticed an epic Baron Davis make from 89 feet. Jon was using a Basketball-Reference feature to find shots, like the Davis one, but it only goes back to 2001. But through the play-by-play logs at stats.NBA.com, you can go back to the 1997 season to find shots by distance: were they any made field-goals that trump Baron?

Unfortunately, from the data I have, the longest made shot in that time period was only 73 feet from Keon Clark. In fact, no shot was even attempted at 89 feet or greater. It may be a long time before his shot is eclipsed. Looking at every backcourt shot from 1997 to 2017 (through Sunday), you get an accuracy of about 2.8 percent. But from 80 feet and further, the accuracy drops to 1.1 percent — only Baron Davis, Vince Carter, and LeBron James have hit shots from those distances, according to my data. Those shots are rare, and they’re nearly impossible to convert. By the way, if you want to access the data, I’ve provided a .csv with every backcourt shot from the play-by-play log here on my github.

Team 3-point defense: A practical application

I’ve opined on the nature of 3-point defense multiple times — as the league becomes even more reliant on the shot, we really can’t ignore its effects either — but theorizing and identifying only goes so far. How about I present some solutions?

For this week, I’ll tackle this problem with basic team stats so that I can I implement the solution in every available season (and league) with 3-pointers. If you have variables like 3-point distance, location (corner versus above the break), and assisted rates for these shots, you should have more accurate and stable results. And, of course, if you have SportVU data, you have some tremendous potential. But you don’t always have that kind of data available, and I would like an improvement to team defensive stats. Not only will you have improved measures of performance for teams, you can have improved individual metrics too. Stats like Box Plus-Minus rely on team ratings, and I could apply the results to my own stats like HBox and Dredge.

Let me begin this with some research background: open 3-point percentage is virtually all noise, and open three’s consist of a huge portion of the 3-pointers teams allow. There’s no pattern to individual 3-point percentage defense either — it’s noise too. Ken Pomeroy found that defenses had little control over opponent 3-pointer percentage in the NCAA as well. But the correlation isn’t completely zero, so there is some degree of control by the defense. It’s usually not true that teams are adept at “choosing” the correct shooters for a low opponent 3-pointer, but 3-pointer can have different difficulty levels and perhaps some teams can control those difficulty levels. For instance, teams that allow fewer outside shots may allow tougher shots from behind the arc in general — there are fewer shots because they’re guarding so well out there. You can see a basic correlation plot below for team data via basketball-reference. Yes, there’s no strong pattern, but there is a tiny correlation; it’s a start.

Another logical argument is that teams that defend shots well in general defend the 3-point line well too — even if they’re just protecting the rim well it can have cascading effects on the rest of the court, erasing easier opportunities. The pattern is much more evident in the graph below. There’s definitely a correlation between those percentages, and it shows a nearly one-to-one relationship between 2-point percentages and 3-point percentages. This means that a team that allows a 2-point percentage five points lower than the league average would expect to see the same disparity for outside shots.

There are more variables of interest too, and I’ll spare you the messy plots. The next step is variable selection through regression. Here’s a rough sketch of my methodology: use ridge regression to build a model for predicting opponent 3-point percentage using season totals. Then test and modify based on how well it actually correlates from stats gathered in the first half of the season to second half opponent 3-point percentage. For the best results, the model data period and the test data period were separated: 1990 to 2010 for the model, and 2011 to 2016 for the test period. From there, I was confident in crafting the model. Curiously, the other major factors of a team’s defense, besides the two mentioned previously, all were significant — FT/FGA and turnover rate — except for defensive rebounding, which does make sense. How should rebounding help you defend outside shots?

Table: correlation of stats in first half of season to opp 3PT% in second half

(2011-16)

The most useful variable, however, was an league average 3PT percentage with totals from the team of interest subtracted — teams can’t play themselves. This means that an outlier squad like Golden State last season would have a lower adjusted league average percentage because their own average is so much higher. You can see how well the model performed in the table above. Remember that this is on data the model has not seen. There’s little correlation between a team’s “defensive” 3-point percent from one half of the season to the next, which is the first number listed[2.]. In fact, just using the league average is more predictive, which says something to the uselessness of opponent 3-point percentage– imagine if you could get better team win predictions by saying everyone will perform at a 50 percent win level for the rest of the season. The model, however, performs substantially better.

From there I have some confidence in sharing model results — these should be much more useful than opponent 3-point percentage. I actually found no predictive advantage in using opponent 3PT% in the first half of the season to predict second half percentage even when mixing with the model’s prediction. You can see the formula below for the prediction model, where the data was built using seasons 1990 to 2016 weighed by 3-point attempts. Note that opponent turnover rate wasn’t included in this version of the model, but it was in most others — it had a fairly weak effect anyway and with the full data set it lost its significant.

Predicted opp. 3PT% = -0.312 +LgAvgAdj3PT% +0.1865295*(Opp2PT% – LgAvg) +1.84*(OppFT/FGA – LgAvg) +0.312*(Opp3PTA per Poss)/(Opp3PTALgAvg per Poss) where %’s are from 0-100

Finally, how can I assure that the formula has some utility? Let’s create a simple adjusted defensive rating and see if it out-performs the basic team defensive rating using the same methods as above (i.e. correlation of stats from one half of the season to the next. It’s as simple as can be too: I look at the points per 100 possessions lost or gained in opponent 3-pointer. There are no further refinements related to, for example, the change in the number of offensive rebounds when assuming different 3-pointer percentages.

Next: Nylon Calculus -- The Western Conference is getting stronger

The table below has a summary with correlation coefficients, as discussed previously. Even though this adjusted defensive rating is way too simplistic, it still performs better than the basic defensive rating. And the correlation of the adjusted defensive rating to itself from the first half to the second half of the season is even stronger, which again suggests that it’s a more stable team measure.

Table: correlation of team defensive ratings in first half of season to  second half

(1997-2016, no 1999)

The tests have shown that the alternatives to “3PT defense” are much stronger, and the effects are significant. We can improve the heavily cited defensive rating, and with teams relying more and more on these shots the adjustment will become even more useful. The consequences are wide-spread — people will use a stat like BPM as ammunition in the Defensive Player of the Year race, but BPM is tied to those defensive ratings. Let’s stop rating defenses based on something they can’t control well.

[1. Here’s a fun new defensive stat: a perimeter defense composite, which shows Wiggins ranked 123rd out of 184 players.]

[2. The correlation coefficient ranges from -1 to 1, where 0 is no relationship, 1 is a perfect relationship, and -1 is a perfect inverse relationship.]