Nylon Calculus Week 10 in Review: Deflections and NBA Star Wars

As we near the end of December, it’s that time again: a new Star Wars movie. Oh and there’s something about a jolly, fat red man named Santa. The 25th has become the NBA’s signature day, and after we’ve all enjoyed all the festivities of that day, from seeing the requisite Finals rematch to the omnipresence of Knicks and Lakers games no matter how good they are, let’s take a look back at the last week in basketball to figure out how we all got here.

Boston’s defensive decline: Regression to the mean strikes back

A few weeks ago, debates were raging about Boston’s defense, which appeared historically great and carried a team beset by roster changes and injuries, filled with rookies, to the best record in the league. There was a lot of rationalizing about why they were so great on defense, and why we should expect their opponent 3-point percentage to hold. But that’s all disintegrated. Look at their defensive splits from before and after Nov. 20 below — it’s a stark contrast. It’s not just their opponent 3-point percentage, which I (correctly) didn’t think would hold; it’s their entire defense.

Table: Boston Celtics season splits

You’d expect defensive stats to get worse as the season goes on because that’s the usual pattern as offensive players get into more of a rhythm and they’re more tuned in, but the drop here is a severe degree. That defensive rebounding decline is eye-opening. Much of that decline has come from Al Horford and Jaylen Brown, and I can’t yet figure out why. They were an atrocious rebounding team last year, so perhaps their true nature is starting to emerge. More commitment to boxing out, and some tweaks from Lord Brad Stevens, can only go so far. Defensive rating usually stabilizes around the 16th game, where the noise becomes less of a factor. But they’ve been unusually volatile this season.

A New DeMar DeRozan

Credit where credit is due: I’ve criticized DeMar DeRozan as much as anyone for his lack of an impact in everywhere but scoring, and how his love of mid-range shots at the expense of 3-pointers depresses his overall value. But he’s setting career highs in shots behind the arc and assists, and by all accounts his defense is better too. His overall numbers are actually better primarily because of higher 2-point shooting percentages, but it’s nice to see him make marginal strides forward in other areas of the game.

The nonlinear awakening

One of the most common arguments people have with NBA teams is whether or not they have a ceiling because of a specific player. If your backcourt consists of Damian Lillard and CJ McCollum, the thinking goes, you can only be so great of a team because your defense will have them as a liability. But that’s some backwards logic. There are always ways to improve your defense even with limited defenders — just ask the Houston Rockets — and you can always make up poor defense with better offense. Functionally, that concern is a related concept though: sometimes interactions have linear effects, and sometimes they’re nonlinear.

For example, rebounding is notoriously nonlinear. You aren’t going to add three guys averaging 10 rebounds a game and suddenly add 30 rebounds to your team — and the same is true when you use rebound percentages. Then there’s spacing, which to the best of my knowledge is actually linear. When team-building, knowing which effects are linear and which are not can be extremely powerful at the margins. If you’re a poor rebounding team that elite rebounder is going to be ever more valuable to you, and you can think about how usage and playmaking are nonlinear when you stagger lineups — perhaps it’s best to let Carmelo Anthony play every minute Russell Westbrook sits, sans garbage time. But at the end of the day, the game is simple: outscore your opponent. You can do that with two poor defenders or none; the means ultimately don’t matter.

LeBron James: Our ideal clone source

I had not seen this article until very recently, and it’s the kind I adore: injuries and biomechanics when landing. The article discusses Derrick Rose and Dennis Smith Jr. but I think we got an even better case study. LeBron James, by age or especially his career minutes load, is performing admirably well, and in some key areas he’s having a career year, which should not happen. How can a man that large with an even larger role on the court be this durable?

Even though he’s a freight train who barrels to the rim, he’s usually in control when landing. If you transfer the energy from the fall intelligently, you can put less stress and shock on your joints. Look at his top dunks for every season he’s had in the clips below — even when he’s clipping defenders and falling backwards he’s still good at bending his knee and bracing himself with both feet. Normally on drives he’s a two-footed lander who sometimes crouches down low when necessary, and when he lands all on one leg it’s bent appropriately and his other leg quickly follows. Biomechanically, he may be the Platonic ideal of an NBA player, and he’s aging gracefully because of it.

The last PER mystery

Earlier this year I broke down PER, but there was still one component that was a bit fuzzy to me: field goals*(2 – factor*team assists/team field goals) where factor = 2/3 – .5*league assists/ league field goals*league free throws/league field goals/2. Let’s explain this term by breaking it into components. Ignore the part with the league numbers for free throws. You get field goals*( 2 – 2/3*team assists/team field goals.) This is just straight points (field goals multiplied by two) subtracting out a proportion based on your team’s assist rate. Basically, since PER gives credit to assisting field goals, you need to debit being assisted. Think about the formula: if your team has no assists, then of course all of your field goals will be unassisted and you will thus get full credit for them. The two-thirds is in there because John Hollinger gives one-thirds of the credit to the player who assisted the shot (it’s two-thirds because it’s being multiplied by field goals, not points.)

So what’s that second component nested inside the factor term? It’s field goals/4*( league AST/league FG*league FT/league FG*team assists/team field goals.) That’s quite the mess. It’s probably best to go back to factor = 2/3 – .5*league assists/ league field goals*league free throws/league field goals/2. You can see that long second part is a modifier of the 2/3rds term. It’s a league adjustment for estimating assisted shots for individual players, but I can’t fathom the details here, and I’ll have to keep searching. Perhaps as Jon Bois suggested sometimes it’s best to leave a few mysterious out there.

A phantom concern: Jason Kidd and foul shots

Last week, Jason Kidd made a curious late-game coaching decision. With 1.4 seconds to go and up by three points with one free throw to go at the line, he instructed Khris Middleton to intentionally miss. He said he was worried about the 4-point play if the Cavaliers were allowed to inbound the ball. This is dumbfounding because if you’re really worried about the 4-point play, then just don’t defend outside shots at all. Plus, intentionally missing opens up the possibility of a heave tying the game, which could lead to a loss; or even more hilariously, a 4-point play deciding the game right there — which is actually more likely because the foul could happen during the rebounding frenzy after the miss.

That’s not Kidd’s first strange decision. Earlier this month up four points with ten seconds left, he elected to intentionally foul the Pistons — and no, it wasn’t Andre Drummond; it was Reggie Jackson. He said he wanted to deny their tactic of the quick two and make it a free-throw shooting contest. That’s … balderdash. You’re stopping the clock, which is definitely what the Pistons want, and sending them to the line is like a quick two points. He didn’t mention anything about the threat of the 3-pointer, which is the only bit of logic you can throw in there. And, lastly, there’s the fun clip from last season where he told Giannis Antetokounmpo to intentionally miss when up three points … with 0.2 seconds left on the clock. He was, again, worried about the four-point play, but you can’t even shoot with two-tenths left; you can only tip it in. Did something happen during his career so that he’s phobic of 4-pointers? Jason Kidd is secretly one of the worst coaches in the league, and we’ll see how long, if ever, the mainstream media takes to realize this.

Revenge of the Thunder

The Oklahoma City Thunder, after weeks of negative think pieces about their three “superstar” arrangement and all their ills, are well past 0.500 basketball now, and they’ve vanquished their clutch problems. This is what I expected a while ago — team clutch stats are largely noise, and they exhibited no signs of teams that win less often in close games. They’ve won most of their close games lately, and they’ve had two overtime wins to boot. This peaked in the last week where they went undefeated, which included a one-point win and a three-point win. I think we can cross out that issue — now they just need to find Russell Westbrook’s shooting mojo.

David West: A late-career return of elite performance

There are a few late-bloomers in the NBA who don’t fully develop until their late-20’s and they peak when they’re, say, 32-years-old. But we usually don’t see a player hit career highs in metrics like PER and BPM when he’s 37-years-old — and by a fair distance too. David West is having the best per possession numbers in his career. His shooting efficiency is through the roof, but some of his defensive numbers are jumping off the page too, as he’s grabbing nearly twice as many steals as his career average and netting three times as many blocks as his average. For a 6-foot-9, 250-pound player in his late-30’s, this is certainly surprising.

Looking at West’s shot tracking data, he’s been getting a lot more open shots with the Warriors —  yeah, big surprise. But as you can see from the plots below, not only is he getting more open shots at the rim, which are basically free points, he’s hitting more of his contested shots. The “open/tight” tracking data isn’t perfect, naturally, and it can miss a lot of nuance like being double-teamed or information on where the defender is, or even if a hand is up to contest. Also, shots don’t neatly fall inside four bins so well. Perhaps the average of his “very tightly” defended shots is 1.5 feet, while before it was closer to 1 foot — that’s enough to influence his field-goal percentage by a significant amount. But it’s still quite a difference for a player nearing 40-years-old.

Again, however, his defensive stats are up too. Playing primarily center for the first time in his career explains the block rate, but not the steal rate or the fact that he’s blocking 3-pointers too. West has been active and mobile this season and deserves a lot of credit for this late career resurgence. This isn’t normal, and we’ll see if this trend hopes up over the rest of the season because it’d become a great entry in the hall of old-man NBA performances, which rarely include non-superstars.

Rogue stats: Deflections

When the NBA introduced hustle stats a couple years ago, many people were excited the league was willing to provide new and hopefully useful statistics. But we already have charges thanks to the play-by-play, defended shots are just a repackaging of SportVU data, and screen assists, while are definitely useful, have less power when you don’t have total screens taken or the shots that were aided by screen assists. Loose balls recovered are definitely interesting, but what about deflections? Is it true that they correlate strongly with steals, rendering them effectively useless? Or do these stats provide essential information we’ve been missing beforehand about the league’s best defenders?

You can see how deflections track with steals in the graph below. The relationship is fairly steady, minus a few outliers. It’s almost exactly two deflections per steal, at least in 2017 where we have the full data available. John Wall led in total deflections followed closely by Draymond Green. Who led in total steals? Yes, John Wall followed closely again by Draymond Green. The biggest outlier, by the way, on that chart is Kelly Oubre Jr. who had a 3.4-to-1 ratio. But we should mark that as a property of Oubre’s playing style or is it just random noise?

Now that we’re in year two of collecting deflections, it is indeed time to test its efficacy in explaining defense. The test here is to compare the ratio of deflections to total steals from 2017 to 2018. Players are filtered out when they have less than 50 deflections in either season. That creates a pool of 84 players, which is a pretty decently sized group when we don’t even have one-and-a-half seasons of data. Unfortunately, the correlation of that ratio from season-to-season is a paltry 0.16, and the p-value when including the previous season’s ratio is near that threshold of 5 percent. That suggests a very slight positive relationship for the ratio, meaning it’s more noise than useful information.

The reason this test is important is that if deflections are largely independent of steals — i.e. orthogonal — then it’s a lot more powerful for player evaluations. After all, if deflections perfectly track with steals, then what use are they? Sadly, there is actually a lot of overlap between deflections and steals. Most steals also get credited as deflections (but not all), which is why the correlation is so strong. That still leaves a large swath of deflections that are independent, but it’s tough to know which ones are because we don’t have the data on which steals are not deflections too.

For another illustration of the correlation here, let’s look at the most extreme players in 2017. The two tables below have the players with the highest and lowest ratios. What’s most striking is that the first group averaged a 2.7 ratio one year, and then fell to near the league average with 2.0 the next. It’s a similar pattern with the second group, though not as severe. The aforementioned Kelly Oubre Jr. is actually holding his ratio this season — let’s see if that remains true over the second half of this season.

Table: Players with the highest deflection-to-steal ratio in 2017

Table: Players with the highest deflection-to-steal ratio in 2017

These patterns would be easier to tease out if we knew which of these steals are deflections too. It’s like we’re trying to solve one equation with two unknowns. We have deflections, steals that aren’t deflections, and steals that include deflections. If a player’s ratio changes, we can’t know if it’s because they’re netting fewer steals without deflections or if they’re changing their deflection (that don’t lead to steals) rate. This would lead to a systemic problem if you were to use the stat because it’d be unfair to players who, say, grab more steals that don’t include deflections.

Additionally, if you used the stat in a player metric like PER or BPM, then you’d be double-counting most steals. It’s like if someone wanted to introduce a new passing stat and 80 percent of current assists were included in the stat. That’s fine for descriptive purposes, if you’re just looking at which players are great at passing, for example, to the corners, but it’s not ideal if you want to add to our current understanding of who the best offensive players are since the information largely overlaps another statistic.

Next: Nylon Notebook -- Ball-movers, black holes and the Warriors blocks

There’s still some value in the stat; it’s better to have it than not. I imagine with another season under its belt the deflection could pop up in a well-built statistical plus-minus model, if one were so inclined. But it could at least be a stabilizing influence for steals since they’re twice as common. If they’re this strongly intermingled, then they can be used together for a more stable steal stat — and steals are after all one of the most valuable stats out there. I’d also say deflections leaders are definitely great defenders, by and large, perhaps even moreso than steals leaders.

Maybe there’s a pattern I’m missing I can catch with another year of data. Or maybe the stat will be tweaked to fix this glaring overlap error. I’ll try to be optimistic. As it stands now, it’s a potentially useful but flawed statistic — which probably makes it an even more genuine NBA stat.