Nylon Calculus Week 21 in Review: V-stats and scorers

Mar 3, 2017; Washington, DC, USA; Toronto Raptors guard DeMar DeRozan (10) celebrates after making a three point field goal against the Washington Wizards in the final minute of the fourth quarter at Verizon Center. The Raptors won 114-106. Mandatory Credit: Geoff Burke-USA TODAY Sports
Mar 3, 2017; Washington, DC, USA; Toronto Raptors guard DeMar DeRozan (10) celebrates after making a three point field goal against the Washington Wizards in the final minute of the fourth quarter at Verizon Center. The Raptors won 114-106. Mandatory Credit: Geoff Burke-USA TODAY Sports /
facebooktwitterreddit

Due to college basketball, some NBA injuries, and a number of teams either tanking or resting, we’re not in the most exciting stretch of the season. But we’ve got some seeds to determine, some teams are still fighting for the playoffs, and there’s a lot still to unpack in terms of distinctions and how we remember the season.

When the media and NBA fans look back at, say, 1993, they usually think of a small group of players and a handful of relevant teams. I’m not entirely sure how this season will be remembered in the future — we have Russell Westbrook’s triple-double, Houston’s electric offense, the Warriors pseudo-slumping with Stephen Curry, the Spurs dominating but post-Tim Duncan — and how we discuss the season as it ends, and during the playoffs themselves, will determine how the season is perceived.

Perhaps the data revolution will change that, and we’ll have a more nuanced view of the league than I’m imagining. But there is something at stake here, even as many NBA teams wave the white flag: history itself. And with that, let’s look back at the last week in the NBA.

The sweet sound of the Jazz in Utah

Looking at the standings and some of the team stats league-wide, there’s a sleeping giant hiding among some of the more frequently discussed teams — the Jazz are fourth in the west, and they’re a hair behind Houston for the fourth-highest adjusted point differential. They’re currently allowing 3.8 points per 100 possessions fewer than the league average on defense, per Basketball-Reference; that’s an elite mark for a team. They’re on their way to homecourt advantage, and they should be considered favorites in the first round.

Read More: An NBA block is like an NFL sack

But I know what people are thinking: the Jazz have no clout and little playoff experience, at least excluding Joe Johnson and Boris Diaw. Most NBA fans probably don’t believe in the team, so they’ll look for whatever sign possible that Utah should not be trusted. But the Jazz actually have some qualities of a team you’d associate with outperforming their regular season stats. They’re a defensive club, and those tend to do marginally better in the playoffs than offensive ones. Also, it’s a top-heavy team, and when they can give extended minutes to their best players, they’ll be even better. I don’t imagine they’ll be of heavy interest to most of the media when the post-season begins, but they deserve a little more respect, at least if they’re healthy.

Projecting youth

Last week I presented a couple methods to predict a young player’s peak performance. It was more of an exploration than an iron-clad statement. As such, I thought I’d provide my data as well as an R script to produce the results for both penalized regression and gradient boosting. You can view that here on my Github page. There’s an R script for “youngplayerprojection.r”, which should be able to run once you have the appropriate packages and the two .csv’s that go with the code. It doesn’t completely show my method, and there are more powerful gradient boosting models out there, but it’s a general overview and it’ll allow anyone to tweak things and experiment with the models. If Nylon Calculus is about anything, it’s about expanding the knowledge of the general NBA-public and sharing statistical tools with everyone. So have fun.

Zach LaVine

Talking about Zach LaVine now is a bit uncomfortable. He had a terrible injury, and he’s an exquisite dunker. But over the past few years, the Wolves have been much better on the court without him, and it doesn’t look any better with Basketball-Reference’s Box Plus-Minus or when you mix plus-minus stats with other stats, a la ESPN’s RPM. I know some people are tired of those metrics, but we can’t ignore that kind of information when it’s been consistent for multiple seasons in different environments (starting versus a bench role, new coach, etc.) Yet there’s this disconnect between conventional wisdom and the advanced stats, and I’m afraid it’ll lead to some unwise decisions in the future for Minnesota.

For starters, Zach is benefiting from the DeMar DeRozan athleticism bias: LaVine is an incredible leaper, and he can move quite fast, but that doesn’t make him a great defender. Thus, people assume he’s at least a decent defender, when in actuality he’s one of the worst in the league. He doesn’t pick up much in the way of countable defensive stats, like steals or blocks, or anything fancier. If you’re stating that his defensive presence is subtler, then it should (eventually) show up in something like plus-minus, and it has not.

For his offensive stats, he’s being seen in a more positive light now that he’s a 3-point bomber with a healthy percentage — it’s the kind of effect that makes even Austin Rivers more appealing. But it’s not enough to hide the deficits in his game, as his playmaking skills have been disappointing so far. Yet he’s still mentioned as a part of the Minnesota core, along with Karl-Anthony Towns and Andrew Wiggins. It’s not entirely a coincidence that the Wolves have played better after his injury despite the loss of a starter — a bit of regression helped too — as their net rating in games before the injury was -1.3 points per 100 possessions and +2.7 points afterwards. Until he turns a corner and improves his defense and passing, I don’t think he’s worth a major investment.

The Westbrook solution

In case you didn’t see the massive new article by Bill Simmons, he spent a few words on Russell Westbrook’s season. Yes, his stats are unprecedented; even Simmons comes to the conclusion that Westbrook is taking a larger share of his team’s offense than the mythic Wilt Chamberlain. But I want to address the core issue: sure, Westbrook’s team is mediocre, but it’s no coincidence that he’s crushing the usage rate record with a mediocre one.

The better your team is, the lower your usage rate will probably be, roughly speaking. If your teammates are not good at creating offense, then if you are a star-scorer you will shoot more often — it’s common sense. Westbrook shot less often when he played with Durant, and his usage rate skyrocketed when Durant was on the bench. So I don’t think this a referendum on Westbrook as a player. It’s partly circumstance, and partly team design — and that’s all combined with an incredibly gifted player with few physical parallels.

Chicago’s old investment

The Chicago Bulls have had a disappointing season, and the Dwyane Wade injury was yet another downer for a franchise without direction. I’ve discussed this before, and maybe it’s more obvious in retrospect, but the logic of bringing in Wade really was thin. It’s not a pure rebuilding move, and it’s not one you make for the future; so what’s it for? Wade is old and his history is filled with injuries — this was his peak season, and Chicago probably won’t make the playoffs. It was nice to see him play for his hometown, sure, but that wasn’t a basketball decision.

Now even Miami, a team that was supposedly taking the short-term hit, is ahead of Chicago in the standings, and they’re the team that opted for future flexibility and youth over the immediate benefits of Wade. I’m not trying to pour salt in a wound here; I just want to discuss a major roster decision. Was it worth it signing Wade, Chicago? I hope a few teams learn from their mistake here. Signing a highly visible veteran is temping, but if he doesn’t fit the team and won’t help you compete, your money should be invested elsewhere.

Living in Skal Labissiere’s world

When DeMarcus Cousins was traded to New Orleans, he left a massive vacuum in his wake. However, this is where Sacramento’s incompetence comes in handy: they drafted an alarming number of center-sized players when they had Cousins, and they now finally have an opportunity to show their value with all the new playing time. One such player is Skal Labissiere, a super skinny late first round pick who’s seen consistent playing time after a number of weeks where he rarely played at all. I don’t recall him being labeled a draft day sleeper or diamond in the rough a year ago. Most projection systems were fairly low on his college stats and overall profile. Surprisingly, and I think many people have forgotten this, heading into the freshman college season he was projected as the number one overall pick by many people.

Skal subsequently fell as a prospect when he struggled in his one year at college. He was a poor rebounder for someone his size, his assist rate was awful, and shot-blocking numbers were nothing special. But he’d also been dealing with some personal issues, and his life hasn’t been stable: he’s originally from Haiti, which he left after the devastating earthquake in 2010 that caused his family’s home to collapse with everyone inside. He was trapped for a while, his legs pinned, and for a few weeks he couldn’t walk. But he soon transferred to high schools in the United States and became a highly coveted prospect.

How is he finding success now? He’s shooting frequently with high shooting percentages from most places, and somehow his rebounding numbers are solid now. I’d also say projecting a player based on a handful of games in college is an imprecise system, so let’s not be shocked when someone outperforms expectations. But let’s temper expectations a bit — what’s driving his impressive PER, which is what a lot of people are citing when they discuss him, are his unsustainably high shooting percentages. He has a high usage rate partly because he’s jacking so many midrange shots. Sure, he was known as an outside shooting prospect, but he’s usually in the low 70’s percentage-wise from the free-throw line and he’s not going to keep shooting over 60 percent from ten feet and out.

You can see his shooting stroke in this clip. He has a high, smooth release, and scouts have always seen him as a decent shooter. He’s great at attacking the basket too — here’s a clip where he grabs an errant offensive rebound and finishes high off the glass inside. In that game against the Spurs, he did well on the boards despite San Antonio’s strength there. For example, here’s a quality rebound he had between two Spurs players. Then there’s this play where he defended well vertically using his length, and grabbed the rebound afterwards. My working theory is that like a lot of young big men the constrained environment of the college game — the ridiculous foul limit, the cramped spacing, the inexperienced guards — limited his development and he had an off year after a rocky few years adapting to a new country and new schools. I don’t think he’s quite as good as some of the numbers suggest right now, but he could be a keeper, and he’s lucky there’s now an opportunity to play in Sacramento.

V-Stats: Efficiency and usage for high-scoring wings

A while ago, I attempted to answer some of the assumptions and stereotypes about legends like David Robinson and Hakeem Olajuwon. Both players put up spectacular numbers, but Robinson’s were even better, by objective standards. However, the rationale was that his game suffered significantly against better teams, while the Dream was relatively better. Besides looking at playoff games, which offer a smaller data sample, I thought I’d try to find a thorough and quantitative method to test these assertions. This is what I call a “v-stat:” look at how a player’s offensive efficiency, or any other stat, varies based on the opposing team’s own defensive rating. Thus, you can quantify how well a player’s offensive game responds to better defenses.

For this article, I’m going to concentrate on a group of high-scoring wings because I imagine what’s what people are most interested in anyhow: Michael Jordan, Kobe Bryant, LeBron James, Allen Iverson, Dwyane Wade, and Clyde Drexler. Also, the stat of interest here is a player’s individual offensive efficiency via Basketball-Reference. I imagine people will counter, But won’t certain players take more shots against better teams, which will then confound the analysis? I’ll supply the same analysis for usage rate too, and you’ll see why I’m running offensive efficiency on its own.

Here’s the basic structure: calculate a player’s relative offensive efficiency for every game via the gamelogs. Relative means that I subtract the season average offensive efficiency from every individual game’s offensive efficiency for the player. Then I do something similar for the opposing team: I subtract the season average for defensive efficiency from the team’s own adjusted defensive efficiency for the corresponding season, e.g. if the league average is 105 points per 100 possessions, then a team with a defensive rating of 100 would be -5.

Then you just predict the player’s relative offensive efficiency with the opponent defensive efficiency. For simplicity, I’m only using weighted linear regression here; but feel free to adjust the methodology as you so desire. Theoretically, the relationship should be 1-to-1; that is, for every point per 100 possessions the defense normally allows in a season, the player should see a net gain of 1 point per 100 of their own possessions.

For the data, I’m pulling from Basketball-Reference, and as a gift I’m supplying the code so you can replicate the results yourself. You can find it on my Github page as the “vstats” file; you’ll need the “teamnames” file as well. Since I’m bridging team opponent stats and individual player stats, I need a key because they both use different team name structures (e.g. CHI versus Chicago Bulls.) There is one major limitation, unfortunately: this only works for every season on the team ratings page, so this is only for 1986 and on (Scraping the main season page is not as easy as it’s been in the past, and I wanted the file to run completely independently as much as possible. You could create your own data by manually grabbing the team stats for previous seasons, however, or perhaps from another kind soul who already has it.) Additionally, the code is flexible enough that you just change the player and season inputs yourself, and it’ll spit out the results just a few seconds later.

I’ve already run the code for the players mentioned previously, and the results are in a summary table below. First of all, for efficiency, a v-stat of 1 is the natural average. Anything higher means the player’s efficiency responds more heavily based on the opponent, and vice versa. Consequently, a high v-stat means a player is expected to lose more of his efficiency than the average player versus an elite defense. You can see why I’m focusing on efficiency though; usage rates are remarkably stable as the defenses change. For most players — the exceptions are LeBron James, Vince Carter, and Michael Jordan — usage rates do not significantly change. This makes sense: efficiency is free to change, but a team’s usage rate is always stuck at 100 percent — someone needs to shoot.

Table: v-stats (variable offensive efficiency and usage)

PlayerSeasonsOff. Eff.DRtg: -5Usage%DRtg: -5
Allen Iverson1999 to 20060.92-4.59-0.040.26
Ray Allen1999 to 20100.96-4.78-0.171.00
LeBron James2006 to 20171.02-5.09-0.211.29
Paul Pierce2001 to 20131.03-5.13-0.080.46
Vince Carter2000 to 20061.03-5.14-0.191.14
Michael Jordan1987 to 19981.07-5.34-0.342.07
Tracy McGrady2001 to 20081.12-5.58-0.020.11
Clyde Drexler1986 to 19951.38-6.890.000.03
Kobe Bryant2000 to 20131.40-6.98-0.110.67
Dwyane Wade2006 to 20141.60-7.98-0.040.26

*DRtg: -5 is the response the variable has to an elite defense (-5 def. rating)

You can see that Allen Iverson has the lowest v-stat for offensive efficiency, but his usage rate essentially doesn’t change against better defenses. Ray Allen, however, shoots more often as defenses improve — and those two are the only ones with a v-stat under 1. For most star scorers, it appears they’re slightly less efficient than expected as their opponent’s defensive strength changes, but they usually shoot more often too. I’m sure people assumed that Jordan would be the king of this stat too, but he’s pretty average when it comes to efficiency; he’s mortal. And yes, LeBron is arguably more impressive here. Michael does shoot more often, but I have to stress something here: the usage rate v-stats even for Jordan are modest.

I don’t see too much of a pattern either with the type of scorer. For some reason, Clyde Drexler, Dwyane Wade, and — please don’t kill me Laker fans — Kobe Bryant look especially poor here. Drexler and Wade are slashers and fast-break nightmares, but so was Iverson, and Kobe’s game at least stylistically was pretty similar to Jordan’s. Those guys are winners too, as they all have titles and both Kobe and Wade are known for their big-game moments, so this was perplexing. You can see a graph of Wade’s games below. You may notice an outlier at the bottom that one could blame for pulling the linear regression in one direction, but that was a six-minute game, the regression was weighted by playing time, and I repeated the analysis without that game. Also, for a comparison I have Jordan’s graph too. It’s not easy to see, but Jordan’s curve is flatter.

I should stress something here, however: this data is noisy. If you look at things season by season, you’ll notice the v-stats jump around a lot. The standard error for offensive efficiency ranged from 0.18 to 0.24, where, remember, the average is around 1. In fact, for most players you can assume the v-stat for efficiency is 1. It’s only statistically significant for players like Wade, who are further away from 1.

Next: Nylon Calculus -- Quantifying the impact of length

Like the dreaded topic of “clutchness,” I believe this is another area where it’s a repeatable skill for NBA players with a few exceptions, provided you have several years of data. For the most part, star scorers behave quite predictably and normally as defenses change. But there’s another test of a scorer’s mettle, and the same methodology here can address the question: do same scorers function differently in the playoffs, even after adjustments? Perhaps we can tackle that issue another week. For now, let’s ponder on why Wade and Iverson are so different, and if it even matters.