This is the week of Memphis asserting their dominance, Houston fading without Dwight Howard, Cleveland discovering their mojo, and the Toronto juggernaut being tested with DeMar DeRozan’s injury. Somehow, the playoff standings are an inverse mockery with the Cavaliers, Bulls, Spurs, and Clippers closer to the bottom with Memphis, Golden State, Houston and Toronto looking like title contenders while surprising teams Sacramento and Milwaukee fight for the playoffs in the real world. It’s still early, of course, but the NBA changes quickly and we forget this when we retroactively change our opinions, some failure of post-hoc analysis where we explain the results after they happened and pretend it’s what we all should have expected. These changes are important and without them the NBA would be a boring league.
How Were We Wrong with Memphis?
Most of the basketball universe saw this as a high-40’s win team including Vegas at 48.5, ESPN at 48, FiveThirtyEight at 48 as well as well, and even my own prediction with 47 wins. That’s a lower win total than last year even though Gasol was injured; however, given their point differential, they played like a 46 win team so projections did indeed see a (small) improvement.
The most common post-hoc explanation is that Memphis was 34-13 when Marc Gasol returned from injury, a 60-win pace, and they are truly an elite team when he’s on the court. However, that excludes their slow start with their big man when they went 3-5 and their point differential wasn’t as impressive in the latter half of the season. If you adjust for strength of schedule, they had a point differential of 2.8 when Gasol was in the game including the playoffs. (If you want to point out that he only played 10 minutes in a blowout loss to San Antonio, I’ll point to the score of the game when he came out and how they were already down 7 points at the beginning of the second quarter.) Gasol, in fact, is fundamentally different than he was during their 34-13 streak last season compared to this season — he had a usage rate of 21.9 during that streak, implying a modest level of activity in the offense, and a shooting efficiency well below average. Meanwhile, his usage this season is 25.8 with exemplary efficiency.
The fact that Memphis is again a behemoth on defense is no surprise, but it’s their newfound offensive might that’s pushing them forward. Some of this might be from unsustainable shooting percentages and they’ve been pretty healthy, but they’re getting more out of their wing positions (i.e. fewer minutes from Prince) and Marc looks like a better player on offense — look at how nimble and aggressive he is in the below clip. We didn’t see those events coming and in retrospect how easily could we have? The west keeps getting more crowded at the top.
Gerald Green’s Self Alley-oop
I just thought everyone should see this play. It’s the famous McGrady self alley-oop off the glass, and it’s a wonder it isn’t used any more often because it’s surprisingly effective. Green’s not the first to do it in an NBA game, of course, and it’s actually been done by a player few would guess: Gallinari. You can see Green’s dunk in the video below. Unfortunately, Phoenix lost the game, though they were without Isaiah Thomas and they usually play a lot better with him.
The State of Basketball Statistics Eight Years Ago
The times have changed — advanced basketball stats have progressed tremendously in just eight years, which makes it all the more curious that PER is still so widely used and trusted. PER was one of the first metrics and hasn’t changed in a long time. Its weights for the variables — the value of the basketball stats it incorporated like rebounds — were more or less just guesses with little empirical research. I’ve talked about this before, but it bears repeating. Plus, I went through his archives and found a lot of interesting early research he was doing with PER and other stats. In this one, he makes an intriguing admission about how he values assists:
Also, assists are valued at one-third, since of the three acts involved in the play – getting open, making the pass and making the shot – the passer only did one.
Usage Rate = (League Pace x 40 x [FTA + TO + (AST/3) + (FTA x 0.44)]) / (Team Pace x Minutes)
If you look at the PER formula, Hollinger gives 2/3 “points” to an assist because it’s (assumed to be) one-third of a made field goal. So yes, let’s be a little more careful of how we deploy PER. It’s an out of date model. (I will note, however, the PER is better than most people think at evaluating player value because it gives a lot of weight to high volume shooters where at the time of its development few people make that adjustment. High volume shooters retain a lot of value and metrics like Win Shares undervalue these players.)
You can also see the development of usage rate, an extremely important stat we still use today. Although the results are roughly the same, Hollinger calculated it differently back then. You can see it in the above equation where assists are thrown in and the output is per 40 minutes. Usage rate now is calculated as the proportion of plays (field goals, turnovers, and free throws) that a player uses out of his team’s total plays available. Yet now we’re putting assists back into usage rate — things may change, but often only back and forth, slowly, across the reaches of time.
Speaking of change, Berri is one of the most vocal critics of plus/minus models, but he wasn’t born with this seething hatred and he actually acknowledged the defensive limitations of his metric:
November 19, 2006 at 4:07 pm
I tried to answer a few of these questions today. Let me offer a few more comments.
Fred… we can measure the ability of a team to play defense, which we incorporate in our measure of Wins Produced, but obviously the statistics do not give us a measure of how each player on a team played defense. In the future I think we can take some of the plus-minus statistics to supplement our model.
And while I’m at it, here’s the first mention of the Kobe Bryant assist that I could find.
The Big Penguin’s Slide
As a young center who’s put up impressive stats his first two seasons coming into the season with a new coach who found success with a similar center years ago, we have high hopes for Andre Drummond but he’s been horrific for most of the season. His shooting efficiency has fallen from one of the best last season to one of the worst even with better free-throw shooting — he’s also shooting a lot more often. His shot chart hasn’t changed too much with a few more attempts in the 3-10 foot range. But his field-goal percentage near the rim has plummeted, which kills a player like him with no range. He’s also still scoring off of offensive rebounds. Per stats.NBA.com, he’s getting 6.7 points per 100 possessions on second chance points, compared to 8.3 last season and 6.1 the season before.
Drummond had a nice stretch of games last week when he looked “back” where his efficiency was like the old Andre we knew, but unfortunately he just suffered through a 1/8 game on Sunday. What’s the cause? Stan Van Gundy and his staff are posting him up with high frequency, force feeding for a play that looks like something out of an Al Jefferson possession. In the three before Sunday, he was being used more in the pick and roll and posted up infrequently. Since he can’t dribble well and is awkward moving too much before he shoots, the results has been very disappointing in the low post. You can see one example from the most recent game below where he has all the grace of a blind baby deer:
He airballs another hook shot attempt, and in retrospect it seems silly to attack Golden State’s defense by trying to post up Bogut with Drummond:
But his post moves are actually fairly decent sometimes; he just doesn’t have the requisite soft touch to convert them:
Then again, by forcing him to be a creator you get ugliness like this:
Do not judge his shooting percentages solely on their own. He’s being forced to become a post player, and it’s hurting his numbers. This appears to be a stealth tanking maneuver, but I imagine Van Gundy is taking the long-term view and developing his skills while he’s young. When the Pistons go back to using Drummond more in pick and roll’s, he’s still a devastating finisher as you can see from his games last week and in the clip below. It doesn’t entirely explain his low percentages, but some of that is noise and some of it may be a confidence issue after being thrust into a weird role.
San Antonio’s Give-and-Go
There’s little to say here: I just really love this play. The Spurs point guard brings up the ball quickly, throws it to Manu (usually) on the wing, sprints to the basket while the defender has little time to react, and Manu then fires a bullet to the point guard for a quick two-points. What’s new then here is that they’re doing this with Cory Joseph like a rite of passage. He’s becoming one of them.
The Other Splash Brothers
Although Klay Thompson and Curry get all the press for outside shooting backcourts, right now Lillard and Wesley Matthews have more total three-pointers and they weren’t too far behind last season. Lillard is basically a Curry-lite, a point guard taking a huge number of three’s with a good percentage and a ton off the dribble. Matthews is one of the more overlooked shooting guards in the league and is following a season where he made 201.
In fact, Lillard is currently ahead of Curry in three-pointers by age or season in league. Most people assume Curry will crush the all-time record and no one else has a chance, but breaking a career totals record is about consistency and health as much as anything, which is where Lillard has an advantage. You can see an example below and why he’s so difficult to guard:
Frontier Line: Steals and Blocks
I imagine that most people think of graphs in advanced stats articles as something to please the masses and to make the ideas presented therein more palatable and less technical. But graphing is an extremely important practice at the research level because it’s easier to see patterns spatially and it’s easier to organize and understand information.
For example, when looking for relationships when building a player metric, I stumbled on an intriguing graph between two simple things: blocks and steals. What you see below is what’s known as a frontlier line and you can see an example of another kind with scoring volume and scoring efficiency here. I would not have guessed that the relationship with steals and blocks is such that there’s a linear relationship for the total amount. In some ways, this should be obvious: the extreme shotblockers like Manute Bol come up with few steals, while the ultimate ballhawks like Chris Paul or McMillan come up with few blocks.
By this method, David Robinson’s 1992 season is the most impressive — no one else had a combination of steals and blocks like he had. But the other members of the 2 BPG/2 SPG club — Gerald Wallace and numerous seasons from Olajuwon — are right there too along with small forward shotblocking wizard Kirilenko and the underrated Bo Outlaw. I’m, however, a bit intrigued by the large gap in the lower right-hand side near the frontline line. I don’t know if this means McMillan is the greatest outlier and the line should be adjusted or if we’re, simply by chance, just lacking some extreme high steal seasons — maybe it’s due to conservative defensive schemes. I’m simply not sure.
Humans are pattern-finding machines, and when we find these non-random patterns in nature it’s aesthetically pleasing. There’s little utility for a graph like this, but it’s beautiful in its own way — stats feel a little less chaotic now.