Kevin Pelton is one of the smartest guys covering the NBA, and his SCHOENE projection system is one of the best predictive tools in NBA analytics. He was nice enough to answer a few questions about the upcoming NBA season.
For the uninitiated, Pelton spent years at Basketball Prospectus, where he developed the SCHOENE system (pronounced SHAY-nee), before moving to ESPN. He's written for 82games, HoopsWorld, and Sports Illustrated, and consulted for the Indiana Pacers; basically, he's a smart dude. Here's Kevin on some of the surprising things in this year's projections—which you can find on Insider—how they actually work, and the new directions that are opening up in basketball analytics.
There are plenty of in-depth explanations, but can you give us a bite-sized rundown of what SCHOENE is and how it works?
Basically, it's a comprehensive NBA projection system where both team and player projections interact and work together. I start by using performance over the last three seasons to generate similar players at the same age, and use these predecessors to estimate the impact of aging. Using my best guess at playing time, I add these projections to team factors to come up with team stat lines that yield projected win totals.
So who should we be looking at this year as the powerhouses in the league, going by SCHOENE—does it mostly break down by conventional wisdom, or are there some more divergent projections?
There's a lot of overlap in terms of the nine teams with legitimate championship expectations. The top four in the East, including three teams close together after Miami, and the top five in the West all generally match conventional wisdom. The disagreement is on the next tier, with SCHOENE saying Golden State and New York [Ed.: 42 and 37 wins, respectively] aren't as close to that top group as generally believed and marking Detroit and Minnesota [50 and 52] as sleepers to join them. The other interesting conclusion from SCHOENE is that five of the top six teams are in the West, with the Heat rounding out that group.
What kinds of players and teams does SCHOENE still struggle with? Were there any genuine surprises this year?
Based on SCHOENE's history and the way the teams are built, I wasn't surprised about the Pistons and the Timberwolves rating where they did, and the Warriors didn't really surprise me. So the Knicks projection was certainly the most surprising to me.
As far as where SCHOENE struggles, there are a couple of areas. Projecting injury-prone players is certainly always tricky, and when several of them are stacked on the same team (most notably Cleveland, but also perhaps Golden State), it makes a big difference. At the team level, projecting defense is inevitably more difficult than offense. For example, SCHOENE assumes that Philadelphia will move toward average with a coaching change and high roster turnover rather than the 76ers likely becoming one of the worst defenses in the league.
Who's the single most frustrating player to try to quantify, for whatever reason?
Greg Oden. To me, he's a complete wild card because he was so phenomenal when he last played, but that was multiple surgeries ago. There's not really a lot of precedent for a player missing this much time due to injury and returning to the league, so the numbers don't help much in his case.
How do the projections do with identifying players who are a poor fit within their given systems/teams?
There are some broad attempts to account for fitting players together in terms of how they distribute plays (and how this in turn affects efficiency), the value of assists and diminishing returns in rebounding, but I wouldn't say you can use SCHOENE to specifically identify bad fits.
Which bad traits in a young player are good indicators for future improvement, a la strikeout rate with Nate Silver's PECOTA? (Is it still mainly turnovers?)
In general, it's the ability to create offense. That manifests itself in usage rate, unassisted shots (not part of SCHOENE) and yes, turnovers. The latter is the most counterintuitive because obviously they are negative in the short term. In the long term, there is some evidence that high-turnover players develop more, presumably because they're trying to make plays and eventually those mistakes will become successes.
What are the big black boxes still left in basketball analytics? Defense has always been the go-to, but SportVU seems to be sorting that out. Is there anything left, or is it just a matter of synthesizing what we have?
As far as things that can't be studied using SportVU, I'd point to the role of coaching, which has barely been quantified at all. We have a pretty good sense of which coaches are better schematically on defense—and that coaches have more impact at the defensive end—but even that is tricky to value, and there's all sorts of other things coaches do like managing rotations that we haven't even begun to quantify.
On the other side, what is just now being brought out by stats, especially the new motion tracking stuff?
Besides providing more information about defense, I think one of the interesting next steps will be using player positioning on the floor to show how players affect their teammates when they don't have the basketball on offense, whether by setting screens, cutting or simply keeping their defenders honest. These contributions have only been valued in a general sense using plus-minus data.
Does anything out there seem to you like it might be snake oil? You can speak broadly, if that helps, but where are the dead ends?
I'd be hesitant to label anything a dead end at this point. We still have too much to learn. In that spirit, I think the snake oil is less about specific ideas and more about being careful to avoid overconfidence in conclusions that isn't supported by the numbers.
What are the early thoughts about how SportVU in every arena will change NBA analytics? Does it somewhat outmode the role of the traditional stat-tracking? Or is it a moment where everyone looks around and says, Oh god, we need even more people on this?
I'd say it will change what teams need slightly. More programmers who can put the data in a usable format will probably be necessary, while analysts who understand the game and can make sense of what they're seeing will likely become more valuable.
You can follow Kevin on Twitter here.