Shot Selection and Team Context

With the season just days away[1. Are we there yet? Are we there yet? Are we there yet?], Win Projection Season is in full swing. As folks like Andrew who put a lot of time and effort into their predictions know, there are a lot of elements which can confound the best models. Stuff like injuries and playing time can be guessed at, but there is still a degree of unpredictability to who goes down when and exactly how coaches employ their rotation. The future unknowns of those sorts of things are simply par for the course.

Another area of great uncertainty is how to deal with changed context. For all the ballyhoo over certain specific[1. Usually Canadian] high-profile instances, we’ve gotten reasonably good at describing how good players were in situ and thus if nothing else changed have a solid basis for predicting how well they will perform in the coming year. However, a league without player movement would be pretty boring and thankfully that isn’t the situation of the modern NBA. So accounting for what happens when players change teams or teammates[2. Any holdovers from 2013-14 to last year on Cleveland’s roster can attest to how the context in which they operated changed from one year to the next by virtue of one or two notable additions.] becomes of utmost importance. This is where the “how’s” that most holistic models by nature can’t address become a problem.

Most knowledgeable basketball fans can speak intelligently one why or how playing with LeBron James made the game easier for any of Tristan Thompson, J.R. Smith, Timo Mozgov and so on[3. Possibly even Kevin Love as well, but other things like a lingering back injury or just odd offensive fit might have been confounding factors.]. But quantifying that ease becomes much harder. How much of what might be termed “improved shot selection” is a result of James drawing more attention? How much is the rest of the Cavs finding themselves with “live grenades” far less often? How much might it represent real improvement in the player’s skill level either in terms of decision-making[4. Stop taking those 20 foot pullups with 14 on the shot clock, Bradley.] or ability to shake free of defenders to allow for easier shots from the same spots? Does coaching or scheme play a role and if so, how much of one? I won’t go so far as to say the breakdown among these factors is unknowable, but it isn’t much known either, at least from a numeric perspective.

That uncertainty is a problem, because inherent in predicting and projecting is accounting for that new context. One of the more heralded free agent moves of the summer was DeMarre Carroll going from the Hawks to the Raptors. Carroll has certainly made himself into a much more than serviceable player, and flashed some strong offensive skills over the last few seasons. But he also feasted on open shots created by the Hawks Voltron of Awesomeness. Toronto, not so much, as he’s going from a team who assisted the highest percentage of shot attempts to the team at the very other end. Isolating just the jumpshots taken by wings on every squad can give a degree of illustration for this change in context:

What is clear is that Atlanta wings took a lot of high value shots[1. Neck-and-neck with Philly in terms of the openness and distribution of shots, but there was an, uhm, slight difference in the talent of the shooters in each case.] whereas the Raptors were alongside offensive juggernauts like the Lakers, Knicks, Wolves and Hornets in terms of (not) generating clean looks for their shooting guards and small forwards. Some of this is undoubtedly talent and preference. DeMar DeRozan and Lou Williams like to shoot contested twos, so Toronto shot a lot of pullup twos. On the other hand, Kyle Korver was among the more “guarded” three point shooters in the league and Atlanta’s overall mix was still very high quality despite Korver taking, and quite often making, heavily contested threes with regularity. It is probably fair to suggest it’s some degree of both, which means Carroll is moving into what looks to be a more difficult shooting environment. Given that change, it’s hard to not foresee some decline in terms of his efficiency. How much? That’s the question of the day.

When speaking of context, it’s not just a player’s teammates, but the opposition as well. A challenge often faced by players moving from bench roles to playing big minutes with the starters is they are playing against starters. A player might be able to beat up on less skilled, less cohesive bench units, but find themselves far less effective against the pick of the opposing litter. There is no particular reason to suspect this difficulty affects all players similarly. As was the case in terms of big men bullying smaller defenders, what might be termed the shape of players’ skill curves varies greatly.  

Looking ahead to this season, a team which might run afoul of this notion is the Boston Celtics. Many of my colleagues at TNC (see also grand overlord Ian Levy here) are quite optimistic about the C’s chances this year. I’m a bit more hesitant, in part because once whichever group of “not that bad” players Brad Stevens picks to start lines up against opposing starters on a nightly basis, “not that bad” suddenly becomes “at a disadvantage nearly every night.” This is particularly concerning when looking at players expected to generate much of Boston’s offense.  

If we break opposing units into two rough categories: Starters — lineups containing four or all five starters, and Bench — groups with two or fewer starters, we can draw some rough comparisons. Certainly, not all starter and bench units are created equal[6. And to be fair, on/off methods like RAPM/RPM do attempt to account for opponent strength. But they do so in a largely linear fashion while what I’m suggesting is there might be something of a tipping point effect.], but as a general rule, starters are tougher opponents than the bench.  

Among players with at least 500 FGA last season, three of the 18 who saw the biggest drops in efficiency when facing starters were Celtics. Two of these three. Evan Turner and Isaiah Thomas[3. Kelly Olynyk is the third.] look to be among the main engines of Boston’s offense. Digging a little deeper, it looks even worse. During Thomas’ time in Boston, when the Celtics ripped off a 14-7 record with Lil Zeke in the lineup, he had a 54.5% eFG% against bench units but shot only 41.5% against starters. For the Celtics to make as big a leap as some are suggesting[5. Cough*NATHAN*Cough.], that’s not going to cut it as the team probably can’t afford to “save” him to only beat up on overmatched reserves, but they can’t afford their main offensive threat to perform that poorly for long stretches.

Another player who faces a big step up in general opponent quality will be Nikola Mirotic, who took almost 2.5 times as many shots against bench units as starters. Oklahoma City appears to be planning on bringing Enes Kanter off the bench, despite his big new contract, which might be prudent given the combination of his defensive limitations and the fact that he saw a fairly steep drop in effectiveness against starter-heavy lineups as well.

Of course, none of this is pre-determined. When the data starts to get sliced this finely, sample sizes become even more of a worry. Further, the two effects identified above often pull in different directions, and there are plenty of other things at work as well. It can be tempting to just throw one’s hands up and ignore all this confusing and contradictory data, but the wiser course of action is to identify the things we are more sure about, while acknowledging just how sizable the unknowns can be.