As a follow-up to the piece on Villanova last week, I thought it would be useful to look not just at what predicts team shooting percentage going forward, but how those inputs vary during the season.
First, let’s look at what team stats predict offensive 3-point and 2-point percentage, respectively.
As mentioned last week, at this point in the season an offense’s 2-point percentage is about as useful as its 3-point percentage in predicting how well it will shoot 3’s going forward. As the season continues, a team’s 3-point percentage history continues to become a more useful predictor. Still, even later in the season, 2-point percentage and free-throw percentage combined are about as good as predicting 3-point percentage as 3-point percentage itself.
Now, you won’t hear anybody say, “Hey look at those great 2-point and free-throw numbers. This team can really shoot it from the outside.” But if you really care about whether a team is a good outside shooting team, then that should be a consideration. Especially early in the season, but even late in the season, too.
There is much less trickery needed to assess 2-point defense. To make predictions, it’s a good idea to try to model the process. It just so happens that 2-point percentage is a pretty decent representation of the process. If a team is at one end of the 2-point spectrum, it probably isn’t a fluke. Even if it’s after three or four games. It’s difficult to go through a 2-point slump because if you’re good at twos, some of the shots you’re taking are virtually unmissable.
Conversely, if you’re not a good 2-point shooting team, then a large chunk of your shots are probably contested from outside the restricted area. You might get hot on those on some nights, but it’s tough. So a lot of the process is already baked into 2-point percentage and the other box score stats aren’t going to provide much additional information to project the future.
With 3-pointers, there are no gimmes, even for Micah Mason. Hence, the possibility for multi-game slumps on the team level, and even longer slumps on the player level. Thus, actual 3-point shooting tells us less about its own process than 2-point shooting.
Things common to optimistic early-season projections for both 2-point and 3-point percentage are 3-point rate and assist percentage. Which suggests that in the absence of other information, offenses having shooters and being able to score off the pass are better than offenses that don’t have those things. Of course, at some point there are enough results (especially with 2-point percentage) that an offense can prove that to be false.
While I was working on this, I couldn’t resist taking at look at what predicts shooting defense.
Predicting future 3-point defense is unsurprisingly, kind of foolish. What’s not shown here is the absolute predictability of each shooting percentage. For example, predictions of 3-point percentage allowed made on January 1 have a correlation of 0.18 with the eventual reality, while the correlation for predictions for 2-point percentage allowed is 0.43.
If you insist on assessing a team’s 3-point defense, don’t start with a team’s past 3-point percentage defense. Look at its 2-point percentage defense. During the middle of the season, both 3-point percentage and 3-point attempt rate actually have a bit value, too. (Though again, predictions of 3-point defense are much more noisy than 2-point defense.)
If opponents are able to make 2’s then they should be shooting more 3’s. And if they aren’t shooting more 3’s then the defense is doing something to prevent good looks. And I think that’s why we see some signal in the middle of the season. But I think the bigger takeaway is that future 3-point defense is much more likely to follow a team’s past 2-point defense than its past 3-point defense.
Which is to say, I’m worried about West Virginia, who currently sports the country’s best 3-point percentage allowed along with a 2-point defense ranked 221st. That 3-point percentage is going to regress to the mean, and based on history, it might actually go through the mean.