The 3-point line is a lottery
02.23.12
Continuing the discussion from yesterday, let’s look at more scatterplots! We’ve looked exclusively at defense so far and in order to get a better understanding of how defense works, we need to look at both sides of the ball. Because clearly the offense has some control over what happens on each possession as well.
To get our bearings, let’s look at something that the defense should have more control over compared to the offense - blocked shots. In all of the subsequent pairs of plots, the offensive plot will be on the left and the defensive version will be on the right. Here’s blocked shot percentage:
This makes some sense. Defensive block percentage is a better predictor of the future than offensive block percentage. Basically, a defense’s ability to block shots is greater than the offense’s ability to avoid them. Anthony Davis and Fab Melo will generally get theirs regardless of opponent. However, the offense does have some skill to avoid blocked shots and, since we ignore three-point attempts in this calculation, I expect most of that ability depends on being tall.
Let’s check out free throw percentage.
Some positive slope for the defense but overall these plots look reasonable.
Now for 2-point percentage. If the defense doesn’t have influence on FT% or 3P%, it better have some influence on this.
Alright, life is starting to make sense. There are nearly equal values for R2 and the slopes of the regression are almost identical. In fact, the defensive slope is just a wee bit greater. This is an equal battle between the offense and the defense. Both clearly have skill at influencing 2P%.
Before we get to the payoff, let’s compare 3PA%.
3PA% is one of the most predictable shooting-related stats around. We’ve already established the defense has influence over this, but a major part of an offense’s style is how often they want to shoot threes, so they have influence, too. More influence than the defense, actually.
Now the payoff: 3P%.
Oh dear. The defensive plot is just a random scattering of data, as has been discussed previously, but the offensive version isn’t much better. If you shot 45% in the first half of the 2011 conference season, you’d be expected to shoot about 35% in the second half. If you shot 25% in the first half, you’d be expected to shoot 33% in the second half. A difference you couldn’t notice with your eyes. I don’t know exactly what implications this has on strategy, but when evenly-matched teams get together, action happening beyond the 3-point line is like a lottery. You take a shot and a third of the time you have success.
Of course, there are games where good shooters get a bunch of open looks and they make more than a third of their attempts. You wouldn’t call that luck. (Although, from the offense’s perspective the shooters enjoyed some good fortune to be left open.) But in the long run, that kind of stuff evens out a lot more than I would have thought. In the end, coaches have to decide how much they want to play the lottery. Sometimes, it will be advantageous for the defense to let the offense play the lottery a bunch and sometimes it won’t. The offense has to consider the same things.
John Beilein loves to have his offense play the lottery while Roy Williams hates it. Tubby Smith is cool with opposing offenses playing scratch-off tickets, while Rick Majerus forbids it. If you’re Shaka Smart and you have to play a series of teams better than you in March, it stands to reason that you’ll be willing to play the lottery on offense all game long, six games in a row.
Each philosophy may be correct given the alternatives inside the arc. There are obviously a lot of factors for a coach to consider in making these decisions, many more than just what a team’s 2-point accuracy might be. Nonetheless, it’s an interesting way to think about how defenses (and offenses) work.
How defense works: an investigation
02.22.12
[Note: the graphs originally posted Friday were slightly in error. The corrected data has been posted and does not change the conclusions drawn. My apologies for the mistake.]
The data from last Friday’s post has got me wondering about a lot of things and I hope it did the same for you. Mainly, how does defense work?
First, let’s try to noodle through an explanation of why a team would have no skill in affecting its opponents’ 3P%. If you are on offense, how do you decide when to take a three-pointer? Clearly, there’s some standard for launching a shot. No coach goes into a game telling his team, “we’re going to take 15 3-pointers tonight, regardless of what the defense gives us!” Players have a threshold for when to shoot.
On nights where the defense makes pressuring the ball a priority and de-emphasizes help defense, there will be fewer looks available that meet the average shooter’s standard. And on those nights, shooters aren’t going to take the same number of three-pointers they always do. Thus, a reasonable theory is that the quality of looks doesn’t change much from night to night.
Essentially what you get with defensive three-point percentage is this: When a shooter is open enough to shoot, how many threes do they make? When viewed this way, you can see why we have something close to a defense-free statistic here. I use that term specifically so that it’s analogous to its cousin term, “tempo-free statistic”. Consider these descriptions:
Points per possession: Describes how often a team scores when it has a chance (i.e. possession). This removes tempo from the qualitative assessment of an offense.
3-point percentage: Describes how accurate a team is from three-point range, given the opportunities it has to shoot. This removes the ability of the defense from the qualitative assessment of shooting ability.
The defense’s primary role in three-point defense is to limit the number of looks an opposing team gets. Perhaps saying that a team has no control over its opponents’ 3P% is a stretch. The obvious counter example would be Syracuse, who allows a lot of three-point attempts and generally keeps its opponents below the national average in 3P%. Maybe they have some skill, but keep in mind that the Orange typically play a weaker than average non-conference schedule, almost exclusively at home, and follow that up playing in a conference that historically does not have great shooters.
Over the last decade, here’s how the Big East ranked among the 32 conferences in 3P% during conference play: 25, 28, 24, 15, 24, 27, 23, 20, 22, 30. (Conference rankings now available to subscribers, BTW.) I have a feeling once we filter out the poor shooting ability of Syracuse’s opponents, we would find their affect on opponents’ 3P% is very small. Admittedly, it’s poor form to merely speculate on this, and I do plan to investigate this more thoroughly in the future.
Duke is probably a stronger counterexample, generally putting up good defensive 3P% (and 3PA%) numbers with half their schedule in a conference that occasionally has good shooter. And it stands to reason that a team’s standard for taking a three-pointer could change based on the opponent. If it’s increasingly difficult to score inside the arc as it is against Syracuse, then taking more difficult three-pointers is a reasonable solution.
For a team that is so suffocating on the perimeter like Duke, maybe shooters lower their standard out of the frustration of just wanting to take shots. Or perhaps teams like Duke and Syracuse that play from ahead the majority of the time enjoy the benefit of their opponents launching more questionable threes in order to catch up. It’s possible these factors exist. However, I think it’s clear from the data that these influences are a lot smaller than conventional wisdom allows.
But Ken, this cannot be right. Shooters obviously make more shots when they are wide open.
No argument here, but how many defenses are leaving shooters wide open very often? For any team, no matter how many attempts are allowed, opponents are taking a mix of wide open shots and contested shots and all flavors in between. Over the course of a few games the average quality of three-point shots tends to even out.
And no coach encourages his team to allow uncontested three-pointers. Actually, there might be one. Chattanooga coach John Shulman is going to go on the Mount Rushmore of wacky coaches (right next to Denver’s Joe Scott) when he retires because his team’s defensive stats suggest he coaches as if the three-point line doesn’t exist. Opponents have consistently taken nearly half of their shot attempts from beyond the arc and yet their 3P% is never terribly far from average.
People that are unaware of 3PA% (which is to say nearly everyone) are missing a very telling statistic that explains a lot of how defense works. It’s infinitely more useful than defensive 3P%, anyways. Can coaches use this to their advantage? I’m not sure, except to say that at first glance I think Shulman is crazy for running the system he does. However, I like having him around. The strategic diversity that college hoops offers is part of what make D-I hoops so much more interesting than the NBA to me. I tend to think K, Randy Bennett, and Rick Majerus have it right, but there are national-championship winning coaches (Jim Boeheim and Tubby Smith are two) that run systems that allow a high amount of opponents’ three-point attempts and I’m guessing they have good reasons.
There’s probably much more utility for this information to be used analytically than strategically. Tomorrow, I’ll publish some plots for other team stats to continue this perspective on how both defense and offense work. Would you believe that on a team level, offenses have surprisingly little control over their own 3P%?
Defense has little control over opponents’ 3P%
02.17.12
I took last season’s conference-only data for every team and split it into two halves. Then I compared each team’s opponents’ 3P% between the first half and second half of the conference season. I did the same for opponents’ three-point attempt percentage (their percentage of field-goal attempts that are from three-point range). The following plots of that data should make it clear that opponents’ three-point accuracy is largely out of a team’s control.
