Does Defense Actually Get Tighter in the Clutch?

facebooktwitterreddit

Apr 27, 2014; Oakland, CA, USA; Los Angeles Clippers guard Jamal Crawford (11) fouls Golden State Warriors forward Andre Iguodala (9) during the second quarter of game four of the first round of the 2014 NBA Playoffs at Oracle Arena. Mandatory Credit: Kelley L Cox-USA TODAY Sports

On the surface, the answer to this question seems obvious—of course it does. Players become more engaged on defense late in a close game, while defenses concentrate on eliminating shots taken at the rim. Previous research has shown that both eFG% and shots near the rim both go down in clutch situations. But is this actually true? With the new SportVU data, we can answer this question. 

In this post, I’m defining clutch slightly differently than it’s normally defined. Unfortunately, we don’t have complete data on the score margin after each shot so we have to use final margin[1. Sadly, there are differences in the game clock of the play by play data and the SportVU shot logs making it harder to merge these files and acquire accurate score margins after each shot.]. This means our modified definition of clutch will be the last 5 minutes of the 4th quarter or OT within a game that ends with a final margin of +/- 5[2. This includes all of OT, which might be a bit dicey but for what it’s worth, the largest final margin of victory in an OT game was 10 points so I think including all of OT should be OK.]. While this is not ideal, it’s actually more restrictive than the typical definition of clutch due to the final margin being +/- 5 as opposed to the score margin being +/- 5[3. A lot of games may end up with larger final margins due to the free throws at the end of the game. Unfortunately, we can’t include most of these games which is why our definition is a bit more restrictive. However, it also means we’re likely to be looking at games that were truly close down the stretch.] which means at the very worst, we’re underestimating the number of shots in clutch time instead of overestimating. Provided the sample size is large enough (which it is), this is better than overestimating and biasing the sample.

With that out of the way, does defense actually get tighter in the clutch?

Well it does. Sort of. The increase in tight and very tight defense isn’t a lot- only about a 3% difference in total. We can also look at this another way: given our definition of clutch, which is the last five minutes of a close game, we can expect a team will have about nine FGAs in those five minutes of clutch time (total field goal attempts / total minutes played *5 minutes). And out of those nine FGAs, if about 3 percent are more tightly contested than normally, that means we’re looking at about 0.27 more FGAs that are contested. Multiply this by the average points per FGA (1.00) and you get a value of 0.27 points, which is approximately the value of the points saved in clutch time versus non-clutch time by playing tighter defense[4. There’s a better way to calculate this but as an approximation, this is good enough.]. In the grand scheme of things, that’s a pretty small difference and so I think it’s safe to say that defense isn’t actually getting any tighter in the clutch.

But wait, numerous studies have shown that eFG% goes down in the clutch, so if that’s the case and teams aren’t playing tighter defense, why does eFG% go down? Further proof of this effect can be seen in the chart below:

So for virtually every level of defense played, the average points per shot goes down. Why? It turns out there are a few different reasons. First, let’s look at the distribution of shots by the shot clock:

In the clutch, more shots are taken later in the shot clock. Is there a particular reason for this? Earlier we saw that defense wasn’t much tighter for shots attempted in the clutch but what is possible is that for the shots not attempted, the defense is tighter. Think about it this way: every time a player receives the ball, he has a few actions he can commit. The player can either choose to dribble the ball, pass the ball, the player can turn the ball over, the player can shoot or the player can wind up at the free throw line if he’s fouled immediately. If a defender is closely guarding the player, most players are more likely to pass or dribble. And at that moment, they are passing up a shot to get a better shot.  And at that same moment, the defense is tighter but the opposing player chose not to shoot instead looking for a better shot. This leads to more shots being taken later in the shot clock as teams are taking a longer time to find shots that they are willing to take.

How about if we look at the efficiency at different levels in the shot clock for clutch time vs. non-clutch time?

Again, we still see that shots are less efficient even when controlling for the level of the shot clock. Let’s also look at attempt rates, broken down by both shot clock and shot defense:

We see that the rate of increase in FGA% in clutch time for shots tightly defended and late in the shot clock is larger than most other types. Conversely, we also see that open shots earlier in the shot clock are harder to find in clutch time compared to non-clutch time.

But is the lower eFG% in clutch time fully explained by the shot clock? What about location? Surely teams try to discourage shots at the rim and threes while encouraging teams to take midrange shots right?

Um, not quite. There’s a large increase in the number of three-point attempts in clutch time compared to every other shot type. It appears that defenses are trying to force teams to shoot further out, instead of trying to eliminate Moreyball. It is possible an abnormal amount of these three-point attempts are coming within the last minute of the game when teams are trying to make up deficits (for example, if a team is down 4 with 40 seconds left) but it’s still surprising to see midrange shots are actually attempted less in clutch time compared to non-clutch time. And even more surprising is that midrange shots see a larger decline in attempt rate compared to shots at the rim and threes.

Let’s also take a look at the efficiency for each of these locations in clutch time vs. non-clutch time:

As expected, points per shot goes down for all locations. But for shots at the rim, the efficiency decline is less. The same goes for three-point shots. Also, we see that threes in clutch time are roughly as efficient as all shots taken in non-clutch time (i.e. a three in clutch time has the same expected value as a league average shot in non-clutch time).

Let’s dig a bit deeper and look at how location rates are distributed by shot defense:

Tightly guarded threes have the largest increase in attempt rate in clutch time vs. non-clutch time. Additionally, we see that open midrange shots are attempted less in clutch time. In a previous article of mine, I had found that contested off the dribble midrange shots were the most attempted shot type despite being the most inefficient. Well, here we see that in clutch time, contested midrange shots are attempted more often than in non-clutch time despite being very inefficient (though the difference is marginal). This is a win for defenses trying to force bad shots.

Finally, let’s look at how location rates are distributed by the shot clock:

Everything is attempted later in the shot clock in clutch time except for threes in the 24-22 second interval, which is a pretty small sample size (50 clutch time threes in the 24-22 second interval). Threes and midrange shots are more likely to be attempted later in the shot clock in clutch time compared to non-clutch time. Defenses play tighter, force opposing shooters to pass up on shots earlier in the shot clock for a potentially better shot later in the shot clock which ends up turning into shots further away from the basket because defenses are more concentrated on protecting the rim.

So what can we conclude from all of this research? As I mentioned early in the article, on shots attempted, defense isn’t much tighter in clutch time compared to non-clutch time. However, shots are generally attempted later in the shot clock in clutch time, possibly due to tighter defense on shots not attempted (open shots earlier in the shot clock are attempted less in clutch time), which in turn leads to players looking for better shots later in the shot clock. Defenses seem to do a better job protecting the rim throughout the shot clock in clutch time[5. While restricted area FGA% late in the shot clock is sightly higher in clutch time, the difference is negligible compared to other shot locations.], which leads to more midrange shots and threes attempted late in the shot clock.