Sunday, April 6, 2014

Shooting Performance - Clutch vs. Garbage Time

James Harden Rockets croppedIn this recent post, I laid out an approach for measuring clutch play in the NBA, using my win probability model as the underlying framework. In that post, I broke down a players win probability added (WPA) contributions into three components: expected WPA, clutch WPA, and garbage time WPA. The clutch WPA component measures the "excess" win probability added (or subtracted) a player has amassed due to performance in clutch situations.

Under that definition, I found that the Rockets' James Harden has the most clutch WPA so far this season. However, for this post, I will take a different approach to measuring clutch play. Rather than focus on aggregate WPA contributions, I will measure shooting performance (i.e. eFG%) in both clutch and garbage time situations, and see how they vary for each player. Are there players that "elevate their game" when the stakes are higher?

Defining Clutch and Garbage Time

Previous attempts to define clutch play have been a bit on the kludge-y side (see NBA.com's "last five minutes, within five points" definition). A win probability model allows for a less arbitrary determination of when a game is in a clutch or garbage time situation (although it is still not without its own drawbacks).

I define a clutch situation to be when the win probability added (WPA) of a given play is greater than the typical WPA for that type of play. To use an example from Friday night's double overtime game between the Heat and the Timberwolves: A turnover usually costs a team -2.1% in win probability. However, with 11 seconds to go and his team up by 1, Chase Budinger turned the ball over, dropping his teams's win probability by 10.8%. 10.8% is greater than 2.1%, so under my definition, this was a clutch situation (and I think most would agree).

Conversely, there's also "garbage time", which I'm defining as plays in which the win probability added is less than the expected WPA for that type of play. Take another game from that same night: the Warriors' 102-69 curb-stomping of the Kings. With 8:32 left in the third quarter, and the Kings down 29-68, Ray McCallum missed a pull up jump shot. A missed field goal typically costs a team -1.4% in win probability, but McCallum's miss only cost the Kings -0.1% (their win probability dropped from 0.7% to 0.6%). This would be classified as a garbage time play (and once again, I think most would agree).

Measuring Clutch Performance for the Top 25

The table below summarizes effective field goal percentage (eFG%) for the top 25 shooters in the NBA (according to field goal attempts). Shooting performance is shown for both clutch and garbage time, and the "Diff" column is the difference between a players clutch eFG% and garbage time eFG%.

Total Garbage Clutch
Player FGA eFG FGA eFG FGA eFG Diff
Carmelo Anthony 1,561 50.4% 11 730 50.8% 9 831 50.0% 13 -0.8% 19
Kevin Durant 1,499 56.9% 2 658 55.5% 4 841 58.1% 2 2.6% 8
DeMar DeRozan 1,318 45.3% 23 580 43.9% 25 738 46.5% 23 2.6% 7
LaMarcus Aldridge 1,314 45.9% 22 572 44.6% 21 742 46.8% 22 2.2% 10
Kevin Love 1,285 53.0% 7 656 54.2% 5 629 51.7% 9 -2.5% 20
Blake Griffin 1,275 53.3% 6 585 52.2% 7 690 54.2% 4 2.0% 12
Stephen Curry 1,266 55.8% 3 530 58.1% 2 736 54.2% 4 -3.9% 25
Paul George 1,263 49.0% 14 581 47.8% 15 682 50.1% 11 2.3% 9
Al Jefferson 1,242 51.1% 9 571 48.1% 14 671 53.7% 6 5.6% 2
John Wall 1,238 48.3% 17 510 46.7% 17 728 49.4% 15 2.7% 6
Damian Lillard 1,226 50.4% 11 556 50.8% 9 670 50.1% 11 -0.7% 18
LeBron James 1,202 61.1% 1 516 59.9% 1 686 62.1% 1 2.2% 10
Josh Smith 1,202 44.1% 25 581 45.5% 20 621 42.8% 25 -2.7% 22
Dirk Nowitzki 1,180 54.4% 4 539 56.1% 3 641 52.9% 8 -3.2% 24
Kyrie Irving 1,176 48.7% 15 542 48.6% 13 634 48.8% 16 0.2% 16
Thaddeus Young 1,169 49.1% 13 589 50.3% 11 580 47.8% 19 -2.5% 20
Monta Ellis 1,159 47.1% 18 467 47.0% 16 692 47.1% 21 0.1% 17
Klay Thompson 1,144 52.3% 8 562 51.5% 8 582 53.0% 7 1.5% 13
Rudy Gay 1,099 48.4% 16 495 46.2% 19 604 50.2% 10 4.0% 5
James Harden 1,090 53.4% 5 503 48.9% 12 587 57.3% 3 8.4% 1
Zach Randolph 1,086 46.3% 21 483 44.0% 24 603 48.1% 18 4.1% 3
Isaiah Thomas 1,069 50.9% 10 467 52.5% 6 602 49.8% 14 -2.7% 22
Jeff Green 1,067 46.5% 20 527 44.4% 22 540 48.5% 17 4.1% 3
Kemba Walker 1,065 44.8% 24 459 44.4% 22 606 45.1% 24 0.7% 15
Bradley Beal 1,049 47.1% 18 487 46.3% 18 562 47.8% 19 1.5% 14

For the most part, there doesn't seem to be dramatic differences in player performance between clutch and garbage time. In fact, according to the Fisher Exact Test, the only player on this table with a statistically significant difference in performance is, you guessed it, Frank Stallone James Harden. Harden's eFG% is a full 8.4% higher in clutch situations. For the Harden fans, this is proof of Harden's ability to step up when his team needs him most. For the Harden haters, just another example of Harden's laziness. Why, he's hardly even trying unless the game is on the line!

Lebron James has slightly better performance in the clutch this season, but not by much. But, as you can see, Lebron really doesn't need to elevate his game during the clutch. He's good at his job, regardless of the situation. Stephen Curry is an interesting case, with poorer performance in clutch situations, despite a reputation for late game heroics. But a 54.2% eFG% in the clutch still puts him at #4 out of the top 25 on this list.

It's not really binary

As I mentioned, the clutch definition above still has some drawbacks, with one being that every single play is pigeonholed into either clutch or garbage time. A more realistic definition would allow for plenty of "normal basketball" situations that are neither clutch nor garbage. One way to achieve this would be to only define clutch if the WPA was in the top quartile, as opposed to just above average. I intend to explore this definition in a future post. Also on my to-do list is to examine the persistence of clutch play. Are there players that consistently elevate their game? Or are we just chasing statistical noise? Tune in next week for the dramatic conclusion.

3 comments:

  1. Assuming you have the data, it seems like a natural thing to try to do would be to regress leverage (i.e. the impact of the next marginal point on win percentage) against the likelihood of the player missing instead of scoring (or getting to the line.) - say with logit regression. (Of course, you may have already tried that, and had no significant results.)

    ReplyDelete
    Replies
    1. I have not tried it, but I like the concept. There may be some selection bias in the results (teams playing their scrubs during garbage time, although I suppose the scrubs would be playing defense too).

      Delete
    2. Yeah, there's definitely going to be some of that. You could always incorporate some minimum WPA per point cut off to clip garbage time. (Since data collection is the hard part here, you could even do both...)

      Delete