Answer in the Form of Another Question: Where Analytics Hits The Wall

facebooktwitterreddit

Nov 8, 2014; Milwaukee, WI, USA; Milwaukee Bucks guard Brandon Knight (11) reacts after hitting the game-winning free throw during the game against the Memphis Grizzlies at BMO Harris Bradley Center. The Bucks beat the Grizzlies 93-92. Mandatory Credit: Benny Sieu-USA TODAY Sports

Over the 1230 games that make up an NBA season, certain situations recur with some frequency. Down two, with about 10 seconds left on the game clock, what do you do?

The math in the situation is pretty straightforward, and indicates going for a three and the win is clearly the correct play. Assuming each team has a roughly 50% chance of winning in overtime, a two pointer has to be more than twice as likely to go in as the three to be the better shot[1. Using completely made up numbers, a 33% 3 gives the team a 33% chance of winning, but a 50% two is only good for 25% chance of winning once the game goes into overtime Add in the desire to maybe save the legs of a team’s ‘closing’ lineup from having to go another five minutes and the three to win or lose right there might be even more desirable.].

Ok, now that’s settled, now how do you go about getting that three? At this point, pure analytics breaks down just a little. Take the end of Saturday’s Bucks’ win over over Memphis. Down two, with eight seconds left, Brandon Knight scored an and-1 to put Milwaukee up one. Did Jason Kidd make a mistake by “going for the tie” and just get lucky? It certainly looked like Milwaukee was trying to use a flare screen to free O.J. Mayo for a three-point look as a first option on the play.

Courtney Lee doesn’t get screened, and Marc Gasol is overplaying the pass to Mayo, effectively shutting down this option:

However, the Bucks now have Knight isolated on Marc Gasol, the potential help defenders all are guarding three-point shooters, so help on a Knight drive is either going to be slow, or there will be a pass available to the open three-point shooter:

As it happened, the help was late, allowing Knight a open lane to the rim, with the foul being a nice bonus:

To some degree, this play indicates the difficulty in implementing the analytically indicated solution; the defense knows the math too. Marc Gasol essentially ignored Zaza Pachulia, recognizing the intended target of the play was Mayo. The “best option” was simply not available because, as Charles Barkley likes to say and I like to quote, “Them guys [Memphis] are getting paid too.”

In broader terms, the general prescription to “go for the win” is dependent on the actual percentages, which are largely unknowable in the moment. In this case, Knight could have “gone for the the win” by pulling up for a drifting three over Gasol, or trying to force a pass to a covered shooter. But even without the benefit of hindsight, that shot probably had less than half the chance of being made than a drive to the basket given the pressure on such heavily contested jumper. This is the point where the communication between the “numbers guy” and the “basketball guy” can break down.

Go for the win? Thanks, tell me how stat boy.

A similar situation arises in the case of the “two-for-one” possession at the end of a quarter. Obviously, two is more than one, and getting an extra possession on the opponent is a small but definite edge. So shooting at around 32 seconds on the game clock is advantageous because it ensures that extra offensive possession. Again, it’s not quite as simple as saying “two is more than one, so throw up any shot you can find at 32 seconds” and go from there.

“Short clock” situations are on the whole far less efficient than “normal” possessions, largely because the element of surprise is gone. The defense knowing WHEN a shot is going to occur has something of an advantage in defending that shot. Further, it becomes much harder to run intricate plays involving combinations of players moving and passing in any degree of synchronicity. There’s simply a lot that can go wrong, and not getting a shot off at all is disastrous. So even though a “better play” correctly executed probably results in a better shot, something simple[2. Like the de rigeur clear-out iso.] might actually be best.

Regardless, that last second shot is rarely a particularly good shot. Last season, even discounting heaves from over 30 feet, shots taken with under five seconds on the game clock had an eFG% of 35 percent, or about 0.7 points per shot. A “normal” possession was worth about 1.06 points.

Why is this important?

By going two-for-one, the possession the opponent gets in between transforms from that “end of quarter” inefficient shot to a more “normal” possession. Is the extra possession worth allowing the opposition a better chance of scoring themselves? It depends on the shot you can get the first time through. If “going two-for-one” translates into dribbling down and jacking up a contested three, it might not be worth the trade off.

That contested three was worth about 0.7 points per shot in 2013/14 (23.4 percent 3FG% on off the dribble threes with a defender 3 feet or closer from the shooter). By going two-for-one with those shots, the offense nets about 1/3 of a point[4. Two shots worth 0.7 points each against the oppositions 1.06 point “normal” possession is a net expected gain of .34 points.]. On the other hand, simply taking the “normal” possession and forcing the opponent to take the low value end of quarter possession is actually about an overall improvement of 2.0 pts per 100/possessions.[5. The 1.06 points from a “normal” possession is worth .36 points more than the .7 point last second shot given up. If anything, this specific example likely underestimates the benefits of NOT going two-for-one in the specific scenario. The opponents offensive efficiency is generally significantly higher on possessions following a defensive rebound than those following made field goal. A poor shot on the first possession means more defensive boards and less picking the ball out of the net. Essentially, an extremely contested early three is more likely to give up a transition bucket the other way.]

This doesn’t mean attempting to get a two-for-one is bad, of course. It simply means the quality of the first shot matters a great deal. Not even attempting to generate a two-for-one look is a bad play, but deciding not to shoot a really terrible shot because two-for-one is probably good strategy. Meanwhile players such as Steph Curry or Kyle Lowry who adept at getting off reasonable shots off the dribble under these late quarter circumstances have a little bit of hidden value in those situations. In any event, the math only answers the what, not the how.

While these are small and discrete circumstances, the same issue with translating the numbers to the floor is extant in the bigger picture as well. The underpinnings of the MoreyBall “layups and threes” strategy are unassailable: higher expected points per shot is better than lower. But what happens when the other team properly adjusts? As I noted last week, the Rockets struggled more than most top teams against good defenses because knowing the right shots to take isn’t a road map to how to find those shots within the confines of the 24 second clock. Is it possible to generate exclusively “MoreyBall approved” shots versus an NBA defense? Perhaps, perhaps not, but that is a question of basketball strategy, tactics and talent, not one of analytics. While using the numbers to ask the right questions, it’s vital to remember in many cases studying the stats provides the intermediate step rather than the final answer.