This story is over 5 years old.


The Horrible Sloan Paper That Would Destroy Basketball

Why would anyone, even a game theorist, suggest that what basketball needs is more intentional fouling at the end of games? It's less about the answer than the question.
March 5, 2015, 5:35pm

"Why did we make the atomic bomb? Well, we figured that we had this uranium, and a theoretical framework, and I just said 'Hey, if it's there, why not? Progress is progress!" - ROBERT OPPENHEIMER (unattributed)

There are many questions you can ask about sports. What makes them beautiful? How do they reflect society, beautiful and un-? What do teams mean to and in their communities? How can brands use them to leverage their products? (Not every question is good.)

The most common question people ask, though, is generally: "How can a team win at sports?" If you're looking to answer this question without consulting mathematical evidence, you're fighting a deer with one hand tied behind your back. Some people think writing about stats is boring. Personally, I am THRILLED, AMPED, PUMPED, every time I see a spreadsheet with that good data. I think projections are neat! What a cool trick, I think.

You say, "Hey I ran a regression on this sports data," and I say "Oh man, it's a good thing this is regressed, because I wouldn't want to read some un-regressed bullcrap, I am an educated person," even though I have NO IDEA how regression works!

I am saying this to set a baseline. I am not scandalized or bored by data. It takes an awful lot to get me to wretch about something I see in a paper at the MIT Sloan Sports Analytics Conference. Which brings me to this, this monstrosity, this Frankenstein paper with a vampire's heart by Franklin Kenter, a PhD Candidate at Rice University, that was unleashed at Sloan last weekend.

The thesis of this paper—which is underwritten by the Ticketmaster corporation, a company that profits from basketball being enjoyable, as well as from a bunch of service fees—is that basketball teams are not spending enough time engaging in the most irritating aspect of a basketball game, and that what basketball needs more of is the grinding start-stop that results from executing the foul game at the end of a close contest. Right now, most NBA teams use the technique to squeeze two or three more possessions out of a game when they're down with less than a minute remaining. No one is particularly fond of this practice. It is a very boring thing to watch.

The conclusion that this is all actually good and smart is reached by modeling basketball using game theory methods. Unlike a statistical method, in which you take a pile of data and observe what it tells you, a game theorem breaks down a system into component parts, and runs every possible outcome of those component parts through a computer algorithm and determines what the most likely victorious strategy will be. The paper makes a point of establishing this distinction, "Beyond describing when to foul in basketball, a key takeaway of this article is that statistics cannot necessarily uncover all optimal strategies in sport. This is especially true for strategies such as the one presented in this article."

Outside of my formulation of a foolproof Settlers of Catan strategy*, I am not a game theorist, so I cannot really criticize the veracity of this paper's methods**. I guess, when I was reading it, I thought it seemed a little reductive. But flattening the complexities of sports into numbers is by definition reductive, and yet those numbers are still often useful and correct. We should probably assume that this paper was written in good faith, with sound methods.

But, all the same, the conclusion reached is so maddening. It's an endorsement of a proposed form of basketball so mind numbing that it seems immoral to even suggest it. "The results of this analysis are summarized in Figure 2 above. The threshold for when to foul in typical NBA play is very pronounced in the model and is a near-linear function of the point spread. In fact, for NBA play, a team trailing by p points should foul with approximately t = 13.82 + 10.32 p seconds remaining." For reference what that would actually look like, consult this chart:

I will regard this suggestion in good faith, and assume that Kenter doesn't think a basketball team would actually start fouling down thirteen with 2:20 remaining. But he does say that, "a team trailing by eight points should start fouling with 90 seconds remaining." A MINUTE AND A HALF OF FOUL-GAMING! "Research suggests that NBA teams should make an arena with 20,000 people inside of it sit around for 15 to 20 minutes while the game is being slightly leveraged," basically.

The paper's analysis suggests that employing an optimal fouling strategy would add one or two expected wins a season, which is a lot when you're considering signing a bench player but maybe not that much when you balance it against turning what ought to be the best part of your product into a nightmare for everyone participating.

This is not the paper's most radically boring conclusion, though. "One of the most surprising outcomes of our model demonstrates that the leading team should foul **more often [**emphasis theirs] than the trailing team." This is a very specific definition of the word "Should," where "Should" means "Should do something to win a game" and "Should" doesn't mean "Should do something for the good of a game, or the people watching the game, or humanity at large."

"For typical NBA play, a team leading by p points should foul with approximately t = -14.99 + 23.06 p seconds remaining," Kenter asserts. Here is what that would look like:

If you were watching a game and a team that was up three with nearly a minute remaining intentionally and repeatedly fouled the opposing ball handler, you would be mad. If a team was leading by five and they fouled with a minute and forty seconds remaining, you would be so mad that you would stop watching the game, walk outside, and set fire to the nearest plant.

This might be a winning strategy, but it is about as cynical and unsporting a strategy as anyone could possibly devise. Why should a team employ a strategy this cynical, except that it might help them win? Wouldn't this violate the spirit of fair play? Isn't this a strategy that encourages teams to abandon basketball and embrace... something that is not like basketball, that is overall less fun and worse? Isn't this—and this is actually not a question so let's rephrase—this is a strategy that would be employed by a hedge fund manager coaching his child's AAU team.

Blessedly, this is not something basketball aficionados need to worry about. Coaches are stubborn weirdos, for better and for worse. It took three decades for coaches to let their players shoot threes en masse, it's hard to imagine a bunch of stubborn older men are going to take to "Stop playing basketball a full two minutes before the game is up" with enthusiasm. Perhaps, someday, teams will employ aggressive late game defenses, and so wind up employing a version of this strategy in some backdoor way. We should all obviously hope to be dead by then.

There is a lot we can learn about sport when we void the games of their context and look at them as sheets of data. The question of whether a particular work is actually valuable in a real-life-game depends on if you can reinsert your conclusions back into the real world. Voros McCracken's radial work with BABIP and the role of luck in hitting was an alien concept that seemed very strange at first to baseball analysts. But when you actually stick it inside real baseball, it's a skeleton for why a player's production fluctuates from year to year. It illuminates, and it works.

This is not that, in every meaningful way. It is, instead, something like the best argument against Sloan and Sloan-ism in general—a weird idea that, when used in real life, would be heinous in multiple ways. So how genuinely valuable is gaming out a strategy that would make the sport both unsporting and unwatchable? How is it that no one, at some point, said, "Oh, also, this would be horrible, if teams actually did things this way."

The paper doesn't even mention the horror show of two teams executing the model's suggestion against each other; the end of a game could go on for a solid, whistle-blasted half hour without any threat of a lead changing! The NBA, in a predictable anti-disruption move that would be very much to its credit, would surely legislate the practice out of the game. It's happened before; the shot clock, most notably, was invented because early NBA games devolved into a blob of keep away dribbling and intentional fouling. That worked, for the sport and everyone who watches it.

In the end, it's not that the paper isn't right. It might be. It's certainly a more convincing stab at strategically optimizing late game foul shooting than the average civilian could come up with. But the proposed solution is an aesthetic nightmare that no self regarding coach would implement. Never mind how unpleasant it would be, it's altogether too abstract to work in actual sports. And, as that's the case, it's not really good sports analysis, or really even analyzing sports at all. It's work that is decontextualizing sports for the sake of decontextualization. There must be better ways to have fun than this.

*Just maximize the number of pips you get on a turn; losing half your resources isn't THAT bad, because you will still probably have enough to build something on your next turn. Also, if your table is cool with it, push the creation of a futures market when you're receiving too much of a given resource on a turn, and realize the the dice are the true rulers of the game and you should be playing Tigris and Euphrates instead.

**Okay, there are two problems that even I can sniff out here. The first is that fouling up three with more than two possessions remaining in a game might make sense from an algorithmic standpoint, but it's extremely risky, and this risk is not accounted for in the paper at all. You are basically volunteering half a possession of ground to the other team. Up three with less than a possession? That's a tenable strategy that some teams already approach in this way, even if it strikes me as unsporting.

The second is the model's shooting percentage structure. The model is built assuming that teams shoot 48% from two point range, 36% from three, and 75% from the line. This is true over the course of a whole game, but in higher leverage situations teams tend to shoot worse from the field, significantly worse from three, and slightly better from the free throw line. The reasons for the first two are all fairly esoteric: panicked strategies, tighter opposing defenses, and looser refereeing are all conceivable assumptions. The third is simpler: teams are often intentionally fouled when they're up, they know it's coming, so they shuttle the ball to their best free throw shooter. I believe accounting for this extra context would make the proposed strategy less effective than the paper suggests, though I concede that I am not a game theorist and I may be misinterpreting the matter.