Thinking Out Loud: Point-Shaving Detection

Those knee-deep in the hoopla of NCAA conference championship tournaments might not have noticed, but Yahoo! Sports reported yesterday that Auburn guard Varez Ward was being investigated by the FBI for point shaving. Auburn, that upstanding institution of higher learning whose athletics program has always followed the straight and narrow before, joins an august group including Boston College, Northwestern, and (my alma mater) Tulane.

The story announced that the FBI has launched an investigation focused around two SEC games. Let's let ESPN.com tell the story:

"Ward scored three points and had six turnovers in the 68-50 loss to Alabama [on Feb. 7], playing 17 minutes. Vegas Insider said Alabama was favored by five points.

"Ward lasted only 19 seconds after coming off the bench in the 56-53 defeat against Arkansas [Jan. 25] before crumpling to the floor. [Auburn head coach Tony] Barbee later said Ward took a knee to the right leg he had injured early in his sophomore season with the Longhorns, when he ruptured his quadriceps tendon on a dunk during pregame warmups. Auburn still covered the 9.5-point spread."

This story is of course juicy and train-wrecky in its own right, but that's not what interests me. Suppose that the FBI isn't able to get a confession out of Ward or some gambling associate. Could someone prove (or at least produce circumstantial evidence) that a player was shaving points using forensic sabermetrics*?

A player in a point-shaving scandal would significantly underperform both his season averages and the average performance allowed by his opponents in the games in question. After all, there's a big difference between Power Forward X (who averages 10 points and 8 rebounds) getting 3 points and 1 rebound in a game against Syracuse, and 3 points and 1 rebound in a game against Stony Brook. Ideally, one could also track bets placed on those games, looking for unusual amounts of action.

This method still has a lot of problems. For one, I wouldn't expect much help from the casinos. That's not to say they wouldn't go out of their way to help make sure the games are clean -- they have a business incentive to do so, after all. Rather, the type of person who'd set up a point-shaving scheme is also the type of person with few scruples about placing illegal bets, either offshore or through a local bookie; these would obviously be much harder to track. And looking just at statistics would produce too many false positives: if you're just looking at statistics, how can you be sure that the player didn't just have a bad night, or was suffering from a nagging injury? On top of that, point-shaving scandals (at least, the ones that went public) are very rare, so it would be tough to build a set of statistics from point-shavers to train any algorithm.

The specific case of Ward is fraught with additional complications. Auburn played three games between Jan. 25 and Feb. 7, in which Ward averaged 29 minutes; did Ward decide not to shave points in those games? Auburn split the two questionable games against the spread; did Ward switch betting sides?

Judging from all this, I'm not sure this type of analysis is possible, but I would love to hear what other people think about it. What would you include/exclude from this algorithm? Are there counter-arguments I'm missing? Has someone actually done some type of analysis along these lines?
_________

*-I like the sound of "forensic sabermetrics". Maybe we could make one of those police procedurals; call it "CSI: Mom's Basement". Someone get David Caruso to return my calls!