Is Point Shaving Really a Problem in the N.C.A.A.?

One problem with the data crunching revolution that's transformed economics over the past decade is that the final results of most any intelligently conducted research can all turn depending on how the numbers are manipulated.

Felix Salmon over at Market Movers points to an NYT op-ed by Wharton forensic economist and major proponent of prediction markets Justin Wolfers on the betting scandal engulfing the N.B.A.

In the piece, Wolfers calls out his study (pdf) last year that argued there was a good chance the N.C.A.A. was rife with illegal point shaving.

First reported (pdf) by the excellent David Leonhardt over at NYT, the paper went on to receive much wider media attention.

(The N.C.A.A. countered that "only" 1.5 percent of players had taken money to play worse. The purist in me was a bit dismayed at that -- even 1.5 percent seems unacceptable. But when you factor in the ridiculous fact that college players aren't financially compensated makes me wonder why the figure isn't higher?)

Much less noticed was a study by Dan Bernhardt of the University of Illinois at Urbana-Champaign and Steven Heston at the University of Maryland which disputed Wolfers' findings.

Wolfers had taken a sample of some 40,000 N.C.A.A. games between 1989 and 2005 and found that 5.5 percent of the teams favored to win by more than 12 points (according to the point spread) likely engaged in point shaving.

Bernhardt and Reutter countered that this was not a result of nefarious activity but just the nature of the game.

Their proof?

"The researchers used two approaches. First, they looked at the changes in the point spread between the opening and closing lines, which was not done in the Wolfers study. The opening line is the point spread that bookmakers initially quote, and the closing line is the spread quoted just before the game begins.

They identified 4,350 games in which the closing spread was wider than the open spread, and 4,956 games in which the spread either remained unchanged or fell.

Their hypothesis was that if the opening line on a favorite team increased between the opening and closing spreads, more money was being bet on the favorite than the underdog, making point shaving implausible. On the other hand, in games where the spread fell between the opening and closing lines, then gamblers betting on the underdog may have induced members of the favored team to intentionally play badly in order to win by less than the spread.

Finding very similar patterns regardless of whether the point spread rose or fell, the researchers report no statistically significant evidence of point shaving."

(Source)

Bernhardt and Heston's second approach was to look at college games in which there was no betting (games between non-elite Division I teams) meaning no incentive to shave points.

For these games, they computed the expected winning margins using some statistical wizardry and compared the results to the situation with gambling.

"The probability discrepancies are greater for games without gambling," the researchers concluded.

They attribute the assymtery in winning margins to two factors:

- Teams ahead by a small margin hold the ball at the end of a game, strategically reducing the number of scoring opportunities for both teams in order to maximize the probability of winning. "This strategy has the effect of reducing the chances of beating a large spread," they noted. "In contrast, a team up by a large margin is sure to win and will not hold the ball, thus raising the frequency of blow outs."

- The impact of foul troubles on strong and weak teams is likely to have a different impact on the final score. "Strong teams tend to be deeper, with a stronger bench, so if the top players get in foul trouble, the team is still likely to win by a small margin," Bernhardt said. "But the same it not true for a weak team. If a weak team gets in foul trouble, it is likely to get blown out because its secondary players are weak."

(Source)

So who is correct here? One clue might be that while Wolfers' paper was published in the well respected American Economic Review, Bernhardt and Heston's is still in working paper mode.

But thanks to the internet, that may not necessarily be a bad thing.