Thursday, February 21, 2013

Using Data to Identify Point Shaving

Using Data to Identify Point Shaving in College Basketball

The book Super Crunchers by Ian Ayres is the perfect read for anyone interested in Big Data.  The book gives an overview of Bid Data applied in many different topics, and how it is changing the world we live in.  Ayres uses a small portion of his book to discuss the findings of Australian economist Justin Wolfers. Wolfers went about to study Las Vegas bookies and how accurate they were in predicting NCAA college basketball games. 

Wolfers had to analyze data from the past 16 years of college basketball, over 44,000 games.  His first visualization of the data was to graph the actual margin of victory relative to the predicted markets point spread. Wolfers found that “the graph was band on a normal bell curve”.  He had discovered that almost 50 percent, actually 50.1 percent, of the time the favored team beat the point spread.

After further analyzing the data, Wolfers looked at the last five minutes of these games.  The favored teams were on track to cover the spread about 50 percent of the time. But, during the last five minutes of the games, the scoring shortfall appeared.  This wasn’t enough to conclude point shaving, but it was a reason to keep digging further into the data.  Wolfers claimed that the shortfall for these teams to cover the spread is statistically significant.   He argues that the 47 percent probability that the team will cover the spread is more than two standard deviations away from the 50 percent probability if it was truly a fair bet.

I think that that Wolfers findings do not have the most sound statistical evidence.  But, with the volume of games that he had to analyze, these small differences represent a major impact in the course of the games.  I think that his initial findings with the spread to the outcome of actual games being almost perfectly normal validate his later findings. 


Sources:


2 comments:

  1. Very interesting topic. My first thoughts are about the last 5 minutes of the game. I have watched college basketball my entire life and am almost certain that the last minute of a basketball game is the longest minute in all of sports. This is generally because it is common strategy for the team that is behind to foul in order to get the ball back sooner. This makes sense then that the team that is favored experiences a scoring shortfall late in the game. This strategy limits the time of posession for the team that is ahead and is also disadventagous if they are poor foul shooters. This strategy could explain the phenomena of a scoring shortfall late in the game as opposed to point shaving.

    ReplyDelete
  2. I was hesitant to blog about this phenomena because I did not believe Wolfers conclusions had any merit. But after thinking about the problem for a couple days, I came up with some of my one reasons to blog this article. The last minute of the game is very long. But the opportunity for scoring is skewed towards the team in the lead, because they are fouled every possession almost immediately. The losing team has less scoring possibilities. Because the scoring dynamic completely changes at the end of the game, I wanted this article to lead into a second experiment. Now that Wolfers has isolated these "shaving games" would looking at a player or teams free throw percentage during the last 5 minutes show a statistical difference between the rest of the game?

    Although his evidence leads lots of room for error. I think that Wolfers findings combined with free throw data could either prove or disprove his original conclusion.

    ReplyDelete