Software Model Predicts College Football Recruits' ChoicesSoftware Model Predicts College Football Recruits' Choices

A team of researchers is using SAS Institute's statistical software to predict which college No. 1 football recruit Terrelle Pryor will choose. Earlier this week the prediction model called for him to pick Ohio State; now it's saying he'll likely choose Penn State. Michigan has dropped to his No. 3 choice. Hmm, we'll see.Can SAS's software really determine Pryor's final choice? The model, created by economists at Mercer University's Stetson School of Business and Economics, had an accuracy rate of 73% as of Feb. 7 based on the final choices of 249 recruits. No. 1 Pryor is the holdout, and his indecision apparently has caused the prediction model to go a bit haywire in the past few days.

The model was developed using SAS software and information provided by Rivals.com, and relies primarily on historical data. It was built on a database that captured characteristics of the choices of 3,395 recruits between 2002 and 2004. A large amount of player and team data was gathered for the task. The researchers then developed a special form of a statistical model known as a probit to try and capture the decision making process of recruits.

The researchers said they were surprised by the results. (I guess I'm not so surprised, since they seem somewhat consistent to me with the emotions of young athlete.) Recruits' decisions weren't based so much on a school's graduation rate, its number of bowl championship appearances, or the number of its players drafted by the NFL. The factors really important to them included whether they had visited a school (there's no question teenagers often make college choices based on their gut feelings after visiting various campuses), the distance from the athlete's hometown (so he can go home to Mom often and revel in his hometown's adulation for its local celebrity), and the size of the stadium (the bigger the stadium, the louder the roar).

So does the model work? Well, predicting the choice of nearly three out of four recruits isn't too bad, and is probably better than the predictions of even the most obsessed and knowledgeable college football junkies. As a Wolverine fan, I'm hoping this is one of the cases in which the model's latest prediction about Pryor is wrong. (And how did the model suddenly change since earlier this week, notably after Pryor told the media his father is pushing hard for him to visit Penn State? The model is based on historical data, not Pryor's latest comments, right? Hmm, again.)

Well, here's my message to Pryor: Michigan has the largest stadium in the country, and serves up a roar that will leave your ears ringing for long after the win. Let a statistical model try to capture that feeling.