Predictions for the Top Recruits as of February 05, 2009

Welcome to the “college football recruiting prediction model” website hosted by the Stetson School of Business and Economics at Mercer University.  This website contains information about an econometric model that has been developed by three economists (Mike DuMond, Allen Lynch and Jennifer Platania) to predict the collegiate choices of high school football players.  

Each year, on the first Wednesday in February, college coaches, administrators, and football fans eagerly await the choices of high school seniors as they sign “letters of intent” with specific colleges.  These letters of intent officially link top ranked high school football players with specific colleges and effectively determine the rosters of college football programs in years to come.  The process of recruiting these players is highly competitive, as schools battle with one another to sign the best of the best in the world of high school football.  Recruiting gurus attempt to forecast the choices of these high school athletes as college football fans desperately search the Internet for bits and pieces of information that might shed light on the choices of these teenage football players.

The importance to a specific college of signing a strong group of athletes on “signing day” cannot be overstated.  College football has become big business in recent years.  The amount of revenue a college or university can generate by fielding a competitive football team can reach staggering levels.  The college football program at the University of Texas at Austin, for example, brought in over $53 million[1] during the 2004-2005 academic year.  Given the potential for such revenue, as well as the general interest in the sport held by rabid college football fans, an econometric model to predict the choices of college becomes useful.

The authors used statistical software developed by SAS along with data provided by Rivals.com (a national website dedicated to college football and recruiting) in the development of this model.  The model was built on a database capturing characteristics and decisions of 3,395 recruited athletes for the three “recruiting seasons” between 2002 and 2004.  On average, each player was choosing from among a group of 4 schools.  A wide array of player and team level data were gathered for this task.  Then, a special form of a probit model was developed to capture, to the best extent possible a statistical equation to capture the decision making process.

We were a bit surprised by the results.  There were a number of factors that we thought would significantly impact the decision of the high school athlete that didn’t.  For example, factors like the school’s graduation rate, the number of Bowl Championship Series (BCS) bowl appearances, the current roster depth at the recruited player’s position, the number of players from a specific college drafted by the NFL, and even the number of national championships won by a particular program don’t systematically influence the decisions of high school athletes.  Surprised?  So were we.  What, then, does matter?  As it turns out the following factors DO significantly impact the decision of high school athletes:

• Whether the athlete made an “official visit” to a specific college                      
• Whether the school is in a BCS conference                             
• The distance from the high school athlete’s hometown to a specific school          
• Whether the recruit is in the same state as a specific school
• The final AP Ranking of a specific school in the previous year of competition            
• The number of conference titles a school has recorded in recent years
• Whether the school is currently under a “bowl ban” for violating NCAA rules                    
• The current number of scholarship reductions a school faces for violating NCAA rules             
• The size of the team’s stadium (measured in terms of seating capacity)
• Whether the school has an on-campus stadium                       
• The current age of the team’s stadium                               

So, in a nutshell, high school athletes prefer winning programs that are
close to home, are in possession of good physical facilities, and are in good
graces with the NCAA.  Interestingly enough however, reduced scholarships increase the likelihood of choosing a particular school, holding all else constant.  This is likely because reduced scholarships imply reduced competition for exposure and playing time in the future.

Does the model predict accurately?  We believe the model predicts quite accurately!  Within the sample used to build the model, we accurately predicted approximately 68 percent of the choices of high school athletes.  (Considering that each player was choosing from a group of, on average, four schools, a 68 percent accuracy rate isn't too bad.)  In 2005, we examined the choices of the top 100 high school prospects and used the model to accurately predict the choices of 71 of these players.  After a brief hiatus in 2006 we came back with more predictions in 2007.  For the top 100 of the 2007 recruiting class, the model accurately predicted the decisions of 72.  Making the accuracy rate of the 2007 class more impressive is the fact that each player in the 2007 class was choosing from an average of 6.2 schools!  In 2008, the model accurately predicted the choices of 73% of the Top 250 recruits in the land.

Information regarding the predicted and actual choices for the top 250 recruits of 2009 (as determined by Rivals.com) appears to the right.  Also to the right, the potential schools and associated probabilities, as determined by the results of the model, for the recruits in this group that have yet to decide are listed.  Of the 245 athletes that have made predictions, we have accurately pegged the decisions of 174, translating into a 71% accuracy rate. 

The results of this model appear in an article published in the February, 2008 edition of The Journal of Sports Economics.  The recruiting prediction model is the subject of a current article authored by Andy Staples in Sports Illustrated (link). Additionally, separate media stories regarding the results of the model have appeared in the Columbus Dispatch, the Atlanta Journal-Constitution and via an Associated Press article written by Paul Nowell in 2005. 

The authors thank Juan Del Valle for his invaluable assistance in the collection of GIS related data.  The authors would also like to thank Kaitlin David for web site development and support.

About the authors:

Currently, the model is predicting at a 71% accuracy rate for the committed players.