Recruiting Predictions & Statistics

Can the world of statistics help predict basketball recruiting results? Jackson Fambrough thinks so. Fambrough developed a model using economic principles that predicts where college basketball recruits will, or should, go to school.

We wanted to learn more about this model - and show it to Inside Carolina readers - so we posed some questions to Fambrough about how he came up with the idea and what type of accuracy it yields. And while this model applies to all Division I schools, we thought it would be interesting to see how it relates to recent UNC recruiting targets.

What is the general overview of your model?
"The model is what economists refer to as a probit (short for probability) model. Probit models predict the probability of something happening given certain variables that are plugged into the model. My model takes variables dealing with three categories (recruit characteristics, school characteristics, and the relationship between the school and recruit) and plugs them into the model to predict where a recruit will or should go to school. Recruit characteristics involve recruit rankings and what position they play. School characteristics include items such as a school's success athletically and academically as well as age/capacity of stadiums. The relationship between the recruit and the school involves the geographical location of the school, if the recruit took an official visit to the school, and the relationship between the recruit and the school's academics.

"Essentially, the model is capable of predicting the chances any school has with any recruit, all you need to have is the data for the characteristics mentioned above."

What made you come up with the idea?
"Growing up I was always a huge fan of college basketball recruiting and I always wanted to try and predict where recruits would end up. I recently graduated from Elon University with a Bachelor's Degree in Economics. During my junior year I had to start thinking about what I wanted to do for my senior thesis. My mentor, Dr. Jennifer Platania, coauthored a paper on an economic model for college football recruiting and it inspired me to see if I could use the some of the principles from her paper to create a foundation for an economic model for college basketball recruiting. A year later I submitted my senior thesis, which ended up being the first version of the model I use today."

"The model is 74% accurate for the Top 150 recruits in the 2006-2013 classes. The model is currently being applied to the classes of 2014 and 2015 and is 77% accurate for the class of 2014 and 100% accurate for the class of 2015. Of the recruits the model got wrong, 37% of them transferred - including the 2012 recruits who have only played one year. If you remove the 2012 class, because they have only played one year, 40% of the recruits the model got wrong transferred from the classes of 2006-2011. The transfer stats coupled with the accuracy rate provide validation to the model being used as a real world application."

How accurate has your model been with the recruits that UNC offered the last few years (2011-2014 classes)?
"For the recruits UNC has offered over the last few years (2011-2014 classes), the model has been close to 76% accurate. Robert Johnson (2014) and Mitch McGary (2012) were two of the recruits UNC offered during this period that the model incorrectly predicted UNC would be their school of choice. Regarding recent UNC transfers, the model was wrong on the Wear twins, Larry Drew II, and Alex Stepheson. The model showed UCLA should've been the school for both the Wear twins and Drew II, the school where they transferred to after leaving UNC.

"Presently, 2014 has turned out to be a fantastic recruiting class for UNC and there is still a chance the class isn't finished yet as Rashad Vaughn still has UNC on his list of possible schools. The model predicts UNLV is the favorite to land Vaughn by 35% if he follows through with his official visit there. Even if he comes to UNC for an official visit, UNLV is still the favorite by close to the same margin."

What are you hoping to accomplish with your model?
"When it comes to my model, there are three main goals I hope to accomplish: 1) help recruits, 2) help programs, and 3) provide fans with insight into the recruiting process.

"In today's recruiting world recruits have a multitude of people trying to influence their decision. These outside influences can lead to a recruit choosing the wrong school, eventually causing the player to seek a transfer. As we have seen the past few years, the number of players transferring has dramatically increased. The model could help a recruit compare schools and determine which one would best fit him without any bias.

"College basketball programs, on the other hand, could use the model to help determine which recruits it has the best chance to land and effectively deploy its efforts and recruiting budget. The model could help prevent a school from wasting time and effort on recruiting someone whom they have a limited chance to land. On the other hand, the model could help a school identify a recruit with whom they think they have a small chance to land, but in fact shows a high probability of success. Using the model, a school could hypothetically lay out its recruiting plan by choosing its targets and setting the priority for each one. This in turn would minimize recruiting costs and lead to an increase in talent coming into the basketball program, leading to more wins, and ultimately providing more revenue to one of the two largest revenue streams of college athletic departments.

"Lastly, there are the fans. As everyone knows, fans can't get enough of college basketball recruiting (as we saw most recently in the Andrew Wiggins saga). Fans always want to know the likelihood a recruit will commit to their school and will look at every tweet, news story, and magic 8-ball trying to get the scoop on where a recruit will end up going to school. I think the model, using numbers and data rather than feelings and leaks, could be the tool that provides them with the best odds of predicting where a recruit will attend school."

Table of UNC Recruiting Targets (2009-2014)

 Year Recruit Predicted Actual 2014 Justin Jackson UNC UNC 2014 Theo Pinson UNC UNC 2014 Joel Berry UNC UNC 2014 Robert Johnson UNC Indiana 2013 Andrew Wiggins Kentucky Kansas 2013 Noah Vonleh Indiana Indiana 2013 Isaiah Hicks UNC UNC 2013 Troy Williams Indiana Indiana 2013 Kennedy Meeks UNC UNC 2013 Bronson Koenig Marquette Wisconsin 2013 Nate Britt UNC UNC 2012 Kaleb Tarczewski Kansas Arizona 2012 Rasheed Sulaimon Duke Duke 2012 Cameron Ridley Texas Texas 2012 Mitch McGary UNC Michigan 2012 Marcus Paige UNC UNC 2012 J.P. Tokoto UNC UNC 2012 Brice Johnson UNC UNC 2012 Adam Woodbury Iowa Iowa 2012 Joel James UNC UNC 2012 Roscoe Allen UNLV Stanford 2011 Austin Rivers Duke Duke 2011 James Michael McAdoo UNC UNC 2011 P.J. Hairston UNC UNC 2011 Cody Zeller Indiana Indiana 2011 Adjehi Baru UNC Charleston 2011 Alex Murphy Duke Duke 2011 Angelo Chol (T) Arizona Arizona 2011 Desmond Hubert UNC UNC 2010 Harrison Barnes UNC UNC 2010 Reggie Bullock UNC UNC 2010 Kendall Marshall UNC UNC 2009 John Henson UNC UNC 2009 Dexter Strickland UNC UNC 2009 Travis Wear (T) UCLA UNC 2009 Davis Wear (T) UCLA UNC 2009 Leslie McDonald UNC UNC 2009 Reeves Nelson UCLA UCLA

(T)=Transferred