Online Program

Return to main conference page
Wednesday, May 29
Data Visualization
Computational Statistics
Machine Learning
Opening Mixer & E-Posters
Wed, May 29, 5:30 PM - 7:00 PM
Grand Ballroom Foyer

Data-Driven College Admissions: Useful Metrics or Numeric Nonsense? (306366)

*Emily Rose Flanagan, University of Washington 
Connie He, University of Washington 
Grace Radford-Tingley, University of Washington 

Keywords: Education, Logistic Regression, Linear Regression, Prediction

With the ever-increasing number of undergraduate students applying to college, it is not only important for admission committees to closely examine application data to retain the best and the brightest, but also to ensure that admitted students are equipped to thrive in their academic environment. Current studies on modeling academic success in college are limited to using either high school GPA or standardized test scores, and frequently reduce notions of academic success down to a binary prediction problem. Furthermore, many of these studies are funded by testing companies who have a vested interest in ensuring the significance of test scores in the model.

Using data from 10,200 freshmen students from the classes of 2016 and 2017 at three University of Washington campuses, we evaluate the effectiveness of the covariates on predicting success, as defined by above a 3.0 GPA, using a logistic regression model. We extend the current literature by examining additional factors such as standardized test score, type of test taken, extension credits (such as AP or IB), transfer credits, and campus, as well as by implementing a continuous weighted least squares linear regression model to predict exact GPAs after the first year. We hope that the results from this investigation will help in choosing from the many qualified applicants vying for admission to the University.