A computer science professor at Dartmouth College, Andrew Campbell, has developed a smartphone app that he says can predict a student's GPA based on their behavior.
The app, called SmartGPA, uses a smartphone's ability to sense a person's activity and passively measures the kind of behavior that affects grade point average. The accelerometer, GPS, and microphone track whether a student is sleeping, studying, partying, or going to class.
The data was compiled from a 30-student, 10-week study inspired by Campbell's surprising finding that there's no correlation between time spent in the classroom and a student's grades, writes Kirk Carapezza of WGBH News. Obviously, students who study more and party less have higher GPAs, but Campbell also found that GPA was affected by face-to-face conversations in the evening and attending class more in the first half of the semester. Interestingly, the number of hours spent partying doesn't matter, but the ability to buckle down and focus later in the semester does.
Campbell said that the prediction is incredibly accurate:
We can predict GPA within 17 hundredths of a point. Even though it's a small cohort of students, the result is statistically significant.
His GPA prediction mechanism was derived from a student health app he made previously called StudentLife, writes Rob Wolfe of the Valley News. It relies only on what it can sense, according to Capital Lifestyle, and doesn't need user input or make use of background information like test scores or IQ.
He plans on using the app not to babysit students, but merely to show them how they manage their time in the hopes they will pay more attention to where their days go. Hopefully this will help them develop good habits that will last them the rest of their academic career.
Campbell said that delivering real information influences student behavior:
You show them their sleep patterns, their study patterns, their partying patterns. In presenting them this information in a visual way, they can be reflective about it.
This fall Campbell will study an introductory psychology course at the University of Texas in Austin which has 800-1,000 students. The students will be split into several groups: a control group for which researchers only collect data, another group in which researchers infer GPA, and a third in which students respond to surveys throughout the testing period.
Some are understandably concerned about the seemingly Orwellian nature of apps like these, writes Anya Kamenetz of NPR, and Campbell admits it's a problem:
It's the Achilles heel of all these applications. It's sensitive data.
Campbell will be presenting his research in September at the UbiComp 2015 conference in Osaka, Japan. The conference rated his paper in the top 5% of submissions because it was "groundbreaking." He said:
It's a general result that might be applicable not only to Dartmouth and other universities, but also to Google and other industries.