Your smartphone knows if you will pass or fail this semester

Tyler Cowan at Marginal Revolution pointed me to this new paper (pdf) by Rui Wang (Dartmouth College), Gabrielle Harari (University of Texas at Austin), Peilin Hao, Zia Zhou and Andrew Campbell (all Dartmouth College). The authors used continuous sensing data (like location, audio) collected from students' smartphones over a semester (apparently over 53GB of data per student!) to infer the times students spent studying (smartphone stationary in a location like the library, with no audio), in conversations (smartphone stationary, with audio), partying (smartphone in a location like a fraternity, with loud audio), etc. They also collected data on the students' personality, stress, sleep, and a bunch of other things.

What did they find? Nothing too surprising, according to the paper:

Our prediction model indicates that students with better grades are more conscientious, study more, experience positive moods across the term but register a drop in positive affect after the midterm point, experience lower levels of stress as the term progresses, are less social in terms of conversations during the evening period, and experience change in their conversation duration patterns later in the term.

In this age of worries about the data that is collected from us, I wonder how they got their 30 students to agree to this data collection exercise (and they must have had a really good story for their institutional review board!). Two takeaways from the paper though:

1. Students' behaviours really do matter for their chances of passing - studying more makes a difference! Though apparently attendance in class does not - that might be an artefact of the small sample size and the homogeneity of the sample. Note that they do find that students who increase their attendance after the mid-point of the semester do perform better; and
2. Student smartphones potentially contain a wealth of unexploited data that could be used to develop study aids. How long before there is an app that prompts you to decrease your time spent partying (or encourages you to party more early in the semester, and less later in the semester)?

Finally, this suggests that perhaps universities could use this data to identify students at risk of failure, and intervene. Provided students let their university have access to continuous sensing data from their phones, of course!

Related reading:

• Could early identification of students at risk of failure be a bad thing?