Turns out that we’re often pretty terrible at reporting on our own habits — how much we’re exercising, when we’re sleeping and feel rested. But what we are good at is engaging with a wide variety of technology 24/7.
“Walking around with our smartphones and wearable devices, the devices actually generate massive digital traces of our behavior in the real world,” said Tim Althoff , an assistant professor at the Paul G. Allen School of Computer Science & Engineering at the University of Washington.
Althoff and his team are tapping into this data in some unexpected ways in order to tease out a better understanding of human health, including insights into sleep and performance, exercise and health inequality, and mental health and counseling.
Althoff shared his research Wednesday at a talk titled “Data Science for Human Well-Being” presented at the Paul G. Allen School’s 2019 Annual Research Showcase .
To gain insight into how sleep cycles correlate with our ability to function , Althoff paired data on the speed at which people type and click in the Bing search engine with sleep monitoring information collected from Microsoft Band, a wearable activity tracker.
He found that typing speed peaked roughly two hours after someone wakes up, which matches generally accepted sleep research looking at when people are most alert and validated the approach as a means for investigating sleep and performance. The study also looked at the effects of insufficient sleep over one or multiple nights on cognitive ability.
In additional research, Althoff has looked at the correlation between keystroke speeds and incidents of car accidents , and he’s investigating the performance of NFL football players and typing speeds.
Althoff is excited by the potential to use information gathered unobtrusively from common consumer devices to gain health insights.
The data is “reported for millions and millions of people and it allows us to study at a massive scale,” he said. “It allows us to study at a granular detail, doing that continuously over long periods of time and getting that at a relatively low cost. This is a big deal.”
Traditionally much of this information is gleaned from self-reporting or in lab experiments with a few subjects over a finite period. In many cases, the scientists ask only binary, yes-and-no questions that miss the range of lived experiences.
There are challenges to collecting and analyzing data gathered from devices and apps — particularly concerns about informed consent for participation and privacy. In some cases, the consent is part of fine print in the terms of service, which Althoff agreed was a less ideal solution. For the initial sleep study, participants were contacted and actively opted into the research.
In another study, Althoff and colleagues partnered with a step-tracking app called Azumio, providing a dataset of 68 million days of physical activity for more than 700,000 people living in 111 countries. They matched Azumio’s exercise data with national obesity rates and with the walkability of U.S. cities. The results found inequality in activity worldwide and that more walkable cities have less inequality in exercise.
In the mental health realm, Althoff used “natural language processing” tools to analyze text conversations in Crisis Text Line, a popular app for people texting with trained counselors. After someone interacted with the app, the researchers followed up to ask if they felt better, worse or the same.
Althoff analyzed the exchanges between the more and less successful counselors to find patterns. He concluded that better counselors shared traits such as providing more personalized, less scripted exchanges, and moved quickly to problem solving to help change a user’s outlook or perspective. The results are already creating a real-world benefit.
“We’re excited,” Althoff said, “because this type of data-driven insights have concretely changed how they train counselors.”
Editors Note: This story was updated to change “concretely” in the final quote from “completely.”