Scroll through your friends’ Instagram feeds and you’ll likely see the sunniest, most charmed versions of their lives: days at the beach, elegant meals, and playful pets. Now, however, new research indicates that even the most charmed imagery might tell a different story.
According to a study published on the electronic preprint service arXiv, machine learning analysis of users’ Instagram accounts can “identify markers of depression.” The study’s authors—Andrew G. Reece, a Ph.D. candidate at Harvard, and Christopher M. Danforth, a professor at the University of Vermont—further claim that their “models outperformed general practitioners’ average diagnostic success rates for depression.”
The MIT Technology Review explains that Reece and Danforth’s findings map strikingly onto conventional associations with depression. The two performed their study on a group of about 170 workers from Amazon’s Mechanical Turk, “of whom around 70 were clinically depressed,” according to the Review. Participants completed “a standard clinical depression survey” and also provided other information, including the dates of their diagnoses where applicable. The algorithm looked for patterns in the images posted by depressed individuals prior to their diagnoses, while also evaluating an array of recent photos from individuals who were not depressed.
While the study looked at a variety of features—including the number of people in the images and the language used to describe them—the most immediate flag may well be the color schemes apparently preferred by those who were depressed: “The researchers found that depressed individuals tend to post images that are bluer, grayer, and darker.” That plays out in the filters that Instagram users impose on their images. The Review writes that while “healthy individuals preferred a filter called Valencia, which lightens photographs,” depressed participants in the study tended to employ one that grayscales their photographs.
While these findings are striking, there may still be reason to hesitate before fully embracing Instagram as a mass diagnostic tool. Perhaps most notably, it’s important to remember that social media rarely serves as an automatic translation of a user’s life and experiences. Given how well the color schemes line up with those stereotypically associated with depression, it’s possible that these images were a conscious form of communication for some users, especially in advance of their diagnoses. In that light, perhaps the algorithm is best at pinpointing those who are already preparing to communicate with others about their mental health.
