Examining the Correlation Between Depression and Social Behavior on Smartphones Through Usage Metadata: Empirical Study

BACKGROUND: As smartphone has been widely used, understanding how depression correlates with social behavior on smartphones can be beneficial for early diagnosis of depression. An enormous amount of research relied on self-report questionnaires, which is not objective. Only recently the increased availability of rich data about human behavior in digital space has provided new perspectives for the investigation of individual differences. OBJECTIVE: The objective of this study was to explore depressed Chinese individuals’ social behavior in digital space through metadata collected via smartphones. METHODS: A total of 120 participants were recruited to carry a smartphone with a metadata collection app (MobileSens). At the end of metadata collection, they were instructed to complete the Center for Epidemiological Studies-Depression Scale (CES-D). We then separated participants into nondepressed and depressed groups based on their scores on CES-D. From the metadata of smartphone usage, we extracted 44 features, including traditional social behaviors such as making calls and sending SMS text messages, and the usage of social apps (eg, WeChat and Sina Weibo, 2 popular social apps in China). The 2-way ANOVA (nondepressed vs depressed × male vs female) and multiple logistic regression analysis were conducted to investigate differences in social behaviors on smartphones among users. RESULTS: The results found depressed users received less calls from contacts (all day: F(1,116)=3.995, P=.048, η(2)=0.033; afternoon: F(1,116)=5.278, P=.02, η(2)=0.044), and used social apps more frequently (all day: F(1,116)=6.801, P=.01, η(2)=0.055; evening: F(1,116)=6.902, P=.01, η(2)=0.056) than nondepressed ones. In the depressed group, females used Weibo more frequently than males (all day: F(1,116)=11.744, P=.001, η(2)=0.092; morning: F(1,116)=9.105, P=.003, η(2)=0.073; afternoon: F(1,116)=14.224, P<.001, η(2)=0.109; evening: F(1,116)=9.052, P=.003, η(2)=0.072). Moreover, usage of social apps in the evening emerged as a predictor of depressive symptoms for all participants (odds ratio [OR] 1.007, 95% CI 1.001-1.013; P=.02) and male (OR 1.013, 95% CI 1.003-1.022; P=.01), and usage of Weibo in the morning emerged as a predictor for female (OR 1.183, 95% CI 1.015-1.378; P=.03). CONCLUSIONS: This paper finds that there exists a certain correlation between depression and social behavior on smartphones. The result may be useful to improve social interaction for depressed individuals in the daily lives and may be insightful for early diagnosis of depression.