Human interactions via internet and mobile apps are increasing and ubiquitous among all walks of life. Young and older adults are prolific in these media, as well as public and private agencies seeking to promote health behaviors and communications. Over 500 million users are registered to Twitter.com, a microblogging and social network service that conveys text messages to a restricted set of followers. Given current new user registration rates, Twitter may reach over one billion users in the next year. Over 50% of nonprofits using social media sites employ Twitter, the second highest used social media application after Facebook. Every second, 2,200 new Tweets appear offering timely broadcasts about users'current states, interests and ideas. Bigdata sources such as Twitter afford new potential in combining "real-time" spatial and social networking information to study causal processes underlying who, where and how individuals are connected. These key features are lacking in the traditional research designs and intervention strategies for health and health-related behaviors. For example, social network analysis (SNA), a strand of systems science, offers a unique lens to understand how human interactions through social network may collectively optimize mental health outcomes in social and geographic "locations." The ability to model attributes of nodes and tie strength makes SNA a powerful descriptive and predictive method. Network analyses can capture integration and fragmentation, concepts integral to prevention system planning. Indeed, novel mixed methods of SNA with machine learning can enhance prediction of flu spread in a large real-world population and future criminal incidents in a real city. The goal of this project is to develop and apply a new class of statistical and causal inference models for human interactions and their impacts on health and health-related behavioral outcomes using network data from social media sources such as Twitter.com by focusing on three research studies that integrate online social media information to understand the roles of human interaction on (a) disease spread, (b) mental health in a hard-to-reach population, and (c) presence and extent of "natural helping" in promoting wellness and reducing norms supporting violence in low-income urban neighborhoods.
This project will impact behavioral and social health sciences, including systems science communities, broadly and specifically by: (a) filling a critical gap in the literature on causal inference models for social network interactions, (b) strengthening the evidence base for how social media is used and affecting health related outcomes and behaviors, and (c) improving community capacity-building in health.
|Chen, Tian; Kowalski, Jeanne; Chen, Rui et al. (2016) Rank-preserving regression: a more robust rank regression model against outliers. Stat Med 35:3333-46|
|Chen, Tian; Wu, Pan; Tang, Wan et al. (2016) Variable selection for distribution-free models for longitudinal zero-inflated count responses. Stat Med 35:2770-85|
|Tang, Wan; Lu, Naiji; Chen, Tian et al. (2015) On performance of parametric and distribution-free models for zero-inflated and over-dispersed count responses. Stat Med 34:3235-45|
|He, H; Wang, W J; Hu, J et al. (2015) Distribution-free Inference of Zero-inated Binomial Data for Longitudinal Studies. J Appl Stat 42:2203-2219|
|Gunzler, D; Tang, W; Lu, N et al. (2014) A class of distribution-free models for longitudinal mediation analysis. Psychometrika 79:543-68|
|He, Hua; Tang, Wan; Wang, Wenjuan et al. (2014) Structural zeroes and zero-inflated models. Shanghai Arch Psychiatry 26:236-42|