In recent years, the Internet has become increasingly interactive and user driven as the availability of social media technologies has grown. Consequently, research has shifted from a focus on mainstream media, such as television and print newspapers, to new social media and mobile applications. Although social media are most often associated with Facebook, YouTube, and Twitter, in practice, social media include a broader range of websites that incorporate tools that facilitate interaction among users and build virtual communities through blogs, chat rooms, newsfeeds, and other communication tools (Boyd 2008). This paper presents two case studies that illustrate social media data analysis on behalf of government agencies to better understand target populations and to shape and evaluate programs and services. The case studies highlight common characteristics of social media data analysis approaches across topic areas. Further, the limitations of social media data and the opportunities and limitations of automated scanning tools used to collect and analyze large amounts of publicly available, unstructured, online data are discussed.

The first case study describes the efforts of the Centers for Medicare & Medicaid (CMS) Office of the Medicare Ombudsman (OMO) to identify systemic beneficiary issues and unintended consequences using social media data analysis. According to Pew Research Center, as of April 2012, over half of American adults ages 65 and older reported using the internet or email. The goal of the OMO environmental scanning demonstration was to develop mechanisms to gather feedback from Medicare beneficiaries, caregivers, and advocates to make recommendations to Medicare to improve the beneficiary experience.

The second case study describes the methodology and results of an Office of National Drug Control Policy-sponsored study that gauged the influence of drugs and alcohol associated online content on youth. Most youth use social media daily and teens are increasingly multitasking across multiple media with estimates as high as10 hours a day (Kaiser 2010; Common Sense Media 2012). Automated social media content scans and virtual observations of blog conversations and social media postings were employed to collect and analyze drug- and alcohol- related interactions occurring online.

A synthesis of limitations and subsequent lessons learned includes:

1. Neither the social media data collected through automated scanning tools nor virtual observations can be considered representative of a specific population or subgroup. Therefore, it is difficult it is to attribute content to specific individuals—oftentimes engagement with social media occurs with near anonymity—and the corresponding lack of information on the coverage of these data, prevent construction of accurate population-based estimates. This results in exploratory and descriptive findings.
2. People use different terms to describe issues and concerns online than policymakers, healthcare providers, or researchers.
3. In content analysis, specific keywords yielded more relevant or precise data. Keyword based searches, thus, evolve in real-time based on findings.
4. Automated scanning must be paired with manual research – and supporting data analysis if available – to understand the context and scope of the information.
5. Automated scanning identifies innovative sources of information that may not be otherwise identified through manual searches.