Social Media As Qualitative Research: Reports of a Data Mining Experiment

Background:  Consumers are routinely broadcasting information about their health, their state of mind, and getting and giving advice about health through social media. Social media offers volumes of rapid, real-time, self-reported and unfiltered information.  

Program background:  We wanted to explore if the unsolicited and easily accessed information posted by consumers on social media sites would correlated with findings from previous rounds of expensive and time-consuming qualitative research. Our topic was parent’s observations of possible poisonings, their course of action and their view of poison control services. We also sought to determine if such channels offered an entry point for public health interventions or promotion opportunities. Using two subscription services that aggregate streams of information from the social and traditional web, hundreds of thousands of blog sites, online discussion forums, Twitter feeds and news articles were searched. Words used in conjunction with “poison control” were analyzed to determine context and ranked for sentiment. Facebook was excluded because it does not allow outside services access to data.

Evaluation Methods and Results:  In a 3 month period “poison control” appeared in 1100 blog posts, 428 comments on blog sites, 860 discussion forums, and 160 Twitter feeds. Nearly 30% of social media mentions using the words “poison” and “control” were unrelated to an experience with a poison control service. Relevant mentions of unintentional exposures or crowdsourcing of advice, largely by parents to other parents, skewed negative in sentiment. This initial finding was in conflict with qualitative data, however, detailed observation of individual posts, easily accomplished using both services, revealed that the actual experience of talking with a poison control service was overwhelmingly positive. Needing the service was what produced negative sentiment. Terms and conversation tags such as “parenting fail”, “mom fail” and “bad parent” were commonly included in these posts. This exact sentiment also surfaced in a series of 10 focus groups previously fielded with parents at.

Conclusions:  The cost approximately of the social media mining services combined was far less than even a tenth the cost of our qualitative research. The posts also afforded opportunities to engage directly with consumers, which helped build a community of parents who were then able to champion poison centers as a resource.

Implications for research and/or practice:  Social media offers significant opportunities for understanding consumer health behavior that are more cost-effective than traditional research methods. Observing health “crowdsourcing” and unfiltered experience also affords opportunities for direct engagement.