WI21-01: Using Online SSDI Conversations to Improve Communication and Outreach

Abstract

Text analysis of data collected from online forum conversations, a form of user-generated content (UGC), reveals that Social Security Disability Insurance (SSDI) applicants and recipients, the “customers,” share concerns and confusion about the application and appeal process rules and policies. Extant research suggests that confusions about how SSDI rules are interpreted and applied significantly contribute to high SSDI rejection and appeal rates. This study attempts to provide insights into designing effective communication strategies to reduce confusion and improve customer service experiences and welfare. Given the size of the data, we first use unsupervised machine learning algorithms to derive topics and model them using epistemic network analysis (ENA) via conversational connections. The resulting ENA provides insights on the structural relationships between different issues surrounding SSDI (e.g., struggles the applicants faced in communicating with and obtaining information from SSA). Taking advantage of the longitudinal nature of the data, we also model trajectory ENAs to investigate how these issues evolve against the backdrop of environmental and policy changes. To provide deeper contextual value through human judgment, we use the derived topics as seed words in nCoder (an automated classifier). The resulting codes can be used in different applications, from analyzing the efficacy of existing policy to providing practical policy recommendations.