Tracing Policy-relevant Information in Social Media: The Case of Twitter before and during the COVID-19 Crisis

Abstract Real-time social media data hold great conceptual promise for research and policymaking, but also face substantial limitations and shortcomings inherent to processing re-purposed data in near-real-time. This paper aims to fill two research gaps important for understanding utility of real-time social media data for policymaking: What policy-relevant information is contained in this data and whether this information changes in periods of abrupt social, economic, and policy change. To do so, this paper focuses on two salient policy areas heavily affected by the lockdown policies responding to the 2020 COVID-19 crisis – early childhood education and care policies, and labor market policies focused on (un)employment. We utilize Twitter data for a four-month period during the first wave of COVID-19 and data for the same four-month period the preceding year. We analyze this data using a novel method combining structural topic models and latent semantic scaling, which allows us to summarize the data in detail and to test for change of content between the period of ‘normalcy’ and period of ‘crisis’. With regards to the first research gap, we show that there is policy-relevant information in Twitter data, but that the majority of our data is of limited relevance, and that the data that is relevant present some challenges and limitations. With regards to the second research gap, we successfully quantify the change in relevant information between periods of ‘normalcy’ and ‘crisis’. We also comment on the practicality and advantages of our approach for leveraging micro-blogging data in near real-time.