From Post to Policy: Using Social Media Data to Inform Decision-Making

Abstract

While researchers and policy-makers traditionally rely on survey methods as they seek to understand preferences, user-generated data on social media—coupled with advanced methods of Natural Language Processing—can, in certain cases, serve as a valid alternative. In this thesis, I introduce a novel data set of global social media content and present a multilingual algorithmic method of text analysis which provides valuable insights into population well-being and public opinion at a global scale. I conduct three validation tests to assess the extent to which metrics computed from social media data are consistent with more traditional methods of measurement such as census population counts, well-being surveys, and political polls. I go on to present two case studies which rely on social media-based metrics. In the first, we evaluate the effect of temperatures on subjective well-being worldwide. We find a non-linear, inverse U-shaped relationship and estimate high-temperature damages in a large selection of countries. In the second, we connect subjective perception of climate events with real estate market outcomes. We find that while objective temperature stress is consistently associated with lower location value, regions where sentiment is most sensitive to climate discomfort are also the ones where these shocks are the strongest. Both empirical studies confirm the strong potential of social media data for policy-makers and researchers alike.