Temperature and temperament: Evidence from Twitter

How do people value their climate? This paper demonstrates a new approach to estimating preferences for nonmarket goods using social media data. I combine more than a billion Twitter updates with natural language processing algorithms to construct a rich panel dataset of expressed sentiment for the United States and six other English-speaking countries around the world. In the U.S., I find consistent and statistically significant declines in expressed sentiment from both hot and cold temperatures. To better understand how preferences may adapt, I document heterogeneity in both regional and seasonal responses. I complete the U.S. analysis with a suite of validation exercises to understand the magnitude of these effects and two methods to estimate willingness-to-pay for climate amenities. Finally, I document similar relationships between temperature and expressed sentiment for four out of the six non-U.S. countries I examine.