Opinions in the Twittersphere

• In Event: Methods for Digital Data

Abstract

Ongoing conversations within the survey and social science communities suggest that Twitter offers a rich source of real-time public opinion data: a way to track opinions in an important segment of the mass public in the digital age. On the other hand, much of the optimism about Twitter as a source of real-time opinion tracking has yet to bear out in the way many expected. A number of studies have shown that Twitter tracks consumer sentiment, candidate support, movie revenues, the stock market, and other phenomena. Yet, others who have attempted to replicate these findings have found mixed results. In this study, we compare a probability-based sample of Twitter users with data from a general population survey along an array of common survey attitude measures. We then use text-modeling and machine learning techniques to explore the value and challenges of inferring the opinions of individuals based on their Twitter data. We highlight the challenges of reliably inferring preferences or attitudes from Twitter data and identify new avenues for more general kinds of opinion estimation.

Authors

• Stefan J. Wojcik, Twitter

• Adam G. Hughes, Pew Research Center