Chapter 5: Why Don’t Tweets Consistently Track Elections? Lessons from Linking Twitter and Survey Data Streams, Josh Pasek and Jake Dailey

This chapter attempts to identify conditions that discriminate election predictions derived from Twitter data with those from more traditional survey-based analysis. Comparing tweets about Obama and McCain in 2008 with data on candidate favorability from a large rolling cross-sectional survey, we test four potential explanations for the discrepancies that emerged in earlier studies. These include the proposals 1) that distinctions in results are principally a function of demographic differences between Twitter users and survey respondents, 2) that Twitter data may capture trends in perceptions of the candidates rather than absolute levels of support, 3) that the sentiment of tweets may operate independently in predicting favorability for each candidate, and 4) that tweets and survey metrics may only come to align toward the end of the election cycle, once there is more general attention to the campaign. By recognizing the dynamics of election campaigns, this chapter identifies consistent results linking social media sentiment with the favorability of both major party candidates in the 2008 election. These correspondences, however, were only consistent in the most intense part of the campaign and were not apparent earlier in the election season.