Significant research resources have been dedicated towards investigating the feasibility of using unstructured data — primarily social media data — to forecast behavior in electoral contexts. Most of these efforts have focused on using social data as a proxy for public opinion data in predicting voting patterns and very few have proven to be effective outside very controlled circumstances. However, the ready availability and sheer volume of social data, combined with the ongoing challenges facing traditional opinion research methods, recommends continued analysis of the utility of unstructured data.
In this AAPOR presentation (May 18, 2019), Ipsos’ Chris Jackson outlines our work using social media analytics, provides a framework for best use of unstructured data moving forward, and discusses specific tools and applications of social to election forecasting.
Since the beginning of 2018, Ipsos tracked over 500 elections in Brazil, Canada, Mexico, and the United States using a combination of survey data collection, statistical modeling, expert consultation, and social media analytics. This experience has highlighted the challenges in using social media data as a substitute for survey data in predicting population behaviors. However, through these elections, we also believe that we have found the appropriate fit for purpose of social data analytics in the electoral context.
We believe social data can be used for three separate purposes. First, social as an indicator of enthusiasm and directionality in popular support. Second, as a measure of the effectiveness of campaign communications and the potential for disinformation messaging. Third, as a supporting data stream into our survey-based MRP models forecasting election results.