Does Reddit know better? The impact of emotions on Presidential approval rates of public opinion polls

This study investigates whether public’s emotional state relates to public opinion polls, in particular the approval rate of Barack Obama during his eight years of presidency of the United States. Given the positive penetration rates of Internet users and the recently released substantial database of Reddit comments, the main objective of this study is to determine to what extent public emotions as measured from a large-scale collection of Reddit comments can be used to track public opinion polls. Despite the frequent use of polls, one must acknowledge three flaws, i.e. inaccurate data, requirement of large amount of respondents and time consuming to accomplish. Therefore, polls are often expensive to conduct. As a reply to these shortcomings, sentiment analysis uses Natural Language Processing techniques (NLP) to determine opinions from online sources. Sentiment analysis tools that automatically extract opinions from textual comments are quick, inexpensive and applicable on a large-scale. Previous research in sentiment analysis mainly focuses on a two-dimensional scale (positive versus negative). Polls are often limited to two-dimensional division (approve versus disapprove). NLP tools can provide a multidimensional view on public opinion by means of emotion analysis. A linear regression model with AR(I)MA errors is fitted to investigate the main question. The answer to the main question is twofold; on one hand, no significant results were found and therefore the study emphasizes that the approach was not sufficient to confirm the ability of public emotion to track presidential approval rates. On the other hand, emotional shifts aligned to remarkable events were captured. As a consequence of this conclusion, I still believe that expanding social media tracking tools with emotional analysis can be beneficial over relative costly and time consuming surveys and public opinion polls. Further research should move beyond the idea of supplementing public opinion polls and instead focus on substituting the polls.