From Total Hits to Unique Visitors Modelfor Election’s Forecasting

Using Internet to predict elections has been a topic of interest for different fields and researchers from Google have showed an approach employing user’s queries on that search engine, but recent studies claim that these kind of techniques could not replace the traditional pools of Electoral Predictions using Social Media Data.

Abstract

Using Internet to predict elections has been a topic of interest for different fields. Researchers from Google have showed an approach employing user’s queries on that search engine [4]. Other site, The Daily Beast, has create an “Election Oracle” [1], scanning 40,000 blogs and social media sites and applying a sentiment analysis to made their predictions. In both these case, the predictions are expressed as a likelihood of winning and not the total amount candidate votes or percent expected. This makes sense because they have applied their methodology to U.S.A elections which are based in a two-party system where one candidate won and the other lose. An multi-party approach was proposed by Tumasjan et al [3], they state that is possible to predict the result by counting the number of Twitter mentions of Political Parties and Candidates. They have tested this idea in 2009 German elections obtaining a similar accuracy of traditional election polls. However, all these methods require an important span of time to be implemented. Moreover, recent studies claim that these kind of techniques could not replace the traditional pools Limits of Electoral Predictions using Social Media Data,[2] setting, among other things, that these algorithms do not offer a methodology to sample data.