Bibliographic review of the electoral forecast through big data.

Abstract

Few studies offer an overview of the level of predictability of big data in political science. This study makes a description of the origin of the information, context, level of error, and statistical prediction that big data uses for electoral forecasting. This bibliographic review was carried out with the Google Scholar search engine. In total, 41 studies were found that met selection criteria, 13 employing computational methods, 19 sentiment analyses, and 4 supervised sentiment analyses. The result of the study revealed that big data is mainly focused on the use of social networks, particularly Twitter's API (Application Programming Interface). Big data was found to be a growing technique that presents electoral forecasts with an average MAE (Mean Absolute Error) of 2.7%. Almost all the publications are made through isolated case studies, without identifying, so far, a general integrative theoretical model. It is concluded that there is limited evidence of the development of political science with the use of big data, especially in Latin America.