Using Demographics in Predicting Election Results with Twitter

Eric Sanders, Michelle de Gier, Antal van den Bosch

Abstract: The results of two Dutch elections are predicted by counting political party mentions from tweets. In an attempt to improve the predictions, gender and age information from the Twitter users is automatically derived and used to adapt the party counts to the demographics in the election turnout. The prediction improves only slightly in one of the elections where the correlation between election outcome and Twitter-based prediction was relatively lower to begin with (0.86 versus 0.97). The relatively inaccurate estimation of Twitter user age may hinder a larger improvement.