On February 2013, over 35 million Italian citizens voted to renew the national Parliament. For the first time in Italian history, Internet played an important and perhaps decisive role as place of political debating. In order to investigate the role that social media played in this electoral campaign, we collected social media mentions from Facebook and Twitter of the main political leaders during the month prior the election. This resulted in dataset composed by more than 2 millions Twitter messages and 25 millions Facebook interactions. Starting from critical multi-level approach, present paper will attempt both to deal with complexity of election predictive models based on online data and to highlight main determinants affecting vote and online mention. All these questions will be answered through the use of large-scale longitudinal data.