The next panel at AoIR 2013 is one which I'm presenting in as well - we've brought together a number of presentations on the use of Twitter in national elections. The first presenter is Luca Rossi, whose focus is on the 2013 Italian election. He and his colleagues have examined activity on Twitter and Facebook during the month before the February election, gathering some 2 million @mentions and finding Facebook content which its own metrics reported some 25 million users talking about.
Is such activity related to the eventual election results at all? Can it predict the election outcome, in fact? This would mean taking the role of opinion polls, which in this election also turned out to be incorrect, partly due to the shifts in the party systems resulting from the rise of the Cinque Stelle party.
One way to attempt such predictions is to link @mentions of specific social media accounts to particular parties - an @mention of @berlusconi_2013 would add to the numbers for the PDL, for example. But the social media interaction trends are also very different from general electoral opinion - some parties were especially strong on Facebook, others on Twitter, due to platform-specific demographics and user cultures.
From an observation of the divergences between polls, @mention and "talking about" trends and the eventual election result it is possible to calculate the error margins for each measure, for each party, and also to track the mean average error over time. In polls, the mean average error tends to decrease towards Election Day, but this is not the case for the Twitter and Facebook data - so we're dealing with a very different kind of data here.
Overall, there is no clear correlation between the social media activity and political surveys; rather, the activity is highly related to mass media coverage. This is often commentary on political developments, rather than an expression of support for specific perspectives. So, social media activity does not appear to be predictive of election results - rather, it is related to broader media dynamics more than the evolution of political opinion.