Polling in the New Era of Italian Politics

In late February, something happened to the Italian government that had never happened before: a hung parliament. After 75 percent of the population turned out to vote, it took two days to tally the results. Now, almost three weeks later, the center right and center left parties remain in a steadfast gridlock. A third party–the so-called 5 Star Movement–received only marginally fewer votes. Which wouldn’t be all that exciting, except that the 5 Star Movement is just a single guy, and that guy is a comedian–Beppe Grillo–who refuses to speak on television or radio, lest his message be skewed. Instead, he blogs and he speaks in the piazzas around Italy.

The hung parliament and the comedian contender mark an odd time for Italy, which will have it’s next election in less than a year, according to people who know about Italian politics (of which I’m not one). Things are changing online and off, and the old ways of doing things are clearly not holding up. That’s even true of the polling methods that were used to predict the outcome of the election in the weeks leading up to it.

In a great article on the Daily Beast, which gives an awesomely readable and accessible account of the ever-so-complicated Italian political climate, Tim Parks recaps the final counts for the four main candidates, Luigi Bersani, Mario Monti, Silvio Berlusconi, and Grillo, respectively:

So 30 percent to grim support of the old workplace; 10 percent to the gentleman upholding the international monetary system; 30 percent to the rich, old guy inviting us all to have a good time; and 25 percent to the wild man who wants to kick ass.

How did the pollsters fair? Not great. They gave 40 percent to Bersani (the old workplace), 20 percent to Monti (the gentleman), less than 20 percent to Berlusconi (the old rich guy), and 15 percent to comedian Beppe.

Northeastern professor Alessandro Vespignani was also watching the election closely, and he and his team at the Laboratory for Modeling Biological and Socio-technical Systems made a few predictions of their own. They got all of them right except for one: Berlusconi.

How did they do it? And why were they off for the old rich guy? Well, it all goes back to that old ways thing I said earlier. Calling people on the phone seems to no longer be the easiest or most efficient way to probe society’s sentiments. In collaboration with researchers at the Institute for Scientific Interchange Foundation in Italy, Vespignani’s team tracked voter’s intentions by what they said on Twitter, instead. And actually, according to one of the project coordinators, Northeastern researcher Nicola Perra, “We did not make predictions. We just studied the raw signal, and there are biases, in geography, age, et cetera.”

On their website, TweetPolitik (it’s all in Italian), the team demonstrates results from three different types of analyses, which they performed in the weeks leading up to the election.

First, they made activity maps using geo-localization data. These show where in the country people tweeting about each of the candidates reside. “Each dot in the maps is a tweet. The transparent circles represent how intense the signal was in each census area,” said Perra. So, the map of Twitter activity about Beppe Grillo looks like this:

Activity map displaying tweets mentioning 5 Start Movement candidate, Beppe Grillo. Image via TweetPolitik.

Activity map displaying tweets mentioning 5 Start Movement candidate, Beppe Grillo. 53% of the tweets came from the northern region of the country. Image via TweetPolitik.

While the one for Berlusconi looks like this:

Activity map showing tweets for Berlusconi. Image via TweetPolitik.

Activity map showing tweets for Berlusconi. Nearly 50% of these tweets came from the North. Image via TweetPolitik.

Conversation maps show the highways of conversation on Twitter for each party. These data came from hashtags associated with the various parties and represent people in different areas tweeting about the same thing:

Conversation map showing Twitter discussions using hashtags associated with Bersani's party, the center left. Image via TweetPolitik.

Conversation map showing Twitter discussions using hashtags associated with Bersani’s party, the center left. Most conversations took place between residents of large metropolitan areas, usually in the North or South of the country. Image via TweetPolitik.

Finally, (and this one is my favorite), the monitoring stream graphic shows how various discussion topics evolved over time in a ten day period surrounding the election.  On February 19th around 7 o’clock, Beppe was a popular subject:

A snapshot of twitter activity by Italian voters in the days leading up to the election on February 24. Image via TweetPolitik.

A snapshot of twitter activity by Italian voters in the days leading up to the election on February 24. Image via TweetPolitik.

All of this is just reporting the signal that they saw in the data. The “predictions” (which weren’t really predictions, but just signals) came from the global share of activity seen for each party. This stuff isn’t presented on the site.

“In summa, some numbers were really close to the final outcome, others were not,” said Perra. “The reasons of the discrepancies are biases. We could have corrected the signals considering age distribution, Twitter penetration in different areas, but we decided not to make it about prediction, at this round.”

But this still leaves the question of why Berlusconi evaded their glance, even if it wasn’t a prediction. That has to do with social phenomena, said Vespignani. Berlusconi promised to repeal an important but hefty tax put in place by the last president, Monti (who actually took over for Berlusconi when he was obviously flailing during the height of the economic crisis). Everyone knew this was a bad idea for the country, but it was also incredibly tantalizing for each individual taxpayer. So, no one in their right mind would claim to support Berlusconi, in person or on Twitter. But behind the safe shield of a voting booth? That’s another story.