Chris Hanretty

UPDATE: There’s a reason academics shouldn’t blog so much. I’ve now realised that the analysis below is mistaken; I had calculated the variance of Callegaro and Gasperoni’s measure incorrectly. The revised picture - which I’m still working on - consequently looks a lot flatter, with little or no change over the last fifteen days of the campaign.

In a search to understand what might happen between now and election day, I’ve taken a look at the polling from the last election. Then, the pollsters got it wrong - either that, or people changed their minds in between the polling embargo and the day of the election.

The answer, I think, is a little bit of both. I’ve replicated the wonderful analysis of Simon Jackman for the 2006 polls, using Italian data from Callegaro and Gasperoni as a starting point, and adding data from the ITANES tracking survey from the 40 days before the election.

Key points?

1. People did change their mind in the last fourteen days, and they shifted to the centre-right;
2. All but five of the 73 polls overpredicted the left’s support;
3. The infamous Penn, Schoen, & Berland poll was not accurate at the time it was taken;
4. The three most accurate pollsters were Piepoli, TNS Abacus, and Ispo; unfortunately TNS Abacus hasn’t provided a poll this time round, and Piepoli has only supplied one
5. The three least accurate pollsters were Tql, Simera, and Brunikmedia. Only Brunikmedia is still providing polls.

Response to Callegaro and Gasperoni, pdf.