Given the large number of companies providing polling information, and concerns over their instrumentalisation expressed two days ago by Nando Pagnoncelli, head of the association of Italian market research companies, we should be cautious about interpreting polling data.

We should also have been more cautious in 2006, when polls showed the centre-left with a solid majority which evaporated. It’s not clear whether this was due to bad polls, or a shift in public opinion during the last fifteen days of the campaign, when polls may not be published.

This paper reviews the accuracy of polls from 2006, comparing them with the final election result. This is not appropriate if you believe that there were shifts in public opinion over the last fifteen days. But, for the record, the authors argue

• that coverage errors (failure to account for large number of voters with no landline but only a mobile phone) and failure to screen for likely voters were the reasons behind the inaccuracy.
• Euromedia and Penn Schroeder Borland were the closest to the final results, despite the “manifestly distorted” graphics used by the latter