In my last post, on December 2nd, the forecast of No vote was 54% for Italy. The result was more negative for Renzi: 59% No.
During last week I have been posting a “regional forecast” based on relative interest in “votare si – no” in the different regions, despite the fact that, to my knowledge, there were not polls published at the regional level. The reason was to test the Google tool in an heterogeneous environment. The Mezzogiorno ( Abruzo, Campania, Puglia, Basilicata, Calabria e Molise) + the islands (Sicily and Sardinia) suffers from much worse macro conditions than the Northen part of the country. Among other things, this includes a big digital gap with respect to the North, as I pointed in my post on December 1st: The southern regions were seen has having a highest probability of a No vote. If the smaller fraction of internet users in the South are not so representative of the population we might be getting biased results here in a larger extent than in the Northen regions.
In fact this is what has happened.
As seen in the enclosed table, the biggest errors, by far, are seen in the South, where the actual rate of No vote so about 25% higher than expected by using Google searches. There is one exception, Abruzzo, where the error was less sizeable.
In my view these results may ask for further research, We see that the error in Lombardy, which almost doubles the size of the other large italian regions (Lazio, Campania, Sicilia), the error is very smal. That maybe a coincidence or may be it is related to the fact that Lombardy is not only the largest but also the most developed region in Italy.
In short the idea of exploiting the regional differences allows for variability and will help in getting a deeper knowledge of behavior of the public when facing choices.