Italian Referendum 2016 – wrapping up

At some point I will do a paper summarizing advantages and shortcoming of the method I first applied with success for the Greek Referendum in 2016. What have we learned from applying it in the Italian Referendum?

  1. Unlike the Greek referendum just looking at no and si was not enough. This means we had to specify something more to properly identify relevant searches.
  2. Looking at the unquoted votare no and votare si did somewhat better but left out many other relevant searches plus we do not know what the unquote ones choose exactly.
  3. Looking at the searches in 2 but quoted i.e. the exact sentence is even more restrictive.
  4. As we monitored 2 we were getting weaker results for NO down the stretch because si means oneself in addition to yes hence we were picking searches like a che ora si puo votare il 4 dicembre. When we cleaned these we got closer: no9
  5. In all cases there was a drop in the last day. Probably people were making up their minds and leaving the search.
  6. Looking at perche votare no/si lost significance towards the end.
  7. Interesting results can be obtained by looking at io voto si -no and io voto no -si.

In all cases search clearly showed results ranging from “yes could not possibly win” to “no wins hands down”.


This entry was posted in Big Data, data, nowcasting, Social Media and tagged , , . Bookmark the permalink.