I am still reluctant to write a political analysis (and I think I will simply let the issue fade out), but I would still like to make some final considerations upon the process I witnessed at SWG during the last weekend – not the election itself, but just an attempt to predict its outcome.

When the dust settled, data has been re-examined in order to compare the prediction (made when the ballot boxes were still open) with actual results:

I have heard and read a lot of hostile comments, especially here (on Mantellini’s blog, but not just there), and I think that there are some lessons to be learned from what happened with exit polls, even if on a lesser scale than in 2006: we – the public – are given a range such as 40%-43%, and I always assumed that the estimate is halfway between the lower and upper bound, that is as if the estimated result for the winning alliance was 43%(±1.5%). But, as Roberto Weber explained in a note he sent us after the real results were available, last minute decisions are taken in this regard – and a lessening of the difference in the last hours (suggesting a converging trend) made them choose to put the estimate (43% for the PdL+Lega+MpA alliance) at the upper end of the 3% interval and not at the middle.

Statistics in experimental science are rarely done with complex tools: the arithmetic mean and the standard deviation are good enough, because data is generally reliable. Psychologists and social researchers need more sophisticated tools, because their data is often poor in quantity and quality: by resorting to online polls, SWG is able to tap into a huge sample (63,000 users) and their results, in my opinion, proved that the job can be done through the web.