Site Meter

Monday, June 14, 2004

Nexus and the Olive Tree

I would have thought that spin stops while votes are being counted (except in Florida in 2000). Given the vague powers of the European parliament, I assume that spinmeisters (masters maitres maestri) will work overtime interpreting the results, but I would have guessed that exit polls and projections were done honestly. It is not clear how one would want to distort them since the final numbers will arrive by tomorrow.

I continued to think this even when Rai (Italian public television) hired prime minister Berlusconi’s favourite polling firm (datamedia) slightly disguised as part of a consortium called Nexus. Representatives of l’Ulivo (the olive tree) the main opposition alliance, objected at the time.

Now I see the outer limits of spin. Yesterday Berlusconi was hammered at the polls. This is no surprise first because he has made clear to Italians what he is and second because he has linked himself to Bush and the invasion of Iraq. However, he and other leaders of Forza Italia (yaay Italy) did not manage to lower expectations far enough and fast enough. The line going in was that anything over 25% would be OK even though the party got almost 30% in the last parliamentary election.

The results from Nexus quickly showed that this wasn’t low enough. They also showed a very clear downward trend so far tonight. I present % vote for Forza Italia, the whole center right coalition and the whole center left coalition in successive predictions.

Forza Italia Center Right Total Center Left Total
Exit Poll 22 % +/ 1.5 % 45.5 +/- 5.5 % 46.5 +/- 5.5 %
First Projection 22.3 % 46.5 % 44.6 %
Second Projection 21.8 % 46.6 % 44.4 %
Third Projection 21.5 % 46.1 % 44.6 %
Fourth Projection 20.7 % 45.4 % 45 %
Fifth Projection 20.5 % 45.3 % 45.6 %
Sixth Projection 20.7 % 44.9 % 45.9 %
Actual vote 20.5 % 44 % 47.4 %
33388 of 65878 precincts reporting

The really weird projections concern two tiny parties representing linguistic minorities, the UV which represents French Speakers in Val d’Aosta and the SVP which represents German speakers in Sud Tirole/Alto Adige. These projections can only be described as Merde/Sheitz, since each party was consistently projected to get 0.0 % of the vote. Clearly the allegedly representative sample of precincts did not include any precincts in Val d’Aosta or the Sud Tirole.
With 33388 of 65878 precincts reporting the SVP had 0.7% of the vote so they can not (even with rounding) fall below 0.3 %. Earlier totals were closer to 1% (they count relatively fast in the Sud Tirole).

Now it seems to me that, in this day and age, there is no need to choose a sample of precincts to make projections. It should be possible to have a data base of vote by precinct from past elections and calculate total votes now divided by votes then for reporting precincts, then to use this factor to project national totals as a multiple of last elections national totals. This won’t work if precinct boundaries change (some might have changed boundaries but I think most don’t). It has the advantage that, since it is automatic, no cheating is possible. It would also avoid offending the Sud Tirolishers

Needless to say center left politicians are irritated (in spite of being pleased about the results). I am irritated too, since a thunder storm interrupted my power supply 9 times while I was posting this and I know that Berlusconi is behind it sometimes (can't wear my tinfoil hat in a thunderstorm you know).

No comments: