I had an interesting chat yesterday evening on how polls are reported by the media. Badly, in most cases.
A good example of this from the FT (surprisingly) came yesterday:
Angela Merkel’s party falls a couple of points in one poll by one polling company (that tends to overestimate the CDU/CSU) becomes a full story on slipping in the polls.
The chat I had yesterday was specifically about Italian polls and how they performed during the last election. The post-election consensus was that they did poorly, but in reality on closer inspection (overestimating the centre-left PD by 4-5 points aside) many of the figures and trends weren’t that poor. The real issue was how they were interpreted (by myself included in some cases) - but, lots of the data was there and should have been emphasised more:
- see Tweet here and final pre-election figures.
- points 4) 5) and 6) here
- this on regional polls - the narrow gap nationally meant that these went to the centre-right, and what should have also been clear pre-election is that even had there been a 5-point gap nationally this wouldn’t had been sufficient to swing several historically centre-right regions to the centre-left.
- 54% chance of no majority emerging in the Senate.
In short, the polls had one clear shortcoming which had a significant impact on the election outcome due to Italy’s voting system (other than the fact that while there were quite a few national polls, the regional coverage of data is very scarce making predictions the more complicated - there were clear limitations within the polling data when looked at post-vote) - yet, many trends were evident but weren’t adequately interpreted by most analysis and coverage.
I’m confident everyone will have learnt a lot and will apply these learnings in time for the next Italian election.
p.s. on this topic I went to a great talk by Nate Silver at the RSA yesterday afternoon which you can listen to here.