Tuesday, May 01, 2007

How could German pollsters get it so wrong?

I have analysed UKPollingReport data for the months before the general election in Germany in 2005. According to the polls the CDU should have got around 42% of the vote whereas it actually got 35.2% just ahead of the SPD’s 34.3%.

According to my regressions Emnid and Infratest under-estimate the projection for the CDU, which would mean that their projections are actually closer to reality. Emnid works with the ARD, whereas Infratest works with N24. This regression controls for the time trend. Forsa, which works with the conservative magazine stern overpredicted the CDU share of the vote even more.

One reason seems to be that some people switched from CDU to FDP, which was under-predicted. Another observation is that over the months we see a declining share for other parties (except the main five) among the polls, whereas in reality this is about 4%.

No comments: