Welcome to European Tribune. It's gone a bit quiet around here these days, but it's still going.
Display:
I'm suspecting that polls are having systematic problems that are not reflected accurately in the probabilities. Apart from the problems of predicting elections (and I would include last UK parliament election in the list) I have also noticed in my work that people increasingly don't answer their phones when you call from a number they don't know and people in comment sections during the US election writing about how if you answered once, pollsters kept calling (without being asked to be in a poll).

These are annecdotes, but could point towards inexpensive and accurate polling over the phone being a thing of the past. Door to door would still probably work, but I don't think anyone does that anymore. Or to put it another way, as a society we traded inexpensive and accurate polling for phone salespeople once caller ID became a standard feature.

by fjallstrom on Sat Apr 22nd, 2017 at 03:26:23 PM EST
All fair points, but it is difficult to know whether such problems bias the sample towards any particular candidate. For the moment, an aggregation of such polls is the best indicator we have of how the election is trending, but my confidence in any predictions based on polls in a tight race is low. Thus any combination of two from the four leading candidates is still possible, and the likelihood of a Macron/Le Pen match-up is only marginally higher than the other match-ups.

Fladem has an interesting chart of average polling error up in a comment on this diary on Booman.

The largest error recorded in the sample period is 3.36%, which would not be enough to lift Fillon or Mélenchon above Macron or Le Pen. So it would be a bit of an upset if either made the second round at this stage.  But as you say, the problems with telephone polling may have been getting gradually worse...

Index of Frank's Diaries

by Frank Schnittger (mail Frankschnittger at hot male dotty communists) on Sat Apr 22nd, 2017 at 04:33:29 PM EST
[ Parent ]
One way to assess the pollsters' work would be to check how they did during the last presidential election in 2012:

During the last week before the first round on 22 April, Hollande was polled at 27-28% (he got 28.6% of the vote), Sarkozy 26-28% with CSA at 24% (he got 27.2% of the vote), Le Pen was at 14-16% during the first half of April but last week polls showed her at 17% (and she ended up with 17.9% of the vote). The largest difference was with Mélenchon who was polled at 13-15% and ended up with only 11.1% of the vote; Bayrou was also polled at 9-11% and received 9.1% of the vote.

All in all, it looks like the polls didn't do too badly, catching the rise of Le Pen in the last week but less so the fall of Mélenchon, and the differences between the polls vs. first round results were around 1% for the first three candidates.

by Bernard on Sat Apr 22nd, 2017 at 06:04:46 PM EST
[ Parent ]

But he also underlines large error margins: FN last polled between 17 and 28 %.

by Bjinse on Sun Apr 23rd, 2017 at 12:43:37 PM EST
[ Parent ]
fjallstrom:
These are annecdotes, but could point towards inexpensive and accurate polling over the phone being a thing of the past. Door to door would still probably work, but I don't think anyone does that anymore.

True enough: there was a subject on TV last week on French pollsters. To a large extent, they no longer do phone calls but run their polls over the net. Their panel members must fill a questionnaire to sort them out by age, gender, occupation etc... and they are regularly solicited by email, which is easier to ignore if you don't feel like filling the web form out (poll companies like Harris also offer goodies and a contest to lure people to answer the poll).

by Bernard on Sat Apr 22nd, 2017 at 06:11:46 PM EST
[ Parent ]

Display:

Occasional Series