European Tribune - Community, Politics & Progress.

Germany: Pollsters' Total Failure (Updated)

by DoDo
Wed Sep 21st, 2005 at 07:37:22 AM EST

We will be discussing the significance of the German election, in many ways, for days to come. Here's an interesting piece from the diaries ~ whataboutbob

As SPIEGEL ON-LINE writes (in German), all five big pollsters failed spectacularly in predicting the result of the German elections. Below a graph summing up the deviations:

During the last elections, Allensbach was the worst - with a sum of errors of 7.2%, then ridiculed as extreme, now even the best is worse. What you can also see reinforced is that the political party-connected institutes (Allensbach and Emnid close to CDU, Forsa to SPD) are the less precise. But as there is a tendency, there was either a false groupthink and skewed questions asked in surveys, or some really strange last-minute developments in voters' minds.

Beyond the now well-known "we don't want a Grand Coalition!" CDU/CSU to FDP swing, there's the around 1% - you can only calculate this - whom they counted for the CDU/CSU instead of the far right, probably a result of not offering the brownies as a choice to the polled people.

Then look at the Greens' consistent underestimation by 1.1% - I can only explain this by a grand swing from CSU (mostly back) to Greens in Bavaria. Which happened, but I'm not sure why and why now - maybe this is an example of traditional voting (see below) not taken into account.

Update [2005-9-21 3:57:43 by DoDo]: I sum up other possible factors suggested so far by others in the comments:

changes in participation: rose in left-voting East Germany, fell in the conservative South (thanks Minerva);
far-right voter hiding: many won't admit their voter intentions, pollsters forgot to foresee that (thanks Saturday);
large numbers of undecideds and no-confidence vote for Merkel (thanks brainwave).

I note that as in the USA, published polls in Germany aren't the unprocessed result of surveys.

In the USA, the survey results are re-weighted by the assumed numbers or party allegiance distribution of registered voters etc.

In Germany, some institutes re-weigh after comparing last election resuts with surveyed people's claims on what they voted for, and some also re-weigh based on assumptions on the power of tradition (i.e., that some people will name a different party than the one they used to vote for to the surveyer, but once in the voter booth, they don't dare to vote differently).

. Home
. About
. Contact
. New User Guide
. FAQ
. ET Editorial Guidelines
. Search
. Search (Google)

Username:
Password:

. Make a new account
. Reset password

Germany: Pollsters' Total Failure (Updated) | 10 comments (10 topical, 0 editorial, 0 hidden)

Re: Germany: Pollsters' Total Failure (none / 0)

Maybe its because I've been working on the computer all day, so missed it in your article...but what are their explanations for being ...so wrong???

"Once in awhile we get shown the light, in the strangest of places, if we look at it right" - Hunter/Garcia

by whataboutbob on Tue Sep 20th, 2005 at 10:12:56 AM EST

New poll (none / 0)

Something I found since at SPIEGEL ONLINE: in a new poll, more than two thirds of voters are dissatisfied with the election result.

Somewhat surprisingly, FDP voters are in the lead with 88%. CDU/CSU voters follow with 81%, 69% of Left party and 68% of Green voters follow, and 'a majority' of SPD voters are last.

On the other hand, regarding my nightmare scenario, in another poll a mere 25% would support repeated elections and 73% oppose, I guess that wouldn't bode well for the CDU/CSU if they pursue a new elections tactic too openly.

As mentioned in other threads, 33% would favor a Grand Coalition, 26% a 'Jamaica coalition' with Greens, FDP and CDU/CSU, less than 20% favore an 'Ampelcoalition' with SPD, Greens and FDP, and a mere 9% a red-red-green coalition.

In there is a Grand Coalition, half want Schröder as chancellor, only 43% want Merkel.

*Traitor*, n.
A benighted individual who perceives an illusory distinction between serving his nation and abetting the criminals who govern it.

by DoDo on Tue Sep 20th, 2005 at 10:13:20 AM EST

Gallup (none / 0)

In the same article: the head of Gallup criticised German counterparts, claiming they failed to ensure representativeness - but I think a failure so far to break into the German market is behind this... (See also my own guesses for explanations for the shifts.)

*Traitor*, n.
A benighted individual who perceives an illusory distinction between serving his nation and abetting the criminals who govern it.

by DoDo on Tue Sep 20th, 2005 at 10:16:08 AM EST
[ Parent ]

Re: Germany: Pollsters' Total Failure (none / 0)

Incorrect sampling and weighting looks like it could explain a lot of this. With so many parties and so much regional diversity, it is probably a lot harder in Germany than in the US, and mistakes in weighting probably become more obvious (in the biparty US, mistakes in estimating Republican allegiance automatically put people in the Democratic column, and vice-versa, creating a better prospect that mistakes on both sides cancel each other out; in Germany, there are a lot more mistakes one could make that do not necessarily offset).

Ambivalence on the right could also explain some of it -- was turnout low in right-wing strongholds? Are there time-frame issues, too (i.e., when were the polls taken)?

National Public Radio in the US the other day claimed there had been a big swing away from the CDU in the last few weeks due to Hurricane Katrina -- it supposedly enabled the left to portray Merkel as a know-nothing, antigovernment Bushite who would leave Germany similarly vulnerable to disaster...

by Minerva on Tue Sep 20th, 2005 at 10:18:57 AM EST

Re: Germany: Pollsters' Total Failure (none / 0)

On your first paragraph: that is rather unlikely, if it would be so, pollsters' predictions would have been off by a similar margin for years. But Germany had four big parties for 25 years, and five of them for 15 years (even if one or two of the smaller was at times below 5%), and could get these four/five rather close to truth previously.

But you mention turnout - a very good idea, come to think of it. I looked up, and indeed you are right: in the (left-voting) East, participation generally increased, in the (conservative-voting) South, it fell. Tough, all of these changes are of a few percents, I'm not sure that explains it all. (As for timezones, Europe is small :-) from the English Channel to the borders of Romania and Ukraine, it is all one timezone: CET, which is GMT+1, or ET+6.)

As for a Katrina effect and campaign in Germany, the first time I hear it - but someone closer to the pulse should decide on that.

*Traitor*, n.
A benighted individual who perceives an illusory distinction between serving his nation and abetting the criminals who govern it.

by DoDo on Tue Sep 20th, 2005 at 10:32:44 AM EST
[ Parent ]

Westerwelle (none / 0)

I added this to the nightmare scenario thread as an update too, but this bit of Schadenfreude worth to repeat: Guido Westerwelle, the current leader of the FDP (who shares much of the blame for turning the FDP an unserious neoliberal party), got only 8.7% in Bonn - 5.5% less than in 2002, and 5% less than list votes for his party in Bonn!...

I admit it is not at all likely, but would be funny if a traffic lights coalition would be formed after a decapitation of the FDP...

*Traitor*, n.
A benighted individual who perceives an illusory distinction between serving his nation and abetting the criminals who govern it.

by DoDo on Tue Sep 20th, 2005 at 10:19:58 AM EST

Isn't the main problem here simply (4.00 / 3)

that one-third of the electorate remained undecided well into election week? I imagine what pollsters do is ignore the undecideds and extrapolate from those respondents that do state a preference. This of course inevitably introduces a special kind of error - respondents that have a preference cannot be representative of those that don't. In a "normal" election, that error can be neglected; in the present case, it could not. To make my point as clear as possible: suppose you saw before the election a poll that said (phantasy numbers, just trying to simplify):

CDU/CSU 40%
SPD 30%
Linke 10%
FDP 10%
Greens 10%

But the data this poll was supposed to reflect would have really looked more like this:

CDU/CSU 28%
SPD 21%
Linke 7%
FDP 7%
Greens 7%
UNDECIDED 30%

And this is exactly what the pollsters should have published.

One more point: I'm highly sceptical of the theory that the massive swing away from the Union was mostly caused by loyal CDU/CSU voters who went for the FDP in order to prevent a grand coalition. First off, this doesn't make too much sense. The only way "lending" votes to the FDP could have helped prevent a big coalition was if the FDP was in danger of falling short of the 5% threshold. But that danger was never apparent in the polls! Secondly, only half of the votes the Union lost actually went to the FDP. Infratest-Dimap projects that the CDU/CSU lost 1,110,000 voters to the FDP and a combined 1,100,000 votes to the Linke, "others" (i.e., mostly, Neo-Nazis), and above all to non-voters. In my opinion, what happened here was quite clearly a no-confidence vote against Merkel.

Damn George Bush! Damn everyone that won't damn George Bush! Damn every one that won't put lights in his window and sit up all night damning George Bush!

by brainwave on Tue Sep 20th, 2005 at 04:32:02 PM EST

Re: Isn't the main problem here simply (none / 1)

...one-third of the electorate remained undecided well into election week? ...In a "normal" election, that error can be neglected; in the present case, it could not.

That may well be. I have the memory of a long line of elections in which the losing side, which also trailed in the polls, sought to keep up hopes by pointing to a large number of undecideds (and various vactors why they would like this party more) - only to end up with the undecideds splitting the same way as the opinionated. But this years' German elections were a truly complex choice for voters.

But the data this poll was supposed to reflect would have really looked more like this:
...
UNDECIDED 30%
And this is exactly what the pollsters should have published.

Fair point! (Actually, this is done in polls in Hungary, because the undecided and refuse-to-answer 'fraction' was always high.)

loyal CDU/CSU voters who went for the FDP in order to prevent a grand coalition. First off, this doesn't make too much sense.

You assume all voters to be as rational as you or me :-) But, it can be rational too: if you are a loose supporter of a right-wing government rather than a staunch CDU/CSU voter, you may think that if a Grand Coalition comes, it is better to have a strong FDP as opposition. At any rate, on election night, ARD showed that support for a grand coalition fell from around 40% to 20% while support for black-yellow changed the opposite way since the last poll, just at the same time as there was the CDU to SPD swing.

I agree it wasn't just the CDU/CSU to FDP swing, in fact that's what I wrote of below the fold. Regarding the Infratest-Dimap voter migration figures, I have to ask whether they concern voter movements in the last week - or voter movements since the 2002 elections. (I suspect the latter.)

*Traitor*, n.
A benighted individual who perceives an illusory distinction between serving his nation and abetting the criminals who govern it.

by DoDo on Wed Sep 21st, 2005 at 03:39:10 AM EST
[ Parent ]

Re: Isn't the main problem here simply (none / 0)

Forgot this:

In my opinion, what happened here was quite clearly a no-confidence vote against Merkel.

That sure played a role in the CDU slump over the last few months and after the TV debate, but in the last few days? On second thought, it very well could: in the undecideds' minds.

*Traitor*, n.
A benighted individual who perceives an illusory distinction between serving his nation and abetting the criminals who govern it.

by DoDo on Wed Sep 21st, 2005 at 03:50:33 AM EST
[ Parent ]

Re: Germany: Pollsters' Total Failure (none / 1)

there's the around 1% whom they counted for the CDU/CSU instead of the far right, probably a result of not offering the brownies as a choice to the polled people

As far as I know, right wing parties are offered as a choice in the polling interviews. But it is still hard to get the numbers of NPD et al. correctly: The pressure of social acceptability (in German social and media science: "soziale Gewünschtheit") causes many right wing voters to hide their real party preference. Because rightist voters, as long as they do not belong to the extremist core, can not afford to become socially stigmated. But normally, pollsters should have a formula for that.

by Saturday (geckes(at)gmx.net) on Wed Sep 21st, 2005 at 03:04:43 AM EST

Germany: Pollsters' Total Failure (Updated) | 10 comments (10 topical, 0 editorial, 0 hidden)

Go to: [ European Tribune Homepage : Top of page : Top of comments ]