Minaret initiative and the pre-vote surveys

I would like to come back to the vote on the Minaret initiative that has taken place on November 29, 2009. The outcome of this vote has been rather unexpected with 57.5 percent of the voters accepting the initiative of the populist right that asked for banning the construction of Minarets in Switzerland. Among other things, the outcome of the vote has come as a surprise, because the pre-vote surveys predicted a clear defeat of the initiative.

In the meantime, the VOX-survey that was, as usual, conducted after the vote, has been published. I have taken a look at the file and discovered to my surprise that even in this survey that has been put in the field immediately after the vote, the results concerning the initiative do not correspond to the outcome of the vote. In fact, according to the survey, only a minority of 48.5 percent of the voters accepted the initiative. Given that there were only 682 interviewees who had participated in the vote and who indicated to have either voted in favor or against the initiative, this outcome is not significantly different from the true outcome. Still, I wondered what might have caused this difference.

My first hunch was that those voters who did not answer the question about the vote or who maintained that they did not know how they had voted, did in fact not want to reveal their vote, and had actually voted in favour of the initiative. The positioning of these voters on the arguments in favour and against the initiative to some extent (although not fully) supports this assumption. However, adding these voters to the supporters’ camp, brings the support for the initiative among the respondents to the survey only up to 50.4 percent.

My second hunch was that the partisans of the right tend to collaborate less in surveys than the partisans of the left. Indeed, the partisans of the SVP are somewhat underrepresented in the survey, if we compare their share with the share the SVP obtained in the last federal elections of 2007 (23.4 percent compared to 28.9 percent). Conversely, the partisans of the SP are overrepresented among the respondents of the survey (27.0 percent compared to 19.5 percent). The Liberals are also somewhat overrepresented, while the Christian-democrats are somewhat underrepresented. The proportions in the survey refer to the voters who have indicated their party attachments. This applies to only 58 percent of the voters. Assuming that the respondents indicating no party attachments are correctly represented and assuming that the partisans of the major parties have been over- or underrepresented as indicated by the comparison with the federal elections brings the outcome of the vote among the survey respondents up to 51.2 percent. In other words, the aspect of partisan over- and underrepresentation does not help much to explain the remaining difference either.

My third hunch was that the difference might be the result of another type of selective participation in the survey – the one due to the voters’ level of education. As the VOX-analysis shows, the level of education has had a very strong impact on the outcome of the vote, with the more highly educated having rejected the initiative much more frequently than the less well educated. It turns out that, compared to the figures of the Federal Office of Statistics for the level of education of the population of Swiss nationality aged 25-64, the survey sample is considerably “overeducated”. Correcting for this education bias of the survey sample, the outcome of the vote in the survey gets up to 56.0 percent in favour of the initiative. This still falls short of the real outcome, but comes very close to it.

I conclude from this analysis that the major reason for the difficulties of the surveys to predict the outcome of the vote is not in the first place that survey respondents do not tell the truth. Nor is it the partisan selection bias in survey participation, as I had expected myself. In fact, the major reason for these difficulties resides in another type of selection bias – the one related to the level of education.


4 thoughts on “Minaret initiative and the pre-vote surveys

  1. In this case, pollsters should easily be able to weigh their dataset according to the distribution of partisanship, education and age, given they also ask these questions.
    To me it comes as a big and disappointing surprise that pollsters (state-paid scientists) do not take into account such biases in their analysis, when we see that the exact term can be predicted more or less accurately, as soon as one “corrects” for sociodemographic variables.

  2. Thank-you very much Mr. Kriesi for this intersting hypothesis. It is the most concrete feedback we got in the last view months. It is also concrete enough to work on it in the future. Two immediate feedbacks.

    We will check first of all wheter the hypothesis is also true for the pre-vote polls. We know some response-problems occur mainly in the VOX-analysis (post-poll) because the interview lasts significantly longer than pre-polls.

    Second we will check wheter the hypothesis fits generally for the VOX-analysis. It isn’t clear wheter this was a unique response-phenomen or if it’s generelly the case. When it is generelly true, we will certainly discus within the VOX-partnership possible adoption to reduce the bias.

    Best regards, Lukas Golder, gfs.bern spokesman

  3. thanks for this reaction. It is, of course, possible that the education-bias applies only to certain proposals. In the particular case of the minaret-initiative, this bias was so important, because the level of education was one of key determinants of the voters’ choices. In cases, where the level of education is not so important for the voting behaviour, this bias is likely to be less relevant.

  4. Pingback: The Minaret Initiative…(again) « Denkerhaus

Leave a Reply to Lukas Golder Cancel reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s