City Research Online

Abstract

Pre-election polls are the most prominent type of surveys. As with any other survey, estimates are only of interest if they do not deviate significantly from the true state of nature. Even though pre-election polls in Germany as well as in other countries repeatedly show noticeably inaccurate results, their failure appears to be quickly forgotten. No comparison considering all available German data on actual election results and the confidence intervals based on pre-election polls has been published. In the study reported here only 69% of confidence intervals covered the election result, whereas statistically 95% would have to be expected. German pre-election polls even just a month ahead are therefore much less accurate than most introductory statistical textbooks would suggest.

Publication Type:	Article
Publisher Keywords:	Pre-Election-Polls, Empirical coverage, Confidence intervals for binomial data, Design effects, Sonntagsfrage
Subjects:	J Political Science > JC Political theory
Departments:	School of Policy & Global Affairs > Sociology & Criminology

Export

Downloads