The Black Day of Public Opinion Research Revisited
Bayesian reanalysis of the Hungarian opinion poll results from 2002
Absztrakt
This paper reanalyzes the forecasts of the 2002 Hungarian parliamentary elections based on a rarely used statistical approach, Bayesian analysis. In contrast to the frequentist approach of statistics, this method enables the researcher to include prior information in the analysis. It may be relevant in the case of the elections under investigation, since although many polling companies had measured a higher support for the Hungarian Socialist Party in the first three months of 2002, before the elections incorrectly they all predicted the victory of the Fidesz-Hungarian Democratic Forum coalition. The empirical analysis is based on two samples used by Ipsos and by Tárki to produce their forecasts published eight days before the elections. The results show that in this particular case the Bayesian analysis is more precise than the frequentist analysis of the same data or the forecasts the same opinion polling companies published.