The use of non-sample information in exit poll surveys in Poland

Abstract

Exit poll is a commonly used tool to predict election outcome in democratic countries. The aims of this survey, however, go beyond the standard prediction which usually loses its value after 1-2 days. Lasting benefits of exit poll result from the possibility of estimating vote distribution in socio-demographic groups, changes of political preferences, the motives for choosing a candidate, etc. No other survey is capable of providing such detailed data with satisfactory precision. Nonetheless, the exit poll accuracy, both in Poland and abroad, often leaves much to be desired. It seems that while conducting the research the non-sample information is not used sufficiently, which could improve the quality and the precision of the survey. The sources of auxiliary variables, which can be used in exit poll, along with the analysis of technical aspects of their acquisition and combination are outlined in this paper. Statistical methods aiming at incorporating the information about those variables to the survey, both at the stage of selecting the sample of precincts and at the stage of forecasting election results are proposed. Developed solutions were subjected to the simulation testing on the parliamentary election to the Sejm 2011 data. The results confirm the possibility of a significant increase in the effectiveness of estimates by improving ‘representativeness’ of a sample and by applying complex estimation of parameters.