표본추출틀에 따른 선거여론조사의 정확성 평가에 대한 연구

Abstract

In order to evaluate the accuracy of the pre-election survey, measure A (Martin et al., 2005), the adjusted odds ratio between the forecasted results of the survey and the actual election result of the two major electoral candidates, is widely used in recent practices. This study focuses on whether the accuracy of the pre-election survey is significantly different from the telephone survey’s sampling frame, and studies whether the bias from the dual frame approach can be different in terms of the mobile and the landline proportion in sampling, using measure A. I analyzed 149 pre-election survey data from the 8th Local Election in 2022. Even with the limited size of the data, there is no distinct and systematic differences between the dual frame approach and the mobile-only approach in terms of bias mean size and bias direction, but the bias case rate of the mobile-only is more high than the dual frame. Furthermore, in the analysis of the mobile-landline proportion optimization, the sampling frame categories with equal to 10% landline proportion resulted in a relatively small mean of absolute bias and so did the category with less than 10% in the mean of bias. Due to the limitation of the data, just one trial of analysis, and other accountable factors, the generalization of these findings is premature. Therefore, additional research analysis with systematic and synthetic methodology is needed to prolong the positive effects of this study.