Advancements of survey design in election polls and surveys 2
|Convenor||Ms Vilma Agalioti-sgompou (ISER University of Essex )|
The recent Ukrainian mass protests for European integration led to the demise of President Yanukovych, occupation of the Crimea and war in the Donbas. In these extremely difficult conditions, Ukraine held early presidential and parliamentary elections. This presentation considers the methodological specifics of conducting exit polls in conditions of the threat to state sovereignty.
We consider steps to ensure the reliability of data in the following stages: 1) precinct selection; 2) respondent selection, and 3) data collection methods - interviews or self-administered questionnaires. We review also possibility of using phone survey during election day to improve the quality of data.
Early national elections of the President (spring) and Parliament (autumn) were held in Ukraine in 2014. If pollster’s forecast of the results of presidential elections were fairly accurate, the prediction of the parliamentary elections was terribly wrong. The presentation will provide results of testing hypothesis about the possible causes of incorrect prediction elections. Several hypothesis were tested, such as: undecided voters behavior, "late swing", spiral of silence impact, likely voters and differential turnout problem, non-response and sampling frame biases etc. Our progress and experience of mathematical modeling of the behavior of undecided voters using discriminant analysis will be
This paper proposes a forecasting model of voter turnout for the upcoming 2015 British General Election at both the constituency and the national level. The suggested model forecasts with a long lead time and a low prediction error. The paper thereby achieves two goals: First, it helps to correct for the overreporting of turnout before the election, and, second, it compares the predictive performance of key variables in several explanatory theories of turnout.
On occasion of the 2015 Spanish local and regional elections, we poll public opinion in three cities of Valencian region through a web questionnaire, using chain sampling to recruit respondents. In a superpopulation model framework, election results of local elections corresponding to the three polled cities as well as outcomes of the regional election are forecasted using the answers of a new specific designed questionnaire which enables to intensively exploit small-area out-of-the sample available data as a mechanism to reduce the bias, mainly of self-selection and coverage, that both collection mode and recruitment process introduce.