Likely Voter Models in a Multiparty System and in Unstable Conditions: A Case Study Using the Czech General Election of 2017
Abstract
Research problem: Likely voter models must deal with two main forms of uncertainty: who will turn out to vote, and for which party or candidate he or she will vote for. Measuring and estimating this decision is even more complicated in multi-party systems where there is a political volatile situation when a new political party or movement emerges in each election.
In our research on elections to the Chamber of Deputies of the Parliament of the Czech Republic in October 2017, we deal with both issues: how to measure intention to go to the polls on election day and how to estimate party choice. Data and Methodology: The data comes from the 'Our Society' research project implemented by the Czech Public Opinion Research Centre (CVVM).
It is a long-term omnibus survey carried out once a month on a sample of approximately 1,000 respondents. Each sample selected by quota procedure is considered representative of the population of the Czech Republic aged 15 years and over.
Data collection is done in face-to-face interviews with a standardised omnibus-type questionnaire. To validate the likely voter model, we fielded a panel survey: the respondents of the October pre-election survey also participated in a post-election CATI survey.
Objectives and results: This paper will examine two key questions: (1) how to measure vote intention, and (2) how to estimate the choice for a particular political party in a multi-party system. We tested different measures of vote intention: a 4-point scale traditionally used in the 'Our Society' survey, a 1 to 10 scale, the Perry-Gallup index, and finally a 4-point scale of vote intention combined with an item measuring respondent's past turnout behaviour.
We investigated how the results changed for each party when we applied different measures of vote intention, using both a cut-off approach to identify voters and non-voters, as well as a probabilistic approach. In order to estimate the choice for a particular political party, we use both a simple question and a probabilistic attribution based on further questions and on a declaration estimating the respondent's reported certainty in voting for a particular party.
Based on our results, we recommend using of a simple measure of the vote intention. With regard to estimating party choice for respondents who are not completely certain redistribution of preferences among all likely parties is the best polling strategy in multi-party systems.