Abstract
Research question: Improving the representativeness of pre-election polls by past vote weighting is a frequent but controversial adjustment (Durrand et al., 2015). The recall of past vote is a strongly correlating variable with the current voting intention and at the same time, the distribution of the past vote is known for the whole population.
However, answers to past vote question are influenced by cognitive biases that can reduce the quality of data: memory failure, reconciliation and social desirability. It is hypothesized that recall reliability is positively associated with the stability of voting behavior.
At the same time, electoral volatility has been increasing in Western Europe over the past two decades (Chiaramonte, Emanuele 2017). The question of this paper is whether past vote weighting is making pre-election polls more accurate and whether the benefits of past vote weighting are decreasing with increasing electoral volatility.
Data and methodology: The paper uses data from approximately 30 pre-election polls from multi-party systems with a pre-election and post-election wave. For each poll, an estimate of election result is calculated with and without past vote weighting adjustment.
To estimate the election result, a variable of reported voting behavior from the post-election wave is used to eliminate the influence of voter hesitation on the accuracy of the estimate. The change in the accuracy of the estimate after applying past vote weighting is compared with the level of electoral volatility.