THE POLLS—REVIEW

Abstract

Abstract The twenty-first century has seen an important transition in survey modes used for electoral polls. This transition has not ended yet. It is thus possible to examine differences between modes used in the same election. Different modes are more or less prone to social desirability and use different sampling frames and recruitment strategies that may lead to differences in estimation. However, available literature does not show systematic and substantial differences between modes. In this article, we examine differences between modes across 15 elections and referendums that took place since 2005 in four countries: Canada, France, the United Kingdom, and the United States. We first assess differences in average estimates, variance, trends, and forecasts. We then pool the data to analyze whether there are differences that apply in all contexts. We conclude that differences between modes vary with context and over time. There are some consistent differences, however, as online polls are less likely to detect movement than are telephone or IVR polls. In a context in which online polls are becoming dominant, citizens may not be provided with a reliable portrait of the state of public opinion. IVR polls tended to be more precise than other polls recently, but they also tended to have a conservative bias. For the future, it will be important to monitor closely new developments in the methodology used for election polls. The presence of multiple modes in pre-election polling and new developments in mixed modes would be beneficial to voters and researchers alike.