The perils of pre-election polling: Election cycles and the exacerbation of measurement error in illiberal regimes

Abstract

Pre-election polls are increasingly being used to forecast and study elections in new and weak democracies yet the tension that surrounds elections in less liberal regimes may make data from such surveys particularly prone to measurement error. Polls conducted before and after the 2011 Ugandan election and data gathered at various points of the election cycles in 34 African countries are compared to show that, before elections, survey respondents become increasingly wary of the interview and its purpose, especially in less liberal regimes. Those who are wary report more support for the incumbent, the incumbent’s policies and the state and effect of wariness on attitudes is significantly higher on surveys that take place immediately before elections. Misreporting increases the estimated level of support for the incumbent, and may also significantly change the apparent correlates of regime support across the election cycle. Ironically, it was found that pre-election polls may be particularly poorly suited to forecasting elections or capturing the effects of campaigns on political attitudes.