How to Survey Turnout? Evidence from a Randomised Wording Experiment in Five Democracies

Researchers studying electoral participation at the individual level often rely on post-election surveys, but the turnout rate is typically overestimated in survey samples (even in probability samples). The usual solution is to ponder observations so that turnout in the weighted sample corresponds to the observed turnout. However, what part of this overestimation is caused by a social desirability bias as opposed to a sampling bias remains an open question. This study presents the results of a wording experiment embedded in nine recent post-election surveys: three national elections (Spain, Switzerland, and France) and six regional elections (Quebec, Ontario, Zurich, Lucerne, Catalonia and Lower Saxony). Half of respondents were asked whether they voted or not. In the other split-sample, social desirability was experimentally softened by exposing respondents to three response options that offered different ways to report that they did not vote.