Voting Not to Tell Others

Abstract

Whereas turnout overreporting in surveys has been a widely studied phenomenon, turnout underreporting –stating that one abstained when in fact one voted– has been generally overlooked. The main reason is that the size of the effect is typically smaller: overreporting easily is in the double digits, whereas underreporters are around 1--2%. Survey data from several elections in New Zealand and Ireland (where turnout validation takes place after electoral surveys) suggest that turnout underreporting is not an conscious choice to hide one's vote but a consequence of genuine bad memory. This paper proposes a theoretical framework to analyze this phenomenon: it is shown that, under very mild conditions, 1.5% of underreporters in fact suggests that 6--10% of respondents are satisficing –i.e., giving inaccurate or random answers that are perceived to be safe or acceptable. This is known to bias against findings: Montecarlo simulations show that the reduction of power can be substantial even when observed underreporting is very small. I provide simple and easily applied solutions on how to address the challenges posed by this particular kind of satisficing, both for survey design and for data analysis.