Following Taylor et al. (2001), we design an induced-value experiment to test for hypothetical bias in a referendum voting mechanism. In our experiment, the level of benefit from the public good increases with the number of Yes votes (conditional on the referendum passing by majority rule). This is intended to introduce uncertainty. In contrast to Taylor et al. (2001), we find evidence of significant hypothetical bias in a referendum voting mechanism, when the level of benefits is uncertain. A cheap-talk treatment is shown to eliminate this bias.