In theory, voter attitudes towards policy changes (e.g., whether to increase the minimum wage) ought to depend on their beliefs about the current level of the relevant policy variable. In this paper, I test this hypothesis using a large-scale (n = 5, 000) and pre-registered survey experiment that spans four different policy areas. The experiment yields four main results. First, voters have both inaccurate and biased beliefs about the levels of the policy variables. Second, voters’ attitudes are remarkably unresponsive to changes in their beliefs about levels: for example, exogenously increasing average beliefs about the top tax rate by ∼8.5 percentage points does not increase the share who want to cut the top tax rate. Third, the observed unresponsiveness cannot be rationalised by a model in which voters form attitudes towards policy changes by comparing actual and preferred policy levels. Fourth, although attitudes are unresponsive to the quantitative information presented, they can be swayed by qualitative arguments.