We show in a novel voting model with costly information acquisition that in equilibrium nobody votes without acquiring information and that the probability of the better alternative winning converges to one as the size of the electorate approaches infinity. In a large-scale internet experiment during the US Presidential Election, we find alarming rates of uninformed voting (>42 percent). The problem is exacerbated in treatments that allow for expressive voting motives and overconfidence (rates up to 56 percent).
Increasing the electorate size substantially raises efficiency as long as uninformed voting is not too biased towards one alternative.