Why do some people go to the polling station, sometimes up to several times a year, while others always prefer to stay at home? This question has launched a wide theoretical debate in both economics and political science, but convincing empirical support for the different models proposed is still rare. The basic rational voting model of Downs (1957) predicts zero participation because each individual vote is extremely unlikely to be pivotal. One prominent modification of this model is the inclusion of a civic duty term into the voter's utility function (Riker and Ordeshook, 1968) which has been the basis of structural ethical voting models such as Coate and Conlin (2004) and Feddersen and Sandroni (2006). Another branch of structural models looks at informational asymmetries among citizens (Feddersen and Pesendorfer, 1996, 1999). This paper tests the implications of these two branches of structural models by exploiting a unique variability in compulsory voting laws in Swiss federal states. By analyzing a newly compiled comparative data set covering the 1900-1950 period, we find large positive effects of the introduction of compulsory voting laws on turnout. Along with the arguably exogenous treatment allocation, several specification and placebo tests lend support to a causal interpretation of this result. The findings of this study lend support to the ethical voting models since citizens do react to compulsory voting laws only if it is enforced with a fee. At the same time, the informational aspect of non-voting is questioned as new voters do not delegate their votes.