It is often argued that electoral vulnerability is critical to constituency responsiveness. We investigate this possibility using different measures of vulnerability, but argue that in the United States the Republican Party may be less responsive than the Democratic Party due to its core constituency and view of representation. We test our hypotheses relying on an innovative research design that exploits referenda in U.S. states to compare legislator voting behavior with voter preferences on exactly the same policy proposal, allowing us to overcome the measurement problems of much previous research. Based on a newly compiled data set of more than 3,000 voting decisions for 818 legislators on 27 referenda, we find high levels of congruence, but that congruence with the median voter is higher for legislators who are running for reelection. We also find that Democrats are more responsive after a close election but that Republicans are not sensitive to electoral margins.