Voters in U.S. legislative elections receive markedly different representation depending on which party’s candidate they elect, and because of the incumbency advantage, the effects of this choice can persist for many years. What are the long-term consequences of these two phenomena? Combining electoral and legislative roll-call data in a dynamic regression-discontinuity design, we find that a “coin-flip” election in a moderate electorate significantly influences representation for more than a decade. Across the U.S. House, the U.S. Senate, and state legislatures, the effects of one election persist for at least a decade in all settings and as much as three decades in some settings. Further results suggest that elected officials do not adapt their roll-call voting to their districts’ preferences over time, and voters do not systematically respond by replacing incumbents.