I develop a theory of dynamic responsiveness that suggests that parties that win elections choose candidates who are more extreme and parties that lose elections choose candidates who are more moderate. Moreover, the size of past victories matters. Close elections yield little change, but landslides yield larger changes in the candidates offered by both parties. I test this theory by analyzing the relationship between Republican vote share in U.S. Senate elections and the ideology of candidates offered in the subsequent election. The results show that Republican (Democratic) victories in past elections yield candidates who are more (less) conservative in subsequent elections, and the effect is proportional to the margin of victory. This suggests that parties or candidates pay attention to past election returns. One major implication is that parties may remain polarized in spite of their responsiveness to the median voter.