Replication Data for: Systematically Dependent Competing Risks and Strategic Retirement

In many applications of survival analysis, the risk of an event occurring for one reason is dependent on the risk of the same event occurring for another reason. For example, when politicians suspect they might lose an election, they may strategically choose to retire. In such situations, the often-used multinomial logit model suffers from bias and underestimates the degree of strategic retirement, for example, to what extent poor prior electoral performance diminishes electoral prospects. To address this problem, the present article proposes a systematically dependent competing-risks (SDCR) model of survival analysis. Unlike the frailty model, the SDCR model can also deal with more than two risks. Monte Carlo simulation demonstrates how much the SDCR model reduces bias. Reanalysis of data on U.S. congressional careers (Box-Steffensmeier and Jones 2004) documents the strategic retirement of representatives, indicating that electoral pressure is more effective at turning out incumbents than previously recognized.