Without a Cure: Split Population Estimators for Heterogeneous Duration Processes.

Do filibusters convey information? In a previous paper we address this question by presenting and testing an incomplete information account of the filibuster. The game-theoretic model produces a discontinuous, non-monotonic hazard rate against which we compare the empirical hazard rate of filibusters. The discontinuity in the predicted hazard results from the presence of different types of filibusters. Existing duration models, however, are unable to account for this heterogeneity when the types are unobserved. In this paper we develop a maximum likelihood estimator that more directly corresponds to the theoretical prediction by modeling the observed duration process as a mixture between related duration processes with different forms of duration dependence. Our estimator differs from related "cure" models since they assume that a subset of observation will never fail; we assume that all observations ultimately fail, but that they exhibit distinct hazards. [ABSTRACT FROM AUTHOR]