Analyzing Dynamic Multiple Spell Durations Using Counting Processes

I propose a counting process approach to analyze multiple spell duration data. These data are doubly stochastic in a sense that both durations for an individual and the number of durations within a fixed period are random variables. I allow unobserved individual heterogeneities enters into the model as fixed effects. In addition, conditional on the individual fixed effect, durations are state dependent. A first-difference transformation is developed to cancel fixed effects, and a minimum distance estimator is re-introduced with simplified proofs. Finite sample properties are investigated in simulations. The approach is applied to study an individual's work absence decisions