EASY ESTIMATION METHODS FOR DISCRETE-TIME DURATION MODELS

Recent literature has emphasized that estimation of duration models should take account of the sampling scheme generating the data, but this cannot be done using the duration modules in currently-available packages. This paper shows that discrete-time duration models can be estimated easily using widely-available, non-specialist, software packages. It focuses on the case where the data are for respondents randomly selected from a stock at one date and then interviewed after some time interval has elapsed, and also shows that these results generalize to other sampling schemes where the likelihood function requires conditioning on survival to a data between sample selection and interview. The findings extend earlier results concerning the easy estimation of duration models which use data from random population samples.