Yamaguchi's data set contains two interesting time-dependent covariates. The first, marriage, has been employed in many studies of labour market behaviour, migration, and of course in demography. Another type of time-dependent covariates may be termed "seasonal effects". Such effects occur, for instance, in the labour market. Another notable example are the college students in the example data set. When investigating the hazards of dropping out of college, Yamaguchi (1991, p. 151) noted that drop out is most likely to occur during the months of May and June. That is, a typical pattern of drop out is that students do not return to college after summer.
In the following, we first remain within the boundaries of conventional Partial Likelihood modelling, incorporating time-dependent covariates in the usual way. In the next step, we shall use the method of episode splitting. Finally, I will discuss a model that seems very appropriate for these data, i.e. a "piecewise constant exponential model".
Last update: 28 Jan 2000