Wed.., September 26, 2007
The talk shows the use and development of nonparametric
functional concepts to model duration time data. We extend
the classical Cox model in various directions. First, we give
up the restriction of proportional hazards and allow covariate
effects to change smoothly and functionally with (duration) time.
Secondly, we include calender time in the model so that
the functional components vary in both time axes, duration time
and calender time, respectively. Finally, we show how unobserved
heterogeneity can be incorporated using frailty effects. For
estimation of the functional component we make use of penalized
spline smoothing. The approach is appealing due to its numerical
efficiency and its link to mixed models. The latter allows for
simple estimation and estimation of the smoothing parameter.
The modeling exercise is demonstrated on unemployment
data with the focus of analyzing the length of unemployment