Übergangsratenmodelle bei intervalldatierten Ereignissen

Abstract

In event history data, events often are not dated exactly but only a time interval is recorded. In contrast, common estimators for transition rate models assume exactly dated events. While the maximum likelihood method can be adapted quite easily to time intervals for the end of episodes when the starting point is exactly known, interval dating for the starting points leads to problems of unobserved heterogeneity. Using a distributional assumption for the starting points, a maximum likelihood estimator is derived for that case. A Monte-Carlo experiment shows that by this approach unbiased estimates can be derived while conventional estimators yield substantially biased estimates if broad time-intervals are used. However, for narrow intervals, estimates based on the assumption of exact dating prove to be rather accurate.