Multilevel risk models for retrospective age-of-onset data: School children's first cigarette

Survival analysis provides a natural conceptual framework for considering r isk. However survey data on age of onset typically possess measurement erro r and sample design complexities that are absent from the usual settings in which survival analysis is applied The authors describe a random effects d iscrete time survival model that addresses these problems. They illustrate its use by an analysis of retrospective report data on the age of onset of smoking from two cross-sectional school-based studies.