The School of Mathematics, Statistics and Computer Science cordially invites interested parties to a Statistics Seminar. Detail are as follows:
DATE: Thursay 21 May
VENUE: Room F47, First Floor, Science Building, PMB Campus
TOPIC: Some Aspects of Multilevel Discrete-Time-To-Event Modelling and Analysis Approaches
PRESENTER: Samuel OM Manda, Biostatistics Research Unit, South African Medical Research Council, Pretoria, South Africa
Traditional survival models are based on the assumption that time is measured on a continuous scale. However, time is often reported and recorded as grouped continuous; typically in months or years, resulting in a large number of tied time points. Estimation of heavily tied failure times by proportional hazards models would generally produce inconsistent estimators. An alternative of breaking ties by slight adjustments introduces accuracy bias into the data. In these situations, discrete time duration models are more appropriate, where indicators of the event of interest at user defined time-intervals are treated as binary responses.
This presentation will start with a review of multilevel discrete-time event models and extensions to handling recurrent events and transitions between states; the events could be of same or different types. The focus will be on how standard statistical software can be adopted for implementing these methods, which have huge potential for investigating evolutions and changes of life-defining events including family formation and dissolution studies. In these studies, the data collected are often longitudinal and are highly complex, where several event types occur repeatedly over time on a subject, inducing possible interdependences between different event processes. We illustrate these methods with applications to childbearing and marriage processes.Posted on