SMABS 2004 Jena University
SMABS 2004 Home Organization About Jena Sponsors Links Contact SMABS Home

European Association of Methodology

Department of methodology and evaluation research

Jena University


© Webmaster

Contributions: Abstract

In previous work by Muthén and Masyn (2003) and Masyn (2003), it has been shown that discrete-time survival models may be estimated using a Latent Class Regression (LCR) framework (see also Vermunt, 1997, 2002). This framework conveniently accommodates many interesting model extensions including, for example, nonparametric specification of unobserved heterogeneity in survival processes across individuals utilizing the latent class structure. What has not yet been fully incorporated into this framework is the possibility of measurement error or bias in the determination of event occurrence and event timing.

Typically, the implicit assumption in survival analysis is that the measurement error or bias only occurs in the event time for those who have experienced the event. For example, in retrospective studies, subjects may report event times as more recent than the actual occurrences - this is often referred to as telescoping. It could be important, however, to also consider that there may be uncertainty in the event occurrence determination itself. For example, consider the effect of projection, where a subject who has thought about the event or wished for the event may be more likely to report having experienced the event when he actually hasn’t. As another example, consider any one of a number of psychopathologies that are treated as directly observable, terminating events while they are actually unobserved but manifest in a set of measured variables, usually clinical diagnostic criteria. This paper presents a way to incorporate error in event determination motivated by a data example involving the timing and duration of postpartum duration.