Parallel Processes and Recall Errors

Event history data collected in retrospective surveys are typically affected by measurement error, and in particular, recall errors. Errors may be related to telescoping or recall decay effects (the inability to recall all the events occurring in the past). The latter are likely to be relevant with long recall periods. The literature on the effect of measurement error in event history analysis (Holt et al, 1991; Torelli, Trivellato, 1993) refers to single spell data and time constant covariates, allowing for measurement error in durations but not in the covariates. In this paper we focus on parallel processes. We investigate the impact of the omission of short episodes from the individual's history, referring to situations where one process is exogenous and the other one endogenous. Interest lays on the effect of the explaining process on the dependent process. A simulation study is developed where a response error mechanism is superimposed on "true histories". Different estimation methods are considered: Cox's partial likelihood and a Weibull parametric model.