Arbeitsbereich

Statistische Demografie

Auf einen Blick Projekte Publikationen Team

Projekt

Modeling Informative Censoring Mechanisms

Jutta Gampe

Ausführliche Beschreibung

Censored data occur whenever exact values cannot be observed but only intervals in which the observations lie are given. A crucial assumption in the analysis of censored observations is that the censoring mechanism is independent of the unobserved value. This is called independent or uninformative censoring.

If this assumption is violated, then the censoring mechanism needs to be part of the statistical model to obtain unbiased estimates. A typical example of informative (interval-) censoring occurs, when observations are not reported as single values but as intervals that express the uncertainty about the exact value. Should this uncertainty vary with the magnitude of the (unobserved) exact value, then the interval length itself carries information about the censored observation, and the censoring mechanism itself is informative. Estimates of individual ages that commonly become wider when the individual is older are a typical example.

Modeling the censoring process either needs strong assumptions about the mechanism or some kind of calibration sample that allows the censoring scheme to be estimated from data. If such a calibration sample is available, then the estimation of the actual distribution or regression model becomes a deconvolution problem, which can be solved, for example, by a Penalized Composite Link Model.

Schlagworte:

Statistik und Mathematik

Das Max-Planck-Institut für demografische Forschung (MPIDR) in Rostock ist eines der international führenden Zentren für Bevölkerungswissenschaft. Es gehört zur Max-Planck-Gesellschaft, einer der weltweit renommiertesten Forschungsgemeinschaften.