Inverse probability weighted estimation in a reliability model with missing data

Amel Mezaouer (1) , Jean-Francois Dupuy (2) , Kamal Boukhetala (3)

1 Faculté des Sciences
2 MIA - Mathématiques, Image et Applications - EA 3165
3 Faculté des Mathématiques

The linear transformation model is a useful regression model for the analysis of reliability data. In this paper, we consider the problems of estimation and testing in this model in a context of missing data. Precisely, we consider the situation where explanatory variables are available for every unit in the experiment sample, while the event times and censoring indicators can only be observed on a subset of the sample. We rely on an inverse probability weighted-type estimation approach for approximating the regression parameter of interest. The theoretical and numerical properties of the resulting estimator are investigated. The proposed approach appears to outperform the classical complete-case analysis.

Mots clés: Inverse probability weighted estimation; Large-sample properties; Linear transformation model; Missing data; Simulation study

Domaines: Méthodologie [stat.ME]

Amel Mezaouer, Jean-Francois Dupuy, Kamal Boukhetala. Inverse probability weighted estimation in a reliability model with missing data. 2011. ⟨hal-00586529⟩