The paper focuses on a Bayesian treatment of measurement error problems and on the question of the specification of the prior distribution of the unknown covariates. It presents a flexible semiparametric model for this distribution based on a mixture of normal distributions with an unknown number of components. Implementation of this prior model as part of a full Bayesian analysis of measurement error problems is described in classical set-ups that are encountered in epidemiological studies: logistic regression between unknown covariates and outcome, with a normal or log-normal error model and a validation group. The feasibility of this combined model is tested and its performance is demonstrated in a simulation study that includes an assessment of the influence of misspecification of the prior distribution of the unknown covariates and a comparison with the semiparametric maximum likelihood method of Roeder, Carroll and Lindsay. Finally, the methodology is illustrated on a data set on coronary heart disease and cholesterol levels in blood.
|Translated title of the contribution||Mixture models in measurement error problems, with reference to epidemiological studies|
|Pages (from-to)||549 - 566 Part 3|
|Journal||Journal of the Royal Statistical Society: Series A|
|Publication status||Published - 2002|
Bibliographical notePublisher: Blackwell Publ Ltd
Other identifier: IDS Number 606NR