Correcting for measurement error in parametric duration models by quasi-likelihood.
Collaborative Research Center 386, Discussion Paper 157
In regression models for duration data it is usually implicitly assumed that all variables are measured and operationalized exactly. If measurement error is present, however, but not taken into account, parameter estimates may be severely biased. The present paper studies measurement error corrected estimation in the context of a huge class of parametric duration models. The proposed quasi-likelihood based method easily allows - as long as no censoring occurs - to deal simultaneously with covariate measurement error as well as with measurement error in the duration itself and yields estimates with sound asymptotic properties. A general formula for the measurement error corrected quasi-score function can be derived, which is valid for most of the commonly used parametric duration models.