Abstract

In most data sources, measurement errors are of major concern. There are many different approaches to tackle these measurement errors, e.g., regression calibration and simex. These approaches need a different type of information regarding the measurement error to correct the analysis. In this talk, we will focus on how robust the different approaches are against structural and random measurement errors in linear regression analysis, when the information of the measurement errors is not fully known. Special focus will be laid on a new method based on robust optimization techniques that needs no distributional assumption on the measurement errors. Within a simulation study, the different approaches are compared under structural and random measurement errors that typically occur in registers and surveys.