ABSTRACT

In this chapter we describe a measurement error bias-correction method that shares the simplicity, generality, and approximate-inference characteristics of regression calibration. As the previous chapter indicated, regression calibration is ideally suited for problems in which the calibration function E(X | W) can be estimated reasonably well and to problems such as generalized linear models. Simulation extrapolation (SIMEX) is ideally suited to problems with additive measurement error, and more generally to any problem in which the measurement error generating process can be imitated on a computer via Monte Carlo methods.