There has been increasing acknowledgment of the importance of measurement error in epidemiology and this leads to a various investigation on alternative methods for correcting the measurement error in exposure. Measurement error may be due to inadequacies in measuring instruments or human error. Errors in variables can also arise even when the measurement process is accurate. It is well known that the regression slope between a response and predictor variable is underestimated when the predictor variable is measured imprecisely. This thesis starts with simple linear nondifferential measurement error model with simulation study as illustration. Two correction methods are considered for simple as well as all multiple logistic regression models with all covariates measured with error. The...[ Read more ]

There has been increasing acknowledgment of the importance of measurement error in epidemiology and this leads to a various investigation on alternative methods for correcting the measurement error in exposure. Measurement error may be due to inadequacies in measuring instruments or human error. Errors in variables can also arise even when the measurement process is accurate. It is well known that the regression slope between a response and predictor variable is underestimated when the predictor variable is measured imprecisely. This thesis starts with simple linear nondifferential measurement error model with simulation study as illustration. Two correction methods are considered for simple as well as all multiple logistic regression models with all covariates measured with error. The first method is the adjustment by a correction factor and the second is the first order Taylor series expansion from modification of the method proposed by Rosner et al. We also explore the effect of measurement error in covariate for a linear model applied a longitudinal data. We derive a slope parameter estimators for two different longitudinal model settings by using expected estimating equation (EEE) and generalized estimating equation (GEE). We assume the parameters in the measurement error model are available and can be estimated from the validation study in this thesis. Simulation will be conducted for some parameters comparison of different methods.