Error in measurement is inevitable in epidemiological study. According to the classical regression model, Y=β₀ + β[minute]T where Y represents the response, e.g. the occurrence of a disease of interest, β represents the dose response parameter which the main epidemioigical study is undertaken to estimate and T represents the true exposure. The non-differential measurement error arises from the fact that the true exposure, T, has not been assessed, but a crude value, R, is available instead. A validation study is undertaken to estimate a correction factor needed to remove the bias in the crude estimates. The correction factor estimated is based on repeated measures of exposure. The unbiasedness holds if the repeated measurements are independent, but it is not always true. We investigate...[ Read more ]

Error in measurement is inevitable in epidemiological study. According to the classical regression model, Y=β₀ + β[minute]T where Y represents the response, e.g. the occurrence of a disease of interest, β represents the dose response parameter which the main epidemioigical study is undertaken to estimate and T represents the true exposure. The non-differential measurement error arises from the fact that the true exposure, T, has not been assessed, but a crude value, R, is available instead. A validation study is undertaken to estimate a correction factor needed to remove the bias in the crude estimates. The correction factor estimated is based on repeated measures of exposure. The unbiasedness holds if the repeated measurements are independent, but it is not always true. We investigate how well the correction factor performs in the univariate and bivariate cases for different values of the underlying parameters based on asymptotic normality. Re-sampling methods, Jackknife and Bootstrap methods, for estimating the correction factor and its variance are introduced in both cases.