Parameter Estimation Of Correlated Outcome And Covariates Measured With Error In Logistic Regression
Abstract
In epidemiological studies, it is common to have binary data as the response variable. Hence, the common model applied to represent the relationship between the binary outcome and risk factors is logistic regression. Problems that commonly occur in binary data are intra-class correlation in the outcome variable and measurement error or misclassification in the covariates/risk factors. An alternative statistical procedures, William's procedure and Generalized Estimating Equation for correlated outcome and direct adjustment for measurement error, are used to handle these problems. The application of these methods is implemented to the Children Nutrient Status Data. The data is obtained from the study conducted in June 2000 - October 2000, involving 402 children under five years old in three locations in Central Java. In the analysis of intra-enumerator data, there is an indication of the presence of intra-correlation between data co llected and enumerator.