STATA MODULE FOR HEALTH STUDIES WITH MEASUREMENT ERROR

The objective of our proposal is the development of STATA-MEM, A software toolkit for the analysis of measurement error models, with specific but restrictive reference to generalized linear models (GLIM's). The problem of measurement error is common to many cancer studies, e.g., studies involving the measurement of nutrition or hormones, as well as more generally to other diseases. To cope with this common and important problem, the biostatistical profession has developed a large and rapidly growing research literature on the analysis of data subject to measurement error. However, there is a glaring lack of commercial software that allows users to actually perform the latest analyses. We will develop such software in STATA, using the easy interface common to STATA, and incorporating the latest result in field. In Phase I, we will demonstrate the feasibility of our plan by implementing two common methods, regression calibration and SIMEX, GLIM's, allowing for additive measurement error, various types of data structures which yield information about the measurement error, and inference both with asymptotic standard error estimates as well as with the bootstrap. Our long-term goals are the add to the methods from Phase I such additional features as instrumental variables, missing data in linear models, nonparametric regression,a nd multiplicative error, mixed models.