An Exploration of the GSIMEX Approach to Modeling Variables with Correlated Measurement Errors in R

• Activity Number: 257
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section for Statistical Programmers and Analysts
• Author(s): Jennifer Weeding*+ and Mark C. Greenwood
• Companies: and Montana State University
• Keywords: Measurement Error ; SIMEX ; GSIMEX

Abstract

Measurement error in regression models can lead to bias in parameter estimates, a loss of power, and mask important features of the data. Measurement errors are commonly ignored due to the lack of software available to implement the methods and the extra information needed to correct for measurement error is not available to the researcher. The SIMEX (Simulation Extrapolation) approach of Cook and Stefanski (1994) is a simulation-based method of inference that accounts for additive measurement errors. Ronning and Roseman (2008) made this approach more general by allowing for the correlation of measurement errors in their approach, GSIMEX. However, this method is currently unavailable in software packages. An implementation of the method in R is discussed. A simulation study is used to explore this approach, followed by an application to a data set that contains measurement error in both the response variable and the explanatory variable.