414 – Contributed Oral Poster Presentations: Section on Bayesian Statistical Science
A Bayesian Approach to Modeling Measurement Errors
Jennifer Weeding
Montana State University
Mark Greenwood
Montana State University
The impacts of ignoring covariate measurement error in regression models have been well documented and include bias in parameter estimates and a loss of power. Measurement error in the response variable has received less attention and correlated measurement errors (between the response and explanatory variables) even less. A Bayesian model that accounts for measurement errors is implemented in the simple linear regression setting, with extensions to the correlated measurement error setting and multiple linear regression setting. A simulation study is used to explore this approach, and allows for comparison to other popular measurement error correction methods (SIMEX, method of moments, ordinary least squares-no correction). The Bayesian measurement error model provided approximately unbiased results in all cases considered and directly provides a corrected estimate of unexplainable random variation.