A regression model for longitudinal change in the presence of measurement error

Abstract

Purpose: The analysis of change in measured variables has become quite popular in studies where data are collected repeatedly over time. The authors describe some of the potential pitfalls in the analysis of change when the variable for change is measured with error. They show that regression analysis is often biased, possibly leading to erroneous results.

Methods: A simple method to correct for measurement error bias in regression models that model change is presented.

Results and conclusions: The two examples illustrate how measurement error can adversely affect an analysis. The bias-corrected approach yields valid results.