|
Abstract:
|
In many large-scale studies, it is difficult to use the most accurate measurement technique for measuring a variable of interest due to cost or logistical reasons. Imperfect instruments may be used to measure such variables in the main study and carefully designed auxiliary studies may be used to calibrate the accurate and imperfect instruments. All the studies, taken together create a variety of patterns of missing data and, thus, can be handled using multiple imputation. However, unlike the standard missing data problem, only the subjects in the main study are used in the ultimate analysis. This paper discusses a variety of modeling and analysis issues related to using the multiple imputation approach. Examples include standard measurement error calibration, mixed mode surveys, combining information from multiple surveys and harmonization of variables to a common standard.
|
