|
Abstract:
|
A recent application in the NHANES survey provides interesting insights into practical applications of the multiple imputation technique. The NHANES is a unique survey with self-reported measures, physical measurements, laboratory measurements and an examination known as Dual X-ray absorptbiometry (DXA). DXA measures body fat and bone density for several regions of the body. While some observations seemed to be missing at random, for many observations the missingness was directly related to the outcome measures. This presentation discusses the numerous practical issues that had to be addressed for this unique data set, including the highly correlated data structure, variable transformations, outlier detection, symmetry of imputed measurements, relationships of imputed variables, lower and upper bounds, and the handling of influential observations.
|
