Abstract:
|
We have collaborated on several surveys and studies in the area of dental research in the past decade. These include studies on the efficacy of a prenatal oral health program for mothers and their children, the use of sealants on molars of school children in Jamaica, the impact of measures to address stress in a Dental College, a comparison of school-based and community-based dental clinics, and a study of simple predictors of carious teeth in children. The data sets involved in these analyses had small amounts of missing data. Missing data can impact analyses through causing bias in and increasing variance of estimators. In these studies, there usually are variables that are correlated with missing outcomes and predictor variables which can be used to build models for imputation of missing values. These models range in type from parametric statistical models to procedures for matching subjects to find donors. Another approach, related to survey post stratification, to addressing missing information is to weight the respondent data within categories of subjects. In this paper, methods of handling missing data are studied through simulations based on contexts of published studies.
|